diff --git a/include/crm/common/strings_internal.h b/include/crm/common/strings_internal.h index c791f9ea3b..9062b0ac5a 100644 --- a/include/crm/common/strings_internal.h +++ b/include/crm/common/strings_internal.h @@ -1,243 +1,242 @@ /* * Copyright 2015-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_STRINGS_INTERNAL__H #define PCMK__CRM_COMMON_STRINGS_INTERNAL__H #include // bool #include // uint32_t, etc. #include // guint, GList, GHashTable #include // PCMK_VALUE_TRUE, PCMK_VALUE_FALSE #include // crm_strdup_printf() #ifdef __cplusplus extern "C" { #endif /* internal constants for generic string functions (from strings.c) */ #define PCMK__PARSE_INT_DEFAULT -1 #define PCMK__PARSE_DBL_DEFAULT -1.0 /* internal generic string functions (from strings.c) */ enum pcmk__str_flags { pcmk__str_none = 0, pcmk__str_casei = 1 << 0, pcmk__str_null_matches = 1 << 1, pcmk__str_regex = 1 << 2, pcmk__str_star_matches = 1 << 3, }; int pcmk__scan_double(const char *text, double *result, const char *default_text, char **end_text); int pcmk__guint_from_hash(GHashTable *table, const char *key, guint default_val, guint *result); -bool pcmk__ends_with(const char *s, const char *match); bool pcmk__ends_with_ext(const char *s, const char *match); char *pcmk__trim(char *str); void pcmk__add_separated_word(GString **list, size_t init_size, const char *word, const char *separator); int pcmk__compress(const char *data, unsigned int length, unsigned int max, char **result, unsigned int *result_len); int pcmk__scan_ll(const char *text, long long *result, long long default_value); int pcmk__scan_min_int(const char *text, int *result, int minimum); int pcmk__scan_port(const char *text, int *port); int pcmk__parse_ll_range(const char *text, long long *start, long long *end); GHashTable *pcmk__strkey_table(GDestroyNotify key_destroy_func, GDestroyNotify value_destroy_func); GHashTable *pcmk__strikey_table(GDestroyNotify key_destroy_func, GDestroyNotify value_destroy_func); GHashTable *pcmk__str_table_dup(GHashTable *old_table); void pcmk__insert_dup(GHashTable *table, const char *name, const char *value); /*! * \internal * \brief Get a string value with a default if NULL * * \param[in] s String to return if non-NULL * \param[in] default_value String (or NULL) to return if \p s is NULL * * \return \p s if \p s is non-NULL, otherwise \p default_value */ static inline const char * pcmk__s(const char *s, const char *default_value) { return (s == NULL)? default_value : s; } /*! * \internal * \brief Create a hash table with integer keys * * \param[in] value_destroy_func Function to free a value * * \return Newly allocated hash table * \note It is the caller's responsibility to free the result, using * g_hash_table_destroy(). */ static inline GHashTable * pcmk__intkey_table(GDestroyNotify value_destroy_func) { return g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL, value_destroy_func); } /*! * \internal * \brief Insert a value into a hash table with integer keys * * \param[in,out] hash_table Table to insert into * \param[in] key Integer key to insert * \param[in] value Value to insert * * \return Whether the key/value was already in the table * \note This has the same semantics as g_hash_table_insert(). If the key * already exists in the table, the old value is freed and replaced. */ static inline gboolean pcmk__intkey_table_insert(GHashTable *hash_table, int key, gpointer value) { return g_hash_table_insert(hash_table, GINT_TO_POINTER(key), value); } /*! * \internal * \brief Look up a value in a hash table with integer keys * * \param[in] hash_table Table to check * \param[in] key Integer key to look for * * \return Value in table for \key (or NULL if not found) */ static inline gpointer pcmk__intkey_table_lookup(GHashTable *hash_table, int key) { return g_hash_table_lookup(hash_table, GINT_TO_POINTER(key)); } /*! * \internal * \brief Remove a key/value from a hash table with integer keys * * \param[in,out] hash_table Table to modify * \param[in] key Integer key of entry to remove * * \return Whether \p key was found and removed from \p hash_table */ static inline gboolean pcmk__intkey_table_remove(GHashTable *hash_table, int key) { return g_hash_table_remove(hash_table, GINT_TO_POINTER(key)); } gboolean pcmk__str_in_list(const gchar *s, const GList *lst, uint32_t flags); bool pcmk__strcase_any_of(const char *s, ...) G_GNUC_NULL_TERMINATED; bool pcmk__str_any_of(const char *s, ...) G_GNUC_NULL_TERMINATED; bool pcmk__char_in_any_str(int ch, ...) G_GNUC_NULL_TERMINATED; int pcmk__strcmp(const char *s1, const char *s2, uint32_t flags); int pcmk__numeric_strcasecmp(const char *s1, const char *s2); char *pcmk__str_copy_as(const char *file, const char *function, uint32_t line, const char *str); /*! * \internal * \brief Copy a string, asserting on failure * * \param[in] str String to copy (can be \c NULL) * * \return Newly allocated copy of \p str, or \c NULL if \p str is \c NULL * * \note The caller is responsible for freeing the return value using \c free(). */ #define pcmk__str_copy(str) pcmk__str_copy_as(__FILE__, __func__, __LINE__, str) void pcmk__str_update(char **str, const char *value); void pcmk__g_strcat(GString *buffer, ...) G_GNUC_NULL_TERMINATED; static inline bool pcmk__str_eq(const char *s1, const char *s2, uint32_t flags) { return pcmk__strcmp(s1, s2, flags) == 0; } // Like pcmk__add_separated_word() but using a space as separator static inline void pcmk__add_word(GString **list, size_t init_size, const char *word) { return pcmk__add_separated_word(list, init_size, word, " "); } /* Correctly displaying singular or plural is complicated; consider "1 node has" * vs. "2 nodes have". A flexible solution is to pluralize entire strings, e.g. * * if (a == 1) { * crm_info("singular message"): * } else { * crm_info("plural message"); * } * * though even that's not sufficient for all languages besides English (if we * ever desire to do translations of output and log messages). But the following * convenience macros are "good enough" and more concise for many cases. */ /* Example: * crm_info("Found %d %s", nentries, * pcmk__plural_alt(nentries, "entry", "entries")); */ #define pcmk__plural_alt(i, s1, s2) (((i) == 1)? (s1) : (s2)) // Example: crm_info("Found %d node%s", nnodes, pcmk__plural_s(nnodes)); #define pcmk__plural_s(i) pcmk__plural_alt(i, "", "s") static inline int pcmk__str_empty(const char *s) { return (s == NULL) || (s[0] == '\0'); } static inline char * pcmk__itoa(int an_int) { return crm_strdup_printf("%d", an_int); } static inline char * pcmk__ftoa(double a_float) { return crm_strdup_printf("%f", a_float); } static inline char * pcmk__ttoa(time_t epoch_time) { return crm_strdup_printf("%lld", (long long) epoch_time); } // note this returns const not allocated static inline const char * pcmk__btoa(bool condition) { return condition? PCMK_VALUE_TRUE : PCMK_VALUE_FALSE; } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_STRINGS_INTERNAL__H diff --git a/lib/common/schemas.c b/lib/common/schemas.c index 27611f9f24..77f51b3225 100644 --- a/lib/common/schemas.c +++ b/lib/common/schemas.c @@ -1,1607 +1,1611 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include // UCHAR_MAX #include #include #include // g_str_has_prefix() #include #include // xmlNode #include // xmlChar #include #include #include #include #include #include #include /* PCMK__XML_LOG_BASE */ #include "crmcommon_private.h" #define SCHEMA_ZERO { .v = { 0, 0 } } #define schema_strdup_printf(prefix, version, suffix) \ crm_strdup_printf(prefix "%u.%u" suffix, (version).v[0], (version).v[1]) typedef struct { xmlRelaxNGPtr rng; xmlRelaxNGValidCtxtPtr valid; xmlRelaxNGParserCtxtPtr parser; } relaxng_ctx_cache_t; static GList *known_schemas = NULL; static bool initialized = false; static bool silent_logging = FALSE; static void G_GNUC_PRINTF(2, 3) xml_log(int priority, const char *fmt, ...) { va_list ap; va_start(ap, fmt); if (silent_logging == FALSE) { /* XXX should not this enable dechunking as well? */ PCMK__XML_LOG_BASE(priority, FALSE, 0, NULL, fmt, ap); } va_end(ap); } static int xml_latest_schema_index(void) { /* This function assumes that pcmk__schema_init() has been called * beforehand, so we have at least two schemas (one real schema and the * "none" schema). * * @COMPAT: The "none" schema is deprecated since 2.1.8. * Update this when we drop that schema. */ return g_list_length(known_schemas) - 2; } /*! * \internal * \brief Return the schema entry of the highest-versioned schema * * \return Schema entry of highest-versioned schema */ static GList * get_highest_schema(void) { /* The highest numerically versioned schema is the one before none * * @COMPAT none is deprecated since 2.1.8 */ GList *entry = pcmk__get_schema("none"); pcmk__assert((entry != NULL) && (entry->prev != NULL)); return entry->prev; } /*! * \internal * \brief Return the name of the highest-versioned schema * * \return Name of highest-versioned schema (or NULL on error) */ const char * pcmk__highest_schema_name(void) { GList *entry = get_highest_schema(); return ((pcmk__schema_t *)(entry->data))->name; } /*! * \internal * \brief Find first entry of highest major schema version series * * \return Schema entry of first schema with highest major version */ GList * pcmk__find_x_0_schema(void) { #if defined(PCMK__UNIT_TESTING) /* If we're unit testing, this can't be static because it'll stick * around from one test run to the next. It needs to be cleared out * every time. */ GList *x_0_entry = NULL; #else static GList *x_0_entry = NULL; #endif pcmk__schema_t *highest_schema = NULL; if (x_0_entry != NULL) { return x_0_entry; } x_0_entry = get_highest_schema(); highest_schema = x_0_entry->data; for (GList *iter = x_0_entry->prev; iter != NULL; iter = iter->prev) { pcmk__schema_t *schema = iter->data; /* We've found a schema in an older major version series. Return * the index of the first one in the same major version series as * the highest schema. */ if (schema->version.v[0] < highest_schema->version.v[0]) { x_0_entry = iter->next; break; } /* We're out of list to examine. This probably means there was only * one major version series, so return the first schema entry. */ if (iter->prev == NULL) { x_0_entry = known_schemas->data; break; } } return x_0_entry; } static inline bool version_from_filename(const char *filename, pcmk__schema_version_t *version) { - if (pcmk__ends_with(filename, ".rng")) { + if (filename == NULL) { + return false; + } + + if (g_str_has_suffix(filename, ".rng")) { return sscanf(filename, "pacemaker-%hhu.%hhu.rng", &(version->v[0]), &(version->v[1])) == 2; } else { return sscanf(filename, "pacemaker-%hhu.%hhu", &(version->v[0]), &(version->v[1])) == 2; } } static int schema_filter(const struct dirent *a) { int rc = 0; pcmk__schema_version_t version = SCHEMA_ZERO; if (strstr(a->d_name, "pacemaker-") != a->d_name) { /* crm_trace("%s - wrong prefix", a->d_name); */ } else if (!pcmk__ends_with_ext(a->d_name, ".rng")) { /* crm_trace("%s - wrong suffix", a->d_name); */ } else if (!version_from_filename(a->d_name, &version)) { /* crm_trace("%s - wrong format", a->d_name); */ } else { /* crm_debug("%s - candidate", a->d_name); */ rc = 1; } return rc; } static int schema_cmp(pcmk__schema_version_t a_version, pcmk__schema_version_t b_version) { for (int i = 0; i < 2; ++i) { if (a_version.v[i] < b_version.v[i]) { return -1; } else if (a_version.v[i] > b_version.v[i]) { return 1; } } return 0; } static int schema_cmp_directory(const struct dirent **a, const struct dirent **b) { pcmk__schema_version_t a_version = SCHEMA_ZERO; pcmk__schema_version_t b_version = SCHEMA_ZERO; if (!version_from_filename(a[0]->d_name, &a_version) || !version_from_filename(b[0]->d_name, &b_version)) { // Shouldn't be possible, but makes static analysis happy return 0; } return schema_cmp(a_version, b_version); } /*! * \internal * \brief Add given schema + auxiliary data to internal bookkeeping. */ static void add_schema(enum pcmk__schema_validator validator, const pcmk__schema_version_t *version, const char *name, GList *transforms) { pcmk__schema_t *schema = NULL; schema = pcmk__assert_alloc(1, sizeof(pcmk__schema_t)); schema->validator = validator; schema->version.v[0] = version->v[0]; schema->version.v[1] = version->v[1]; schema->transforms = transforms; // schema->schema_index is set after all schemas are loaded and sorted if (version->v[0] || version->v[1]) { schema->name = schema_strdup_printf("pacemaker-", *version, ""); } else { schema->name = pcmk__str_copy(name); } known_schemas = g_list_prepend(known_schemas, schema); } static void wrap_libxslt(bool finalize) { static xsltSecurityPrefsPtr secprefs; int ret = 0; /* security framework preferences */ if (!finalize) { pcmk__assert(secprefs == NULL); secprefs = xsltNewSecurityPrefs(); ret = xsltSetSecurityPrefs(secprefs, XSLT_SECPREF_WRITE_FILE, xsltSecurityForbid) | xsltSetSecurityPrefs(secprefs, XSLT_SECPREF_CREATE_DIRECTORY, xsltSecurityForbid) | xsltSetSecurityPrefs(secprefs, XSLT_SECPREF_READ_NETWORK, xsltSecurityForbid) | xsltSetSecurityPrefs(secprefs, XSLT_SECPREF_WRITE_NETWORK, xsltSecurityForbid); if (ret != 0) { return; } } else { xsltFreeSecurityPrefs(secprefs); secprefs = NULL; } /* cleanup only */ if (finalize) { xsltCleanupGlobals(); } } /*! * \internal * \brief Check whether a directory entry matches the upgrade XSLT pattern * * \param[in] entry Directory entry whose filename to check * * \return 1 if the entry's filename is of the form * upgrade-X.Y-ORDER.xsl with each number in the range 0 to * 255, or 0 otherwise */ static int transform_filter(const struct dirent *entry) { const char *re = NULL; unsigned int major = 0; unsigned int minor = 0; unsigned int order = 0; /* Each number is an unsigned char, which is 1 to 3 digits long. (Pacemaker * requires an 8-bit char via a configure test.) */ re = "upgrade-[[:digit:]]{1,3}\\.[[:digit:]]{1,3}-[[:digit:]]{1,3}\\.xsl"; if (!pcmk__str_eq(entry->d_name, re, pcmk__str_regex)) { return 0; } /* Performance isn't critical here and this is simpler than range-checking * within the regex */ if (sscanf(entry->d_name, "upgrade-%u.%u-%u.xsl", &major, &minor, &order) != 3) { return 0; } if ((major > UCHAR_MAX) || (minor > UCHAR_MAX) || (order > UCHAR_MAX)) { return 0; } return 1; } /*! * \internal * \brief Compare transform files based on the version strings in their names * * This is a crude version comparison that relies on the specific structure of * these filenames. * * \retval -1 if \p entry1 sorts before \p entry2 * \retval 0 if \p entry1 sorts equal to \p entry2 * \retval 1 if \p entry1 sorts after \p entry2 * * \note The GNU \c versionsort() function would be perfect here, but it's not * portable. */ static int compare_transforms(const struct dirent **entry1, const struct dirent **entry2) { unsigned char major1 = 0; unsigned char major2 = 0; unsigned char minor1 = 0; unsigned char minor2 = 0; unsigned char order1 = 0; unsigned char order2 = 0; // If these made it through the filter, they should be of the right format CRM_LOG_ASSERT(sscanf((*entry1)->d_name, "upgrade-%hhu.%hhu-%hhu.xsl", &major1, &minor1, &order1) == 3); CRM_LOG_ASSERT(sscanf((*entry2)->d_name, "upgrade-%hhu.%hhu-%hhu.xsl", &major2, &minor2, &order2) == 3); if (major1 < major2) { return -1; } else if (major1 > major2) { return 1; } if (minor1 < minor2) { return -1; } else if (minor1 > minor2) { return 1; } if (order1 < order2) { return -1; } else if (order1 > order2) { return 1; } return 0; } /*! * \internal * \brief Free a list of XSLT transform struct dirent objects * * \param[in,out] data List to free */ static void free_transform_list(void *data) { g_list_free_full((GList *) data, free); } /*! * \internal * \brief Load names of upgrade XSLT stylesheets from a directory into a table * * Stylesheets must have names of the form "upgrade-X.Y-order.xsl", where: * * X is the schema major version * * Y is the schema minor version * * ORDER is the order in which the stylesheet occurs in the transform pipeline * * \param[in] dir Directory containing XSLT stylesheets * * \return Table with schema version as key and \c GList of associated transform * files (as struct dirent) as value */ static GHashTable * load_transforms_from_dir(const char *dir) { struct dirent **namelist = NULL; GHashTable *transforms = NULL; int num_matches = scandir(dir, &namelist, transform_filter, compare_transforms); if (num_matches < 0) { int rc = errno; crm_warn("Could not load transforms from %s: %s", dir, pcmk_rc_str(rc)); goto done; } transforms = pcmk__strkey_table(free, free_transform_list); for (int i = 0; i < num_matches; i++) { pcmk__schema_version_t version = SCHEMA_ZERO; unsigned char order = 0; // Placeholder only if (sscanf(namelist[i]->d_name, "upgrade-%hhu.%hhu-%hhu.xsl", &(version.v[0]), &(version.v[1]), &order) == 3) { char *version_s = crm_strdup_printf("%hhu.%hhu", version.v[0], version.v[1]); GList *list = g_hash_table_lookup(transforms, version_s); if (list == NULL) { /* Prepend is more efficient. However, there won't be many of * these, and we want them to remain sorted by version. It's not * worth reversing all the lists at the end. * * Avoid calling g_hash_table_insert() if the list already * exists. Otherwise free_transform_list() gets called on it. */ list = g_list_append(list, namelist[i]); g_hash_table_insert(transforms, version_s, list); } else { list = g_list_append(list, namelist[i]); free(version_s); } } else { // Sanity only, should never happen thanks to transform_filter() free(namelist[i]); } } done: free(namelist); return transforms; } void pcmk__load_schemas_from_dir(const char *dir) { int lpc, max; struct dirent **namelist = NULL; GHashTable *transforms = NULL; max = scandir(dir, &namelist, schema_filter, schema_cmp_directory); if (max < 0) { int rc = errno; crm_warn("Could not load schemas from %s: %s", dir, pcmk_rc_str(rc)); goto done; } // Look for any upgrade transforms in the same directory transforms = load_transforms_from_dir(dir); for (lpc = 0; lpc < max; lpc++) { pcmk__schema_version_t version = SCHEMA_ZERO; if (version_from_filename(namelist[lpc]->d_name, &version)) { char *version_s = crm_strdup_printf("%hhu.%hhu", version.v[0], version.v[1]); char *orig_key = NULL; GList *transform_list = NULL; if (transforms != NULL) { // The schema becomes the owner of transform_list g_hash_table_lookup_extended(transforms, version_s, (gpointer *) &orig_key, (gpointer *) &transform_list); g_hash_table_steal(transforms, version_s); } add_schema(pcmk__schema_validator_rng, &version, NULL, transform_list); free(version_s); free(orig_key); } else { // Shouldn't be possible, but makes static analysis happy crm_warn("Skipping schema '%s': could not parse version", namelist[lpc]->d_name); } } for (lpc = 0; lpc < max; lpc++) { free(namelist[lpc]); } done: free(namelist); if (transforms != NULL) { g_hash_table_destroy(transforms); } } static gint schema_sort_GCompareFunc(gconstpointer a, gconstpointer b) { const pcmk__schema_t *schema_a = a; const pcmk__schema_t *schema_b = b; // @COMPAT The "none" schema is deprecated since 2.1.8 if (pcmk__str_eq(schema_a->name, PCMK_VALUE_NONE, pcmk__str_none)) { return 1; } else if (pcmk__str_eq(schema_b->name, PCMK_VALUE_NONE, pcmk__str_none)) { return -1; } else { return schema_cmp(schema_a->version, schema_b->version); } } /*! * \internal * \brief Sort the list of known schemas such that all pacemaker-X.Y are in * version order, then "none" * * This function should be called whenever additional schemas are loaded using * \c pcmk__load_schemas_from_dir(), after the initial sets in * \c pcmk__schema_init(). */ void pcmk__sort_schemas(void) { known_schemas = g_list_sort(known_schemas, schema_sort_GCompareFunc); } /*! * \internal * \brief Load pacemaker schemas into cache * * \note This currently also serves as an entry point for the * generic initialization of the libxslt library. */ void pcmk__schema_init(void) { if (!initialized) { const char *remote_schema_dir = pcmk__remote_schema_dir(); char *base = pcmk__xml_artefact_root(pcmk__xml_artefact_ns_legacy_rng); const pcmk__schema_version_t zero = SCHEMA_ZERO; int schema_index = 0; initialized = true; wrap_libxslt(false); pcmk__load_schemas_from_dir(base); pcmk__load_schemas_from_dir(remote_schema_dir); free(base); // @COMPAT Deprecated since 2.1.8 add_schema(pcmk__schema_validator_none, &zero, PCMK_VALUE_NONE, NULL); /* add_schema() prepends items to the list, so in the simple case, this * just reverses the list. However if there were any remote schemas, * sorting is necessary. */ pcmk__sort_schemas(); // Now set the schema indexes and log the final result for (GList *iter = known_schemas; iter != NULL; iter = iter->next) { pcmk__schema_t *schema = iter->data; crm_debug("Loaded schema %d: %s", schema_index, schema->name); schema->schema_index = schema_index++; } } } static bool validate_with_relaxng(xmlDocPtr doc, xmlRelaxNGValidityErrorFunc error_handler, void *error_handler_context, const char *relaxng_file, relaxng_ctx_cache_t **cached_ctx) { int rc = 0; bool valid = true; relaxng_ctx_cache_t *ctx = NULL; CRM_CHECK(doc != NULL, return false); CRM_CHECK(relaxng_file != NULL, return false); if (cached_ctx && *cached_ctx) { ctx = *cached_ctx; } else { crm_debug("Creating RNG parser context"); ctx = pcmk__assert_alloc(1, sizeof(relaxng_ctx_cache_t)); ctx->parser = xmlRelaxNGNewParserCtxt(relaxng_file); CRM_CHECK(ctx->parser != NULL, goto cleanup); if (error_handler) { xmlRelaxNGSetParserErrors(ctx->parser, (xmlRelaxNGValidityErrorFunc) error_handler, (xmlRelaxNGValidityWarningFunc) error_handler, error_handler_context); } else { xmlRelaxNGSetParserErrors(ctx->parser, (xmlRelaxNGValidityErrorFunc) fprintf, (xmlRelaxNGValidityWarningFunc) fprintf, stderr); } ctx->rng = xmlRelaxNGParse(ctx->parser); CRM_CHECK(ctx->rng != NULL, crm_err("Could not find/parse %s", relaxng_file); goto cleanup); ctx->valid = xmlRelaxNGNewValidCtxt(ctx->rng); CRM_CHECK(ctx->valid != NULL, goto cleanup); if (error_handler) { xmlRelaxNGSetValidErrors(ctx->valid, (xmlRelaxNGValidityErrorFunc) error_handler, (xmlRelaxNGValidityWarningFunc) error_handler, error_handler_context); } else { xmlRelaxNGSetValidErrors(ctx->valid, (xmlRelaxNGValidityErrorFunc) fprintf, (xmlRelaxNGValidityWarningFunc) fprintf, stderr); } } rc = xmlRelaxNGValidateDoc(ctx->valid, doc); if (rc > 0) { valid = false; } else if (rc < 0) { crm_err("Internal libxml error during validation"); } cleanup: if (cached_ctx) { *cached_ctx = ctx; } else { if (ctx->parser != NULL) { xmlRelaxNGFreeParserCtxt(ctx->parser); } if (ctx->valid != NULL) { xmlRelaxNGFreeValidCtxt(ctx->valid); } if (ctx->rng != NULL) { xmlRelaxNGFree(ctx->rng); } free(ctx); } return valid; } static void free_schema(gpointer data) { pcmk__schema_t *schema = data; relaxng_ctx_cache_t *ctx = NULL; switch (schema->validator) { case pcmk__schema_validator_none: // not cached break; case pcmk__schema_validator_rng: // cached ctx = (relaxng_ctx_cache_t *) schema->cache; if (ctx == NULL) { break; } if (ctx->parser != NULL) { xmlRelaxNGFreeParserCtxt(ctx->parser); } if (ctx->valid != NULL) { xmlRelaxNGFreeValidCtxt(ctx->valid); } if (ctx->rng != NULL) { xmlRelaxNGFree(ctx->rng); } free(ctx); schema->cache = NULL; break; } free(schema->name); g_list_free_full(schema->transforms, free); free(schema); } /*! * \internal * \brief Clean up global memory associated with XML schemas */ void pcmk__schema_cleanup(void) { if (known_schemas != NULL) { g_list_free_full(known_schemas, free_schema); known_schemas = NULL; } initialized = false; wrap_libxslt(true); } /*! * \internal * \brief Get schema list entry corresponding to a schema name * * \param[in] name Name of schema to get * * \return Schema list entry corresponding to \p name, or NULL if unknown */ GList * pcmk__get_schema(const char *name) { if (name == NULL) { return NULL; } for (GList *iter = known_schemas; iter != NULL; iter = iter->next) { pcmk__schema_t *schema = iter->data; if (pcmk__str_eq(name, schema->name, pcmk__str_none)) { return iter; } } return NULL; } /*! * \internal * \brief Compare two schema version numbers given the schema names * * \param[in] schema1 Name of first schema to compare * \param[in] schema2 Name of second schema to compare * * \return Standard comparison result (negative integer if \p schema1 has the * lower version number, positive integer if \p schema1 has the higher * version number, of 0 if the version numbers are equal) */ int pcmk__cmp_schemas_by_name(const char *schema1_name, const char *schema2_name) { GList *entry1 = pcmk__get_schema(schema1_name); GList *entry2 = pcmk__get_schema(schema2_name); if (entry1 == NULL) { return (entry2 == NULL)? 0 : -1; } else if (entry2 == NULL) { return 1; } else { pcmk__schema_t *schema1 = entry1->data; pcmk__schema_t *schema2 = entry2->data; return schema1->schema_index - schema2->schema_index; } } static bool validate_with(xmlNode *xml, pcmk__schema_t *schema, xmlRelaxNGValidityErrorFunc error_handler, void *error_handler_context) { bool valid = false; char *file = NULL; relaxng_ctx_cache_t **cache = NULL; if (schema == NULL) { return false; } if (schema->validator == pcmk__schema_validator_none) { return true; } file = pcmk__xml_artefact_path(pcmk__xml_artefact_ns_legacy_rng, schema->name); crm_trace("Validating with %s (type=%d)", pcmk__s(file, "missing schema"), schema->validator); switch (schema->validator) { case pcmk__schema_validator_rng: cache = (relaxng_ctx_cache_t **) &(schema->cache); valid = validate_with_relaxng(xml->doc, error_handler, error_handler_context, file, cache); break; default: crm_err("Unknown validator type: %d", schema->validator); break; } free(file); return valid; } static bool validate_with_silent(xmlNode *xml, pcmk__schema_t *schema) { bool rc, sl_backup = silent_logging; silent_logging = TRUE; rc = validate_with(xml, schema, (xmlRelaxNGValidityErrorFunc) xml_log, GUINT_TO_POINTER(LOG_ERR)); silent_logging = sl_backup; return rc; } bool pcmk__validate_xml(xmlNode *xml_blob, const char *validation, xmlRelaxNGValidityErrorFunc error_handler, void *error_handler_context) { GList *entry = NULL; pcmk__schema_t *schema = NULL; CRM_CHECK((xml_blob != NULL) && (xml_blob->doc != NULL), return false); if (validation == NULL) { validation = pcmk__xe_get(xml_blob, PCMK_XA_VALIDATE_WITH); } pcmk__warn_if_schema_deprecated(validation); entry = pcmk__get_schema(validation); if (entry == NULL) { pcmk__config_err("Cannot validate CIB with %s " PCMK_XA_VALIDATE_WITH " (manually edit to use a known schema)", ((validation == NULL)? "missing" : "unknown")); return false; } schema = entry->data; return validate_with(xml_blob, schema, error_handler, error_handler_context); } /*! * \internal * \brief Validate XML using its configured schema (and send errors to logs) * * \param[in] xml XML to validate * * \return true if XML validates, otherwise false */ bool pcmk__configured_schema_validates(xmlNode *xml) { return pcmk__validate_xml(xml, NULL, (xmlRelaxNGValidityErrorFunc) xml_log, GUINT_TO_POINTER(LOG_ERR)); } /* With this arrangement, an attempt to identify the message severity as explicitly signalled directly from XSLT is performed in rather a smart way (no reliance on formatting string + arguments being always specified as ["%s", purposeful_string], as it can also be ["%s: %s", some_prefix, purposeful_string] etc. so every argument pertaining %s specifier is investigated), and if such a mark found, the respective level is determined and, when the messages are to go to the native logs, the mark itself gets dropped (by the means of string shift). NOTE: whether the native logging is the right sink is decided per the ctx parameter -- NULL denotes this case, otherwise it carries a pointer to the numeric expression of the desired target logging level (messages with higher level will be suppressed) NOTE: on some architectures, this string shift may not have any effect, but that's an acceptable tradeoff The logging level for not explicitly designated messages (suspicious, likely internal errors or some runaways) is LOG_WARNING. */ static void G_GNUC_PRINTF(2, 3) cib_upgrade_err(void *ctx, const char *fmt, ...) { va_list ap, aq; char *arg_cur; bool found = false; const char *fmt_iter = fmt; uint8_t msg_log_level = LOG_WARNING; /* default for runaway messages */ const unsigned * log_level = (const unsigned *) ctx; enum { escan_seennothing, escan_seenpercent, } scan_state = escan_seennothing; va_start(ap, fmt); va_copy(aq, ap); while (!found && *fmt_iter != '\0') { /* while casing schema borrowed from libqb:qb_vsnprintf_serialize */ switch (*fmt_iter++) { case '%': if (scan_state == escan_seennothing) { scan_state = escan_seenpercent; } else if (scan_state == escan_seenpercent) { scan_state = escan_seennothing; } break; case 's': if (scan_state == escan_seenpercent) { size_t prefix_len = 0; scan_state = escan_seennothing; arg_cur = va_arg(aq, char *); if (arg_cur == NULL) { break; } else if (g_str_has_prefix(arg_cur, "WARNING: ")) { prefix_len = sizeof("WARNING: ") - 1; msg_log_level = LOG_WARNING; } else if (g_str_has_prefix(arg_cur, "INFO: ")) { prefix_len = sizeof("INFO: ") - 1; msg_log_level = LOG_INFO; } else if (g_str_has_prefix(arg_cur, "DEBUG: ")) { prefix_len = sizeof("DEBUG: ") - 1; msg_log_level = LOG_DEBUG; } else { break; } found = true; if (ctx == NULL) { memmove(arg_cur, arg_cur + prefix_len, strlen(arg_cur + prefix_len) + 1); } } break; case '#': case '-': case ' ': case '+': case '\'': case 'I': case '.': case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': case '*': break; case 'l': case 'z': case 't': case 'j': case 'd': case 'i': case 'o': case 'u': case 'x': case 'X': case 'e': case 'E': case 'f': case 'F': case 'g': case 'G': case 'a': case 'A': case 'c': case 'p': if (scan_state == escan_seenpercent) { (void) va_arg(aq, void *); /* skip forward */ scan_state = escan_seennothing; } break; default: scan_state = escan_seennothing; break; } } if (log_level != NULL) { /* intention of the following offset is: cibadmin -V -> start showing INFO labelled messages */ if (*log_level + 4 >= msg_log_level) { vfprintf(stderr, fmt, ap); } } else { PCMK__XML_LOG_BASE(msg_log_level, TRUE, 0, "CIB upgrade: ", fmt, ap); } va_end(aq); va_end(ap); } /*! * \internal * \brief Apply a single XSL transformation to given XML * * \param[in] xml XML to transform * \param[in] transform XSL name * \param[in] to_logs If false, certain validation errors will be sent to * stderr rather than logged * * \return Transformed XML on success, otherwise NULL */ static xmlNode * apply_transformation(const xmlNode *xml, const char *transform, gboolean to_logs) { char *xform = NULL; xmlNode *out = NULL; xmlDocPtr res = NULL; xsltStylesheet *xslt = NULL; xform = pcmk__xml_artefact_path(pcmk__xml_artefact_ns_legacy_xslt, transform); /* for capturing, e.g., what's emitted via */ if (to_logs) { xsltSetGenericErrorFunc(NULL, cib_upgrade_err); } else { xsltSetGenericErrorFunc(&crm_log_level, cib_upgrade_err); } xslt = xsltParseStylesheetFile((const xmlChar *) xform); CRM_CHECK(xslt != NULL, goto cleanup); /* Caller allocates private data for final result document. Intermediate * result documents are temporary and don't need private data. */ res = xsltApplyStylesheet(xslt, xml->doc, NULL); CRM_CHECK(res != NULL, goto cleanup); xsltSetGenericErrorFunc(NULL, NULL); /* restore default one */ out = xmlDocGetRootElement(res); cleanup: if (xslt) { xsltFreeStylesheet(xslt); } free(xform); return out; } /*! * \internal * \brief Perform all transformations needed to upgrade XML to next schema * * \param[in] input_xml XML to transform * \param[in] schema_index Index of schema that successfully validates * \p original_xml * \param[in] to_logs If false, certain validation errors will be sent to * stderr rather than logged * * \return XML result of schema transforms if successful, otherwise NULL */ static xmlNode * apply_upgrade(const xmlNode *input_xml, int schema_index, gboolean to_logs) { pcmk__schema_t *schema = g_list_nth_data(known_schemas, schema_index); pcmk__schema_t *upgraded_schema = g_list_nth_data(known_schemas, schema_index + 1); xmlNode *old_xml = NULL; xmlNode *new_xml = NULL; xmlRelaxNGValidityErrorFunc error_handler = NULL; pcmk__assert((schema != NULL) && (upgraded_schema != NULL)); if (to_logs) { error_handler = (xmlRelaxNGValidityErrorFunc) xml_log; } for (GList *iter = schema->transforms; iter != NULL; iter = iter->next) { const struct dirent *entry = iter->data; const char *transform = entry->d_name; crm_debug("Upgrading schema from %s to %s: applying XSL transform %s", schema->name, upgraded_schema->name, transform); new_xml = apply_transformation(input_xml, transform, to_logs); pcmk__xml_free(old_xml); if (new_xml == NULL) { crm_err("XSL transform %s failed, aborting upgrade", transform); return NULL; } input_xml = new_xml; old_xml = new_xml; } // Final result document from upgrade pipeline needs private data pcmk__xml_new_private_data((xmlNode *) new_xml->doc); // Ensure result validates with its new schema if (!validate_with(new_xml, upgraded_schema, error_handler, GUINT_TO_POINTER(LOG_ERR))) { crm_err("Schema upgrade from %s to %s failed: " "XSL transform pipeline produced an invalid configuration", schema->name, upgraded_schema->name); crm_log_xml_debug(new_xml, "bad-transform-result"); pcmk__xml_free(new_xml); return NULL; } crm_info("Schema upgrade from %s to %s succeeded", schema->name, upgraded_schema->name); return new_xml; } /*! * \internal * \brief Get the schema list entry corresponding to XML configuration * * \param[in] xml CIB XML to check * * \return List entry of schema configured in \p xml */ static GList * get_configured_schema(const xmlNode *xml) { const char *schema_name = pcmk__xe_get(xml, PCMK_XA_VALIDATE_WITH); pcmk__warn_if_schema_deprecated(schema_name); return pcmk__get_schema(schema_name); } /*! * \brief Update CIB XML to latest schema that validates it * * \param[in,out] xml XML to update (may be freed and replaced * after being transformed) * \param[in] max_schema_name If not NULL, do not update \p xml to any * schema later than this one * \param[in] transform If false, do not update \p xml to any schema * that requires an XSL transform * \param[in] to_logs If false, certain validation errors will be * sent to stderr rather than logged * * \return Standard Pacemaker return code */ int pcmk__update_schema(xmlNode **xml, const char *max_schema_name, bool transform, bool to_logs) { int max_stable_schemas = xml_latest_schema_index(); int max_schema_index = 0; int rc = pcmk_rc_ok; GList *entry = NULL; pcmk__schema_t *best_schema = NULL; pcmk__schema_t *original_schema = NULL; xmlRelaxNGValidityErrorFunc error_handler = to_logs ? (xmlRelaxNGValidityErrorFunc) xml_log : NULL; CRM_CHECK((xml != NULL) && (*xml != NULL) && ((*xml)->doc != NULL), return EINVAL); if (max_schema_name != NULL) { GList *max_entry = pcmk__get_schema(max_schema_name); if (max_entry != NULL) { pcmk__schema_t *max_schema = max_entry->data; max_schema_index = max_schema->schema_index; } } if ((max_schema_index < 1) || (max_schema_index > max_stable_schemas)) { max_schema_index = max_stable_schemas; } entry = get_configured_schema(*xml); if (entry == NULL) { return pcmk_rc_cib_corrupt; } original_schema = entry->data; if (original_schema->schema_index >= max_schema_index) { return pcmk_rc_ok; } for (; entry != NULL; entry = entry->next) { pcmk__schema_t *current_schema = entry->data; xmlNode *upgrade = NULL; if (current_schema->schema_index > max_schema_index) { break; } if (!validate_with(*xml, current_schema, error_handler, GUINT_TO_POINTER(LOG_ERR))) { crm_debug("Schema %s does not validate", current_schema->name); if (best_schema != NULL) { /* we've satisfied the validation, no need to check further */ break; } rc = pcmk_rc_schema_validation; continue; // Try again with the next higher schema } crm_debug("Schema %s validates", current_schema->name); rc = pcmk_rc_ok; best_schema = current_schema; if (current_schema->schema_index == max_schema_index) { break; // No further transformations possible } // coverity[null_field] The index check ensures entry->next is not NULL if (!transform || (current_schema->transforms == NULL) || validate_with_silent(*xml, entry->next->data)) { /* The next schema either doesn't require a transform or validates * successfully even without the transform. Skip the transform and * try the next schema with the same XML. */ continue; } upgrade = apply_upgrade(*xml, current_schema->schema_index, to_logs); if (upgrade == NULL) { /* The transform failed, so this schema can't be used. Later * schemas are unlikely to validate, but try anyway until we * run out of options. */ rc = pcmk_rc_transform_failed; } else { best_schema = current_schema; pcmk__xml_free(*xml); *xml = upgrade; } } if ((best_schema != NULL) && (best_schema->schema_index > original_schema->schema_index)) { crm_info("%s the configuration schema to %s", (transform? "Transformed" : "Upgraded"), best_schema->name); crm_xml_add(*xml, PCMK_XA_VALIDATE_WITH, best_schema->name); } return rc; } int pcmk_update_configured_schema(xmlNode **xml) { return pcmk__update_configured_schema(xml, true); } /*! * \brief Update XML from its configured schema to the latest major series * * \param[in,out] xml XML to update * \param[in] to_logs If false, certain validation errors will be * sent to stderr rather than logged * * \return Standard Pacemaker return code */ int pcmk__update_configured_schema(xmlNode **xml, bool to_logs) { pcmk__schema_t *x_0_schema = pcmk__find_x_0_schema()->data; pcmk__schema_t *original_schema = NULL; GList *entry = NULL; if (xml == NULL) { return EINVAL; } entry = get_configured_schema(*xml); if (entry == NULL) { return pcmk_rc_cib_corrupt; } original_schema = entry->data; if (original_schema->schema_index < x_0_schema->schema_index) { // Current configuration schema is not acceptable, try to update xmlNode *converted = NULL; const char *new_schema_name = NULL; pcmk__schema_t *schema = NULL; entry = NULL; converted = pcmk__xml_copy(NULL, *xml); if (pcmk__update_schema(&converted, NULL, true, to_logs) == pcmk_rc_ok) { new_schema_name = pcmk__xe_get(converted, PCMK_XA_VALIDATE_WITH); entry = pcmk__get_schema(new_schema_name); } schema = (entry == NULL)? NULL : entry->data; if ((schema == NULL) || (schema->schema_index < x_0_schema->schema_index)) { // Updated configuration schema is still not acceptable if ((schema == NULL) || (schema->schema_index < original_schema->schema_index)) { // We couldn't validate any schema at all if (to_logs) { pcmk__config_err("Cannot upgrade configuration (claiming " "%s schema) to at least %s because it " "does not validate with any schema from " "%s to the latest", original_schema->name, x_0_schema->name, original_schema->name); } else { fprintf(stderr, "Cannot upgrade configuration (claiming " "%s schema) to at least %s because it " "does not validate with any schema from " "%s to the latest\n", original_schema->name, x_0_schema->name, original_schema->name); } } else { // We updated configuration successfully, but still too low if (to_logs) { pcmk__config_err("Cannot upgrade configuration (claiming " "%s schema) to at least %s because it " "would not upgrade past %s", original_schema->name, x_0_schema->name, pcmk__s(new_schema_name, "unspecified version")); } else { fprintf(stderr, "Cannot upgrade configuration (claiming " "%s schema) to at least %s because it " "would not upgrade past %s\n", original_schema->name, x_0_schema->name, pcmk__s(new_schema_name, "unspecified version")); } } pcmk__xml_free(converted); converted = NULL; return pcmk_rc_transform_failed; } else { // Updated configuration schema is acceptable pcmk__xml_free(*xml); *xml = converted; if (schema->schema_index < xml_latest_schema_index()) { if (to_logs) { pcmk__config_warn("Configuration with %s schema was " "internally upgraded to acceptable (but " "not most recent) %s", original_schema->name, schema->name); } } else if (to_logs) { crm_info("Configuration with %s schema was internally " "upgraded to latest version %s", original_schema->name, schema->name); } } } else if (!to_logs) { pcmk__schema_t *none_schema = NULL; entry = pcmk__get_schema(PCMK_VALUE_NONE); pcmk__assert((entry != NULL) && (entry->data != NULL)); none_schema = entry->data; if (original_schema->schema_index >= none_schema->schema_index) { // @COMPAT the none schema is deprecated since 2.1.8 fprintf(stderr, "Schema validation of configuration is " "disabled (support for " PCMK_XA_VALIDATE_WITH " set to \"" PCMK_VALUE_NONE "\" is deprecated" " and will be removed in a future release)\n"); } } return pcmk_rc_ok; } /*! * \internal * \brief Return a list of all schema files and any associated XSLT files * later than the given one * \brief Return a list of all schema versions later than the given one * * \param[in] schema The schema to compare against (for example, * "pacemaker-3.1.rng" or "pacemaker-3.1") * * \note The caller is responsible for freeing both the returned list and * the elements of the list */ GList * pcmk__schema_files_later_than(const char *name) { GList *lst = NULL; pcmk__schema_version_t ver; if (!version_from_filename(name, &ver)) { return lst; } for (GList *iter = g_list_nth(known_schemas, xml_latest_schema_index()); iter != NULL; iter = iter->prev) { pcmk__schema_t *schema = iter->data; if (schema_cmp(ver, schema->version) != -1) { continue; } for (GList *iter2 = g_list_last(schema->transforms); iter2 != NULL; iter2 = iter2->prev) { const struct dirent *entry = iter2->data; lst = g_list_prepend(lst, pcmk__str_copy(entry->d_name)); } lst = g_list_prepend(lst, crm_strdup_printf("%s.rng", schema->name)); } return lst; } static void append_href(xmlNode *xml, void *user_data) { GList **list = user_data; char *href = pcmk__xe_get_copy(xml, "href"); if (href == NULL) { return; } *list = g_list_prepend(*list, href); } static void external_refs_in_schema(GList **list, const char *contents) { /* local-name()= is needed to ignore the xmlns= setting at the top of * the XML file. Otherwise, the xpath query will always return nothing. */ const char *search = "//*[local-name()='externalRef'] | //*[local-name()='include']"; xmlNode *xml = pcmk__xml_parse(contents); pcmk__xpath_foreach_result(xml->doc, search, append_href, list); pcmk__xml_free(xml); } static int read_file_contents(const char *file, char **contents) { int rc = pcmk_rc_ok; char *path = NULL; - if (pcmk__ends_with(file, ".rng")) { + if (g_str_has_suffix(file, ".rng")) { path = pcmk__xml_artefact_path(pcmk__xml_artefact_ns_legacy_rng, file); } else { path = pcmk__xml_artefact_path(pcmk__xml_artefact_ns_legacy_xslt, file); } rc = pcmk__file_contents(path, contents); free(path); return rc; } static void add_schema_file_to_xml(xmlNode *parent, const char *file, GList **already_included) { char *contents = NULL; char *path = NULL; xmlNode *file_node = NULL; GList *includes = NULL; int rc = pcmk_rc_ok; /* If we already included this file, don't do so again. */ if (g_list_find_custom(*already_included, file, (GCompareFunc) strcmp) != NULL) { return; } /* Ensure whatever file we were given has a suffix we know about. If not, * just assume it's an RNG file. */ - if (!pcmk__ends_with(file, ".rng") && !pcmk__ends_with(file, ".xsl")) { + if (!g_str_has_suffix(file, ".rng") && !g_str_has_suffix(file, ".xsl")) { path = crm_strdup_printf("%s.rng", file); } else { path = pcmk__str_copy(file); } rc = read_file_contents(path, &contents); if (rc != pcmk_rc_ok || contents == NULL) { crm_warn("Could not read schema file %s: %s", file, pcmk_rc_str(rc)); free(path); return; } /* Create a new node with the contents of the file * as a CDATA block underneath it. */ file_node = pcmk__xe_create(parent, PCMK__XE_FILE); crm_xml_add(file_node, PCMK_XA_PATH, path); *already_included = g_list_prepend(*already_included, path); xmlAddChild(file_node, xmlNewCDataBlock(parent->doc, (const xmlChar *) contents, strlen(contents))); /* Scan the file for any or nodes and build up * a list of the files they reference. */ external_refs_in_schema(&includes, contents); /* For each referenced file, recurse to add it (and potentially anything it * references, ...) to the XML. */ for (GList *iter = includes; iter != NULL; iter = iter->next) { add_schema_file_to_xml(parent, iter->data, already_included); } free(contents); g_list_free_full(includes, free); } /*! * \internal * \brief Add an XML schema file and all the files it references as children * of a given XML node * * \param[in,out] parent The parent XML node * \param[in] name The schema version to compare against * (for example, "pacemaker-3.1" or "pacemaker-3.1.rng") * \param[in,out] already_included A list of names that have already been added * to the parent node. * * \note The caller is responsible for freeing both the returned list and * the elements of the list */ void pcmk__build_schema_xml_node(xmlNode *parent, const char *name, GList **already_included) { xmlNode *schema_node = pcmk__xe_create(parent, PCMK__XA_SCHEMA); crm_xml_add(schema_node, PCMK_XA_VERSION, name); add_schema_file_to_xml(schema_node, name, already_included); if (schema_node->children == NULL) { // Not needed if empty. May happen if name was invalid, for example. pcmk__xml_free(schema_node); } } /*! * \internal * \brief Return the directory containing any extra schema files that a * Pacemaker Remote node fetched from the cluster */ const char * pcmk__remote_schema_dir(void) { const char *dir = pcmk__env_option(PCMK__ENV_REMOTE_SCHEMA_DIRECTORY); if (pcmk__str_empty(dir)) { return PCMK__REMOTE_SCHEMA_DIR; } return dir; } /*! * \internal * \brief Warn if a given validation schema is deprecated * * \param[in] Schema name to check */ void pcmk__warn_if_schema_deprecated(const char *schema) { /* @COMPAT Disabling validation is deprecated since 2.1.8, but * resource-agents' ocf-shellfuncs (at least as of 4.15.1) uses it */ if (pcmk__str_eq(schema, PCMK_VALUE_NONE, pcmk__str_none)) { pcmk__config_warn("Support for " PCMK_XA_VALIDATE_WITH "='%s' is " "deprecated and will be removed in a future release " "without the possibility of upgrades (manually edit " "to use a supported schema)", schema); } } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include gboolean cli_config_update(xmlNode **xml, int *best_version, gboolean to_logs) { int rc = pcmk__update_configured_schema(xml, to_logs); if (best_version != NULL) { const char *name = pcmk__xe_get(*xml, PCMK_XA_VALIDATE_WITH); if (name == NULL) { *best_version = -1; } else { GList *entry = pcmk__get_schema(name); pcmk__schema_t *schema = (entry == NULL)? NULL : entry->data; *best_version = (schema == NULL)? -1 : schema->schema_index; } } return (rc == pcmk_rc_ok)? TRUE: FALSE; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/strings.c b/lib/common/strings.c index d149855dca..af8e1bb8ac 100644 --- a/lib/common/strings.c +++ b/lib/common/strings.c @@ -1,1326 +1,1292 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // DBL_MIN #include #include #include /*! * \internal * \brief Scan a long long integer from a string * * \param[in] text String to scan * \param[out] result If not NULL, where to store scanned value * \param[in] default_value Value to use if text is NULL or invalid * \param[out] end_text If not NULL, where to store pointer to first * non-integer character * * \return Standard Pacemaker return code (\c pcmk_rc_ok on success, * \c pcmk_rc_bad_input on failed string conversion due to invalid * input, or \c ERANGE if outside long long range) * \note Sets \c errno on error */ static int scan_ll(const char *text, long long *result, long long default_value, char **end_text) { long long local_result = default_value; char *local_end_text = NULL; int rc = pcmk_rc_ok; errno = 0; if (text != NULL) { local_result = strtoll(text, &local_end_text, 10); if (errno == ERANGE) { rc = errno; crm_debug("Integer parsed from '%s' was clipped to %lld", text, local_result); } else if (local_end_text == text) { rc = pcmk_rc_bad_input; local_result = default_value; crm_debug("Could not parse integer from '%s' (using %lld instead): " "No digits found", text, default_value); } else if (errno != 0) { rc = errno; local_result = default_value; crm_debug("Could not parse integer from '%s' (using %lld instead): " "%s", text, default_value, pcmk_rc_str(rc)); } if ((end_text == NULL) && !pcmk__str_empty(local_end_text)) { crm_debug("Characters left over after parsing '%s': '%s'", text, local_end_text); } errno = rc; } if (end_text != NULL) { *end_text = local_end_text; } if (result != NULL) { *result = local_result; } return rc; } /*! * \internal * \brief Scan a long long integer value from a string * * \param[in] text The string to scan (may be NULL) * \param[out] result Where to store result (or NULL to ignore) * \param[in] default_value Value to use if text is NULL or invalid * * \return Standard Pacemaker return code */ int pcmk__scan_ll(const char *text, long long *result, long long default_value) { long long local_result = default_value; int rc = scan_ll(text, &local_result, default_value, NULL); if (result != NULL) { *result = local_result; } return rc; } /*! * \internal * \brief Scan an integer value from a string, constrained to a minimum * * \param[in] text The string to scan (may be NULL) * \param[out] result Where to store result (or NULL to ignore) * \param[in] minimum Value to use as default and minimum * * \return Standard Pacemaker return code * \note If the value is larger than the maximum integer, EOVERFLOW will be * returned and \p result will be set to the maximum integer. */ int pcmk__scan_min_int(const char *text, int *result, int minimum) { int rc; long long result_ll; rc = pcmk__scan_ll(text, &result_ll, (long long) minimum); if (result_ll < (long long) minimum) { crm_warn("Clipped '%s' to minimum acceptable value %d", text, minimum); result_ll = (long long) minimum; } else if (result_ll > INT_MAX) { crm_warn("Clipped '%s' to maximum integer %d", text, INT_MAX); result_ll = (long long) INT_MAX; rc = EOVERFLOW; } if (result != NULL) { *result = (int) result_ll; } return rc; } /*! * \internal * \brief Scan a TCP port number from a string * * \param[in] text The string to scan * \param[out] port Where to store result (or NULL to ignore) * * \return Standard Pacemaker return code * \note \p port will be -1 if \p text is NULL or invalid */ int pcmk__scan_port(const char *text, int *port) { long long port_ll; int rc = pcmk__scan_ll(text, &port_ll, -1LL); if (rc != pcmk_rc_ok) { crm_warn("'%s' is not a valid port: %s", text, pcmk_rc_str(rc)); } else if ((text != NULL) // wasn't default or invalid && ((port_ll < 0LL) || (port_ll > 65535LL))) { crm_warn("Ignoring port specification '%s' " "not in valid range (0-65535)", text); rc = (port_ll < 0LL)? pcmk_rc_before_range : pcmk_rc_after_range; port_ll = -1LL; } if (port != NULL) { *port = (int) port_ll; } return rc; } /*! * \internal * \brief Scan a double-precision floating-point value from a string * * \param[in] text The string to parse * \param[out] result Parsed value on success, or * \c PCMK__PARSE_DBL_DEFAULT on error * \param[in] default_text Default string to parse if \p text is * \c NULL * \param[out] end_text If not \c NULL, where to store a pointer * to the position immediately after the * value * * \return Standard Pacemaker return code (\c pcmk_rc_ok on success, * \c EINVAL on failed string conversion due to invalid input, * \c EOVERFLOW on arithmetic overflow, \c pcmk_rc_underflow * on arithmetic underflow, or \c errno from \c strtod() on * other parse errors) */ int pcmk__scan_double(const char *text, double *result, const char *default_text, char **end_text) { int rc = pcmk_rc_ok; char *local_end_text = NULL; pcmk__assert(result != NULL); *result = PCMK__PARSE_DBL_DEFAULT; text = (text != NULL) ? text : default_text; if (text == NULL) { rc = EINVAL; crm_debug("No text and no default conversion value supplied"); } else { errno = 0; *result = strtod(text, &local_end_text); if (errno == ERANGE) { /* * Overflow: strtod() returns +/- HUGE_VAL and sets errno to * ERANGE * * Underflow: strtod() returns "a value whose magnitude is * no greater than the smallest normalized * positive" double. Whether ERANGE is set is * implementation-defined. */ const char *over_under; if (QB_ABS(*result) > DBL_MIN) { rc = EOVERFLOW; over_under = "over"; } else { rc = pcmk_rc_underflow; over_under = "under"; } crm_debug("Floating-point value parsed from '%s' would %sflow " "(using %g instead)", text, over_under, *result); } else if (errno != 0) { rc = errno; // strtod() set *result = 0 on parse failure *result = PCMK__PARSE_DBL_DEFAULT; crm_debug("Could not parse floating-point value from '%s' (using " "%.1f instead): %s", text, PCMK__PARSE_DBL_DEFAULT, pcmk_rc_str(rc)); } else if (local_end_text == text) { // errno == 0, but nothing was parsed rc = EINVAL; *result = PCMK__PARSE_DBL_DEFAULT; crm_debug("Could not parse floating-point value from '%s' (using " "%.1f instead): No digits found", text, PCMK__PARSE_DBL_DEFAULT); } else if (QB_ABS(*result) <= DBL_MIN) { /* * errno == 0 and text was parsed, but value might have * underflowed. * * ERANGE might not be set for underflow. Check magnitude * of *result, but also make sure the input number is not * actually zero (0 <= DBL_MIN is not underflow). * * This check must come last. A parse failure in strtod() * also sets *result == 0, so a parse failure would match * this test condition prematurely. */ for (const char *p = text; p != local_end_text; p++) { if (strchr("0.eE", *p) == NULL) { rc = pcmk_rc_underflow; crm_debug("Floating-point value parsed from '%s' would " "underflow (using %g instead)", text, *result); break; } } } else { crm_trace("Floating-point value parsed successfully from " "'%s': %g", text, *result); } if ((end_text == NULL) && !pcmk__str_empty(local_end_text)) { crm_debug("Characters left over after parsing '%s': '%s'", text, local_end_text); } } if (end_text != NULL) { *end_text = local_end_text; } return rc; } /*! * \internal * \brief Parse a guint from a string stored in a hash table * * \param[in] table Hash table to search * \param[in] key Hash table key to use to retrieve string * \param[in] default_val What to use if key has no entry in table * \param[out] result If not NULL, where to store parsed integer * * \return Standard Pacemaker return code */ int pcmk__guint_from_hash(GHashTable *table, const char *key, guint default_val, guint *result) { const char *value; long long value_ll; int rc = pcmk_rc_ok; CRM_CHECK((table != NULL) && (key != NULL), return EINVAL); if (result != NULL) { *result = default_val; } value = g_hash_table_lookup(table, key); if (value == NULL) { return pcmk_rc_ok; } rc = pcmk__scan_ll(value, &value_ll, 0LL); if (rc != pcmk_rc_ok) { crm_warn("Using default (%u) for %s because '%s' is not a " "valid integer: %s", default_val, key, value, pcmk_rc_str(rc)); return rc; } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { crm_warn("Using default (%u) for %s because '%s' is not in valid range", default_val, key, value); return ERANGE; } if (result != NULL) { *result = (guint) value_ll; } return pcmk_rc_ok; } /*! * \brief Parse a time+units string and return milliseconds equivalent * * \param[in] input String with a nonnegative number and optional unit * (optionally with whitespace before and/or after the * number). If missing, the unit defaults to seconds. * * \return Milliseconds corresponding to string expression, or * \c PCMK__PARSE_INT_DEFAULT on error */ long long crm_get_msec(const char *input) { char *units = NULL; // Do not free; will point to part of input long long multiplier = 1000; long long divisor = 1; long long msec = PCMK__PARSE_INT_DEFAULT; int rc = pcmk_rc_ok; if (input == NULL) { return PCMK__PARSE_INT_DEFAULT; } // Skip initial whitespace while (isspace(*input)) { input++; } rc = scan_ll(input, &msec, PCMK__PARSE_INT_DEFAULT, &units); if ((rc == ERANGE) && (msec > 0)) { crm_warn("'%s' will be clipped to %lld", input, msec); } else if ((rc != pcmk_rc_ok) || (msec < 0)) { crm_warn("'%s' is not a valid time duration: %s", input, ((rc == pcmk_rc_ok)? "Negative" : pcmk_rc_str(rc))); return PCMK__PARSE_INT_DEFAULT; } /* If the number is a decimal, scan_ll() reads only the integer part. Skip * any remaining digits or decimal characters. * * @COMPAT Well-formed and malformed decimals are both accepted inputs. For * example, "3.14 ms" and "3.1.4 ms" are treated the same as "3ms" and * parsed successfully. At a compatibility break, decide if this is still * desired. */ while (isdigit(*units) || (*units == '.')) { units++; } // Skip any additional whitespace after the number while (isspace(*units)) { units++; } /* @COMPAT Use exact comparisons. Currently, we match too liberally, and the * second strncasecmp() in each case is redundant. */ if ((*units == '\0') || (strncasecmp(units, "s", 1) == 0) || (strncasecmp(units, "sec", 3) == 0)) { multiplier = 1000; divisor = 1; } else if ((strncasecmp(units, "ms", 2) == 0) || (strncasecmp(units, "msec", 4) == 0)) { multiplier = 1; divisor = 1; } else if ((strncasecmp(units, "us", 2) == 0) || (strncasecmp(units, "usec", 4) == 0)) { multiplier = 1; divisor = 1000; } else if ((strncasecmp(units, "m", 1) == 0) || (strncasecmp(units, "min", 3) == 0)) { multiplier = 60 * 1000; divisor = 1; } else if ((strncasecmp(units, "h", 1) == 0) || (strncasecmp(units, "hr", 2) == 0)) { multiplier = 60 * 60 * 1000; divisor = 1; } else { // Invalid units return PCMK__PARSE_INT_DEFAULT; } // Apply units, capping at LLONG_MAX if (msec > (LLONG_MAX / multiplier)) { return LLONG_MAX; } return (msec * multiplier) / divisor; } /*! * \brief Parse milliseconds from a Pacemaker interval specification * * \param[in] input Pacemaker time interval specification (a bare number * of seconds; a number with a unit, optionally with * whitespace before and/or after the number; or an ISO * 8601 duration) * \param[out] result_ms Where to store milliseconds equivalent of \p input on * success (limited to the range of an unsigned integer), * or 0 if \p input is \c NULL or invalid * * \return Standard Pacemaker return code (specifically, \c pcmk_rc_ok if * \p input is valid or \c NULL, and \c EINVAL otherwise) */ int pcmk_parse_interval_spec(const char *input, guint *result_ms) { long long msec = PCMK__PARSE_INT_DEFAULT; int rc = pcmk_rc_ok; if (input == NULL) { msec = 0; goto done; } if (input[0] == 'P') { crm_time_t *period_s = crm_time_parse_duration(input); if (period_s != NULL) { msec = crm_time_get_seconds(period_s); msec = QB_MIN(msec, G_MAXUINT / 1000) * 1000; crm_time_free(period_s); } } else { msec = crm_get_msec(input); } if (msec < 0) { crm_warn("Using 0 instead of invalid interval specification '%s'", input); msec = 0; rc = EINVAL; } done: if (result_ms != NULL) { *result_ms = (msec >= G_MAXUINT)? G_MAXUINT : (guint) msec; } return rc; } gboolean crm_is_true(const char *s) { gboolean ret = FALSE; return (crm_str_to_boolean(s, &ret) < 0)? FALSE : ret; } int crm_str_to_boolean(const char *s, int *ret) { if (s == NULL) { return -1; } if (pcmk__strcase_any_of(s, PCMK_VALUE_TRUE, "on", "yes", "y", "1", NULL)) { if (ret != NULL) { *ret = TRUE; } return 1; } if (pcmk__strcase_any_of(s, PCMK_VALUE_FALSE, PCMK_VALUE_OFF, "no", "n", "0", NULL)) { if (ret != NULL) { *ret = FALSE; } return 1; } return -1; } /*! * \internal * \brief Replace any trailing newlines in a string with \0's * * \param[in,out] str String to trim * * \return \p str */ char * pcmk__trim(char *str) { int len; if (str == NULL) { return str; } for (len = strlen(str) - 1; len >= 0 && str[len] == '\n'; len--) { str[len] = '\0'; } return str; } -static inline bool -ends_with(const char *s, const char *match, bool as_extension) -{ - if (pcmk__str_empty(match)) { - return true; - } else if (s == NULL) { - return false; - } else { - size_t slen, mlen; - - /* Besides as_extension, we could also check - !strchr(&match[1], match[0]) but that would be inefficient. - */ - if (as_extension) { - s = strrchr(s, match[0]); - return (s == NULL)? false : !strcmp(s, match); - } - - mlen = strlen(match); - slen = strlen(s); - return ((slen >= mlen) && !strcmp(s + slen - mlen, match)); - } -} - -/*! - * \internal - * \brief Check whether a string ends with a certain sequence - * - * \param[in] s String to check - * \param[in] match Sequence to match against end of \p s - * - * \return \c true if \p s ends case-sensitively with match, \c false otherwise - * \note pcmk__ends_with_ext() can be used if the first character of match - * does not recur in match. - */ -bool -pcmk__ends_with(const char *s, const char *match) -{ - return ends_with(s, match, false); -} - /*! * \internal * \brief Check whether a string ends with a certain "extension" * * \param[in] s String to check * \param[in] match Extension to match against end of \p s, that is, * its first character must not occur anywhere * in the rest of that very sequence (example: file * extension where the last dot is its delimiter, * e.g., ".html"); incorrect results may be * returned otherwise. * * \return \c true if \p s ends (verbatim, i.e., case sensitively) * with "extension" designated as \p match (including empty * string), \c false otherwise * - * \note Main incentive to prefer this function over \c pcmk__ends_with() + * \note Main incentive to prefer this function over \c g_str_has_suffix() * where possible is the efficiency (at the cost of added * restriction on \p match as stated; the complexity class * remains the same, though: BigO(M+N) vs. BigO(M+2N)). */ bool pcmk__ends_with_ext(const char *s, const char *match) { - return ends_with(s, match, true); + if (pcmk__str_empty(match)) { + return true; + } else if (s == NULL) { + return false; + } else { + s = strrchr(s, match[0]); + return (s == NULL)? false : !strcmp(s, match); + } } /*! * \internal * \brief Create a hash of a string suitable for use with GHashTable * * \param[in] v String to hash * * \return A hash of \p v compatible with g_str_hash() before glib 2.28 * \note glib changed their hash implementation: * * https://gitlab.gnome.org/GNOME/glib/commit/354d655ba8a54b754cb5a3efb42767327775696c * * Note that the new g_str_hash is presumably a *better* hash (it's actually * a correct implementation of DJB's hash), but we need to preserve existing * behaviour, because the hash key ultimately determines the "sort" order * when iterating through GHashTables, which affects allocation of scores to * clone instances when iterating through allowed nodes. It (somehow) also * appears to have some minor impact on the ordering of a few pseudo_event IDs * in the transition graph. */ static guint pcmk__str_hash(gconstpointer v) { const signed char *p; guint32 h = 0; for (p = v; *p != '\0'; p++) h = (h << 5) - h + *p; return h; } /*! * \internal * \brief Create a hash table with case-sensitive strings as keys * * \param[in] key_destroy_func Function to free a key * \param[in] value_destroy_func Function to free a value * * \return Newly allocated hash table * \note It is the caller's responsibility to free the result, using * g_hash_table_destroy(). */ GHashTable * pcmk__strkey_table(GDestroyNotify key_destroy_func, GDestroyNotify value_destroy_func) { return g_hash_table_new_full(pcmk__str_hash, g_str_equal, key_destroy_func, value_destroy_func); } /*! * \internal * \brief Insert string copies into a hash table as key and value * * \param[in,out] table Hash table to add to * \param[in] name String to add a copy of as key * \param[in] value String to add a copy of as value * * \note This asserts on invalid arguments or memory allocation failure. */ void pcmk__insert_dup(GHashTable *table, const char *name, const char *value) { pcmk__assert((table != NULL) && (name != NULL)); g_hash_table_insert(table, pcmk__str_copy(name), pcmk__str_copy(value)); } /* used with hash tables where case does not matter */ static gboolean pcmk__strcase_equal(gconstpointer a, gconstpointer b) { return pcmk__str_eq((const char *)a, (const char *)b, pcmk__str_casei); } static guint pcmk__strcase_hash(gconstpointer v) { const signed char *p; guint32 h = 0; for (p = v; *p != '\0'; p++) h = (h << 5) - h + g_ascii_tolower(*p); return h; } /*! * \internal * \brief Create a hash table with case-insensitive strings as keys * * \param[in] key_destroy_func Function to free a key * \param[in] value_destroy_func Function to free a value * * \return Newly allocated hash table * \note It is the caller's responsibility to free the result, using * g_hash_table_destroy(). */ GHashTable * pcmk__strikey_table(GDestroyNotify key_destroy_func, GDestroyNotify value_destroy_func) { return g_hash_table_new_full(pcmk__strcase_hash, pcmk__strcase_equal, key_destroy_func, value_destroy_func); } static void copy_str_table_entry(gpointer key, gpointer value, gpointer user_data) { if (key && value && user_data) { pcmk__insert_dup((GHashTable *) user_data, (const char *) key, (const char *) value); } } /*! * \internal * \brief Copy a hash table that uses dynamically allocated strings * * \param[in,out] old_table Hash table to duplicate * * \return New hash table with copies of everything in \p old_table * \note This assumes the hash table uses dynamically allocated strings -- that * is, both the key and value free functions are free(). */ GHashTable * pcmk__str_table_dup(GHashTable *old_table) { GHashTable *new_table = NULL; if (old_table) { new_table = pcmk__strkey_table(free, free); g_hash_table_foreach(old_table, copy_str_table_entry, new_table); } return new_table; } /*! * \internal * \brief Add a word to a string list of words * * \param[in,out] list Pointer to current string list (may not be \p NULL) * \param[in] init_size \p list will be initialized to at least this size, * if it needs initialization (if 0, use GLib's default * initial string size) * \param[in] word String to add to \p list (\p list will be * unchanged if this is \p NULL or the empty string) * \param[in] separator String to separate words in \p list * (a space will be used if this is NULL) * * \note \p word may contain \p separator, though that would be a bad idea if * the string needs to be parsed later. */ void pcmk__add_separated_word(GString **list, size_t init_size, const char *word, const char *separator) { pcmk__assert(list != NULL); if (pcmk__str_empty(word)) { return; } if (*list == NULL) { if (init_size > 0) { *list = g_string_sized_new(init_size); } else { *list = g_string_new(NULL); } } if ((*list)->len == 0) { // Don't add a separator before the first word in the list separator = ""; } else if (separator == NULL) { // Default to space-separated separator = " "; } g_string_append(*list, separator); g_string_append(*list, word); } /*! * \internal * \brief Compress data * * \param[in] data Data to compress * \param[in] length Number of characters of data to compress * \param[in] max Maximum size of compressed data (or 0 to estimate) * \param[out] result Where to store newly allocated compressed result * \param[out] result_len Where to store actual compressed length of result * * \return Standard Pacemaker return code */ int pcmk__compress(const char *data, unsigned int length, unsigned int max, char **result, unsigned int *result_len) { int rc; char *compressed = NULL; char *uncompressed = strdup(data); #ifdef CLOCK_MONOTONIC struct timespec after_t; struct timespec before_t; #endif if (max == 0) { max = (length * 1.01) + 601; // Size guaranteed to hold result } #ifdef CLOCK_MONOTONIC clock_gettime(CLOCK_MONOTONIC, &before_t); #endif compressed = pcmk__assert_alloc((size_t) max, sizeof(char)); *result_len = max; rc = BZ2_bzBuffToBuffCompress(compressed, result_len, uncompressed, length, CRM_BZ2_BLOCKS, 0, CRM_BZ2_WORK); rc = pcmk__bzlib2rc(rc); free(uncompressed); if (rc != pcmk_rc_ok) { crm_err("Compression of %d bytes failed: %s " QB_XS " rc=%d", length, pcmk_rc_str(rc), rc); free(compressed); return rc; } #ifdef CLOCK_MONOTONIC clock_gettime(CLOCK_MONOTONIC, &after_t); crm_trace("Compressed %d bytes into %d (ratio %d:1) in %.0fms", length, *result_len, length / (*result_len), (after_t.tv_sec - before_t.tv_sec) * 1000 + (after_t.tv_nsec - before_t.tv_nsec) / 1e6); #else crm_trace("Compressed %d bytes into %d (ratio %d:1)", length, *result_len, length / (*result_len)); #endif *result = compressed; return pcmk_rc_ok; } char * crm_strdup_printf(char const *format, ...) { va_list ap; int len = 0; char *string = NULL; va_start(ap, format); len = vasprintf(&string, format, ap); pcmk__assert(len > 0); va_end(ap); return string; } /*! * \internal * \brief Parse a range specification string * * A valid range specification string can be in any of the following forms, * where \c "X", \c "Y", and \c "Z" are nonnegative integers that fit into a * long long variable: * * "X-Y" * * "X-" * * "-Y" * * "Z" * * In the list above, \c "X" is the start value and \c "Y" is the end value of * the range. Either the start value or the end value, but not both, can be * empty. \c "Z", a single integer with no \c '-' character, is both the start * value and the end value of its range. * * If the start value or end value is empty, then the parsed result stored in * \p *start or \p *end (respectively) is \c PCMK__PARSE_INT_DEFAULT after a * successful parse. * * If the specification string consists of only a single number, then the same * value is stored in both \p *start and \p *end on a successful parse. * * \param[in] text String to parse * \param[out] start Where to store start value (can be \c NULL) * \param[out] end Where to store end value (can be \c NULL) * * \return Standard Pacemaker return code * * \note The values stored in \p *start and \p *end are undefined if the return * value is not \c pcmk_rc_ok. */ int pcmk__parse_ll_range(const char *text, long long *start, long long *end) { int rc = pcmk_rc_ok; long long local_start = 0; long long local_end = 0; gchar **split = NULL; guint length = 0; const gchar *start_s = NULL; const gchar *end_s = NULL; // Do not free char *remainder = NULL; if (start == NULL) { start = &local_start; } if (end == NULL) { end = &local_end; } *start = PCMK__PARSE_INT_DEFAULT; *end = PCMK__PARSE_INT_DEFAULT; if (pcmk__str_empty(text)) { rc = ENODATA; goto done; } split = g_strsplit(text, "-", 2); length = g_strv_length(split); start_s = split[0]; if (length == 2) { end_s = split[1]; } if (pcmk__str_empty(start_s) && pcmk__str_empty(end_s)) { rc = pcmk_rc_bad_input; goto done; } if (!pcmk__str_empty(start_s)) { rc = scan_ll(start_s, start, PCMK__PARSE_INT_DEFAULT, &remainder); if (rc != pcmk_rc_ok) { goto done; } if (!pcmk__str_empty(remainder)) { rc = pcmk_rc_bad_input; goto done; } } if (length == 1) { // String contains only a single number, which is both start and end *end = *start; goto done; } if (!pcmk__str_empty(end_s)) { rc = scan_ll(end_s, end, PCMK__PARSE_INT_DEFAULT, &remainder); if ((rc == pcmk_rc_ok) && !pcmk__str_empty(remainder)) { rc = pcmk_rc_bad_input; } } done: g_strfreev(split); return rc; } /*! * \internal * \brief Find a string in a list of strings * * \note This function takes the same flags and has the same behavior as * pcmk__str_eq(). * * \note No matter what input string or flags are provided, an empty * list will always return FALSE. * * \param[in] s String to search for * \param[in] lst List to search * \param[in] flags A bitfield of pcmk__str_flags to modify operation * * \return \c TRUE if \p s is in \p lst, or \c FALSE otherwise */ gboolean pcmk__str_in_list(const gchar *s, const GList *lst, uint32_t flags) { for (const GList *ele = lst; ele != NULL; ele = ele->next) { if (pcmk__str_eq(s, ele->data, flags)) { return TRUE; } } return FALSE; } static bool str_any_of(const char *s, va_list args, uint32_t flags) { if (s == NULL) { return false; } while (1) { const char *ele = va_arg(args, const char *); if (ele == NULL) { break; } else if (pcmk__str_eq(s, ele, flags)) { return true; } } return false; } /*! * \internal * \brief Is a string a member of a list of strings? * * \param[in] s String to search for in \p ... * \param[in] ... Strings to compare \p s against. The final string * must be NULL. * * \note The comparison is done case-insensitively. The function name is * meant to be reminiscent of strcasecmp. * * \return \c true if \p s is in \p ..., or \c false otherwise */ bool pcmk__strcase_any_of(const char *s, ...) { va_list ap; bool rc; va_start(ap, s); rc = str_any_of(s, ap, pcmk__str_casei); va_end(ap); return rc; } /*! * \internal * \brief Is a string a member of a list of strings? * * \param[in] s String to search for in \p ... * \param[in] ... Strings to compare \p s against. The final string * must be NULL. * * \note The comparison is done taking case into account. * * \return \c true if \p s is in \p ..., or \c false otherwise */ bool pcmk__str_any_of(const char *s, ...) { va_list ap; bool rc; va_start(ap, s); rc = str_any_of(s, ap, pcmk__str_none); va_end(ap); return rc; } /*! * \internal * \brief Sort strings, with numeric portions sorted numerically * * Sort two strings case-insensitively like strcasecmp(), but with any numeric * portions of the string sorted numerically. This is particularly useful for * node names (for example, "node10" will sort higher than "node9" but lower * than "remotenode9"). * * \param[in] s1 First string to compare (must not be NULL) * \param[in] s2 Second string to compare (must not be NULL) * * \retval -1 \p s1 comes before \p s2 * \retval 0 \p s1 and \p s2 are equal * \retval 1 \p s1 comes after \p s2 */ int pcmk__numeric_strcasecmp(const char *s1, const char *s2) { pcmk__assert((s1 != NULL) && (s2 != NULL)); while (*s1 && *s2) { if (isdigit(*s1) && isdigit(*s2)) { // If node names contain a number, sort numerically char *end1 = NULL; char *end2 = NULL; long num1 = strtol(s1, &end1, 10); long num2 = strtol(s2, &end2, 10); // allow ordering e.g. 007 > 7 size_t len1 = end1 - s1; size_t len2 = end2 - s2; if (num1 < num2) { return -1; } else if (num1 > num2) { return 1; } else if (len1 < len2) { return -1; } else if (len1 > len2) { return 1; } s1 = end1; s2 = end2; } else { // Compare non-digits case-insensitively int lower1 = tolower(*s1); int lower2 = tolower(*s2); if (lower1 < lower2) { return -1; } else if (lower1 > lower2) { return 1; } ++s1; ++s2; } } if (!*s1 && *s2) { return -1; } else if (*s1 && !*s2) { return 1; } return 0; } /*! * \internal * \brief Sort strings. * * This is your one-stop function for string comparison. By default, this * function works like \p g_strcmp0. That is, like \p strcmp but a \p NULL * string sorts before a non-NULL string. * * The \p pcmk__str_none flag produces the default behavior. Behavior can be * changed with various flags: * * - \p pcmk__str_regex - The second string is a regular expression that the * first string will be matched against. * - \p pcmk__str_casei - By default, comparisons are done taking case into * account. This flag makes comparisons case- * insensitive. This can be combined with * \p pcmk__str_regex. * - \p pcmk__str_null_matches - If one string is \p NULL and the other is not, * still return \p 0. * - \p pcmk__str_star_matches - If one string is \p "*" and the other is not, * still return \p 0. * * \param[in] s1 First string to compare * \param[in] s2 Second string to compare, or a regular expression to * match if \p pcmk__str_regex is set * \param[in] flags A bitfield of \p pcmk__str_flags to modify operation * * \retval negative \p s1 is \p NULL or comes before \p s2 * \retval 0 \p s1 and \p s2 are equal, or \p s1 is found in \p s2 if * \c pcmk__str_regex is set * \retval positive \p s2 is \p NULL or \p s1 comes after \p s2, or \p s2 * is an invalid regular expression, or \p s1 was not found * in \p s2 if \p pcmk__str_regex is set. */ int pcmk__strcmp(const char *s1, const char *s2, uint32_t flags) { /* If this flag is set, the second string is a regex. */ if (pcmk_is_set(flags, pcmk__str_regex)) { regex_t r_patt; int reg_flags = REG_EXTENDED | REG_NOSUB; int regcomp_rc = 0; int rc = 0; if (s1 == NULL || s2 == NULL) { return 1; } if (pcmk_is_set(flags, pcmk__str_casei)) { reg_flags |= REG_ICASE; } regcomp_rc = regcomp(&r_patt, s2, reg_flags); if (regcomp_rc != 0) { rc = 1; crm_err("Bad regex '%s' for update: %s", s2, strerror(regcomp_rc)); } else { rc = regexec(&r_patt, s1, 0, NULL, 0); regfree(&r_patt); if (rc != 0) { rc = 1; } } return rc; } /* If the strings are the same pointer, return 0 immediately. */ if (s1 == s2) { return 0; } /* If this flag is set, return 0 if either (or both) of the input strings * are NULL. If neither one is NULL, we need to continue and compare * them normally. */ if (pcmk_is_set(flags, pcmk__str_null_matches)) { if (s1 == NULL || s2 == NULL) { return 0; } } /* Handle the cases where one is NULL and the str_null_matches flag is not set. * A NULL string always sorts to the beginning. */ if (s1 == NULL) { return -1; } else if (s2 == NULL) { return 1; } /* If this flag is set, return 0 if either (or both) of the input strings * are "*". If neither one is, we need to continue and compare them * normally. */ if (pcmk_is_set(flags, pcmk__str_star_matches)) { if (strcmp(s1, "*") == 0 || strcmp(s2, "*") == 0) { return 0; } } if (pcmk_is_set(flags, pcmk__str_casei)) { return strcasecmp(s1, s2); } else { return strcmp(s1, s2); } } /*! * \internal * \brief Copy a string, asserting on failure * * \param[in] file File where \p function is located * \param[in] function Calling function * \param[in] line Line within \p file * \param[in] str String to copy (can be \c NULL) * * \return Newly allocated copy of \p str, or \c NULL if \p str is \c NULL * * \note The caller is responsible for freeing the return value using \c free(). */ char * pcmk__str_copy_as(const char *file, const char *function, uint32_t line, const char *str) { if (str != NULL) { char *result = strdup(str); if (result == NULL) { crm_abort(file, function, line, "Out of memory", FALSE, TRUE); crm_exit(CRM_EX_OSERR); } return result; } return NULL; } /*! * \internal * \brief Update a dynamically allocated string with a new value * * Given a dynamically allocated string and a new value for it, if the string * is different from the new value, free the string and replace it with either a * newly allocated duplicate of the value or NULL as appropriate. * * \param[in,out] str Pointer to dynamically allocated string * \param[in] value New value to duplicate (or NULL) * * \note The caller remains responsibile for freeing \p *str. */ void pcmk__str_update(char **str, const char *value) { if ((str != NULL) && !pcmk__str_eq(*str, value, pcmk__str_none)) { free(*str); *str = pcmk__str_copy(value); } } /*! * \internal * \brief Append a list of strings to a destination \p GString * * \param[in,out] buffer Where to append the strings (must not be \p NULL) * \param[in] ... A NULL-terminated list of strings * * \note This tends to be more efficient than a single call to * \p g_string_append_printf(). */ void pcmk__g_strcat(GString *buffer, ...) { va_list ap; pcmk__assert(buffer != NULL); va_start(ap, buffer); while (true) { const char *ele = va_arg(ap, const char *); if (ele == NULL) { break; } g_string_append(buffer, ele); } va_end(ap); } diff --git a/lib/common/tests/strings/Makefile.am b/lib/common/tests/strings/Makefile.am index 0d64003eb2..63985a3fd3 100644 --- a/lib/common/tests/strings/Makefile.am +++ b/lib/common/tests/strings/Makefile.am @@ -1,40 +1,39 @@ # # Copyright 2020-2025 the Pacemaker project contributors # # The version control history for this file may have further details. # # This source code is licensed under the GNU General Public License version 2 # or later (GPLv2+) WITHOUT ANY WARRANTY. # include $(top_srcdir)/mk/common.mk include $(top_srcdir)/mk/tap.mk include $(top_srcdir)/mk/unittest.mk # Add "_test" to the end of all test program names to simplify .gitignore. check_PROGRAMS = crm_get_msec_test \ crm_is_true_test \ crm_str_to_boolean_test \ pcmk__add_word_test \ pcmk__btoa_test \ pcmk__compress_test \ - pcmk__ends_with_test \ pcmk__g_strcat_test \ pcmk__guint_from_hash_test \ pcmk__numeric_strcasecmp_test \ pcmk__parse_ll_range_test \ pcmk__s_test \ pcmk__scan_double_test \ pcmk__scan_ll_test \ pcmk__scan_min_int_test \ pcmk__scan_port_test \ pcmk__str_any_of_test \ pcmk__str_in_list_test \ pcmk__str_table_dup_test \ pcmk__str_update_test \ pcmk__strcmp_test \ pcmk__strkey_table_test \ pcmk__strikey_table_test \ pcmk__trim_test TESTS = $(check_PROGRAMS) diff --git a/lib/common/tests/strings/pcmk__ends_with_test.c b/lib/common/tests/strings/pcmk__ends_with_test.c deleted file mode 100644 index 75035714fa..0000000000 --- a/lib/common/tests/strings/pcmk__ends_with_test.c +++ /dev/null @@ -1,57 +0,0 @@ -/* - * Copyright 2021 the Pacemaker project contributors - * - * The version control history for this file may have further details. - * - * This source code is licensed under the GNU General Public License version 2 - * or later (GPLv2+) WITHOUT ANY WARRANTY. - */ - -#include - -#include - -static void -bad_input(void **state) { - assert_false(pcmk__ends_with(NULL, "xyz")); - - assert_true(pcmk__ends_with(NULL, NULL)); - assert_true(pcmk__ends_with(NULL, "")); - assert_true(pcmk__ends_with("", NULL)); - assert_true(pcmk__ends_with("", "")); - assert_true(pcmk__ends_with("abc", NULL)); - assert_true(pcmk__ends_with("abc", "")); -} - -static void -ends_with(void **state) { - assert_true(pcmk__ends_with("abc", "abc")); - assert_true(pcmk__ends_with("abc", "bc")); - assert_true(pcmk__ends_with("abc", "c")); - assert_true(pcmk__ends_with("abcbc", "bc")); - - assert_false(pcmk__ends_with("abc", "def")); - assert_false(pcmk__ends_with("abc", "defg")); - assert_false(pcmk__ends_with("abc", "bcd")); - assert_false(pcmk__ends_with("abc", "ab")); - - assert_false(pcmk__ends_with("abc", "BC")); -} - -static void -ends_with_ext(void **state) { - assert_true(pcmk__ends_with_ext("ab.c", ".c")); - assert_true(pcmk__ends_with_ext("ab.cb.c", ".c")); - - assert_false(pcmk__ends_with_ext("ab.c", ".def")); - assert_false(pcmk__ends_with_ext("ab.c", ".defg")); - assert_false(pcmk__ends_with_ext("ab.c", ".cd")); - assert_false(pcmk__ends_with_ext("ab.c", "ab")); - - assert_false(pcmk__ends_with_ext("ab.c", ".C")); -} - -PCMK__UNIT_TEST(NULL, NULL, - cmocka_unit_test(bad_input), - cmocka_unit_test(ends_with), - cmocka_unit_test(ends_with_ext)) diff --git a/lib/common/xml.c b/lib/common/xml.c index 30383f9ba9..6e5484daaa 100644 --- a/lib/common/xml.c +++ b/lib/common/xml.c @@ -1,1858 +1,1858 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include // uint32_t #include #include #include #include // stat(), S_ISREG, etc. #include #include // gboolean, GString #include // xmlNode, etc. #include // xmlChar, xmlGetUTF8Char() #include #include #include // PCMK__XML_LOG_BASE, etc. #include "crmcommon_private.h" //! libxml2 supports only XML version 1.0, at least as of libxml2-2.12.5 #define XML_VERSION ((const xmlChar *) "1.0") /*! * \internal * \brief Get a string representation of an XML element type for logging * * \param[in] type XML element type * * \return String representation of \p type */ const char * pcmk__xml_element_type_text(xmlElementType type) { static const char *const element_type_names[] = { [XML_ELEMENT_NODE] = "element", [XML_ATTRIBUTE_NODE] = "attribute", [XML_TEXT_NODE] = "text", [XML_CDATA_SECTION_NODE] = "CDATA section", [XML_ENTITY_REF_NODE] = "entity reference", [XML_ENTITY_NODE] = "entity", [XML_PI_NODE] = "PI", [XML_COMMENT_NODE] = "comment", [XML_DOCUMENT_NODE] = "document", [XML_DOCUMENT_TYPE_NODE] = "document type", [XML_DOCUMENT_FRAG_NODE] = "document fragment", [XML_NOTATION_NODE] = "notation", [XML_HTML_DOCUMENT_NODE] = "HTML document", [XML_DTD_NODE] = "DTD", [XML_ELEMENT_DECL] = "element declaration", [XML_ATTRIBUTE_DECL] = "attribute declaration", [XML_ENTITY_DECL] = "entity declaration", [XML_NAMESPACE_DECL] = "namespace declaration", [XML_XINCLUDE_START] = "XInclude start", [XML_XINCLUDE_END] = "XInclude end", }; // Assumes the numeric values of the indices are in ascending order if ((type < XML_ELEMENT_NODE) || (type > XML_XINCLUDE_END)) { return "unrecognized type"; } return element_type_names[type]; } /*! * \internal * \brief Apply a function to each XML node in a tree (pre-order, depth-first) * * \param[in,out] xml XML tree to traverse * \param[in,out] fn Function to call for each node (returns \c true to * continue traversing the tree or \c false to stop) * \param[in,out] user_data Argument to \p fn * * \return \c false if any \p fn call returned \c false, or \c true otherwise * * \note This function is recursive. */ bool pcmk__xml_tree_foreach(xmlNode *xml, bool (*fn)(xmlNode *, void *), void *user_data) { if (xml == NULL) { return true; } if (!fn(xml, user_data)) { return false; } for (xml = pcmk__xml_first_child(xml); xml != NULL; xml = pcmk__xml_next(xml)) { if (!pcmk__xml_tree_foreach(xml, fn, user_data)) { return false; } } return true; } void pcmk__xml_set_parent_flags(xmlNode *xml, uint64_t flags) { for (; xml != NULL; xml = xml->parent) { xml_node_private_t *nodepriv = xml->_private; if (nodepriv != NULL) { pcmk__set_xml_flags(nodepriv, flags); } } } /*! * \internal * \brief Set flags for an XML document * * \param[in,out] doc XML document * \param[in] flags Group of enum pcmk__xml_flags */ void pcmk__xml_doc_set_flags(xmlDoc *doc, uint32_t flags) { if (doc != NULL) { xml_doc_private_t *docpriv = doc->_private; pcmk__set_xml_flags(docpriv, flags); } } /*! * \internal * \brief Check whether the given flags are set for an XML document * * \param[in] doc XML document to check * \param[in] flags Group of enum pcmk__xml_flags * * \return \c true if all of \p flags are set for \p doc, or \c false otherwise */ bool pcmk__xml_doc_all_flags_set(const xmlDoc *doc, uint32_t flags) { if (doc != NULL) { xml_doc_private_t *docpriv = doc->_private; return (docpriv != NULL) && pcmk_all_flags_set(docpriv->flags, flags); } return false; } // Mark document, element, and all element's parents as changed void pcmk__mark_xml_node_dirty(xmlNode *xml) { if (xml == NULL) { return; } pcmk__xml_doc_set_flags(xml->doc, pcmk__xf_dirty); pcmk__xml_set_parent_flags(xml, pcmk__xf_dirty); } /*! * \internal * \brief Clear flags on an XML node * * \param[in,out] xml XML node whose flags to reset * \param[in,out] user_data Ignored * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ bool pcmk__xml_reset_node_flags(xmlNode *xml, void *user_data) { xml_node_private_t *nodepriv = xml->_private; if (nodepriv != NULL) { nodepriv->flags = pcmk__xf_none; } return true; } /*! * \internal * \brief Set the \c pcmk__xf_dirty and \c pcmk__xf_created flags on an XML node * * \param[in,out] xml Node whose flags to set * \param[in] user_data Ignored * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool mark_xml_dirty_created(xmlNode *xml, void *user_data) { xml_node_private_t *nodepriv = xml->_private; if (nodepriv != NULL) { pcmk__set_xml_flags(nodepriv, pcmk__xf_dirty|pcmk__xf_created); } return true; } /*! * \internal * \brief Mark an XML tree as dirty and created, and mark its parents dirty * * Also mark the document dirty. * * \param[in,out] xml Tree to mark as dirty and created */ static void mark_xml_tree_dirty_created(xmlNode *xml) { pcmk__assert(xml != NULL); if (!pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_tracking)) { // Tracking is disabled for entire document return; } // Mark all parents and document dirty pcmk__mark_xml_node_dirty(xml); pcmk__xml_tree_foreach(xml, mark_xml_dirty_created, NULL); } // Free an XML object previously marked as deleted static void free_deleted_object(void *data) { if(data) { pcmk__deleted_xml_t *deleted_obj = data; g_free(deleted_obj->path); free(deleted_obj); } } // Free and NULL user, ACLs, and deleted objects in an XML node's private data static void reset_xml_private_data(xml_doc_private_t *docpriv) { if (docpriv != NULL) { pcmk__assert(docpriv->check == PCMK__XML_DOC_PRIVATE_MAGIC); pcmk__str_update(&(docpriv->acl_user), NULL); if (docpriv->acls != NULL) { pcmk__free_acls(docpriv->acls); docpriv->acls = NULL; } if(docpriv->deleted_objs) { g_list_free_full(docpriv->deleted_objs, free_deleted_object); docpriv->deleted_objs = NULL; } } } /*! * \internal * \brief Allocate and initialize private data for an XML node * * \param[in,out] node XML node whose private data to initialize * \param[in] user_data Ignored * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool new_private_data(xmlNode *node, void *user_data) { bool tracking = false; CRM_CHECK(node != NULL, return true); if (node->_private != NULL) { return true; } tracking = pcmk__xml_doc_all_flags_set(node->doc, pcmk__xf_tracking); switch (node->type) { case XML_DOCUMENT_NODE: { xml_doc_private_t *docpriv = pcmk__assert_alloc(1, sizeof(xml_doc_private_t)); docpriv->check = PCMK__XML_DOC_PRIVATE_MAGIC; node->_private = docpriv; } break; case XML_ELEMENT_NODE: case XML_ATTRIBUTE_NODE: case XML_COMMENT_NODE: { xml_node_private_t *nodepriv = pcmk__assert_alloc(1, sizeof(xml_node_private_t)); nodepriv->check = PCMK__XML_NODE_PRIVATE_MAGIC; node->_private = nodepriv; if (tracking) { pcmk__set_xml_flags(nodepriv, pcmk__xf_dirty|pcmk__xf_created); } for (xmlAttr *iter = pcmk__xe_first_attr(node); iter != NULL; iter = iter->next) { new_private_data((xmlNode *) iter, user_data); } } break; case XML_TEXT_NODE: case XML_DTD_NODE: case XML_CDATA_SECTION_NODE: return true; default: CRM_LOG_ASSERT(node->type == XML_ELEMENT_NODE); return true; } if (tracking) { pcmk__mark_xml_node_dirty(node); } return true; } /*! * \internal * \brief Free private data for an XML node * * \param[in,out] node XML node whose private data to free * \param[in] user_data Ignored * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool free_private_data(xmlNode *node, void *user_data) { CRM_CHECK(node != NULL, return true); if (node->_private == NULL) { return true; } if (node->type == XML_DOCUMENT_NODE) { reset_xml_private_data((xml_doc_private_t *) node->_private); } else { xml_node_private_t *nodepriv = node->_private; pcmk__assert(nodepriv->check == PCMK__XML_NODE_PRIVATE_MAGIC); for (xmlAttr *iter = pcmk__xe_first_attr(node); iter != NULL; iter = iter->next) { free_private_data((xmlNode *) iter, user_data); } } free(node->_private); node->_private = NULL; return true; } /*! * \internal * \brief Allocate and initialize private data recursively for an XML tree * * \param[in,out] node XML node whose private data to initialize */ void pcmk__xml_new_private_data(xmlNode *xml) { pcmk__xml_tree_foreach(xml, new_private_data, NULL); } /*! * \internal * \brief Free private data recursively for an XML tree * * \param[in,out] node XML node whose private data to free */ void pcmk__xml_free_private_data(xmlNode *xml) { pcmk__xml_tree_foreach(xml, free_private_data, NULL); } /*! * \internal * \brief Return ordinal position of an XML node among its siblings * * \param[in] xml XML node to check * \param[in] ignore_if_set Don't count siblings with this flag set * * \return Ordinal position of \p xml (starting with 0) */ int pcmk__xml_position(const xmlNode *xml, enum pcmk__xml_flags ignore_if_set) { int position = 0; for (const xmlNode *cIter = xml; cIter->prev; cIter = cIter->prev) { xml_node_private_t *nodepriv = ((xmlNode*)cIter->prev)->_private; if (!pcmk_is_set(nodepriv->flags, ignore_if_set)) { position++; } } return position; } /*! * \internal * \brief Remove all attributes marked as deleted from an XML node * * \param[in,out] xml XML node whose deleted attributes to remove * \param[in,out] user_data Ignored * * \return \c true (to continue traversing the tree) * * \note This is compatible with \c pcmk__xml_tree_foreach(). */ static bool commit_attr_deletions(xmlNode *xml, void *user_data) { pcmk__xml_reset_node_flags(xml, NULL); pcmk__xe_remove_matching_attrs(xml, true, pcmk__marked_as_deleted, NULL); return true; } /*! * \internal * \brief Finalize all pending changes to an XML document and reset private data * * Clear the ACL user and all flags, unpacked ACLs, and deleted node records for * the document; clear all flags on each node in the tree; and delete any * attributes that are marked for deletion. * * \param[in,out] doc XML document * * \note When change tracking is enabled, "deleting" an attribute simply marks * it for deletion (using \c pcmk__xf_deleted) until changes are * committed. Freeing a node (using \c pcmk__xml_free()) adds a deleted * node record (\c pcmk__deleted_xml_t) to the node's document before * freeing it. * \note This function clears all flags, not just flags that indicate changes. * In particular, note that it clears the \c pcmk__xf_tracking flag, thus * disabling tracking. */ void pcmk__xml_commit_changes(xmlDoc *doc) { xml_doc_private_t *docpriv = NULL; if (doc == NULL) { return; } docpriv = doc->_private; if (docpriv == NULL) { return; } if (pcmk_is_set(docpriv->flags, pcmk__xf_dirty)) { pcmk__xml_tree_foreach(xmlDocGetRootElement(doc), commit_attr_deletions, NULL); } reset_xml_private_data(docpriv); docpriv->flags = pcmk__xf_none; } /*! * \internal * \brief Create a new XML document * * \return Newly allocated XML document (guaranteed not to be \c NULL) * * \note The caller is responsible for freeing the return value using * \c pcmk__xml_free_doc(). */ xmlDoc * pcmk__xml_new_doc(void) { xmlDoc *doc = xmlNewDoc(XML_VERSION); pcmk__mem_assert(doc); pcmk__xml_new_private_data((xmlNode *) doc); return doc; } /*! * \internal * \brief Free a new XML document * * \param[in,out] doc XML document to free */ void pcmk__xml_free_doc(xmlDoc *doc) { if (doc != NULL) { pcmk__xml_free_private_data((xmlNode *) doc); xmlFreeDoc(doc); } } /*! * \internal * \brief Check whether the first character of a string is an XML NameStartChar * * See https://www.w3.org/TR/xml/#NT-NameStartChar. * * This is almost identical to libxml2's \c xmlIsDocNameStartChar(), but they * don't expose it as part of the public API. * * \param[in] utf8 UTF-8 encoded string * \param[out] len If not \c NULL, where to store size in bytes of first * character in \p utf8 * * \return \c true if \p utf8 begins with a valid XML NameStartChar, or \c false * otherwise */ bool pcmk__xml_is_name_start_char(const char *utf8, int *len) { int c = 0; int local_len = 0; if (len == NULL) { len = &local_len; } /* xmlGetUTF8Char() abuses the len argument. At call time, it must be set to * "the minimum number of bytes present in the sequence... to assure the * next character is completely contained within the sequence." It's similar * to the "n" in the strn*() functions. However, this doesn't make any sense * for null-terminated strings, and there's no value that indicates "keep * going until '\0'." So we set it to 4, the max number of bytes in a UTF-8 * character. * * At return, it's set to the actual number of bytes in the char, or 0 on * error. */ *len = 4; // Note: xmlGetUTF8Char() assumes a 32-bit int c = xmlGetUTF8Char((const xmlChar *) utf8, len); if (c < 0) { GString *buf = g_string_sized_new(32); for (int i = 0; (i < 4) && (utf8[i] != '\0'); i++) { g_string_append_printf(buf, " 0x%.2X", utf8[i]); } crm_info("Invalid UTF-8 character (bytes:%s)", (pcmk__str_empty(buf->str)? " " : buf->str)); g_string_free(buf, TRUE); return false; } return (c == '_') || (c == ':') || ((c >= 'a') && (c <= 'z')) || ((c >= 'A') && (c <= 'Z')) || ((c >= 0xC0) && (c <= 0xD6)) || ((c >= 0xD8) && (c <= 0xF6)) || ((c >= 0xF8) && (c <= 0x2FF)) || ((c >= 0x370) && (c <= 0x37D)) || ((c >= 0x37F) && (c <= 0x1FFF)) || ((c >= 0x200C) && (c <= 0x200D)) || ((c >= 0x2070) && (c <= 0x218F)) || ((c >= 0x2C00) && (c <= 0x2FEF)) || ((c >= 0x3001) && (c <= 0xD7FF)) || ((c >= 0xF900) && (c <= 0xFDCF)) || ((c >= 0xFDF0) && (c <= 0xFFFD)) || ((c >= 0x10000) && (c <= 0xEFFFF)); } /*! * \internal * \brief Check whether the first character of a string is an XML NameChar * * See https://www.w3.org/TR/xml/#NT-NameChar. * * This is almost identical to libxml2's \c xmlIsDocNameChar(), but they don't * expose it as part of the public API. * * \param[in] utf8 UTF-8 encoded string * \param[out] len If not \c NULL, where to store size in bytes of first * character in \p utf8 * * \return \c true if \p utf8 begins with a valid XML NameChar, or \c false * otherwise */ bool pcmk__xml_is_name_char(const char *utf8, int *len) { int c = 0; int local_len = 0; if (len == NULL) { len = &local_len; } // See comment regarding len in pcmk__xml_is_name_start_char() *len = 4; // Note: xmlGetUTF8Char() assumes a 32-bit int c = xmlGetUTF8Char((const xmlChar *) utf8, len); if (c < 0) { GString *buf = g_string_sized_new(32); for (int i = 0; (i < 4) && (utf8[i] != '\0'); i++) { g_string_append_printf(buf, " 0x%.2X", utf8[i]); } crm_info("Invalid UTF-8 character (bytes:%s)", (pcmk__str_empty(buf->str)? " " : buf->str)); g_string_free(buf, TRUE); return false; } return ((c >= 'a') && (c <= 'z')) || ((c >= 'A') && (c <= 'Z')) || ((c >= '0') && (c <= '9')) || (c == '_') || (c == ':') || (c == '-') || (c == '.') || (c == 0xB7) || ((c >= 0xC0) && (c <= 0xD6)) || ((c >= 0xD8) && (c <= 0xF6)) || ((c >= 0xF8) && (c <= 0x2FF)) || ((c >= 0x300) && (c <= 0x36F)) || ((c >= 0x370) && (c <= 0x37D)) || ((c >= 0x37F) && (c <= 0x1FFF)) || ((c >= 0x200C) && (c <= 0x200D)) || ((c >= 0x203F) && (c <= 0x2040)) || ((c >= 0x2070) && (c <= 0x218F)) || ((c >= 0x2C00) && (c <= 0x2FEF)) || ((c >= 0x3001) && (c <= 0xD7FF)) || ((c >= 0xF900) && (c <= 0xFDCF)) || ((c >= 0xFDF0) && (c <= 0xFFFD)) || ((c >= 0x10000) && (c <= 0xEFFFF)); } /*! * \internal * \brief Sanitize a string so it is usable as an XML ID * * An ID must match the Name production as defined here: * https://www.w3.org/TR/xml/#NT-Name. * * Convert an invalid start character to \c '_'. Convert an invalid character * after the start character to \c '.'. * * \param[in,out] id String to sanitize */ void pcmk__xml_sanitize_id(char *id) { bool valid = true; int len = 0; // If id is empty or NULL, there's no way to make it a valid XML ID pcmk__assert(!pcmk__str_empty(id)); /* @TODO Suppose there are two strings and each has an invalid ID character * in the same position. The strings are otherwise identical. Both strings * will be sanitized to the same valid ID, which is incorrect. * * The caller is responsible for ensuring the sanitized ID does not already * exist in a given XML document before using it, if uniqueness is desired. */ valid = pcmk__xml_is_name_start_char(id, &len); CRM_CHECK(len > 0, return); // UTF-8 encoding error if (!valid) { *id = '_'; for (int i = 1; i < len; i++) { id[i] = '.'; } } for (id += len; *id != '\0'; id += len) { valid = pcmk__xml_is_name_char(id, &len); CRM_CHECK(len > 0, return); // UTF-8 encoding error if (!valid) { for (int i = 0; i < len; i++) { id[i] = '.'; } } } } /*! * \internal * \brief Free an XML tree without ACL checks or change tracking * * \param[in,out] xml XML node to free */ void pcmk__xml_free_node(xmlNode *xml) { pcmk__xml_free_private_data(xml); xmlUnlinkNode(xml); xmlFreeNode(xml); } /*! * \internal * \brief Free an XML tree if ACLs allow; track deletion if tracking is enabled * * If \p node is the root of its document, free the entire document. * * \param[in,out] node XML node to free * \param[in] position Position of \p node among its siblings for change * tracking (negative to calculate automatically if * needed) * * \return Standard Pacemaker return code */ static int free_xml_with_position(xmlNode *node, int position) { xmlDoc *doc = NULL; xml_node_private_t *nodepriv = NULL; if (node == NULL) { return pcmk_rc_ok; } doc = node->doc; nodepriv = node->_private; if ((doc != NULL) && (xmlDocGetRootElement(doc) == node)) { /* @TODO Should we check ACLs first? Otherwise it seems like we could * free the root element without write permission. */ pcmk__xml_free_doc(doc); return pcmk_rc_ok; } if (!pcmk__check_acl(node, NULL, pcmk__xf_acl_write)) { pcmk__if_tracing( { GString *xpath = pcmk__element_xpath(node); qb_log_from_external_source(__func__, __FILE__, "Cannot remove %s %x", LOG_TRACE, __LINE__, 0, xpath->str, nodepriv->flags); g_string_free(xpath, TRUE); }, {} ); return EACCES; } if (pcmk__xml_doc_all_flags_set(node->doc, pcmk__xf_tracking) && !pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { xml_doc_private_t *docpriv = doc->_private; GString *xpath = pcmk__element_xpath(node); if (xpath != NULL) { pcmk__deleted_xml_t *deleted_obj = NULL; crm_trace("Deleting %s %p from %p", xpath->str, node, doc); deleted_obj = pcmk__assert_alloc(1, sizeof(pcmk__deleted_xml_t)); deleted_obj->path = g_string_free(xpath, FALSE); deleted_obj->position = -1; // Record the position only for XML comments for now if (node->type == XML_COMMENT_NODE) { if (position >= 0) { deleted_obj->position = position; } else { deleted_obj->position = pcmk__xml_position(node, pcmk__xf_skip); } } docpriv->deleted_objs = g_list_append(docpriv->deleted_objs, deleted_obj); pcmk__xml_doc_set_flags(node->doc, pcmk__xf_dirty); } } pcmk__xml_free_node(node); return pcmk_rc_ok; } /*! * \internal * \brief Free an XML tree if ACLs allow; track deletion if tracking is enabled * * If \p xml is the root of its document, free the entire document. * * \param[in,out] xml XML node to free */ void pcmk__xml_free(xmlNode *xml) { free_xml_with_position(xml, -1); } /*! * \internal * \brief Make a deep copy of an XML node under a given parent * * \param[in,out] parent XML element that will be the copy's parent (\c NULL * to create a new XML document with the copy as root) * \param[in] src XML node to copy * * \return Deep copy of \p src, or \c NULL if \p src is \c NULL */ xmlNode * pcmk__xml_copy(xmlNode *parent, xmlNode *src) { xmlNode *copy = NULL; if (src == NULL) { return NULL; } if (parent == NULL) { xmlDoc *doc = NULL; // The copy will be the root element of a new document pcmk__assert(src->type == XML_ELEMENT_NODE); doc = pcmk__xml_new_doc(); copy = xmlDocCopyNode(src, doc, 1); pcmk__mem_assert(copy); xmlDocSetRootElement(doc, copy); } else { copy = xmlDocCopyNode(src, parent->doc, 1); pcmk__mem_assert(copy); xmlAddChild(parent, copy); } pcmk__xml_new_private_data(copy); return copy; } /*! * \internal * \brief Remove XML text nodes from specified XML and all its children * * \param[in,out] xml XML to strip text from */ void pcmk__strip_xml_text(xmlNode *xml) { xmlNode *iter = xml->children; while (iter) { xmlNode *next = iter->next; switch (iter->type) { case XML_TEXT_NODE: pcmk__xml_free_node(iter); break; case XML_ELEMENT_NODE: /* Search it */ pcmk__strip_xml_text(iter); break; default: /* Leave it */ break; } iter = next; } } /*! * \internal * \brief Check whether a string has XML special characters that must be escaped * * See \c pcmk__xml_escape() and \c pcmk__xml_escape_type for more details. * * \param[in] text String to check * \param[in] type Type of escaping * * \return \c true if \p text has special characters that need to be escaped, or * \c false otherwise */ bool pcmk__xml_needs_escape(const char *text, enum pcmk__xml_escape_type type) { if (text == NULL) { return false; } while (*text != '\0') { switch (type) { case pcmk__xml_escape_text: switch (*text) { case '<': case '>': case '&': return true; case '\n': case '\t': break; default: if (g_ascii_iscntrl(*text)) { return true; } break; } break; case pcmk__xml_escape_attr: switch (*text) { case '<': case '>': case '&': case '"': return true; default: if (g_ascii_iscntrl(*text)) { return true; } break; } break; case pcmk__xml_escape_attr_pretty: switch (*text) { case '\n': case '\r': case '\t': case '"': return true; default: break; } break; default: // Invalid enum value pcmk__assert(false); break; } text = g_utf8_next_char(text); } return false; } /*! * \internal * \brief Replace special characters with their XML escape sequences * * \param[in] text Text to escape * \param[in] type Type of escaping * * \return Newly allocated string equivalent to \p text but with special * characters replaced with XML escape sequences (or \c NULL if \p text * is \c NULL). If \p text is not \c NULL, the return value is * guaranteed not to be \c NULL. * * \note There are libxml functions that purport to do this: * \c xmlEncodeEntitiesReentrant() and \c xmlEncodeSpecialChars(). * However, their escaping is incomplete. See: * https://discourse.gnome.org/t/intended-use-of-xmlencodeentitiesreentrant-vs-xmlencodespecialchars/19252 * \note The caller is responsible for freeing the return value using * \c g_free(). */ gchar * pcmk__xml_escape(const char *text, enum pcmk__xml_escape_type type) { GString *copy = NULL; if (text == NULL) { return NULL; } copy = g_string_sized_new(strlen(text)); while (*text != '\0') { // Don't escape any non-ASCII characters if ((*text & 0x80) != 0) { size_t bytes = g_utf8_next_char(text) - text; g_string_append_len(copy, text, bytes); text += bytes; continue; } switch (type) { case pcmk__xml_escape_text: switch (*text) { case '<': g_string_append(copy, PCMK__XML_ENTITY_LT); break; case '>': g_string_append(copy, PCMK__XML_ENTITY_GT); break; case '&': g_string_append(copy, PCMK__XML_ENTITY_AMP); break; case '\n': case '\t': g_string_append_c(copy, *text); break; default: if (g_ascii_iscntrl(*text)) { g_string_append_printf(copy, "&#x%.2X;", *text); } else { g_string_append_c(copy, *text); } break; } break; case pcmk__xml_escape_attr: switch (*text) { case '<': g_string_append(copy, PCMK__XML_ENTITY_LT); break; case '>': g_string_append(copy, PCMK__XML_ENTITY_GT); break; case '&': g_string_append(copy, PCMK__XML_ENTITY_AMP); break; case '"': g_string_append(copy, PCMK__XML_ENTITY_QUOT); break; default: if (g_ascii_iscntrl(*text)) { g_string_append_printf(copy, "&#x%.2X;", *text); } else { g_string_append_c(copy, *text); } break; } break; case pcmk__xml_escape_attr_pretty: switch (*text) { case '"': g_string_append(copy, "\\\""); break; case '\n': g_string_append(copy, "\\n"); break; case '\r': g_string_append(copy, "\\r"); break; case '\t': g_string_append(copy, "\\t"); break; default: g_string_append_c(copy, *text); break; } break; default: // Invalid enum value pcmk__assert(false); break; } text = g_utf8_next_char(text); } return g_string_free(copy, FALSE); } /*! * \internal * \brief Add an XML attribute to a node, marked as deleted * * When calculating XML changes, we need to know when an attribute has been * deleted. Add the attribute back to the new XML, so that we can check the * removal against ACLs, and mark it as deleted for later removal after * differences have been calculated. * * \param[in,out] new_xml XML to modify * \param[in] element Name of XML element that changed (for logging) * \param[in] attr_name Name of attribute that was deleted * \param[in] old_value Value of attribute that was deleted */ static void mark_attr_deleted(xmlNode *new_xml, const char *element, const char *attr_name, const char *old_value) { xml_doc_private_t *docpriv = new_xml->doc->_private; xmlAttr *attr = NULL; xml_node_private_t *nodepriv; /* Restore the old value (without setting dirty flag recursively upwards or * checking ACLs) */ pcmk__clear_xml_flags(docpriv, pcmk__xf_tracking); crm_xml_add(new_xml, attr_name, old_value); pcmk__set_xml_flags(docpriv, pcmk__xf_tracking); // Reset flags (so the attribute doesn't appear as newly created) attr = xmlHasProp(new_xml, (const xmlChar *) attr_name); nodepriv = attr->_private; nodepriv->flags = 0; // Check ACLs and mark restored value for later removal pcmk__xa_remove(attr, false); crm_trace("XML attribute %s=%s was removed from %s", attr_name, old_value, element); } /* * \internal * \brief Check ACLs for a changed XML attribute */ static void mark_attr_changed(xmlNode *new_xml, const char *element, const char *attr_name, const char *old_value) { xml_doc_private_t *docpriv = new_xml->doc->_private; char *vcopy = pcmk__xe_get_copy(new_xml, attr_name); crm_trace("XML attribute %s was changed from '%s' to '%s' in %s", attr_name, old_value, vcopy, element); // Restore the original value (without checking ACLs) pcmk__clear_xml_flags(docpriv, pcmk__xf_tracking); crm_xml_add(new_xml, attr_name, old_value); pcmk__set_xml_flags(docpriv, pcmk__xf_tracking); // Change it back to the new value, to check ACLs crm_xml_add(new_xml, attr_name, vcopy); free(vcopy); } /*! * \internal * \brief Mark an XML attribute as having changed position * * \param[in,out] new_xml XML to modify * \param[in] element Name of XML element that changed (for logging) * \param[in,out] old_attr Attribute that moved, in original XML * \param[in,out] new_attr Attribute that moved, in \p new_xml * \param[in] p_old Ordinal position of \p old_attr in original XML * \param[in] p_new Ordinal position of \p new_attr in \p new_xml */ static void mark_attr_moved(xmlNode *new_xml, const char *element, xmlAttr *old_attr, xmlAttr *new_attr, int p_old, int p_new) { xml_node_private_t *nodepriv = new_attr->_private; crm_trace("XML attribute %s moved from position %d to %d in %s", old_attr->name, p_old, p_new, element); // Mark document, element, and all element's parents as changed pcmk__mark_xml_node_dirty(new_xml); // Mark attribute as changed pcmk__set_xml_flags(nodepriv, pcmk__xf_dirty|pcmk__xf_moved); nodepriv = (p_old > p_new)? old_attr->_private : new_attr->_private; pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); } /*! * \internal * \brief Calculate differences in all previously existing XML attributes * * \param[in,out] old_xml Original XML to compare * \param[in,out] new_xml New XML to compare */ static void xml_diff_old_attrs(xmlNode *old_xml, xmlNode *new_xml) { xmlAttr *attr_iter = pcmk__xe_first_attr(old_xml); while (attr_iter != NULL) { const char *name = (const char *) attr_iter->name; xmlAttr *old_attr = attr_iter; xmlAttr *new_attr = xmlHasProp(new_xml, attr_iter->name); const char *old_value = pcmk__xml_attr_value(attr_iter); attr_iter = attr_iter->next; if (new_attr == NULL) { mark_attr_deleted(new_xml, (const char *) old_xml->name, name, old_value); } else { xml_node_private_t *nodepriv = new_attr->_private; int new_pos = pcmk__xml_position((xmlNode*) new_attr, pcmk__xf_skip); int old_pos = pcmk__xml_position((xmlNode*) old_attr, pcmk__xf_skip); const char *new_value = pcmk__xe_get(new_xml, name); // This attribute isn't new pcmk__clear_xml_flags(nodepriv, pcmk__xf_created); if (strcmp(new_value, old_value) != 0) { mark_attr_changed(new_xml, (const char *) old_xml->name, name, old_value); } else if ((old_pos != new_pos) && !pcmk__xml_doc_all_flags_set(new_xml->doc, pcmk__xf_ignore_attr_pos |pcmk__xf_tracking)) { /* pcmk__xf_tracking is always set by pcmk__xml_mark_changes() * before this function is called, so only the * pcmk__xf_ignore_attr_pos check is truly relevant. */ mark_attr_moved(new_xml, (const char *) old_xml->name, old_attr, new_attr, old_pos, new_pos); } } } } /*! * \internal * \brief Check all attributes in new XML for creation * * For each of a given XML element's attributes marked as newly created, accept * (and mark as dirty) or reject the creation according to ACLs. * * \param[in,out] new_xml XML to check */ static void mark_created_attrs(xmlNode *new_xml) { xmlAttr *attr_iter = pcmk__xe_first_attr(new_xml); while (attr_iter != NULL) { xmlAttr *new_attr = attr_iter; xml_node_private_t *nodepriv = attr_iter->_private; attr_iter = attr_iter->next; if (pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { const char *attr_name = (const char *) new_attr->name; crm_trace("Created new attribute %s=%s in %s", attr_name, pcmk__xml_attr_value(new_attr), new_xml->name); /* Check ACLs (we can't use the remove-then-create trick because it * would modify the attribute position). */ if (pcmk__check_acl(new_xml, attr_name, pcmk__xf_acl_write)) { pcmk__mark_xml_attr_dirty(new_attr); } else { // Creation was not allowed, so remove the attribute pcmk__xa_remove(new_attr, true); } } } } /*! * \internal * \brief Calculate differences in attributes between two XML nodes * * \param[in,out] old_xml Original XML to compare * \param[in,out] new_xml New XML to compare */ static void xml_diff_attrs(xmlNode *old_xml, xmlNode *new_xml) { // Cleared later if attributes are not really new for (xmlAttr *attr = pcmk__xe_first_attr(new_xml); attr != NULL; attr = attr->next) { xml_node_private_t *nodepriv = attr->_private; pcmk__set_xml_flags(nodepriv, pcmk__xf_created); } xml_diff_old_attrs(old_xml, new_xml); mark_created_attrs(new_xml); } /*! * \internal * \brief Add a deleted object record for an old XML child if ACLs allow * * This is intended to be called for a child of an old XML element that is not * present as a child of a new XML element. * * Add a temporary copy of the old child to the new XML. Then check whether ACLs * would have allowed the deletion of that element. If so, add a deleted object * record for it to the new XML's document, and set the \c pcmk__xf_skip flag on * the old child. * * The temporary copy is removed before returning. The new XML and all of its * ancestors will have the \c pcmk__xf_dirty flag set because of the creation, * however. * * \param[in,out] old_child Child of old XML * \param[in,out] new_parent New XML that does not contain \p old_child */ static void mark_child_deleted(xmlNode *old_child, xmlNode *new_parent) { int pos = pcmk__xml_position(old_child, pcmk__xf_skip); // Re-create the child element so we can check ACLs xmlNode *candidate = pcmk__xml_copy(new_parent, old_child); // Clear flags on new child and its children pcmk__xml_tree_foreach(candidate, pcmk__xml_reset_node_flags, NULL); // free_xml_with_position() will check whether ACLs allow the deletion pcmk__apply_acl(xmlDocGetRootElement(candidate->doc)); /* Try to remove the child again (which will track it in document's * deleted_objs on success) */ if (free_xml_with_position(candidate, pos) != pcmk_rc_ok) { // ACLs denied deletion in free_xml_with_position. Free candidate here. pcmk__xml_free_node(candidate); } pcmk__set_xml_flags((xml_node_private_t *) old_child->_private, pcmk__xf_skip); } /*! * \internal * \brief Mark a new child as moved and set \c pcmk__xf_skip as appropriate * * \param[in,out] old_child Child of old XML * \param[in,out] new_child Child of new XML that matches \p old_child * \param[in] old_pos Position of \p old_child among its siblings * \param[in] new_pos Position of \p new_child among its siblings */ static void mark_child_moved(xmlNode *old_child, xmlNode *new_child, int old_pos, int new_pos) { const char *id_s = pcmk__s(pcmk__xe_id(new_child), ""); xmlNode *new_parent = new_child->parent; xml_node_private_t *nodepriv = new_child->_private; crm_trace("Child element %s with " PCMK_XA_ID "='%s' moved from position " "%d to %d under %s", new_child->name, id_s, old_pos, new_pos, new_parent->name); pcmk__mark_xml_node_dirty(new_parent); pcmk__set_xml_flags(nodepriv, pcmk__xf_moved); /* @TODO Figure out and document why we skip the old child in future * position calculations if the old position is higher, and skip the new * child in future position calculations if the new position is higher. This * goes back to d028b52, and there's no explanation in the commit message. */ if (old_pos > new_pos) { nodepriv = old_child->_private; } pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); } /*! * \internal * \brief Check whether a new XML child comment matches an old XML child comment * * Two comments match if they have the same position among their siblings and * the same contents. * * If \p new_comment has the \c pcmk__xf_skip flag set, then it is automatically * considered not to match. * * \param[in] old_comment Old XML child element * \param[in] new_comment New XML child element * * \retval \c true if \p new_comment matches \p old_comment * \retval \c false otherwise */ static bool new_comment_matches(const xmlNode *old_comment, const xmlNode *new_comment) { xml_node_private_t *nodepriv = new_comment->_private; if (pcmk_is_set(nodepriv->flags, pcmk__xf_skip)) { /* @TODO Should we also return false if old_comment has pcmk__xf_skip * set? This preserves existing behavior at time of writing. */ return false; } if (pcmk__xml_position(old_comment, pcmk__xf_skip) != pcmk__xml_position(new_comment, pcmk__xf_skip)) { return false; } return pcmk__xc_matches(old_comment, new_comment); } /*! * \internal * \brief Check whether a new XML child element matches an old XML child element * * Two elements match if they have the same name and the same ID. (Both IDs can * be \c NULL.) * * For XML attributes other than \c PCMK_XA_ID, we can treat a value change as * an in-place modification. However, when Pacemaker applies a patchset, it uses * the \c PCMK_XA_ID attribute to find the node to update (modify, delete, or * move). If we treat two nodes with different \c PCMK_XA_ID attributes as * matching and then mark that attribute as changed, it can cause this lookup to * fail. * * There's unlikely to ever be much practical reason to treat elements with * different IDs as a change. Unless that changes, we'll treat them as a * mismatch. * * \param[in] old_element Old XML child element * \param[in] new_element New XML child element * * \retval \c true if \p new_element matches \p old_element * \retval \c false otherwise */ static bool new_element_matches(const xmlNode *old_element, const xmlNode *new_element) { return pcmk__xe_is(new_element, (const char *) old_element->name) && pcmk__str_eq(pcmk__xe_id(old_element), pcmk__xe_id(new_element), pcmk__str_none); } /*! * \internal * \brief Check whether a new XML child node matches an old XML child node * * Node types must be the same in order to match. * * For comments, a match is a comment at the same position with the same * content. * * For elements, a match is an element with the same name and the same ID. (Both * IDs can be \c NULL.) * * For other node types, there is no match. * * \param[in] old_child Child of old XML * \param[in] new_child Child of new XML * * \retval \c true if \p new_child matches \p old_child * \retval \c false otherwise */ static bool new_child_matches(const xmlNode *old_child, const xmlNode *new_child) { if (old_child->type != new_child->type) { return false; } switch (old_child->type) { case XML_COMMENT_NODE: return new_comment_matches(old_child, new_child); case XML_ELEMENT_NODE: return new_element_matches(old_child, new_child); default: return false; } } /*! * \internal * \brief Find matching XML node pairs between old and new XML's children * * A node that is part of a matching pair gets its _private:match * member set to the matching node. * * \param[in,out] old_xml Old XML * \param[in,out] new_xml New XML */ static void find_matching_children(xmlNode *old_xml, xmlNode *new_xml) { for (xmlNode *old_child = pcmk__xml_first_child(old_xml); old_child != NULL; old_child = pcmk__xml_next(old_child)) { xml_node_private_t *old_nodepriv = old_child->_private; if ((old_nodepriv == NULL) || (old_nodepriv->match != NULL)) { // Can't process, or we already found a match for this old child continue; } for (xmlNode *new_child = pcmk__xml_first_child(new_xml); new_child != NULL; new_child = pcmk__xml_next(new_child)) { xml_node_private_t *new_nodepriv = new_child->_private; if ((new_nodepriv == NULL) || (new_nodepriv->match != NULL)) { /* Can't process, or this new child already matched some old * child */ continue; } if (new_child_matches(old_child, new_child)) { old_nodepriv->match = new_child; new_nodepriv->match = old_child; break; } } } } /*! * \internal * \brief Mark changes between two XML trees * * Set flags in a new XML tree to indicate changes relative to an old XML tree. * * \param[in,out] old_xml XML before changes * \param[in,out] new_xml XML after changes * * \note This may set \c pcmk__xf_skip on parts of \p old_xml. */ void pcmk__xml_mark_changes(xmlNode *old_xml, xmlNode *new_xml) { /* This function may set the xml_node_private_t:match member on children of * old_xml and new_xml, but it clears that member before returning. * * @TODO Ensure we handle (for example, by copying) or reject user-created * XML that is missing xml_node_private_t at top level or in any children. * Similarly, check handling of node types for which we don't create private * data. For now, we'll skip them in the loops below. */ CRM_CHECK((old_xml != NULL) && (new_xml != NULL), return); if ((old_xml->_private == NULL) || (new_xml->_private == NULL)) { return; } pcmk__xml_doc_set_flags(new_xml->doc, pcmk__xf_tracking); xml_diff_attrs(old_xml, new_xml); find_matching_children(old_xml, new_xml); // Process matches (changed children) and deletions for (xmlNode *old_child = pcmk__xml_first_child(old_xml); old_child != NULL; old_child = pcmk__xml_next(old_child)) { xml_node_private_t *nodepriv = old_child->_private; xmlNode *new_child = NULL; if (nodepriv == NULL) { continue; } if (nodepriv->match == NULL) { // No match in new XML means the old child was deleted mark_child_deleted(old_child, new_xml); continue; } /* Fetch the match and clear old_child->_private's match member. * new_child->_private's match member is handled in the new_xml loop. */ new_child = nodepriv->match; nodepriv->match = NULL; pcmk__assert(old_child->type == new_child->type); if (old_child->type == XML_COMMENT_NODE) { // Comments match only if their positions and contents match continue; } pcmk__xml_mark_changes(old_child, new_child); } /* Mark unmatched new children as created, and mark matched new children as * moved if their positions changed. Grab the next new child in advance, * since new_child may get freed in the loop body. */ for (xmlNode *new_child = pcmk__xml_first_child(new_xml), *next = pcmk__xml_next(new_child); new_child != NULL; new_child = next, next = pcmk__xml_next(new_child)) { xml_node_private_t *nodepriv = new_child->_private; if (nodepriv == NULL) { continue; } if (nodepriv->match != NULL) { /* Fetch the match and clear new_child->_private's match member. Any * changes were marked in the old_xml loop. Mark the move. * * We might be able to mark the move earlier, when we mark changes * for matches in the old_xml loop, consolidating both actions. We'd * have to think about whether the timing of setting the * pcmk__xf_skip flag makes any difference. */ xmlNode *old_child = nodepriv->match; int old_pos = pcmk__xml_position(old_child, pcmk__xf_skip); int new_pos = pcmk__xml_position(new_child, pcmk__xf_skip); if (old_pos != new_pos) { mark_child_moved(old_child, new_child, old_pos, new_pos); } nodepriv->match = NULL; continue; } // No match in old XML means the new child is newly created pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); mark_xml_tree_dirty_created(new_child); // Check whether creation was allowed (may free new_child) pcmk__apply_creation_acl(new_child, true); } } char * pcmk__xml_artefact_root(enum pcmk__xml_artefact_ns ns) { static const char *base = NULL; char *ret = NULL; if (base == NULL) { base = pcmk__env_option(PCMK__ENV_SCHEMA_DIRECTORY); } if (pcmk__str_empty(base)) { base = PCMK_SCHEMA_DIR; } switch (ns) { case pcmk__xml_artefact_ns_legacy_rng: case pcmk__xml_artefact_ns_legacy_xslt: ret = strdup(base); break; case pcmk__xml_artefact_ns_base_rng: case pcmk__xml_artefact_ns_base_xslt: ret = crm_strdup_printf("%s/base", base); break; default: crm_err("XML artefact family specified as %u not recognized", ns); } return ret; } static char * find_artefact(enum pcmk__xml_artefact_ns ns, const char *path, const char *filespec) { char *ret = NULL; switch (ns) { case pcmk__xml_artefact_ns_legacy_rng: case pcmk__xml_artefact_ns_base_rng: - if (pcmk__ends_with(filespec, ".rng")) { + if (g_str_has_suffix(filespec, ".rng")) { ret = crm_strdup_printf("%s/%s", path, filespec); } else { ret = crm_strdup_printf("%s/%s.rng", path, filespec); } break; case pcmk__xml_artefact_ns_legacy_xslt: case pcmk__xml_artefact_ns_base_xslt: - if (pcmk__ends_with(filespec, ".xsl")) { + if (g_str_has_suffix(filespec, ".xsl")) { ret = crm_strdup_printf("%s/%s", path, filespec); } else { ret = crm_strdup_printf("%s/%s.xsl", path, filespec); } break; default: crm_err("XML artefact family specified as %u not recognized", ns); } return ret; } char * pcmk__xml_artefact_path(enum pcmk__xml_artefact_ns ns, const char *filespec) { struct stat sb; char *base = pcmk__xml_artefact_root(ns); char *ret = NULL; ret = find_artefact(ns, base, filespec); free(base); if (stat(ret, &sb) != 0 || !S_ISREG(sb.st_mode)) { const char *remote_schema_dir = pcmk__remote_schema_dir(); free(ret); ret = find_artefact(ns, remote_schema_dir, filespec); } return ret; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include // xmlCleanupParser() #include xmlNode * copy_xml(xmlNode *src) { xmlDoc *doc = pcmk__xml_new_doc(); xmlNode *copy = NULL; copy = xmlDocCopyNode(src, doc, 1); pcmk__mem_assert(copy); xmlDocSetRootElement(doc, copy); pcmk__xml_new_private_data(copy); return copy; } void crm_xml_init(void) { pcmk__schema_init(); } void crm_xml_cleanup(void) { pcmk__schema_cleanup(); xmlCleanupParser(); } void pcmk_free_xml_subtree(xmlNode *xml) { pcmk__xml_free_node(xml); } void free_xml(xmlNode *child) { pcmk__xml_free(child); } void crm_xml_sanitize_id(char *id) { char *c; for (c = id; *c; ++c) { switch (*c) { case ':': case '#': *c = '.'; } } } bool xml_tracking_changes(xmlNode *xml) { return (xml != NULL) && pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_tracking); } bool xml_document_dirty(xmlNode *xml) { return (xml != NULL) && pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_dirty); } void xml_accept_changes(xmlNode *xml) { if (xml != NULL) { pcmk__xml_commit_changes(xml->doc); } } void xml_track_changes(xmlNode *xml, const char *user, xmlNode *acl_source, bool enforce_acls) { if (xml == NULL) { return; } pcmk__xml_commit_changes(xml->doc); crm_trace("Tracking changes%s to %p", (enforce_acls? " with ACLs" : ""), xml); pcmk__xml_doc_set_flags(xml->doc, pcmk__xf_tracking); if (enforce_acls) { if (acl_source == NULL) { acl_source = xml; } pcmk__xml_doc_set_flags(xml->doc, pcmk__xf_acl_enabled); pcmk__unpack_acl(acl_source, xml, user); pcmk__apply_acl(xml); } } void xml_calculate_changes(xmlNode *old_xml, xmlNode *new_xml) { CRM_CHECK((old_xml != NULL) && (new_xml != NULL) && pcmk__xe_is(old_xml, (const char *) new_xml->name) && pcmk__str_eq(pcmk__xe_id(old_xml), pcmk__xe_id(new_xml), pcmk__str_none), return); if (!pcmk__xml_doc_all_flags_set(new_xml->doc, pcmk__xf_tracking)) { // Ensure tracking has a clean start (pcmk__xml_mark_changes() enables) pcmk__xml_commit_changes(new_xml->doc); } pcmk__xml_mark_changes(old_xml, new_xml); } void xml_calculate_significant_changes(xmlNode *old_xml, xmlNode *new_xml) { CRM_CHECK((old_xml != NULL) && (new_xml != NULL) && pcmk__xe_is(old_xml, (const char *) new_xml->name) && pcmk__str_eq(pcmk__xe_id(old_xml), pcmk__xe_id(new_xml), pcmk__str_none), return); /* BUG: If pcmk__xf_tracking is not set for new_xml when this function is * called, then we unset pcmk__xf_ignore_attr_pos via * pcmk__xml_commit_changes(). Since this function is about to be * deprecated, it's not worth fixing this and changing the user-facing * behavior. */ pcmk__xml_doc_set_flags(new_xml->doc, pcmk__xf_ignore_attr_pos); if (!pcmk__xml_doc_all_flags_set(new_xml->doc, pcmk__xf_tracking)) { // Ensure tracking has a clean start (pcmk__xml_mark_changes() enables) pcmk__xml_commit_changes(new_xml->doc); } pcmk__xml_mark_changes(old_xml, new_xml); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c index e6cff3e26c..d293683ccf 100644 --- a/lib/pacemaker/pcmk_graph_producer.c +++ b/lib/pacemaker/pcmk_graph_producer.c @@ -1,1108 +1,1108 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "libpacemaker_private.h" // Convenience macros for logging action properties #define action_type_str(flags) \ (pcmk_is_set((flags), pcmk__action_pseudo)? "pseudo-action" : "action") #define action_optional_str(flags) \ (pcmk_is_set((flags), pcmk__action_optional)? "optional" : "required") #define action_runnable_str(flags) \ (pcmk_is_set((flags), pcmk__action_runnable)? "runnable" : "unrunnable") #define action_node_str(a) \ (((a)->node == NULL)? "no node" : (a)->node->priv->name) /*! * \internal * \brief Add an XML node tag for a specified ID * * \param[in] id Node UUID to add * \param[in,out] xml Parent XML tag to add to */ static xmlNode* add_node_to_xml_by_id(const char *id, xmlNode *xml) { xmlNode *node_xml; node_xml = pcmk__xe_create(xml, PCMK_XE_NODE); crm_xml_add(node_xml, PCMK_XA_ID, id); return node_xml; } /*! * \internal * \brief Add an XML node tag for a specified node * * \param[in] node Node to add * \param[in,out] xml XML to add node to */ static void add_node_to_xml(const pcmk_node_t *node, void *xml) { add_node_to_xml_by_id(node->priv->id, (xmlNode *) xml); } /*! * \internal * \brief Count (optionally add to XML) nodes needing maintenance state update * * \param[in,out] xml Parent XML tag to add to, if any * \param[in] scheduler Scheduler data * * \return Count of nodes added * \note Only Pacemaker Remote nodes are considered currently */ static int add_maintenance_nodes(xmlNode *xml, const pcmk_scheduler_t *scheduler) { xmlNode *maintenance = NULL; int count = 0; if (xml != NULL) { maintenance = pcmk__xe_create(xml, PCMK__XE_MAINTENANCE); } for (const GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = iter->data; if (!pcmk__is_pacemaker_remote_node(node)) { continue; } if ((node->details->maintenance && !pcmk_is_set(node->priv->flags, pcmk__node_remote_maint)) || (!node->details->maintenance && pcmk_is_set(node->priv->flags, pcmk__node_remote_maint))) { if (maintenance != NULL) { crm_xml_add(add_node_to_xml_by_id(node->priv->id, maintenance), PCMK__XA_NODE_IN_MAINTENANCE, (node->details->maintenance? "1" : "0")); } count++; } } crm_trace("%s %d nodes in need of maintenance mode update in state", ((maintenance == NULL)? "Counted" : "Added"), count); return count; } /*! * \internal * \brief Add pseudo action with nodes needing maintenance state update * * \param[in,out] scheduler Scheduler data */ static void add_maintenance_update(pcmk_scheduler_t *scheduler) { pcmk_action_t *action = NULL; if (add_maintenance_nodes(NULL, scheduler) != 0) { action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, scheduler); pcmk__set_action_flags(action, pcmk__action_always_in_graph); } } /*! * \internal * \brief Add XML with nodes that an action is expected to bring down * * If a specified action is expected to bring any nodes down, add an XML block * with their UUIDs. When a node is lost, this allows the controller to * determine whether it was expected. * * \param[in,out] xml Parent XML tag to add to * \param[in] action Action to check for downed nodes */ static void add_downed_nodes(xmlNode *xml, const pcmk_action_t *action) { CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL), return); if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { /* Shutdown makes the action's node down */ xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->node->priv->id, downed); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* Fencing makes the action's node and any hosted guest nodes down */ const char *fence = g_hash_table_lookup(action->meta, PCMK__META_STONITH_ACTION); if (pcmk__is_fencing_action(fence)) { xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->node->priv->id, downed); pe_foreach_guest_node(action->node->priv->scheduler, action->node, add_node_to_xml, downed); } } else if ((action->rsc != NULL) && pcmk_is_set(action->rsc->flags, pcmk__rsc_is_remote_connection) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Stopping a remote connection resource makes connected node down, * unless it's part of a migration */ GList *iter; pcmk_action_t *input; bool migrating = false; for (iter = action->actions_before; iter != NULL; iter = iter->next) { input = ((pcmk__related_action_t *) iter->data)->action; if ((input->rsc != NULL) && pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none) && pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { migrating = true; break; } } if (!migrating) { xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->rsc->id, downed); } } } /*! * \internal * \brief Create a transition graph operation key for a clone action * * \param[in] action Clone action * \param[in] interval_ms Action interval in milliseconds * * \return Newly allocated string with transition graph operation key */ static char * clone_op_key(const pcmk_action_t *action, guint interval_ms) { if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_operation"); return pcmk__notify_key(action->rsc->priv->history_id, n_type, n_task); } return pcmk__op_key(action->rsc->priv->history_id, pcmk__s(action->cancel_task, action->task), interval_ms); } /*! * \internal * \brief Add node details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] xml Transition graph action XML for \p action */ static void add_node_details(const pcmk_action_t *action, xmlNode *xml) { pcmk_node_t *router_node = pcmk__connection_host_for_action(action); crm_xml_add(xml, PCMK__META_ON_NODE, action->node->priv->name); crm_xml_add(xml, PCMK__META_ON_NODE_UUID, action->node->priv->id); if (router_node != NULL) { crm_xml_add(xml, PCMK__XA_ROUTER_NODE, router_node->priv->name); } } /*! * \internal * \brief Add resource details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_resource_details(const pcmk_action_t *action, xmlNode *action_xml) { xmlNode *rsc_xml = NULL; const char *attr_list[] = { PCMK_XA_CLASS, PCMK_XA_PROVIDER, PCMK_XA_TYPE, }; /* If a resource is locked to a node via PCMK_OPT_SHUTDOWN_LOCK, mark its * actions so the controller can preserve the lock when the action * completes. */ if (pcmk__action_locks_rsc_to_node(action)) { crm_xml_add_ll(action_xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) action->rsc->priv->lock_time); } // List affected resource rsc_xml = pcmk__xe_create(action_xml, (const char *) action->rsc->priv->xml->name); if (pcmk_is_set(action->rsc->flags, pcmk__rsc_removed) && (action->rsc->priv->history_id != NULL)) { /* Use the numbered instance name here, because if there is more * than one instance on a node, we need to make sure the command * goes to the right one. * * This is important even for anonymous clones, because the clone's * unique meta-attribute might have just been toggled from on to * off. */ crm_debug("Using orphan clone name %s instead of history ID %s", action->rsc->id, action->rsc->priv->history_id); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->priv->history_id); crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } else if (!pcmk_is_set(action->rsc->flags, pcmk__rsc_unique)) { const char *xml_id = pcmk__xe_id(action->rsc->priv->xml); crm_debug("Using anonymous clone name %s for %s (aka %s)", xml_id, action->rsc->id, action->rsc->priv->history_id); /* ID is what we'd like client to use * LONG_ID is what they might know it as instead * * LONG_ID is only strictly needed /here/ during the * transition period until all nodes in the cluster * are running the new software /and/ have rebooted * once (meaning that they've only ever spoken to a DC * supporting this feature). (@TODO The effect of removing this on * regression tests suggests that it is still needed for unique clones) * * If anyone toggles the unique flag to 'on', the * 'instance free' name will correspond to an orphan * and fall into the clause above instead */ crm_xml_add(rsc_xml, PCMK_XA_ID, xml_id); if ((action->rsc->priv->history_id != NULL) && !pcmk__str_eq(xml_id, action->rsc->priv->history_id, pcmk__str_none)) { crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->priv->history_id); } else { crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } } else { pcmk__assert(action->rsc->priv->history_id == NULL); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->id); } for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) { crm_xml_add(rsc_xml, attr_list[lpc], g_hash_table_lookup(action->rsc->priv->meta, attr_list[lpc])); } } /*! * \internal * \brief Add action attributes to transition graph action XML * * \param[in,out] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_action_attributes(pcmk_action_t *action, xmlNode *action_xml) { xmlNode *args_xml = NULL; pcmk_resource_t *rsc = action->rsc; /* We create free-standing XML to start, so we can sort the attributes * before adding it to action_xml, which keeps the scheduler regression * test graphs comparable. */ args_xml = pcmk__xe_create(action_xml, PCMK__XE_ATTRIBUTES); crm_xml_add(args_xml, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); g_hash_table_foreach(action->extra, hash2field, args_xml); if ((rsc != NULL) && (action->node != NULL)) { // Get the resource instance attributes, evaluated properly for node GHashTable *params = pe_rsc_params(rsc, action->node, rsc->priv->scheduler); pcmk__substitute_remote_addr(rsc, params); g_hash_table_foreach(params, hash2smartfield, args_xml); } else if ((rsc != NULL) && (rsc->priv->variant <= pcmk__rsc_variant_primitive)) { GHashTable *params = pe_rsc_params(rsc, NULL, rsc->priv->scheduler); g_hash_table_foreach(params, hash2smartfield, args_xml); } g_hash_table_foreach(action->meta, hash2metafield, args_xml); if (rsc != NULL) { pcmk_resource_t *parent = rsc; while (parent != NULL) { parent->priv->cmds->add_graph_meta(parent, args_xml); parent = parent->priv->parent; } pcmk__add_guest_meta_to_xml(args_xml, action); } pcmk__xe_sort_attrs(args_xml); } /*! * \internal * \brief Create the transition graph XML for a scheduled action * * \param[in,out] parent Parent XML element to add action to * \param[in,out] action Scheduled action * \param[in] skip_details If false, add action details as sub-elements * \param[in] scheduler Scheduler data */ static void create_graph_action(xmlNode *parent, pcmk_action_t *action, bool skip_details, const pcmk_scheduler_t *scheduler) { bool needs_node_info = true; bool needs_maintenance_info = false; xmlNode *action_xml = NULL; if ((action == NULL) || (scheduler == NULL)) { return; } // Create the top-level element based on task if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* All fences need node info; guest node fences are pseudo-events */ if (pcmk_is_set(action->flags, pcmk__action_pseudo)) { action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT); } else { action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); } } else if (pcmk__str_any_of(action->task, PCMK_ACTION_DO_SHUTDOWN, PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) { action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); } else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { // CIB-only clean-up for shutdown locks action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); crm_xml_add(action_xml, PCMK__XA_MODE, PCMK__VALUE_CIB); } else if (pcmk_is_set(action->flags, pcmk__action_pseudo)) { if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES, pcmk__str_none)) { needs_maintenance_info = true; } action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT); needs_node_info = false; } else { action_xml = pcmk__xe_create(parent, PCMK__XE_RSC_OP); } crm_xml_add_int(action_xml, PCMK_XA_ID, action->id); crm_xml_add(action_xml, PCMK_XA_OPERATION, action->task); if ((action->rsc != NULL) && (action->rsc->priv->history_id != NULL)) { char *clone_key = NULL; guint interval_ms; if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } clone_key = clone_op_key(action, interval_ms); crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, clone_key); crm_xml_add(action_xml, "internal_" PCMK__XA_OPERATION_KEY, action->uuid); free(clone_key); } else { crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, action->uuid); } if (needs_node_info && (action->node != NULL)) { add_node_details(action, action_xml); pcmk__insert_dup(action->meta, PCMK__META_ON_NODE, action->node->priv->name); pcmk__insert_dup(action->meta, PCMK__META_ON_NODE_UUID, action->node->priv->id); } if (skip_details) { return; } if ((action->rsc != NULL) && !pcmk_is_set(action->flags, pcmk__action_pseudo)) { // This is a real resource action, so add resource details add_resource_details(action, action_xml); } /* List any attributes in effect */ add_action_attributes(action, action_xml); /* List any nodes this action is expected to make down */ if (needs_node_info && (action->node != NULL)) { add_downed_nodes(action_xml, action); } if (needs_maintenance_info) { add_maintenance_nodes(action_xml, scheduler); } } /*! * \internal * \brief Check whether an action should be added to the transition graph * * \param[in,out] action Action to check * * \return true if action should be added to graph, otherwise false */ static bool should_add_action_to_graph(pcmk_action_t *action) { if (!pcmk_is_set(action->flags, pcmk__action_runnable)) { crm_trace("Ignoring action %s (%d): unrunnable", action->uuid, action->id); return false; } if (pcmk_is_set(action->flags, pcmk__action_optional) && !pcmk_is_set(action->flags, pcmk__action_always_in_graph)) { crm_trace("Ignoring action %s (%d): optional", action->uuid, action->id); return false; } /* Actions for unmanaged resources should be excluded from the graph, * with the exception of monitors and cancellation of recurring monitors. */ if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed) && !pcmk__str_eq(action->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { const char *interval_ms_s; /* A cancellation of a recurring monitor will get here because the task * is cancel rather than monitor, but the interval can still be used to * recognize it. The interval has been normalized to milliseconds by * this point, so a string comparison is sufficient. */ interval_ms_s = g_hash_table_lookup(action->meta, PCMK_META_INTERVAL); if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) { crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)", action->uuid, action->id, action->rsc->id); return false; } } /* Always add pseudo-actions, fence actions, and shutdown actions (already * determined to be required and runnable by this point) */ if (pcmk_is_set(action->flags, pcmk__action_pseudo) || pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH, PCMK_ACTION_DO_SHUTDOWN, NULL)) { return true; } if (action->node == NULL) { pcmk__sched_err(action->scheduler, "Skipping action %s (%d) " "because it was not assigned to a node (bug?)", action->uuid, action->id); pcmk__log_action("Unassigned", action, false); return false; } if (pcmk_is_set(action->flags, pcmk__action_on_dc)) { crm_trace("Action %s (%d) should be dumped: " "can run on DC instead of %s", action->uuid, action->id, pcmk__node_name(action->node)); } else if (pcmk__is_guest_or_bundle_node(action->node) && !pcmk_is_set(action->node->priv->flags, pcmk__node_remote_reset)) { crm_trace("Action %s (%d) should be dumped: " "assuming will be runnable on guest %s", action->uuid, action->id, pcmk__node_name(action->node)); } else if (!action->node->details->online) { pcmk__sched_err(action->scheduler, "Skipping action %s (%d) " "because it was scheduled for offline node (bug?)", action->uuid, action->id); pcmk__log_action("Offline node", action, false); return false; } else if (action->node->details->unclean) { pcmk__sched_err(action->scheduler, "Skipping action %s (%d) " "because it was scheduled for unclean node (bug?)", action->uuid, action->id); pcmk__log_action("Unclean node", action, false); return false; } return true; } /*! * \internal * \brief Check whether an ordering's flags can change an action * * \param[in] ordering Ordering to check * * \return true if ordering has flags that can change an action, false otherwise */ static bool ordering_can_change_actions(const pcmk__related_action_t *ordering) { return pcmk_any_flags_set(ordering->flags, ~(pcmk__ar_then_implies_first_graphed |pcmk__ar_first_implies_then_graphed |pcmk__ar_ordered)); } /*! * \internal * \brief Check whether an action input should be in the transition graph * * \param[in] action Action to check * \param[in,out] input Action input to check * * \return true if input should be in graph, false otherwise * \note This function may not only check an input, but disable it under certian * circumstances (load or anti-colocation orderings that are not needed). */ static bool should_add_input_to_graph(const pcmk_action_t *action, pcmk__related_action_t *input) { if (input->graphed) { return true; } if (input->flags == pcmk__ar_none) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering disabled", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk__action_runnable) && !ordering_can_change_actions(input)) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional and input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk__action_runnable) && pcmk_is_set(input->flags, pcmk__ar_min_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "minimum number of instances required but input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->flags, pcmk__ar_unmigratable_then_blocks) && !pcmk_is_set(input->action->flags, pcmk__action_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input blocked if 'then' unmigratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->flags, pcmk__ar_if_first_unmigratable) && pcmk_is_set(input->action->flags, pcmk__action_migratable)) { crm_trace("Ignoring %s (%d) input %s (%d): ordering applies " "only if input is unmigratable, but it is migratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if ((input->flags == pcmk__ar_ordered) && pcmk_is_set(input->action->flags, pcmk__action_migratable) - && pcmk__ends_with(input->action->uuid, "_stop_0")) { + && g_str_has_suffix(input->action->uuid, "_stop_0")) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional but stop in migration", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (input->flags == pcmk__ar_if_on_same_node_or_target) { pcmk_node_t *input_node = input->action->node; if ((action->rsc != NULL) && pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { pcmk_node_t *assigned = action->rsc->priv->assigned_node; /* For load_stopped -> migrate_to orderings, we care about where * the resource has been assigned, not where migrate_to will be * executed. */ if (!pcmk__same_node(input_node, assigned)) { crm_trace("Ignoring %s (%d) input %s (%d): " "migration target %s is not same as input node %s", action->uuid, action->id, input->action->uuid, input->action->id, (assigned? assigned->priv->name : ""), (input_node? input_node->priv->name : "")); input->flags = pcmk__ar_none; return false; } } else if (!pcmk__same_node(input_node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, (action->node? action->node->priv->name : ""), (input_node? input_node->priv->name : "")); input->flags = pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering optional", action->uuid, action->id, input->action->uuid, input->action->id); input->flags = pcmk__ar_none; return false; } } else if (input->flags == pcmk__ar_if_required_on_same_node) { if (input->action->node && action->node && !pcmk__same_node(input->action->node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, pcmk__node_name(action->node), pcmk__node_name(input->action->node)); input->flags = pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk__action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): optional", action->uuid, action->id, input->action->uuid, input->action->id); input->flags = pcmk__ar_none; return false; } } else if (input->action->rsc && input->action->rsc != action->rsc && pcmk_is_set(input->action->rsc->flags, pcmk__rsc_failed) && !pcmk_is_set(input->action->rsc->flags, pcmk__rsc_managed) - && pcmk__ends_with(input->action->uuid, "_stop_0") + && g_str_has_suffix(input->action->uuid, "_stop_0") && pcmk__is_clone(action->rsc)) { crm_warn("Ignoring requirement that %s complete before %s:" " unmanaged failed resources cannot prevent clone shutdown", input->action->uuid, action->uuid); return false; } else if (pcmk_is_set(input->action->flags, pcmk__action_optional) && !pcmk_any_flags_set(input->action->flags, pcmk__action_always_in_graph |pcmk__action_added_to_graph) && !should_add_action_to_graph(input->action)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input optional", action->uuid, action->id, input->action->uuid, input->action->id); return false; } crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x", action->uuid, action->id, action_type_str(input->action->flags), input->action->uuid, input->action->id, action_node_str(input->action), action_runnable_str(input->action->flags), action_optional_str(input->action->flags), input->flags); return true; } /*! * \internal * \brief Check whether an ordering creates an ordering loop * * \param[in] init_action "First" action in ordering * \param[in] action Callers should always set this the same as * \p init_action (this function may use a different * value for recursive calls) * \param[in,out] input Action wrapper for "then" action in ordering * * \return true if the ordering creates a loop, otherwise false */ bool pcmk__graph_has_loop(const pcmk_action_t *init_action, const pcmk_action_t *action, pcmk__related_action_t *input) { bool has_loop = false; if (pcmk_is_set(input->action->flags, pcmk__action_detect_loop)) { crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->priv->name : "", action->uuid, action->node? action->node->priv->name : "", input->flags); return false; } // Don't need to check inputs that won't be used if (!should_add_input_to_graph(action, input)) { return false; } if (input->action == init_action) { crm_debug("Input loop found in %s@%s ->...-> %s@%s", action->uuid, action->node? action->node->priv->name : "", init_action->uuid, init_action->node? init_action->node->priv->name : ""); return true; } pcmk__set_action_flags(input->action, pcmk__action_detect_loop); crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)" "for graph loop with %s@%s ", action->uuid, action->node? action->node->priv->name : "", input->action->uuid, input->action->node? input->action->node->priv->name : "", input->flags, init_action->uuid, init_action->node? init_action->node->priv->name : ""); // Recursively check input itself for loops for (GList *iter = input->action->actions_before; iter != NULL; iter = iter->next) { if (pcmk__graph_has_loop(init_action, input->action, (pcmk__related_action_t *) iter->data)) { // Recursive call already logged a debug message has_loop = true; break; } } pcmk__clear_action_flags(input->action, pcmk__action_detect_loop); if (!has_loop) { crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->priv->name : "", action->uuid, action->node? action->node->priv->name : "", input->flags); } return has_loop; } /*! * \internal * \brief Create a synapse XML element for a transition graph * * \param[in] action Action that synapse is for * \param[in,out] scheduler Scheduler data containing graph * * \return Newly added XML element for new graph synapse */ static xmlNode * create_graph_synapse(const pcmk_action_t *action, pcmk_scheduler_t *scheduler) { int synapse_priority = 0; xmlNode *syn = pcmk__xe_create(scheduler->priv->graph, PCMK__XE_SYNAPSE); crm_xml_add_int(syn, PCMK_XA_ID, scheduler->priv->synapse_count++); if (action->rsc != NULL) { synapse_priority = action->rsc->priv->priority; } if (action->priority > synapse_priority) { synapse_priority = action->priority; } if (synapse_priority > 0) { crm_xml_add_int(syn, PCMK__XA_PRIORITY, synapse_priority); } return syn; } /*! * \internal * \brief Add an action to the transition graph XML if appropriate * * \param[in,out] data Action to possibly add * \param[in,out] user_data Scheduler data * * \note This will de-duplicate the action inputs, meaning that the * pcmk__related_action_t:type flags can no longer be relied on to retain * their original settings. That means this MUST be called after * pcmk__apply_orderings() is complete, and nothing after this should rely * on those type flags. (For example, some code looks for type equal to * some flag rather than whether the flag is set, and some code looks for * particular combinations of flags -- such code must be done before * pcmk__create_graph().) */ static void add_action_to_graph(gpointer data, gpointer user_data) { pcmk_action_t *action = (pcmk_action_t *) data; pcmk_scheduler_t *scheduler = (pcmk_scheduler_t *) user_data; xmlNode *syn = NULL; xmlNode *set = NULL; xmlNode *in = NULL; /* If we haven't already, de-duplicate inputs (even if we won't be adding * the action to the graph, so that crm_simulate's dot graphs don't have * duplicates). */ if (!pcmk_is_set(action->flags, pcmk__action_inputs_deduplicated)) { pcmk__deduplicate_action_inputs(action); pcmk__set_action_flags(action, pcmk__action_inputs_deduplicated); } if (pcmk_is_set(action->flags, pcmk__action_added_to_graph) || !should_add_action_to_graph(action)) { return; // Already added, or shouldn't be } pcmk__set_action_flags(action, pcmk__action_added_to_graph); crm_trace("Adding action %d (%s%s%s) to graph", action->id, action->uuid, ((action->node == NULL)? "" : " on "), ((action->node == NULL)? "" : action->node->priv->name)); syn = create_graph_synapse(action, scheduler); set = pcmk__xe_create(syn, PCMK__XE_ACTION_SET); in = pcmk__xe_create(syn, PCMK__XE_INPUTS); create_graph_action(set, action, false, scheduler); for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *input = lpc->data; if (should_add_input_to_graph(action, input)) { xmlNode *input_xml = pcmk__xe_create(in, PCMK__XE_TRIGGER); input->graphed = true; create_graph_action(input_xml, input->action, true, scheduler); } } } static int transition_id = 0; /*! * \internal * \brief Log a message after calculating a transition * * \param[in] scheduler Scheduler data * \param[in] filename Where transition input is stored */ void pcmk__log_transition_summary(const pcmk_scheduler_t *scheduler, const char *filename) { if (pcmk_is_set(scheduler->flags, pcmk__sched_processing_error) || pcmk__config_has_error) { crm_err("Calculated transition %d (with errors)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else if (pcmk_is_set(scheduler->flags, pcmk__sched_processing_warning) || pcmk__config_has_warning) { crm_warn("Calculated transition %d (with warnings)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else { crm_notice("Calculated transition %d%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } if (pcmk__config_has_error) { crm_notice("Configuration errors found during scheduler processing," " please run \"crm_verify -L\" to identify issues"); } } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc) { GList *iter = NULL; pcmk__assert(rsc != NULL); pcmk__rsc_trace(rsc, "Adding actions for %s to graph", rsc->id); // First add the resource's own actions g_list_foreach(rsc->priv->actions, add_action_to_graph, rsc->priv->scheduler); // Then recursively add its children's actions (appropriate to variant) for (iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; child_rsc->priv->cmds->add_actions_to_graph(child_rsc); } } /*! * \internal * \brief Create a transition graph with all cluster actions needed * * \param[in,out] scheduler Scheduler data */ void pcmk__create_graph(pcmk_scheduler_t *scheduler) { GList *iter = NULL; const char *value = NULL; long long limit = 0LL; GHashTable *config_hash = scheduler->priv->options; int rc = pcmk_rc_ok; transition_id++; crm_trace("Creating transition graph %d", transition_id); scheduler->priv->graph = pcmk__xe_create(NULL, PCMK__XE_TRANSITION_GRAPH); value = pcmk__cluster_option(config_hash, PCMK_OPT_CLUSTER_DELAY); crm_xml_add(scheduler->priv->graph, PCMK_OPT_CLUSTER_DELAY, value); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); crm_xml_add(scheduler->priv->graph, PCMK_OPT_STONITH_TIMEOUT, value); crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_STOP_OFFSET, PCMK_VALUE_INFINITY); if (pcmk_is_set(scheduler->flags, pcmk__sched_start_failure_fatal)) { crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_START_OFFSET, PCMK_VALUE_INFINITY); } else { crm_xml_add(scheduler->priv->graph, PCMK__XA_FAILED_START_OFFSET, "1"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_BATCH_LIMIT); crm_xml_add(scheduler->priv->graph, PCMK_OPT_BATCH_LIMIT, value); crm_xml_add_int(scheduler->priv->graph, "transition_id", transition_id); value = pcmk__cluster_option(config_hash, PCMK_OPT_MIGRATION_LIMIT); rc = pcmk__scan_ll(value, &limit, 0LL); if (rc != pcmk_rc_ok) { crm_warn("Ignoring invalid value '%s' for " PCMK_OPT_MIGRATION_LIMIT ": %s", value, pcmk_rc_str(rc)); } else if (limit > 0) { crm_xml_add(scheduler->priv->graph, PCMK_OPT_MIGRATION_LIMIT, value); } if (scheduler->priv->recheck_by > 0) { char *recheck_epoch = NULL; recheck_epoch = crm_strdup_printf("%llu", (unsigned long long) scheduler->priv->recheck_by); crm_xml_add(scheduler->priv->graph, "recheck-by", recheck_epoch); free(recheck_epoch); } /* The following code will de-duplicate action inputs, so nothing past this * should rely on the action input type flags retaining their original * values. */ // Add resource actions to graph for (iter = scheduler->priv->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; pcmk__rsc_trace(rsc, "Processing actions for %s", rsc->id); rsc->priv->cmds->add_actions_to_graph(rsc); } // Add pseudo-action for list of nodes with maintenance state update add_maintenance_update(scheduler); // Add non-resource (node) actions for (iter = scheduler->priv->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; if ((action->rsc != NULL) && (action->node != NULL) && action->node->details->shutdown && !pcmk_is_set(action->rsc->flags, pcmk__rsc_maintenance) && !pcmk_any_flags_set(action->flags, pcmk__action_optional|pcmk__action_runnable) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Eventually we should just ignore the 'fence' case, but for now * it's the best way to detect (in CTS) when CIB resource updates * are being lost. */ if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate) || (scheduler->no_quorum_policy == pcmk_no_quorum_ignore)) { const bool managed = pcmk_is_set(action->rsc->flags, pcmk__rsc_managed); const bool failed = pcmk_is_set(action->rsc->flags, pcmk__rsc_failed); crm_crit("Cannot %s %s because of %s:%s%s (%s)", action->node->details->unclean? "fence" : "shut down", pcmk__node_name(action->node), action->rsc->id, (managed? " blocked" : " unmanaged"), (failed? " failed" : ""), action->uuid); } } add_action_to_graph((gpointer) action, (gpointer) scheduler); } crm_log_xml_trace(scheduler->priv->graph, "graph"); } diff --git a/lib/pacemaker/pcmk_resource.c b/lib/pacemaker/pcmk_resource.c index 6a3e50ae57..c783ac499d 100644 --- a/lib/pacemaker/pcmk_resource.c +++ b/lib/pacemaker/pcmk_resource.c @@ -1,253 +1,253 @@ /* * Copyright 2021-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include // xmlNode #include #include #include #include #include #include #include // Search path for resource operation history (takes node name and resource ID) #define XPATH_OP_HISTORY "//" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" static xmlNode * best_op(const pcmk_resource_t *rsc, const pcmk_node_t *node) { char *xpath = NULL; xmlNode *history = NULL; xmlNode *best = NULL; bool best_effective_op = false; guint best_interval = 0; bool best_failure = false; const char *best_digest = NULL; // Find node's resource history xpath = crm_strdup_printf(XPATH_OP_HISTORY, node->priv->name, rsc->id); history = pcmk__xpath_find_one(rsc->priv->scheduler->input->doc, xpath, LOG_NEVER); free(xpath); // Examine each history entry for (xmlNode *lrm_rsc_op = pcmk__xe_first_child(history, PCMK__XE_LRM_RSC_OP, NULL, NULL); lrm_rsc_op != NULL; lrm_rsc_op = pcmk__xe_next(lrm_rsc_op, PCMK__XE_LRM_RSC_OP)) { const char *digest = pcmk__xe_get(lrm_rsc_op, PCMK__XA_OP_RESTART_DIGEST); guint interval_ms = 0; const char *task = pcmk__xe_get(lrm_rsc_op, PCMK_XA_OPERATION); bool effective_op = false; - bool failure = pcmk__ends_with(pcmk__xe_id(lrm_rsc_op), - "_last_failure_0"); + const char *id = pcmk__xe_id(lrm_rsc_op); + bool failure = (id != NULL) && g_str_has_suffix(id, "_last_failure_0"); pcmk__xe_get_guint(lrm_rsc_op, PCMK_META_INTERVAL, &interval_ms); effective_op = interval_ms == 0 && pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_PROMOTE, PCMK_ACTION_MIGRATE_FROM, NULL); if (best == NULL) { goto is_best; } if (best_effective_op) { // Do not use an ineffective op if there's an effective one. if (!effective_op) { continue; } // Do not use an ineffective non-recurring op if there's a recurring one } else if (best_interval != 0 && !effective_op && interval_ms == 0) { continue; } // Do not use last failure if there's a successful one. if (!best_failure && failure) { continue; } // Do not use an op without a restart digest if there's one with. if (best_digest != NULL && digest == NULL) { continue; } // Do not use an older op if there's a newer one. if (pe__is_newer_op(best, lrm_rsc_op) > 0) { continue; } is_best: best = lrm_rsc_op; best_effective_op = effective_op; best_interval = interval_ms; best_failure = failure; best_digest = digest; } return best; } /*! * \internal * \brief Remove a resource * * \param[in,out] cib An open connection to the CIB * \param[in] cib_opts Options to use in the CIB operation call * \param[in] rsc_id Resource to remove * \param[in] rsc_type Type of the resource ("primitive", "group", etc.) * * \return Standard Pacemaker return code */ int pcmk__resource_delete(cib_t *cib, uint32_t cib_opts, const char *rsc_id, const char *rsc_type) { int rc = pcmk_rc_ok; xmlNode *msg_data = NULL; if (cib == NULL) { return ENOTCONN; } if (rsc_id == NULL || rsc_type == NULL) { return EINVAL; } msg_data = pcmk__xe_create(NULL, rsc_type); crm_xml_add(msg_data, PCMK_XA_ID, rsc_id); rc = cib->cmds->remove(cib, PCMK_XE_RESOURCES, msg_data, cib_opts); rc = pcmk_legacy2rc(rc); pcmk__xml_free(msg_data); return rc; } int pcmk_resource_delete(xmlNodePtr *xml, const char *rsc_id, const char *rsc_type) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; uint32_t cib_opts = cib_sync_call; cib_t *cib = NULL; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } cib = cib_new(); if (cib == NULL) { rc = pcmk_rc_cib_corrupt; goto done; } rc = cib->cmds->signon(cib, crm_system_name, cib_command); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { goto done; } rc = pcmk__resource_delete(cib, cib_opts, rsc_id, rsc_type); done: if (cib != NULL) { cib__clean_up_connection(&cib); } pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml); return rc; } /*! * \internal * \brief Calculate and output resource operation digests * * \param[in,out] out Output object * \param[in,out] rsc Resource to calculate digests for * \param[in] node Node whose operation history should be used * \param[in] overrides Hash table of configuration parameters to override * * \return Standard Pacemaker return code */ int pcmk__resource_digests(pcmk__output_t *out, pcmk_resource_t *rsc, const pcmk_node_t *node, GHashTable *overrides) { const char *task = NULL; xmlNode *xml_op = NULL; pcmk__op_digest_t *digests = NULL; guint interval_ms = 0; int rc = pcmk_rc_ok; if ((out == NULL) || (rsc == NULL) || (node == NULL)) { return EINVAL; } if (!pcmk__is_primitive(rsc)) { // Only primitives get operation digests return EOPNOTSUPP; } // Find XML of operation history to use xml_op = best_op(rsc, node); // Generate an operation key if (xml_op != NULL) { task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION); pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &interval_ms); } if (task == NULL) { // Assume start if no history is available task = PCMK_ACTION_START; interval_ms = 0; } // Calculate and show digests digests = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op, overrides, true, rsc->priv->scheduler); rc = out->message(out, "digests", rsc, node, task, interval_ms, digests); pe__free_digests(digests); return rc; } int pcmk_resource_digests(xmlNodePtr *xml, pcmk_resource_t *rsc, const pcmk_node_t *node, GHashTable *overrides) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pcmk__register_lib_messages(out); rc = pcmk__resource_digests(out, rsc, node, overrides); pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml); return rc; } diff --git a/lib/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c index f2bc1a432c..b5b5af5c13 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,1713 +1,1717 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* This file is intended for code usable with both clone instances and bundle * replica containers. */ #include + +#include // g_str_has_suffix() #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is allowed to run an instance * * \param[in] instance Clone instance or bundle container to check * \param[in] node Node to check * \param[in] max_per_node Maximum number of instances allowed to run on a node * * \return true if \p node is allowed to run \p instance, otherwise false */ static bool can_run_instance(const pcmk_resource_t *instance, const pcmk_node_t *node, int max_per_node) { pcmk_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pcmk__rsc_removed)) { pcmk__rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pcmk__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pcmk__rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pcmk__node_name(node)); return false; } allowed_node = pcmk__top_allowed_node(instance, node); if (allowed_node == NULL) { crm_warn("%s cannot run on %s: node not allowed", instance->id, pcmk__node_name(node)); return false; } if (allowed_node->assign->score < 0) { pcmk__rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pcmk__node_name(node), pcmk_readable_score(allowed_node->assign->score)); return false; } if (allowed_node->assign->count >= max_per_node) { pcmk__rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pcmk__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pcmk__rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pcmk__node_name(node), allowed_node->assign->count); return true; } /*! * \internal * \brief Ban a clone instance or bundle replica from unavailable allowed nodes * * \param[in,out] instance Clone instance or bundle replica to ban * \param[in] max_per_node Maximum instances allowed to run on a node */ static void ban_unavailable_allowed_nodes(pcmk_resource_t *instance, int max_per_node) { if (instance->priv->allowed_nodes != NULL) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, instance->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!can_run_instance(instance, node, max_per_node)) { pcmk__rsc_trace(instance, "Banning %s from unavailable node %s", instance->id, pcmk__node_name(node)); node->assign->score = -PCMK_SCORE_INFINITY; for (GList *child_iter = instance->priv->children; child_iter != NULL; child_iter = child_iter->next) { pcmk_resource_t *child = child_iter->data; pcmk_node_t *child_node = NULL; child_node = g_hash_table_lookup(child->priv->allowed_nodes, node->priv->id); if (child_node != NULL) { pcmk__rsc_trace(instance, "Banning %s child %s " "from unavailable node %s", instance->id, child->id, pcmk__node_name(node)); child_node->assign->score = -PCMK_SCORE_INFINITY; } } } } } } /*! * \internal * \brief Create a hash table with a single node in it * * \param[in] node Node to copy into new table * * \return Newly created hash table containing a copy of \p node * \note The caller is responsible for freeing the result with * g_hash_table_destroy(). */ static GHashTable * new_node_table(pcmk_node_t *node) { GHashTable *table = pcmk__strkey_table(NULL, pcmk__free_node_copy); node = pe__copy_node(node); g_hash_table_insert(table, (gpointer) node->priv->id, node); return table; } /*! * \internal * \brief Apply a resource's parent's colocation scores to a node table * * \param[in] rsc Resource whose colocations should be applied * \param[in,out] nodes Node table to apply colocations to */ static void apply_parent_colocations(const pcmk_resource_t *rsc, GHashTable **nodes) { GList *colocations = pcmk__this_with_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->primary; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; other->priv->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_default); } g_list_free(colocations); colocations = pcmk__with_this_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->dependent; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (!pcmk__colocation_has_influence(colocation, rsc)) { continue; } other->priv->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_nonnegative); } g_list_free(colocations); } /*! * \internal * \brief Compare clone or bundle instances based on colocation scores * * Determine the relative order in which two clone or bundle instances should be * assigned to nodes, considering the scores of colocation constraints directly * or indirectly involving them. * * \param[in] instance1 First instance to compare * \param[in] instance2 Second instance to compare * * \return A negative number if \p instance1 should be assigned first, * a positive number if \p instance2 should be assigned first, * or 0 if assignment order doesn't matter */ static int cmp_instance_by_colocation(const pcmk_resource_t *instance1, const pcmk_resource_t *instance2) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; pcmk_node_t *current_node1 = pcmk__current_node(instance1); pcmk_node_t *current_node2 = pcmk__current_node(instance2); GHashTable *colocated_scores1 = NULL; GHashTable *colocated_scores2 = NULL; pcmk__assert((instance1 != NULL) && (instance1->priv->parent != NULL) && (instance2 != NULL) && (instance2->priv->parent != NULL) && (current_node1 != NULL) && (current_node2 != NULL)); // Create node tables initialized with each node colocated_scores1 = new_node_table(current_node1); colocated_scores2 = new_node_table(current_node2); // Apply parental colocations apply_parent_colocations(instance1, &colocated_scores1); apply_parent_colocations(instance2, &colocated_scores2); // Find original nodes again, with scores updated for colocations node1 = g_hash_table_lookup(colocated_scores1, current_node1->priv->id); node2 = g_hash_table_lookup(colocated_scores2, current_node2->priv->id); // Compare nodes by updated scores if (node1->assign->score < node2->assign->score) { crm_trace("Assign %s (%d on %s) after %s (%d on %s)", instance1->id, node1->assign->score, pcmk__node_name(node1), instance2->id, node2->assign->score, pcmk__node_name(node2)); rc = 1; } else if (node1->assign->score > node2->assign->score) { crm_trace("Assign %s (%d on %s) before %s (%d on %s)", instance1->id, node1->assign->score, pcmk__node_name(node1), instance2->id, node2->assign->score, pcmk__node_name(node2)); rc = -1; } g_hash_table_destroy(colocated_scores1); g_hash_table_destroy(colocated_scores2); return rc; } /*! * \internal * \brief Check whether a resource or any of its children are failed * * \param[in] rsc Resource to check * * \return true if \p rsc or any of its children are failed, otherwise false */ static bool did_fail(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { return true; } for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { if (did_fail((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in,out] node Node to check (will be set NULL if not allowed) * * \return true if *node is either NULL or allowed for \p rsc, otherwise false */ static bool node_is_allowed(const pcmk_resource_t *rsc, pcmk_node_t **node) { if (*node != NULL) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->priv->allowed_nodes, (*node)->priv->id); if ((allowed == NULL) || (allowed->assign->score < 0)) { pcmk__rsc_trace(rsc, "%s: current location (%s) is unavailable", rsc->id, pcmk__node_name(*node)); *node = NULL; return false; } } return true; } /*! * \internal * \brief Compare two clone or bundle instances' instance numbers * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a's instance number is lower, * a positive number if \p b's instance number is lower, * or 0 if their instance numbers are the same */ gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b) { const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; char *div1 = NULL; char *div2 = NULL; pcmk__assert((instance1 != NULL) && (instance2 != NULL)); // Clone numbers are after a colon, bundle numbers after a dash div1 = strrchr(instance1->id, ':'); if (div1 == NULL) { div1 = strrchr(instance1->id, '-'); } div2 = strrchr(instance2->id, ':'); if (div2 == NULL) { div2 = strrchr(instance2->id, '-'); } pcmk__assert((div1 != NULL) && (div2 != NULL)); return (gint) (strtol(div1 + 1, NULL, 10) - strtol(div2 + 1, NULL, 10)); } /*! * \internal * \brief Compare clone or bundle instances according to assignment order * * Compare two clone or bundle instances according to the order they should be * assigned to nodes, preferring (in order): * * - Active instance that is less multiply active * - Instance that is not active on a disallowed node * - Instance with higher configured priority * - Active instance whose current node can run resources * - Active instance whose parent is allowed on current node * - Active instance whose current node has fewer other instances * - Active instance * - Instance that isn't failed * - Instance whose colocations result in higher score on current node * - Instance with lower ID in lexicographic order * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a should be assigned first, * a positive number if \p b should be assigned first, * or 0 if assignment order doesn't matter */ gint pcmk__cmp_instance(gconstpointer a, gconstpointer b) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; unsigned int nnodes1 = 0; unsigned int nnodes2 = 0; bool can1 = true; bool can2 = true; const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; pcmk__assert((instance1 != NULL) && (instance2 != NULL)); node1 = instance1->priv->fns->active_node(instance1, &nnodes1, NULL); node2 = instance2->priv->fns->active_node(instance2, &nnodes2, NULL); /* If both instances are running and at least one is multiply * active, prefer instance that's running on fewer nodes. */ if ((nnodes1 > 0) && (nnodes2 > 0)) { if (nnodes1 < nnodes2) { crm_trace("Assign %s (active on %d) before %s (active on %d): " "less multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return -1; } else if (nnodes1 > nnodes2) { crm_trace("Assign %s (active on %d) after %s (active on %d): " "more multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return 1; } } /* An instance that is either inactive or active on an allowed node is * preferred over an instance that is active on a no-longer-allowed node. */ can1 = node_is_allowed(instance1, &node1); can2 = node_is_allowed(instance2, &node2); if (can1 && !can2) { crm_trace("Assign %s before %s: not active on a disallowed node", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: active on a disallowed node", instance1->id, instance2->id); return 1; } // Prefer instance with higher configured priority if (instance1->priv->priority > instance2->priv->priority) { crm_trace("Assign %s before %s: priority (%d > %d)", instance1->id, instance2->id, instance1->priv->priority, instance2->priv->priority); return -1; } else if (instance1->priv->priority < instance2->priv->priority) { crm_trace("Assign %s after %s: priority (%d < %d)", instance1->id, instance2->id, instance1->priv->priority, instance2->priv->priority); return 1; } // Prefer active instance if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: inactive", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: active", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: active", instance1->id, instance2->id); return -1; } // Prefer instance whose current node can run resources can1 = pcmk__node_available(node1, false, false); can2 = pcmk__node_available(node2, false, false); if (can1 && !can2) { crm_trace("Assign %s before %s: current node can run resources", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: current node can't run resources", instance1->id, instance2->id); return 1; } // Prefer instance whose parent is allowed to run on instance's current node node1 = pcmk__top_allowed_node(instance1, node1); node2 = pcmk__top_allowed_node(instance2, node2); if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: " "parent not allowed on either instance's current node", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: parent not allowed on current node", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: parent allowed on current node", instance1->id, instance2->id); return -1; } // Prefer instance whose current node is running fewer other instances if (node1->assign->count < node2->assign->count) { crm_trace("Assign %s before %s: fewer active instances on current node", instance1->id, instance2->id); return -1; } else if (node1->assign->count > node2->assign->count) { crm_trace("Assign %s after %s: more active instances on current node", instance1->id, instance2->id); return 1; } // Prefer instance that isn't failed can1 = did_fail(instance1); can2 = did_fail(instance2); if (!can1 && can2) { crm_trace("Assign %s before %s: not failed", instance1->id, instance2->id); return -1; } else if (can1 && !can2) { crm_trace("Assign %s after %s: failed", instance1->id, instance2->id); return 1; } // Prefer instance with higher cumulative colocation score on current node rc = cmp_instance_by_colocation(instance1, instance2); if (rc != 0) { return rc; } // Prefer instance with lower instance number rc = pcmk__cmp_instance_number(instance1, instance2); if (rc < 0) { crm_trace("Assign %s before %s: instance number", instance1->id, instance2->id); } else if (rc > 0) { crm_trace("Assign %s after %s: instance number", instance1->id, instance2->id); } else { crm_trace("No assignment preference for %s vs. %s", instance1->id, instance2->id); } return rc; } /*! * \internal * \brief Increment the parent's instance count after assigning an instance * * An instance's parent tracks how many instances have been assigned to each * node via its pcmk_node_t:count member. After assigning an instance to a node, * find the corresponding node in the parent's allowed table and increment it. * * \param[in,out] instance Instance whose parent to update * \param[in] assigned_to Node to which the instance was assigned */ static void increment_parent_count(pcmk_resource_t *instance, const pcmk_node_t *assigned_to) { pcmk_node_t *allowed = NULL; if (assigned_to == NULL) { return; } allowed = pcmk__top_allowed_node(instance, assigned_to); if (allowed == NULL) { /* The instance is allowed on the node, but its parent isn't. This * shouldn't be possible if the resource is managed, and we won't be * able to limit the number of instances assigned to the node. */ CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pcmk__rsc_managed)); } else { allowed->assign->count++; } } /*! * \internal * \brief Assign an instance to a node * * \param[in,out] instance Clone instance or bundle replica container * \param[in] prefer If not NULL, attempt early assignment to this * node, if still the best choice; otherwise, * perform final assignment * \param[in] max_per_node Assign at most this many instances to one node * * \return Node to which \p instance is assigned */ static const pcmk_node_t * assign_instance(pcmk_resource_t *instance, const pcmk_node_t *prefer, int max_per_node) { pcmk_node_t *chosen = NULL; pcmk__rsc_trace(instance, "Assigning %s (preferring %s)", instance->id, ((prefer == NULL)? "no node" : prefer->priv->name)); if (pcmk_is_set(instance->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return NULL; } ban_unavailable_allowed_nodes(instance, max_per_node); // Failed early assignments are reversible (stop_if_fail=false) chosen = instance->priv->cmds->assign(instance, prefer, (prefer == NULL)); increment_parent_count(instance, chosen); return chosen; } /*! * \internal * \brief Try to assign an instance to its current node early * * \param[in] rsc Clone or bundle being assigned (for logs only) * \param[in] instance Clone instance or bundle replica container * \param[in] current Instance's current node * \param[in] max_per_node Maximum number of instances per node * \param[in] available Number of instances still available for assignment * * \return \c true if \p instance was successfully assigned to its current node, * or \c false otherwise */ static bool assign_instance_early(const pcmk_resource_t *rsc, pcmk_resource_t *instance, const pcmk_node_t *current, int max_per_node, int available) { const pcmk_node_t *chosen = NULL; int reserved = 0; pcmk_resource_t *parent = instance->priv->parent; GHashTable *allowed_orig = NULL; GHashTable *allowed_orig_parent = parent->priv->allowed_nodes; const pcmk_node_t *allowed_node = NULL; pcmk__rsc_trace(instance, "Trying to assign %s to its current node %s", instance->id, pcmk__node_name(current)); allowed_node = g_hash_table_lookup(instance->priv->allowed_nodes, current->priv->id); if (!pcmk__node_available(allowed_node, true, false)) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); return false; } /* On each iteration, if instance gets assigned to a node other than its * current one, we reserve one instance for the chosen node, unassign * instance, restore instance's original node tables, and try again. This * way, instances are proportionally assigned to nodes based on preferences, * but shuffling of specific instances is minimized. If a node will be * assigned instances at all, it preferentially receives instances that are * currently active there. * * parent->private->allowed_nodes tracks the number of instances assigned to * each node. If a node already has max_per_node instances assigned, * ban_unavailable_allowed_nodes() marks it as unavailable. * * In the end, we restore the original parent->private->allowed_nodes to * undo the changes to counts during tentative assignments. If we * successfully assigned an instance to its current node, we increment that * node's counter. */ // Back up the allowed node tables of instance and its children recursively pcmk__copy_node_tables(instance, &allowed_orig); // Update instances-per-node counts in a scratch table parent->priv->allowed_nodes = pcmk__copy_node_table(allowed_orig_parent); while (reserved < available) { chosen = assign_instance(instance, current, max_per_node); if (pcmk__same_node(chosen, current)) { // Successfully assigned to current node break; } // Assignment updates scores, so restore to original state pcmk__rsc_debug(instance, "Rolling back node scores for %s", instance->id); pcmk__restore_node_tables(instance, allowed_orig); if (chosen == NULL) { // Assignment failed, so give up pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); pcmk__set_rsc_flags(instance, pcmk__rsc_unassigned); break; } // We prefer more strongly to assign an instance to the chosen node pcmk__rsc_debug(instance, "Not assigning %s to current node %s: %s is better", instance->id, pcmk__node_name(current), pcmk__node_name(chosen)); // Reserve one instance for the chosen node and try again if (++reserved >= available) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: " "other assignments are more important", instance->id, pcmk__node_name(current)); } else { pcmk__rsc_debug(instance, "Reserved an instance of %s for %s. Retrying " "assignment of %s to %s", rsc->id, pcmk__node_name(chosen), instance->id, pcmk__node_name(current)); } // Clear this assignment (frees chosen); leave instance counts in parent pcmk__unassign_resource(instance); chosen = NULL; } g_hash_table_destroy(allowed_orig); // Restore original instances-per-node counts g_hash_table_destroy(parent->priv->allowed_nodes); parent->priv->allowed_nodes = allowed_orig_parent; if (chosen == NULL) { // Couldn't assign instance to current node return false; } pcmk__rsc_trace(instance, "Assigned %s to current node %s", instance->id, pcmk__node_name(current)); increment_parent_count(instance, chosen); return true; } /*! * \internal * \brief Reset the node counts of a resource's allowed nodes to zero * * \param[in,out] rsc Resource to reset * * \return Number of nodes that are available to run resources */ static unsigned int reset_allowed_node_counts(pcmk_resource_t *rsc) { unsigned int available_nodes = 0; pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->assign->count = 0; if (pcmk__node_available(node, false, false)) { available_nodes++; } } return available_nodes; } /*! * \internal * \brief Check whether an instance has a preferred node * * \param[in] instance Clone instance or bundle replica container * \param[in] optimal_per_node Optimal number of instances per node * * \return Instance's current node if still available, otherwise NULL */ static const pcmk_node_t * preferred_node(const pcmk_resource_t *instance, int optimal_per_node) { const pcmk_node_t *node = NULL; const pcmk_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned if ((instance->priv->active_nodes == NULL) || !pcmk_is_set(instance->flags, pcmk__rsc_unassigned) || pcmk_is_set(instance->flags, pcmk__rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pcmk__current_node(instance); if (!pcmk__node_available(node, true, false)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early (unavailable)", instance->id, pcmk__node_name(node)); return NULL; } // Check whether node already has optimal number of instances assigned parent_node = pcmk__top_allowed_node(instance, node); if ((parent_node != NULL) && (parent_node->assign->count >= optimal_per_node)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pcmk__node_name(node)); return NULL; } return node; } /*! * \internal * \brief Assign collective instances to nodes * * \param[in,out] collective Clone or bundle resource being assigned * \param[in,out] instances List of clone instances or bundle containers * \param[in] max_total Maximum instances to assign in total * \param[in] max_per_node Maximum instances to assign to any one node */ void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node) { // Reuse node count to track number of assigned instances unsigned int available_nodes = reset_allowed_node_counts(collective); int optimal_per_node = 0; int assigned = 0; GList *iter = NULL; pcmk_resource_t *instance = NULL; const pcmk_node_t *current = NULL; if (available_nodes > 0) { optimal_per_node = max_total / available_nodes; } if (optimal_per_node < 1) { optimal_per_node = 1; } pcmk__rsc_debug(collective, "Assigning up to %d %s instance%s to up to %u node%s " "(at most %d per host, %d optimal)", max_total, collective->id, pcmk__plural_s(max_total), available_nodes, pcmk__plural_s(available_nodes), max_per_node, optimal_per_node); // Assign as many instances as possible to their current location for (iter = instances; (iter != NULL) && (assigned < max_total); iter = iter->next) { int available = max_total - assigned; instance = iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } current = preferred_node(instance, optimal_per_node); if ((current != NULL) && assign_instance_early(collective, instance, current, max_per_node, available)) { assigned++; } } pcmk__rsc_trace(collective, "Assigned %d of %d instance%s to current node", assigned, max_total, pcmk__plural_s(max_total)); for (iter = instances; iter != NULL; iter = iter->next) { instance = (pcmk_resource_t *) iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } if (instance->priv->active_nodes != NULL) { current = pcmk__current_node(instance); if (pcmk__top_allowed_node(instance, current) == NULL) { const char *unmanaged = ""; if (!pcmk_is_set(instance->flags, pcmk__rsc_managed)) { unmanaged = "Unmanaged resource "; } crm_notice("%s%s is running on %s which is no longer allowed", unmanaged, instance->id, pcmk__node_name(current)); } } if (assigned >= max_total) { pcmk__rsc_debug(collective, "Not assigning %s because maximum %d instances " "already assigned", instance->id, max_total); resource_location(instance, NULL, -PCMK_SCORE_INFINITY, "collective_limit_reached", collective->priv->scheduler); } else if (assign_instance(instance, NULL, max_per_node) != NULL) { assigned++; } } pcmk__rsc_debug(collective, "Assigned %d of %d possible instance%s of %s", assigned, max_total, pcmk__plural_s(max_total), collective->id); } enum instance_state { instance_starting = (1 << 0), instance_stopping = (1 << 1), /* This indicates that some instance is restarting. It's not the same as * instance_starting|instance_stopping, which would indicate that some * instance is starting, and some instance (not necessarily the same one) is * stopping. */ instance_restarting = (1 << 2), instance_active = (1 << 3), instance_all = instance_starting|instance_stopping |instance_restarting|instance_active, }; /*! * \internal * \brief Check whether an instance is active, starting, and/or stopping * * \param[in] instance Clone instance or bundle replica container * \param[in,out] state Whether any instance is starting, stopping, etc. */ static void check_instance_state(const pcmk_resource_t *instance, uint32_t *state) { const GList *iter = NULL; uint32_t instance_state = 0; // State of just this instance // No need to check further if all conditions have already been detected if (pcmk_all_flags_set(*state, instance_all)) { return; } // If instance is a collective (a cloned group), check its children instead if (instance->priv->variant > pcmk__rsc_variant_primitive) { for (iter = instance->priv->children; (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); iter = iter->next) { check_instance_state((const pcmk_resource_t *) iter->data, state); } return; } // If we get here, instance is a primitive if (instance->priv->active_nodes != NULL) { instance_state |= instance_active; } // Check each of the instance's actions for runnable start or stop for (iter = instance->priv->actions; (iter != NULL) && !pcmk_all_flags_set(instance_state, instance_starting |instance_stopping); iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; const bool optional = pcmk_is_set(action->flags, pcmk__action_optional); if (pcmk__str_eq(PCMK_ACTION_START, action->task, pcmk__str_none)) { if (!optional && pcmk_is_set(action->flags, pcmk__action_runnable)) { pcmk__rsc_trace(instance, "Instance is starting due to %s", action->uuid); instance_state |= instance_starting; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } else if (pcmk__str_eq(PCMK_ACTION_STOP, action->task, pcmk__str_none)) { /* Only stop actions can be pseudo-actions for primitives. That * indicates that the node they are on is being fenced, so the stop * is implied rather than actually executed. */ if (!optional && pcmk_any_flags_set(action->flags, pcmk__action_pseudo |pcmk__action_runnable)) { pcmk__rsc_trace(instance, "Instance is stopping due to %s", action->uuid); instance_state |= instance_stopping; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } } if (pcmk_all_flags_set(instance_state, instance_starting|instance_stopping)) { instance_state |= instance_restarting; } *state |= instance_state; } /*! * \internal * \brief Create actions for collective resource instances * * \param[in,out] collective Clone or bundle resource to create actions for * \param[in,out] instances List of clone instances or bundle containers */ void pcmk__create_instance_actions(pcmk_resource_t *collective, GList *instances) { uint32_t state = 0; pcmk_action_t *stop = NULL; pcmk_action_t *stopped = NULL; pcmk_action_t *start = NULL; pcmk_action_t *started = NULL; pcmk__rsc_trace(collective, "Creating collective instance actions for %s", collective->id); // Create actions for each instance appropriate to its variant for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->priv->cmds->create_actions(instance); check_instance_state(instance, &state); } // Create pseudo-actions for rsc start and started start = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_START, !pcmk_is_set(state, instance_starting), true); started = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_RUNNING, !pcmk_is_set(state, instance_starting), false); started->priority = PCMK_SCORE_INFINITY; if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pcmk__set_action_flags(started, pcmk__action_runnable); } // Create pseudo-actions for rsc stop and stopped stop = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOP, !pcmk_is_set(state, instance_stopping), true); stopped = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOPPED, !pcmk_is_set(state, instance_stopping), true); stopped->priority = PCMK_SCORE_INFINITY; if (!pcmk_is_set(state, instance_restarting)) { pcmk__set_action_flags(stop, pcmk__action_migratable); } if (pcmk__is_clone(collective)) { pe__create_clone_notif_pseudo_ops(collective, start, started, stop, stopped); } } /*! * \internal * \brief Get a list of clone instances or bundle replica containers * * \param[in] rsc Clone or bundle resource * * \return Clone instances if \p rsc is a clone, or a newly created list of * \p rsc's replica containers if \p rsc is a bundle * \note The caller must call free_instance_list() on the result when the list * is no longer needed. */ static inline GList * get_instance_list(const pcmk_resource_t *rsc) { if (pcmk__is_bundle(rsc)) { return pe__bundle_containers(rsc); } else { return rsc->priv->children; } } /*! * \internal * \brief Free any memory created by get_instance_list() * * \param[in] rsc Clone or bundle resource passed to get_instance_list() * \param[in,out] list Return value of get_instance_list() for \p rsc */ static inline void free_instance_list(const pcmk_resource_t *rsc, GList *list) { if (list != rsc->priv->children) { g_list_free(list); } } /*! * \internal * \brief Check whether an instance is compatible with a role and node * * \param[in] instance Clone instance or bundle replica container * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return true if \p instance is compatible with \p node and \p role, * otherwise false */ bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current) { pcmk_node_t *instance_node = NULL; CRM_CHECK((instance != NULL) && (node != NULL), return false); if ((role != pcmk_role_unknown) && (role != instance->priv->fns->state(instance, current))) { pcmk__rsc_trace(instance, "%s is not a compatible instance (role is not %s)", instance->id, pcmk_role_text(role)); return false; } if (!is_set_recursive(instance, pcmk__rsc_blocked, true)) { uint32_t target = pcmk__rsc_node_assigned; if (current) { target = pcmk__rsc_node_current; } // We only want instances that haven't failed instance_node = instance->priv->fns->location(instance, NULL, target); } if (instance_node == NULL) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(not assigned to a node)", instance->id); return false; } if (!pcmk__same_node(instance_node, node)) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(assigned to %s not %s)", instance->id, pcmk__node_name(instance_node), pcmk__node_name(node)); return false; } return true; } #define display_role(r) \ (((r) == pcmk_role_unknown)? "matching" : pcmk_role_text(r)) /*! * \internal * \brief Find an instance that matches a given resource by node and role * * \param[in] match_rsc Resource that instance must match (for logging only) * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return \p rsc instance matching \p node and \p role if any, otherwise NULL */ static pcmk_resource_t * find_compatible_instance_on_node(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, const pcmk_node_t *node, enum rsc_role_e role, bool current) { GList *instances = NULL; instances = get_instance_list(rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; if (pcmk__instance_matches(instance, node, role, current)) { pcmk__rsc_trace(match_rsc, "Found %s %s instance %s compatible with %s on %s", display_role(role), rsc->id, instance->id, match_rsc->id, pcmk__node_name(node)); free_instance_list(rsc, instances); // Only frees list, not contents return instance; } } free_instance_list(rsc, instances); pcmk__rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s", display_role(role), rsc->id, match_rsc->id, pcmk__node_name(node)); return NULL; } /*! * \internal * \brief Find a clone instance or bundle container compatible with a resource * * \param[in] match_rsc Resource that instance must match * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return Compatible (by \p role and \p match_rsc location) instance of \p rsc * if any, otherwise NULL */ pcmk_resource_t * pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current) { pcmk_resource_t *instance = NULL; GList *nodes = NULL; const pcmk_node_t *node = NULL; GHashTable *allowed_nodes = match_rsc->priv->allowed_nodes; uint32_t target = pcmk__rsc_node_assigned; if (current) { target = pcmk__rsc_node_current; } // If match_rsc has a node, check only that node node = match_rsc->priv->fns->location(match_rsc, NULL, target); if (node != NULL) { return find_compatible_instance_on_node(match_rsc, rsc, node, role, current); } // Otherwise check for an instance matching any of match_rsc's allowed nodes nodes = pcmk__sort_nodes(g_hash_table_get_values(allowed_nodes), NULL); for (GList *iter = nodes; (iter != NULL) && (instance == NULL); iter = iter->next) { instance = find_compatible_instance_on_node(match_rsc, rsc, (pcmk_node_t *) iter->data, role, current); } if (instance == NULL) { pcmk__rsc_debug(rsc, "No %s instance found compatible with %s", rsc->id, match_rsc->id); } g_list_free(nodes); return instance; } /*! * \internal * \brief Unassign an instance if mandatory ordering has no interleave match * * \param[in] first 'First' action in an ordering * \param[in] then 'Then' action in an ordering * \param[in,out] then_instance 'Then' instance that has no interleave match * \param[in] type Group of enum pcmk__action_relation_flags * \param[in] current If true, "then" action is stopped or demoted * * \return true if \p then_instance was unassigned, otherwise false */ static bool unassign_if_mandatory(const pcmk_action_t *first, const pcmk_action_t *then, pcmk_resource_t *then_instance, uint32_t type, bool current) { // Allow "then" instance to go down even without an interleave match if (current) { pcmk__rsc_trace(then->rsc, "%s has no instance to order before stopping " "or demoting %s", first->rsc->id, then_instance->id); /* If the "first" action must be runnable, but there is no "first" * instance, the "then" instance must not be allowed to come up. */ } else if (pcmk_any_flags_set(type, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then)) { pcmk__rsc_info(then->rsc, "Inhibiting %s from being active " "because there is no %s instance to interleave", then_instance->id, first->rsc->id); return pcmk__assign_resource(then_instance, NULL, true, true); } return false; } /*! * \internal * \brief Find first matching action for a clone instance or bundle container * * \param[in] action Action in an interleaved ordering * \param[in] instance Clone instance or bundle container being interleaved * \param[in] action_name Action to look for * \param[in] node If not NULL, require action to be on this node * \param[in] for_first If true, \p instance is the 'first' resource in the * ordering, otherwise it is the 'then' resource * * \return First action for \p instance (or in some cases if \p instance is a * bundle container, its containerized resource) that matches * \p action_name and \p node if any, otherwise NULL */ static pcmk_action_t * find_instance_action(const pcmk_action_t *action, const pcmk_resource_t *instance, const char *action_name, const pcmk_node_t *node, bool for_first) { const pcmk_resource_t *rsc = NULL; pcmk_action_t *matching_action = NULL; /* If instance is a bundle container, sometimes we should interleave the * action for the container itself, and sometimes for the containerized * resource. * * For example, given "start bundle A then bundle B", B likely requires the * service inside A's container to be active, rather than just the * container, so we should interleave the action for A's containerized * resource. On the other hand, it's possible B's container itself requires * something from A, so we should interleave the action for B's container. * * Essentially, for 'first', we should use the containerized resource for * everything except stop, and for 'then', we should use the container for * everything except promote and demote (which can only be performed on the * containerized resource). */ if ((for_first && !pcmk__str_any_of(action->task, PCMK_ACTION_STOP, PCMK_ACTION_STOPPED, NULL)) || (!for_first && pcmk__str_any_of(action->task, PCMK_ACTION_PROMOTE, PCMK_ACTION_PROMOTED, PCMK_ACTION_DEMOTE, PCMK_ACTION_DEMOTED, NULL))) { rsc = pe__get_rsc_in_container(instance); } if (rsc == NULL) { rsc = instance; // No containerized resource, use instance itself } else { node = NULL; // Containerized actions are on bundle-created guest } matching_action = find_first_action(rsc->priv->actions, NULL, action_name, node); if (matching_action != NULL) { return matching_action; } if (pcmk_is_set(instance->flags, pcmk__rsc_removed) || pcmk__is_down_action(action_name)) { crm_trace("No %s action found for %s%s", action_name, pcmk_is_set(instance->flags, pcmk__rsc_removed)? "orphan " : "", instance->id); } else { crm_err("No %s action found for %s to interleave (bug?)", action_name, instance->id); } return NULL; } /*! * \internal * \brief Get the original action name of a bundle or clone action * * Given an action for a bundle or clone, get the original action name, * mapping notify to the action being notified, and if the instances are * primitives, mapping completion actions to the action that was completed * (for example, stopped to stop). * * \param[in] action Clone or bundle action to check * * \return Original action name for \p action */ static const char * orig_action_name(const pcmk_action_t *action) { // Any instance will do const pcmk_resource_t *instance = action->rsc->priv->children->data; char *action_type = NULL; const char *action_name = action->task; enum pcmk__action_type orig_task = pcmk__action_unspecified; if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { // action->uuid is RSC_(confirmed-){pre,post}_notify_ACTION_INTERVAL CRM_CHECK(parse_op_key(action->uuid, NULL, &action_type, NULL), return pcmk__action_text(pcmk__action_unspecified)); action_name = strstr(action_type, "_notify_"); CRM_CHECK(action_name != NULL, return pcmk__action_text(pcmk__action_unspecified)); action_name += strlen("_notify_"); } orig_task = get_complex_task(instance, action_name); free(action_type); return pcmk__action_text(orig_task); } /*! * \internal * \brief Update two interleaved actions according to an ordering between them * * Given information about an ordering of two interleaved actions, update the * actions' flags (and runnable_before members if appropriate) as appropriate * for the ordering. Effects may cascade to other orderings involving the * actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_interleaved_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t filter, uint32_t type) { GList *instances = NULL; uint32_t changed = pcmk__updated_none; const char *orig_first_task = orig_action_name(first); // Stops and demotes must be interleaved with instance on current node - bool current = pcmk__ends_with(first->uuid, "_" PCMK_ACTION_STOPPED "_0") - || pcmk__ends_with(first->uuid, - "_" PCMK_ACTION_DEMOTED "_0"); + bool current = (first->uuid != NULL) + && (g_str_has_suffix(first->uuid, + "_" PCMK_ACTION_STOPPED "_0") + || g_str_has_suffix(first->uuid, + "_" PCMK_ACTION_DEMOTED "_0")); // Update the specified actions for each "then" instance individually instances = get_instance_list(then->rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *first_instance = NULL; pcmk_resource_t *then_instance = iter->data; pcmk_action_t *first_action = NULL; pcmk_action_t *then_action = NULL; // Find a "first" instance to interleave with this "then" instance first_instance = pcmk__find_compatible_instance(then_instance, first->rsc, pcmk_role_unknown, current); if (first_instance == NULL) { // No instance can be interleaved if (unassign_if_mandatory(first, then, then_instance, type, current)) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); } continue; } first_action = find_instance_action(first, first_instance, orig_first_task, node, true); if (first_action == NULL) { continue; } then_action = find_instance_action(then, then_instance, then->task, node, false); if (then_action == NULL) { continue; } if (order_actions(first_action, then_action, type)) { pcmk__set_updated_flags(changed, first, pcmk__updated_first|pcmk__updated_then); } changed |= then_instance->priv->cmds->update_ordered_actions( first_action, then_action, node, first_instance->priv->cmds->action_flags(first_action, node), filter, type, then->rsc->priv->scheduler); } free_instance_list(then->rsc, instances); return changed; } /*! * \internal * \brief Check whether two actions in an ordering can be interleaved * * \param[in] first 'First' action in the ordering * \param[in] then 'Then' action in the ordering * * \return true if \p first and \p then can be interleaved, otherwise false */ static bool can_interleave_actions(const pcmk_action_t *first, const pcmk_action_t *then) { bool interleave = false; pcmk_resource_t *rsc = NULL; if ((first->rsc == NULL) || (then->rsc == NULL)) { crm_trace("Not interleaving %s with %s: not resource actions", first->uuid, then->uuid); return false; } if (first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s: same resource", first->uuid, then->uuid); return false; } if ((first->rsc->priv->variant < pcmk__rsc_variant_clone) || (then->rsc->priv->variant < pcmk__rsc_variant_clone)) { crm_trace("Not interleaving %s with %s: not clones or bundles", first->uuid, then->uuid); return false; } - if (pcmk__ends_with(then->uuid, "_stop_0") - || pcmk__ends_with(then->uuid, "_demote_0")) { + if (g_str_has_suffix(then->uuid, "_stop_0") + || g_str_has_suffix(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave = crm_is_true(g_hash_table_lookup(rsc->priv->meta, PCMK_META_INTERLEAVE)); pcmk__rsc_trace(rsc, "'%s then %s' will %sbe interleaved (based on %s)", first->uuid, then->uuid, (interleave? "" : "not "), rsc->id); return interleave; } /*! * \internal * \brief Update non-interleaved instance actions according to an ordering * * Given information about an ordering of two non-interleaved actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] instance Clone instance or bundle container * \param[in,out] first "First" action in ordering * \param[in] then "Then" action in ordering (for \p instance's parent) * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_noninterleaved_actions(pcmk_resource_t *instance, pcmk_action_t *first, const pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type) { pcmk_action_t *instance_action = NULL; pcmk_scheduler_t *scheduler = instance->priv->scheduler; uint32_t instance_flags = 0; uint32_t changed = pcmk__updated_none; // Check whether instance has an equivalent of "then" action instance_action = find_first_action(instance->priv->actions, NULL, then->task, node); if (instance_action == NULL) { return changed; } // Check whether action is runnable instance_flags = instance->priv->cmds->action_flags(instance_action, node); if (!pcmk_is_set(instance_flags, pcmk__action_runnable)) { return changed; } // If so, update actions for the instance changed = instance->priv->cmds->update_ordered_actions(first, instance_action, node, flags, filter, type, scheduler); // Propagate any changes to later actions if (pcmk_is_set(changed, pcmk__updated_then)) { for (GList *after_iter = instance_action->actions_after; after_iter != NULL; after_iter = after_iter->next) { pcmk__related_action_t *after = after_iter->data; pcmk__update_action_for_orderings(after->action, scheduler); } } return changed; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two clone or bundle actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { pcmk__assert((first != NULL) && (then != NULL) && (scheduler != NULL)); if (then->rsc == NULL) { return pcmk__updated_none; } else if (can_interleave_actions(first, then)) { return update_interleaved_actions(first, then, node, filter, type); } else { uint32_t changed = pcmk__updated_none; GList *instances = get_instance_list(then->rsc); // Update actions for the clone or bundle resource itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the 'then' clone instances or bundle containers individually for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = iter->data; changed |= update_noninterleaved_actions(instance, first, then, node, flags, filter, type); } free_instance_list(then->rsc, instances); return changed; } } #define pe__clear_action_summary_flags(flags, action, flag) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action summary", action->rsc->id, \ flags, flag, #flag); \ } while (0) /*! * \internal * \brief Return action flags for a given clone or bundle action * * \param[in,out] action Action for a clone or bundle * \param[in] instances Clone instances or bundle containers * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node) { bool any_runnable = false; const char *action_name = orig_action_name(action); // Set original assumptions (optional and runnable may be cleared below) uint32_t flags = pcmk__action_optional |pcmk__action_runnable |pcmk__action_pseudo; for (const GList *iter = instances; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; const pcmk_node_t *instance_node = NULL; pcmk_action_t *instance_action = NULL; uint32_t instance_flags; // Node is relevant only to primitive instances if (pcmk__is_primitive(instance)) { instance_node = node; } instance_action = find_first_action(instance->priv->actions, NULL, action_name, instance_node); if (instance_action == NULL) { pcmk__rsc_trace(action->rsc, "%s has no %s action on %s", instance->id, action_name, pcmk__node_name(node)); continue; } pcmk__rsc_trace(action->rsc, "%s has %s for %s on %s", instance->id, instance_action->uuid, action_name, pcmk__node_name(node)); instance_flags = instance->priv->cmds->action_flags(instance_action, node); // If any instance action is mandatory, so is the collective action if (pcmk_is_set(flags, pcmk__action_optional) && !pcmk_is_set(instance_flags, pcmk__action_optional)) { pcmk__rsc_trace(instance, "%s is mandatory because %s is", action->uuid, instance_action->uuid); pe__clear_action_summary_flags(flags, action, pcmk__action_optional); pcmk__clear_action_flags(action, pcmk__action_optional); } // If any instance action is runnable, so is the collective action if (pcmk_is_set(instance_flags, pcmk__action_runnable)) { any_runnable = true; } } if (!any_runnable) { pcmk__rsc_trace(action->rsc, "%s is not runnable because no instance can run %s", action->uuid, action_name); pe__clear_action_summary_flags(flags, action, pcmk__action_runnable); if (node == NULL) { pcmk__clear_action_flags(action, pcmk__action_runnable); } } return flags; } diff --git a/lib/pacemaker/pcmk_sched_probes.c b/lib/pacemaker/pcmk_sched_probes.c index bda90cee19..a84e382572 100644 --- a/lib/pacemaker/pcmk_sched_probes.c +++ b/lib/pacemaker/pcmk_sched_probes.c @@ -1,905 +1,905 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Add the expected result to a newly created probe * * \param[in,out] probe Probe action to add expected result to * \param[in] rsc Resource that probe is for * \param[in] node Node that probe will run on */ static void add_expected_result(pcmk_action_t *probe, const pcmk_resource_t *rsc, const pcmk_node_t *node) { // Check whether resource is currently active on node pcmk_node_t *running = pe_find_node_id(rsc->priv->active_nodes, node->priv->id); // The expected result is what we think the resource's current state is if (running == NULL) { pe__add_action_expected_result(probe, CRM_EX_NOT_RUNNING); } else if (rsc->priv->orig_role == pcmk_role_promoted) { pe__add_action_expected_result(probe, CRM_EX_PROMOTED); } } /*! * \internal * \brief Create any needed robes on a node for a list of resources * * \param[in,out] rscs List of resources to create probes for * \param[in,out] node Node to create probes on * * \return true if any probe was created, otherwise false */ bool pcmk__probe_resource_list(GList *rscs, pcmk_node_t *node) { bool any_created = false; for (GList *iter = rscs; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (rsc->priv->cmds->create_probe(rsc, node)) { any_created = true; } } return any_created; } /*! * \internal * \brief Order one resource's start after another's start-up probe * * \param[in,out] rsc1 Resource that might get start-up probe * \param[in] rsc2 Resource that might be started */ static void probe_then_start(pcmk_resource_t *rsc1, pcmk_resource_t *rsc2) { const pcmk_node_t *rsc1_node = rsc1->priv->assigned_node; if ((rsc1_node != NULL) && (g_hash_table_lookup(rsc1->priv->probed_nodes, rsc1_node->priv->id) == NULL)) { pcmk__new_ordering(rsc1, pcmk__op_key(rsc1->id, PCMK_ACTION_MONITOR, 0), NULL, rsc2, pcmk__op_key(rsc2->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, rsc1->priv->scheduler); } } /*! * \internal * \brief Check whether a guest resource will stop * * \param[in] node Guest node to check * * \return true if guest resource will likely stop, otherwise false */ static bool guest_resource_will_stop(const pcmk_node_t *node) { const pcmk_resource_t *guest_rsc = NULL; const pcmk_node_t *guest_node = NULL; guest_rsc = node->priv->remote->priv->launcher; guest_node = guest_rsc->priv->assigned_node; /* Ideally, we'd check whether the guest has a required stop, but that * information doesn't exist yet, so approximate it ... */ return pcmk_is_set(node->priv->flags, pcmk__node_remote_reset) || node->details->unclean || pcmk_is_set(guest_rsc->flags, pcmk__rsc_failed) || (guest_rsc->priv->next_role == pcmk_role_stopped) // Guest is moving || ((guest_rsc->priv->orig_role > pcmk_role_stopped) && (guest_node != NULL) && pcmk__find_node_in_list(guest_rsc->priv->active_nodes, guest_node->priv->name) == NULL); } /*! * \internal * \brief Create a probe action for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return Newly created probe action */ static pcmk_action_t * probe_action(pcmk_resource_t *rsc, pcmk_node_t *node) { pcmk_action_t *probe = NULL; char *key = pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0); crm_debug("Scheduling probe of %s %s on %s", pcmk_role_text(rsc->priv->orig_role), rsc->id, pcmk__node_name(node)); probe = custom_action(rsc, key, PCMK_ACTION_MONITOR, node, FALSE, rsc->priv->scheduler); pcmk__clear_action_flags(probe, pcmk__action_optional); pcmk__order_vs_unfence(rsc, node, probe, pcmk__ar_ordered); add_expected_result(probe, rsc, node); return probe; } /*! * \internal * \brief Create probes for a resource on a node, if needed * * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__probe_rsc_on_node(pcmk_resource_t *rsc, pcmk_node_t *node) { uint32_t flags = pcmk__ar_ordered; pcmk_action_t *probe = NULL; pcmk_node_t *allowed = NULL; pcmk_resource_t *top = uber_parent(rsc); const char *reason = NULL; pcmk__assert((rsc != NULL) && (node != NULL)); if (!pcmk_is_set(rsc->priv->scheduler->flags, pcmk__sched_probe_resources)) { reason = "start-up probes are disabled"; goto no_probe; } if (pcmk__is_pacemaker_remote_node(node)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { reason = "Pacemaker Remote nodes cannot run stonith agents"; goto no_probe; } else if (pcmk__is_guest_or_bundle_node(node) && pe__resource_contains_guest_node(rsc->priv->scheduler, rsc)) { reason = "guest nodes cannot run resources containing guest nodes"; goto no_probe; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { reason = "Pacemaker Remote nodes cannot host remote connections"; goto no_probe; } } // If this is a collective resource, probes are created for its children if (rsc->priv->children != NULL) { return pcmk__probe_resource_list(rsc->priv->children, node); } if ((rsc->priv->launcher != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { reason = "resource is inside a container"; goto no_probe; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { reason = "resource is orphaned"; goto no_probe; } else if (g_hash_table_lookup(rsc->priv->probed_nodes, node->priv->id) != NULL) { reason = "resource state is already known"; goto no_probe; } allowed = g_hash_table_lookup(rsc->priv->allowed_nodes, node->priv->id); if (pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes) || pcmk_is_set(top->flags, pcmk__rsc_exclusive_probes)) { // Exclusive discovery is enabled ... if (allowed == NULL) { // ... but this node is not allowed to run the resource reason = "resource has exclusive discovery but is not allowed " "on node"; goto no_probe; } else if (allowed->assign->probe_mode != pcmk__probe_exclusive) { // ... but no constraint marks this node for discovery of resource reason = "resource has exclusive discovery but is not enabled " "on node"; goto no_probe; } } if (allowed == NULL) { allowed = node; } if (allowed->assign->probe_mode == pcmk__probe_never) { reason = "node has discovery disabled"; goto no_probe; } if (pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *guest = node->priv->remote->priv->launcher; if (guest->priv->orig_role == pcmk_role_stopped) { // The guest is stopped, so we know no resource is active there reason = "node's guest is stopped"; probe_then_start(guest, top); goto no_probe; } else if (guest_resource_will_stop(node)) { reason = "node's guest will stop"; // Order resource start after guest stop (in case it's restarting) pcmk__new_ordering(guest, pcmk__op_key(guest->id, PCMK_ACTION_STOP, 0), NULL, top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, rsc->priv->scheduler); goto no_probe; } } // We've eliminated all cases where a probe is not needed, so now it is probe = probe_action(rsc, node); /* Below, we will order the probe relative to start or reload. If this is a * clone instance, the start or reload is for the entire clone rather than * just the instance. Otherwise, the start or reload is for the resource * itself. */ if (!pcmk__is_clone(top)) { top = rsc; } /* Prevent a start if the resource can't be probed, but don't cause the * resource or entire clone to stop if already active. */ if (!pcmk_is_set(probe->flags, pcmk__action_runnable) && (top->priv->active_nodes == NULL)) { pcmk__set_relation_flags(flags, pcmk__ar_unrunnable_first_blocks); } // Start or reload after probing the resource pcmk__new_ordering(rsc, NULL, probe, top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL, flags, rsc->priv->scheduler); pcmk__new_ordering(rsc, NULL, probe, top, reload_key(rsc), NULL, pcmk__ar_ordered, rsc->priv->scheduler); return true; no_probe: pcmk__rsc_trace(rsc, "Skipping probe for %s on %s because %s", rsc->id, node->priv->id, reason); return false; } /*! * \internal * \brief Check whether a probe should be ordered before another action * * \param[in] probe Probe action to check * \param[in] then Other action to check * * \return true if \p probe should be ordered before \p then, otherwise false */ static bool probe_needed_before_action(const pcmk_action_t *probe, const pcmk_action_t *then) { // Probes on a node are performed after unfencing it, not before if (pcmk__str_eq(then->task, PCMK_ACTION_STONITH, pcmk__str_none) && pcmk__same_node(probe->node, then->node)) { const char *op = g_hash_table_lookup(then->meta, PCMK__META_STONITH_ACTION); if (pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) { return false; } } // Probes should be done on a node before shutting it down if (pcmk__str_eq(then->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none) && (probe->node != NULL) && (then->node != NULL) && !pcmk__same_node(probe->node, then->node)) { return false; } // Otherwise probes should always be done before any other action return true; } /*! * \internal * \brief Add implicit "probe then X" orderings for "stop then X" orderings * * If the state of a resource is not known yet, a probe will be scheduled, * expecting a "not running" result. If the probe fails, a stop will not be * scheduled until the next transition. Thus, if there are ordering constraints * like "stop this resource then do something else that's not for the same * resource", add implicit "probe this resource then do something" equivalents * so the relation is upheld until we know whether a stop is needed. * * \param[in,out] scheduler Scheduler data */ static void add_probe_orderings_for_stops(pcmk_scheduler_t *scheduler) { for (GList *iter = scheduler->priv->ordering_constraints; iter != NULL; iter = iter->next) { pcmk__action_relation_t *order = iter->data; uint32_t order_flags = pcmk__ar_ordered; GList *probes = NULL; GList *then_actions = NULL; pcmk_action_t *first = NULL; pcmk_action_t *then = NULL; // Skip disabled orderings if (order->flags == pcmk__ar_none) { continue; } // Skip non-resource orderings, and orderings for the same resource if ((order->rsc1 == NULL) || (order->rsc1 == order->rsc2)) { continue; } // Skip invalid orderings (shouldn't be possible) first = order->action1; then = order->action2; if (((first == NULL) && (order->task1 == NULL)) || ((then == NULL) && (order->task2 == NULL))) { continue; } // Skip orderings for first actions other than stop if ((first != NULL) && !pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) { continue; } else if ((first == NULL) - && !pcmk__ends_with(order->task1, - "_" PCMK_ACTION_STOP "_0")) { + && !g_str_has_suffix(order->task1, + "_" PCMK_ACTION_STOP "_0")) { continue; } /* Do not imply a probe ordering for a resource inside of a stopping * launcher. Otherwise, it might introduce a transition loop, since a * probe could be scheduled after the launcher starts again. */ if ((order->rsc2 != NULL) && (order->rsc1->priv->launcher == order->rsc2)) { if ((then != NULL) && pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) { continue; } else if ((then == NULL) - && pcmk__ends_with(order->task2, - "_" PCMK_ACTION_STOP "_0")) { + && g_str_has_suffix(order->task2, + "_" PCMK_ACTION_STOP "_0")) { continue; } } // Preserve certain order options for future filtering if (pcmk_is_set(order->flags, pcmk__ar_if_first_unmigratable)) { pcmk__set_relation_flags(order_flags, pcmk__ar_if_first_unmigratable); } if (pcmk_is_set(order->flags, pcmk__ar_if_on_same_node)) { pcmk__set_relation_flags(order_flags, pcmk__ar_if_on_same_node); } // Preserve certain order types for future filtering if ((order->flags == pcmk__ar_if_required_on_same_node) || (order->flags == pcmk__ar_if_on_same_node_or_target)) { order_flags = order->flags; } // List all scheduled probes for the first resource probes = pe__resource_actions(order->rsc1, NULL, PCMK_ACTION_MONITOR, FALSE); if (probes == NULL) { // There aren't any continue; } // List all relevant "then" actions if (then != NULL) { then_actions = g_list_prepend(NULL, then); } else if (order->rsc2 != NULL) { then_actions = find_actions(order->rsc2->priv->actions, order->task2, NULL); if (then_actions == NULL) { // There aren't any g_list_free(probes); continue; } } crm_trace("Implying 'probe then' orderings for '%s then %s' " "(id=%d, type=%.6x)", ((first == NULL)? order->task1 : first->uuid), ((then == NULL)? order->task2 : then->uuid), order->id, order->flags); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data; for (GList *then_iter = then_actions; then_iter != NULL; then_iter = then_iter->next) { pcmk_action_t *then = (pcmk_action_t *) then_iter->data; if (probe_needed_before_action(probe, then)) { order_actions(probe, then, order_flags); } } } g_list_free(then_actions); g_list_free(probes); } } /*! * \internal * \brief Add necessary orderings between probe and starts of clone instances * * , in additon to the ordering with the parent resource added upon creating * the probe. * * \param[in,out] probe Probe as 'first' action in an ordering * \param[in,out] after 'then' action wrapper in the ordering */ static void add_start_orderings_for_probe(pcmk_action_t *probe, pcmk__related_action_t *after) { uint32_t flags = pcmk__ar_ordered|pcmk__ar_unrunnable_first_blocks; /* Although the ordering between the probe of the clone instance and the * start of its parent has been added in pcmk__probe_rsc_on_node(), we * avoided enforcing `pcmk__ar_unrunnable_first_blocks` order type for that * as long as any of the clone instances are running to prevent them from * being unexpectedly stopped. * * On the other hand, we still need to prevent any inactive instances from * starting unless the probe is runnable so that we don't risk starting too * many instances before we know the state on all nodes. */ if ((after->action->rsc->priv->variant <= pcmk__rsc_variant_group) || pcmk_is_set(probe->flags, pcmk__action_runnable) // The order type is already enforced for its parent. || pcmk_is_set(after->flags, pcmk__ar_unrunnable_first_blocks) || (pe__const_top_resource(probe->rsc, false) != after->action->rsc) || !pcmk__str_eq(after->action->task, PCMK_ACTION_START, pcmk__str_none)) { return; } crm_trace("Adding probe start orderings for 'unrunnable %s@%s " "then instances of %s@%s'", probe->uuid, pcmk__node_name(probe->node), after->action->uuid, pcmk__node_name(after->action->node)); for (GList *then_iter = after->action->actions_after; then_iter != NULL; then_iter = then_iter->next) { pcmk__related_action_t *then = then_iter->data; if ((then->action->rsc->priv->active_nodes != NULL) || (pe__const_top_resource(then->action->rsc, false) != after->action->rsc) || !pcmk__str_eq(then->action->task, PCMK_ACTION_START, pcmk__str_none)) { continue; } crm_trace("Adding probe start ordering for 'unrunnable %s@%s " "then %s@%s' (type=%#.6x)", probe->uuid, pcmk__node_name(probe->node), then->action->uuid, pcmk__node_name(then->action->node), flags); /* Prevent the instance from starting if the instance can't, but don't * cause any other intances to stop if already active. */ order_actions(probe, then->action, flags); } return; } /*! * \internal * \brief Order probes before restarts and re-promotes * * If a given ordering is a "probe then start" or "probe then promote" ordering, * add an implicit "probe then stop/demote" ordering in case the action is part * of a restart/re-promote, and do the same recursively for all actions ordered * after the "then" action. * * \param[in,out] probe Probe as 'first' action in an ordering * \param[in,out] after 'then' action in the ordering */ static void add_restart_orderings_for_probe(pcmk_action_t *probe, pcmk_action_t *after) { GList *iter = NULL; bool interleave = false; pcmk_resource_t *compatible_rsc = NULL; // Validate that this is a resource probe followed by some action if ((after == NULL) || (probe == NULL) || !pcmk__is_primitive(probe->rsc) || !pcmk__str_eq(probe->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { return; } // Avoid running into any possible loop if (pcmk_is_set(after->flags, pcmk__action_detect_loop)) { return; } pcmk__set_action_flags(after, pcmk__action_detect_loop); crm_trace("Adding probe restart orderings for '%s@%s then %s@%s'", probe->uuid, pcmk__node_name(probe->node), after->uuid, pcmk__node_name(after->node)); /* Add restart orderings if "then" is for a different primitive. * Orderings for collective resources will be added later. */ if (pcmk__is_primitive(after->rsc) && (probe->rsc != after->rsc)) { GList *then_actions = NULL; if (pcmk__str_eq(after->task, PCMK_ACTION_START, pcmk__str_none)) { then_actions = pe__resource_actions(after->rsc, NULL, PCMK_ACTION_STOP, FALSE); } else if (pcmk__str_eq(after->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { then_actions = pe__resource_actions(after->rsc, NULL, PCMK_ACTION_DEMOTE, FALSE); } for (iter = then_actions; iter != NULL; iter = iter->next) { pcmk_action_t *then = (pcmk_action_t *) iter->data; // Skip pseudo-actions (for example, those implied by fencing) if (!pcmk_is_set(then->flags, pcmk__action_pseudo)) { order_actions(probe, then, pcmk__ar_ordered); } } g_list_free(then_actions); } /* Detect whether "then" is an interleaved clone action. For these, we want * to add orderings only for the relevant instance. */ if ((after->rsc != NULL) && (after->rsc->priv->variant > pcmk__rsc_variant_group)) { interleave = crm_is_true(g_hash_table_lookup(after->rsc->priv->meta, PCMK_META_INTERLEAVE)); if (interleave) { compatible_rsc = pcmk__find_compatible_instance(probe->rsc, after->rsc, pcmk_role_unknown, false); } } /* Now recursively do the same for all actions ordered after "then". This * also handles collective resources since the collective action will be * ordered before its individual instances' actions. */ for (iter = after->actions_after; iter != NULL; iter = iter->next) { pcmk__related_action_t *after_wrapper = iter->data; const pcmk_resource_t *chained_rsc = NULL; /* pcmk__ar_first_implies_then is the reason why a required A.start * implies/enforces B.start to be required too, which is the cause of * B.restart/re-promote. * * Not sure about pcmk__ar_first_implies_same_node_then though. It's now * only used for unfencing case, which tends to introduce transition * loops... */ if (!pcmk_is_set(after_wrapper->flags, pcmk__ar_first_implies_then)) { /* The order type between a group/clone and its child such as * B.start-> B_child.start is: * pcmk__ar_then_implies_first_graphed * |pcmk__ar_unrunnable_first_blocks * * Proceed through the ordering chain and build dependencies with * its children. */ if ((after->rsc == NULL) || (after->rsc->priv->variant < pcmk__rsc_variant_group) || (probe->rsc->priv->parent == after->rsc) || (after_wrapper->action->rsc == NULL)) { continue; } chained_rsc = after_wrapper->action->rsc; if ((chained_rsc->priv->variant > pcmk__rsc_variant_group) || (after->rsc != chained_rsc->priv->parent)) { continue; } /* Proceed to the children of a group or a non-interleaved clone. * For an interleaved clone, proceed only to the relevant child. */ if ((after->rsc->priv->variant > pcmk__rsc_variant_group) && interleave && ((compatible_rsc == NULL) || (compatible_rsc != chained_rsc))) { continue; } } crm_trace("Recursively adding probe restart orderings for " "'%s@%s then %s@%s' (type=%#.6x)", after->uuid, pcmk__node_name(after->node), after_wrapper->action->uuid, pcmk__node_name(after_wrapper->action->node), after_wrapper->flags); add_restart_orderings_for_probe(probe, after_wrapper->action); } } /*! * \internal * \brief Clear the tracking flag on all scheduled actions * * \param[in,out] scheduler Scheduler data */ static void clear_actions_tracking_flag(pcmk_scheduler_t *scheduler) { for (GList *iter = scheduler->priv->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; pcmk__clear_action_flags(action, pcmk__action_detect_loop); } } /*! * \internal * \brief Add start and restart orderings for probes scheduled for a resource * * \param[in,out] data Resource whose probes should be ordered * \param[in] user_data Unused */ static void add_start_restart_orderings_for_rsc(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; GList *probes = NULL; // For collective resources, order each instance recursively if (!pcmk__is_primitive(rsc)) { g_list_foreach(rsc->priv->children, add_start_restart_orderings_for_rsc, NULL); return; } // Find all probes for given resource probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE); // Add probe restart orderings for each probe found for (GList *iter = probes; iter != NULL; iter = iter->next) { pcmk_action_t *probe = (pcmk_action_t *) iter->data; for (GList *then_iter = probe->actions_after; then_iter != NULL; then_iter = then_iter->next) { pcmk__related_action_t *then = then_iter->data; add_start_orderings_for_probe(probe, then); add_restart_orderings_for_probe(probe, then->action); clear_actions_tracking_flag(rsc->priv->scheduler); } } g_list_free(probes); } /*! * \internal * \brief Add "A then probe B" orderings for "A then B" orderings * * \param[in,out] scheduler Scheduler data * * \note This function is currently disabled (see next comment). */ static void order_then_probes(pcmk_scheduler_t *scheduler) { #if 0 /* Given an ordering "A then B", we would prefer to wait for A to be started * before probing B. * * For example, if A is a filesystem which B can't even run without, it * would be helpful if the author of B's agent could assume that A is * running before B.monitor will be called. * * However, we can't _only_ probe after A is running, otherwise we wouldn't * detect the state of B if A could not be started. We can't even do an * opportunistic version of this, because B may be moving: * * A.stop -> A.start -> B.probe -> B.stop -> B.start * * and if we add B.stop -> A.stop here, we get a loop: * * A.stop -> A.start -> B.probe -> B.stop -> A.stop * * We could kill the "B.probe -> B.stop" dependency, but that could mean * stopping B "too" soon, because B.start must wait for the probe, and * we don't want to stop B if we can't start it. * * We could add the ordering only if A is an anonymous clone with * clone-max == node-max (since we'll never be moving it). However, we could * still be stopping one instance at the same time as starting another. * * The complexity of checking for allowed conditions combined with the ever * narrowing use case suggests that this code should remain disabled until * someone gets smarter. */ for (GList *iter = scheduler->priv->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; pcmk_action_t *start = NULL; GList *actions = NULL; GList *probes = NULL; actions = pe__resource_actions(rsc, NULL, PCMK_ACTION_START, FALSE); if (actions) { start = actions->data; g_list_free(actions); } if (start == NULL) { crm_debug("No start action for %s", rsc->id); continue; } probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE); for (actions = start->actions_before; actions != NULL; actions = actions->next) { pcmk__related_action_t *before = actions->data; pcmk_action_t *first = before->action; pcmk_resource_t *first_rsc = first->rsc; if (first->required_runnable_before) { for (GList *clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) { before = clone_actions->data; crm_trace("Testing '%s then %s' for %s", first->uuid, before->action->uuid, start->uuid); pcmk__assert(before->action->rsc != NULL); first_rsc = before->action->rsc; break; } } else if (!pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) { crm_trace("Not a start op %s for %s", first->uuid, start->uuid); } if (first_rsc == NULL) { continue; } else if (pe__const_top_resource(first_rsc, false) == pe__const_top_resource(start->rsc, false)) { crm_trace("Same parent %s for %s", first_rsc->id, start->uuid); continue; } else if (!pcmk__is_clone(pe__const_top_resource(first_rsc, false))) { crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid); continue; } crm_debug("Applying %s before %s", first->uuid, start->uuid); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data; crm_debug("Ordering %s before %s", first->uuid, probe->uuid); order_actions(first, probe, pcmk__ar_ordered); } } } #endif } void pcmk__order_probes(pcmk_scheduler_t *scheduler) { // Add orderings for "probe then X" g_list_foreach(scheduler->priv->resources, add_start_restart_orderings_for_rsc, NULL); add_probe_orderings_for_stops(scheduler); order_then_probes(scheduler); } /*! * \internal * \brief Schedule any probes needed * * \param[in,out] scheduler Scheduler data * * \note This may also schedule fencing of failed remote nodes. */ void pcmk__schedule_probes(pcmk_scheduler_t *scheduler) { // Schedule probes on each node in the cluster as needed for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; if (!node->details->online) { // Don't probe offline nodes if (pcmk__is_failed_remote_node(node)) { pe_fence_node(scheduler, node, "the connection is unrecoverable", FALSE); } continue; } if (node->details->unclean) { // Don't probe nodes that need fencing continue; } if (!pcmk_is_set(node->priv->flags, pcmk__node_probes_allowed)) { // The user requested that probes not be done on this node continue; } // Probe each resource in the cluster on this node, as needed pcmk__probe_resource_list(scheduler->priv->resources, node); } } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 26a8c97ec4..5b0a7360e2 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,5119 +1,5120 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include // xmlNode #include // xmlXPathObject, etc. #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); // A (parsed) resource action history entry struct action_history { pcmk_resource_t *rsc; // Resource that history is for pcmk_node_t *node; // Node that history is for xmlNode *xml; // History entry XML // Parsed from entry XML const char *id; // XML ID of history entry const char *key; // Operation key of action const char *task; // Action name const char *exit_reason; // Exit reason given for result guint interval_ms; // Action interval int call_id; // Call ID of action int expected_exit_status; // Expected exit status of action int exit_status; // Actual exit status of action int execution_status; // Execution status of action }; /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(scheduler, option, flag) do { \ GHashTable *config_hash = (scheduler)->priv->options; \ const char *scf_value = pcmk__cluster_option(config_hash, (option)); \ \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } else { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum pcmk__on_fail *failed); static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node); static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler); static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler); static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Check whether a node is a dangling guest node * * \param[in] node Node to check * * \return true if \p node had a Pacemaker Remote connection resource with a * launcher that was removed from the CIB, otherwise false. */ static bool is_dangling_guest_node(pcmk_node_t *node) { return pcmk__is_pacemaker_remote_node(node) && (node->priv->remote != NULL) && (node->priv->remote->priv->launcher == NULL) && pcmk_is_set(node->priv->remote->flags, pcmk__rsc_removed_launched); } /*! * \brief Schedule a fence action for a node * * \param[in,out] scheduler Scheduler data * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider * \c PCMK_OPT_PRIORITY_FENCING_DELAY */ void pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); if (pcmk__is_guest_or_bundle_node(node)) { // Fence a guest or bundle node by marking its launcher as failed pcmk_resource_t *rsc = node->priv->remote->priv->launcher; if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", pcmk__node_name(node), reason, rsc->id); } else { pcmk__sched_warn(scheduler, "Guest node %s will be fenced " "(by recovering its guest resource %s): %s", pcmk__node_name(node), rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ pcmk__set_node_flags(node, pcmk__node_remote_reset); pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", pcmk__node_name(node), reason); pcmk__set_rsc_flags(node->priv->remote, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } else if (pcmk__is_remote_node(node)) { pcmk_resource_t *rsc = node->priv->remote; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", pcmk__node_name(node), reason); } else if (!pcmk_is_set(node->priv->flags, pcmk__node_remote_reset)) { pcmk__set_node_flags(node, pcmk__node_remote_reset); pcmk__sched_warn(scheduler, "Remote node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean", reason); } else { pcmk__sched_warn(scheduler, "Cluster node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \ "[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \ "and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \ "//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \ "|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \ "/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler) { xmlXPathObject *result = NULL; if (!pcmk_is_set(scheduler->flags, flag)) { result = pcmk__xpath_search(scheduler->input->doc, xpath); if (pcmk__xpath_num_results(result) > 0) { pcmk__set_scheduler_flags(scheduler, flag); } xmlXPathFreeObject(result); } } gboolean unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler) { const char *value = NULL; GHashTable *config_hash = pcmk__strkey_table(free, free); const pcmk_rule_input_t rule_input = { .now = scheduler->priv->now, }; scheduler->priv->options = config_hash; pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_input, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, scheduler); pcmk__validate_cluster_options(config_hash); set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES, pcmk__sched_probe_resources); if (!pcmk_is_set(scheduler->flags, pcmk__sched_probe_resources)) { crm_info("Startup probes: disabled (dangerous)"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is nonzero"); pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_fencing); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pcmk__sched_enable_unfencing, XPATH_ENABLE_UNFENCING, scheduler); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->priv->fence_timeout_ms)); crm_debug("Default fencing action timeout: %s", pcmk__readable_interval(scheduler->priv->fence_timeout_ms)); set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED, pcmk__sched_fencing_enabled); if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { crm_debug("STONITH of failed nodes is enabled"); } else { crm_debug("STONITH of failed nodes is disabled"); } scheduler->priv->fence_action = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION); crm_trace("STONITH will %s nodes", scheduler->priv->fence_action); set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING, pcmk__sched_concurrent_fencing); if (pcmk_is_set(scheduler->flags, pcmk__sched_concurrent_fencing)) { crm_debug("Concurrent fencing is enabled"); } else { crm_debug("Concurrent fencing is disabled"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY); if (value) { pcmk_parse_interval_spec(value, &(scheduler->priv->priority_fencing_ms)); crm_trace("Priority fencing delay is %s", pcmk__readable_interval(scheduler->priv->priority_fencing_ms)); } set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES, pcmk__sched_stop_all); crm_debug("Stop all active resources: %s", pcmk__flag_text(scheduler->flags, pcmk__sched_stop_all)); set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER, pcmk__sched_symmetric_cluster); if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_ignore; } else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_freeze; } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_demote; } else if (pcmk__strcase_any_of(value, PCMK_VALUE_FENCE, PCMK_VALUE_FENCE_LEGACY, NULL)) { if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { int do_panic = 0; pcmk__xe_get_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) { scheduler->no_quorum_policy = pcmk_no_quorum_fence; } else { crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop': cluster has never had quorum"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop' because fencing is disabled"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { scheduler->no_quorum_policy = pcmk_no_quorum_stop; } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: crm_debug("On loss of quorum: " "Freeze resources that require quorum"); break; case pcmk_no_quorum_stop: crm_debug("On loss of quorum: " "Stop resources that require quorum"); break; case pcmk_no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case pcmk_no_quorum_fence: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case pcmk_no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES, pcmk__sched_stop_removed_resources); if (pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) { crm_trace("Orphan resources are stopped"); } else { crm_trace("Orphan resources are ignored"); } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS, pcmk__sched_cancel_removed_actions); if (pcmk_is_set(scheduler->flags, pcmk__sched_cancel_removed_actions)) { crm_trace("Orphan resource actions are stopped"); } else { crm_trace("Orphan resource actions are ignored"); } set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE, pcmk__sched_in_maintenance); crm_trace("Maintenance mode: %s", pcmk__flag_text(scheduler->flags, pcmk__sched_in_maintenance)); set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL, pcmk__sched_start_failure_fatal); if (pcmk_is_set(scheduler->flags, pcmk__sched_start_failure_fatal)) { crm_trace("Start failures are always fatal"); } else { crm_trace("Start failures are handled by failcount"); } if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING, pcmk__sched_startup_fencing); } if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pcmk__warn_once(pcmk__wo_blind, "Blind faith: not fencing unseen nodes"); } pe__unpack_node_health_scores(scheduler); scheduler->priv->placement_strategy = pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY); crm_trace("Placement strategy: %s", scheduler->priv->placement_strategy); set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK, pcmk__sched_shutdown_lock); if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) { value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &(scheduler->priv->shutdown_lock_ms)); crm_trace("Resources will be locked to nodes that were cleanly " "shut down (locks expire after %s)", pcmk__readable_interval(scheduler->priv->shutdown_lock_ms)); } else { crm_trace("Resources will not be locked to nodes that were cleanly " "shut down"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->priv->node_pending_ms)); if (scheduler->priv->node_pending_ms == 0U) { crm_trace("Do not fence pending nodes"); } else { crm_trace("Fence pending nodes after %s", pcmk__readable_interval(scheduler->priv->node_pending_ms)); } set_config_flag(scheduler, PCMK_OPT_FENCE_REMOTE_WITHOUT_QUORUM, pcmk__sched_fence_remote_no_quorum); if (pcmk_is_set(scheduler->flags, pcmk__sched_fence_remote_no_quorum)) { crm_trace("Pacemaker Remote nodes may be fenced without quorum"); } else { crm_trace("Pacemaker Remote nodes require quorum to be fenced"); } return TRUE; } /*! * \internal * \brief Create a new node object in scheduler data * * \param[in] id ID of new node * \param[in] uname Name of new node * \param[in] type Type of new node * \param[in] score Score of new node * \param[in,out] scheduler Scheduler data * * \return Newly created node object * \note The returned object is part of the scheduler data and should not be * freed separately. */ pcmk_node_t * pe_create_node(const char *id, const char *uname, const char *type, int score, pcmk_scheduler_t *scheduler) { enum pcmk__node_variant variant = pcmk__node_variant_cluster; pcmk_node_t *new_node = NULL; if (pcmk_find_node(scheduler, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } if (pcmk__str_eq(type, PCMK_VALUE_MEMBER, pcmk__str_null_matches|pcmk__str_casei)) { variant = pcmk__node_variant_cluster; } else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) { variant = pcmk__node_variant_remote; } else { pcmk__config_err("Ignoring node %s with unrecognized type '%s'", pcmk__s(uname, "without name"), type); return NULL; } new_node = calloc(1, sizeof(pcmk_node_t)); if (new_node == NULL) { pcmk__sched_err(scheduler, "Could not allocate memory for node %s", uname); return NULL; } new_node->assign = calloc(1, sizeof(struct pcmk__node_assignment)); new_node->details = calloc(1, sizeof(struct pcmk__node_details)); new_node->priv = calloc(1, sizeof(pcmk__node_private_t)); if ((new_node->assign == NULL) || (new_node->details == NULL) || (new_node->priv == NULL)) { free(new_node->assign); free(new_node->details); free(new_node->priv); free(new_node); pcmk__sched_err(scheduler, "Could not allocate memory for node %s", uname); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->assign->score = score; new_node->priv->id = id; new_node->priv->name = uname; new_node->priv->flags = pcmk__node_probes_allowed; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->running_rsc = NULL; new_node->priv->scheduler = scheduler; new_node->priv->variant = variant; new_node->priv->attrs = pcmk__strkey_table(free, free); new_node->priv->utilization = pcmk__strkey_table(free, free); new_node->priv->digest_cache = pcmk__strkey_table(free, pe__free_digests); if (pcmk__is_pacemaker_remote_node(new_node)) { pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "remote"); pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_remote_nodes); } else { pcmk__insert_dup(new_node->priv->attrs, CRM_ATTR_KIND, "cluster"); } scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node, pe__cmp_node_name); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = pcmk__xe_id(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; // @TODO This doesn't handle rules or id-ref for (attr_set = pcmk__xe_first_child(xml_obj, PCMK_XE_META_ATTRIBUTES, NULL, NULL); attr_set != NULL; attr_set = pcmk__xe_next(attr_set, PCMK_XE_META_ATTRIBUTES)) { for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL); attr != NULL; attr = pcmk__xe_next(attr, NULL)) { const char *value = pcmk__xe_get(attr, PCMK_XA_VALUE); const char *name = pcmk__xe_get(attr, PCMK_XA_NAME); if (name == NULL) { // Sanity continue; } if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) { remote_name = value; } else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) { remote_server = value; } else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) { remote_port = value; } else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) { connect_timeout = value; } else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) { remote_allow_migrate = value; } else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->priv->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node) { if ((new_node->priv->variant == pcmk__node_variant_remote) && (new_node->priv->remote == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(scheduler->flags, pcmk__sched_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } } gboolean unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; pcmk_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; for (xml_obj = pcmk__xe_first_child(xml_nodes, PCMK_XE_NODE, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, PCMK_XE_NODE)) { int score = 0; int rc = pcmk__xe_get_score(xml_obj, PCMK_XA_SCORE, &score, 0); new_node = NULL; id = pcmk__xe_get(xml_obj, PCMK_XA_ID); uname = pcmk__xe_get(xml_obj, PCMK_XA_UNAME); type = pcmk__xe_get(xml_obj, PCMK_XA_TYPE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_NODE "> entry in configuration without id"); continue; } if (rc != pcmk_rc_ok) { // Not possible with schema validation enabled pcmk__config_warn("Using 0 as score for node %s " "because '%s' is not a valid score: %s", pcmk__s(uname, "without name"), pcmk__xe_get(xml_obj, PCMK_XA_SCORE), pcmk_rc_str(rc)); } new_node = pe_create_node(id, uname, type, score, scheduler); if (new_node == NULL) { return FALSE; } handle_startup_fencing(scheduler, new_node); add_node_attrs(xml_obj, new_node, FALSE, scheduler); crm_trace("Done with node %s", pcmk__xe_get(xml_obj, PCMK_XA_UNAME)); } return TRUE; } static void unpack_launcher(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *launcher_id = NULL; if (rsc->priv->children != NULL) { g_list_foreach(rsc->priv->children, (GFunc) unpack_launcher, scheduler); return; } launcher_id = g_hash_table_lookup(rsc->priv->meta, PCMK__META_CONTAINER); if ((launcher_id != NULL) && !pcmk__str_eq(launcher_id, rsc->id, pcmk__str_none)) { pcmk_resource_t *launcher = pe_find_resource(scheduler->priv->resources, launcher_id); if (launcher != NULL) { rsc->priv->launcher = launcher; launcher->priv->launched = g_list_append(launcher->priv->launched, rsc); pcmk__rsc_trace(rsc, "Resource %s's launcher is %s", rsc->id, launcher_id); } else { pcmk__config_err("Resource %s: Unknown " PCMK__META_CONTAINER " %s", rsc->id, launcher_id); } } } gboolean unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = pcmk__xe_id(xml_obj); /* The pcmk_find_node() check ensures we don't iterate over an * expanded node that has already been added to the node list */ if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found remote node %s defined by resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, 0, scheduler); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, 0, scheduler); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2, NULL)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, pcmk__xe_id(xml_obj2), pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, 0, scheduler); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc) { pcmk_node_t *remote_node = NULL; if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) { return; } if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pcmk_find_node(scheduler, new_rsc->id); CRM_CHECK(remote_node != NULL, return); pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s", new_rsc->id, pcmk__node_name(remote_node)); remote_node->priv->remote = new_rsc; if (new_rsc->priv->launcher == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(scheduler, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ pcmk__insert_dup(remote_node->priv->attrs, CRM_ATTR_KIND, "container"); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] scheduler Scheduler data * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when pe__unpack_resource() calls resource_location() */ gboolean unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; GList *gIter = NULL; scheduler->priv->templates = pcmk__strkey_table(free, pcmk__free_idref); for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj, NULL)) { pcmk_resource_t *new_rsc = NULL; const char *id = pcmk__xe_id(xml_obj); if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring <%s> resource without ID", xml_obj->name); continue; } if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) { if (g_hash_table_lookup_extended(scheduler->priv->templates, id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ pcmk__insert_dup(scheduler->priv->templates, id, NULL); } continue; } crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id); if (pe__unpack_resource(xml_obj, &new_rsc, NULL, scheduler) == pcmk_rc_ok) { scheduler->priv->resources = g_list_append(scheduler->priv->resources, new_rsc); pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", xml_obj->name, id); } } for (gIter = scheduler->priv->resources; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; unpack_launcher(rsc, scheduler); link_rsc2remotenode(scheduler, rsc); } scheduler->priv->resources = g_list_sort(scheduler->priv->resources, pe__cmp_rsc_priority); if (pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) { /* Ignore */ } else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the " PCMK_OPT_STONITH_ENABLED " option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } /*! * \internal * \brief Validate the levels in a fencing topology * * \param[in] xml \c PCMK_XE_FENCING_TOPOLOGY element */ void pcmk__validate_fencing_topology(const xmlNode *xml) { if (xml == NULL) { return; } CRM_CHECK(pcmk__xe_is(xml, PCMK_XE_FENCING_TOPOLOGY), return); for (const xmlNode *level = pcmk__xe_first_child(xml, PCMK_XE_FENCING_LEVEL, NULL, NULL); level != NULL; level = pcmk__xe_next(level, PCMK_XE_FENCING_LEVEL)) { const char *id = pcmk__xe_id(level); int index = 0; if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring fencing level without ID"); continue; } if (pcmk__xe_get_int(level, PCMK_XA_INDEX, &index) != pcmk_rc_ok) { pcmk__config_err("Ignoring fencing level %s with invalid index", id); continue; } if ((index < ST__LEVEL_MIN) || (index > ST__LEVEL_MAX)) { pcmk__config_err("Ignoring fencing level %s with out-of-range " "index %d", id, index); } } } gboolean unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler) { xmlNode *xml_tag = NULL; scheduler->priv->tags = pcmk__strkey_table(free, pcmk__free_idref); for (xml_tag = pcmk__xe_first_child(xml_tags, PCMK_XE_TAG, NULL, NULL); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag, PCMK_XE_TAG)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = pcmk__xe_id(xml_tag); if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID, (const char *) xml_tag->name); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag, PCMK_XE_OBJ_REF, NULL, NULL); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref, PCMK_XE_OBJ_REF)) { const char *obj_ref = pcmk__xe_id(xml_obj_ref); if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID, xml_obj_ref->name, tag_id); continue; } pcmk__add_idref(scheduler->priv->tags, tag_id, obj_ref); } } return TRUE; } /*! * \internal * \brief Unpack a ticket state entry * * \param[in] xml_ticket XML ticket state to unpack * \param[in,out] userdata Scheduler data * * \return pcmk_rc_ok (to always continue unpacking further entries) */ static int unpack_ticket_state(xmlNode *xml_ticket, void *userdata) { pcmk_scheduler_t *scheduler = userdata; const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pcmk__ticket_t *ticket = NULL; ticket_id = pcmk__xe_id(xml_ticket); if (pcmk__str_empty(ticket_id)) { return pcmk_rc_ok; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(scheduler->priv->ticket_constraints, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, scheduler); if (ticket == NULL) { return pcmk_rc_ok; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = pcmk__xml_attr_value(xIter); if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) { continue; } pcmk__insert_dup(ticket->state, prop_name, prop_value); } granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED); if (granted && crm_is_true(granted)) { pcmk__set_ticket_flags(ticket, pcmk__ticket_granted); crm_info("We have ticket '%s'", ticket->id); } else { pcmk__clear_ticket_flags(ticket, pcmk__ticket_granted); crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED); if (last_granted) { long long last_granted_ll = 0LL; int rc = pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); if (rc != pcmk_rc_ok) { crm_warn("Using %lld instead of invalid " PCMK_XA_LAST_GRANTED " value '%s' in state for ticket %s: %s", last_granted_ll, last_granted, ticket->id, pcmk_rc_str(rc)); } ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY); if (standby && crm_is_true(standby)) { pcmk__set_ticket_flags(ticket, pcmk__ticket_standby); if (pcmk_is_set(ticket->flags, pcmk__ticket_granted)) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { pcmk__clear_ticket_flags(ticket, pcmk__ticket_standby); } crm_trace("Done with ticket state for %s", ticket_id); return pcmk_rc_ok; } static void unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = NULL; pcmk_resource_t *rsc = NULL; int maint = 0; if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { return; } if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) { return; } crm_trace("Processing Pacemaker Remote node %s", pcmk__node_name(this_node)); pcmk__scan_min_int(pcmk__xe_get(state, PCMK__XA_NODE_IN_MAINTENANCE), &maint, 0); if (maint) { pcmk__set_node_flags(this_node, pcmk__node_remote_maint); } else { pcmk__clear_node_flags(this_node, pcmk__node_remote_maint); } rsc = this_node->priv->remote; if (!pcmk_is_set(this_node->priv->flags, pcmk__node_remote_reset)) { this_node->details->unclean = FALSE; pcmk__set_node_flags(this_node, pcmk__node_seen); } attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, this_node, TRUE, scheduler); if (pe__shutdown_requested(this_node)) { crm_info("%s is shutting down", pcmk__node_name(this_node)); this_node->details->shutdown = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current)) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) { crm_info("%s is in maintenance mode", pcmk__node_name(this_node)); this_node->details->maintenance = TRUE; } discovery = pcmk__node_attr(this_node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__warn_once(pcmk__wo_rdisc_enabled, "Support for the " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " node attribute is deprecated and will be removed" " (and behave as 'true') in a future release."); if (pcmk__is_remote_node(this_node) && !pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute on Pacemaker Remote node %s" " because fencing is disabled", pcmk__node_name(this_node)); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("%s has resource discovery disabled", pcmk__node_name(this_node)); pcmk__clear_node_flags(this_node, pcmk__node_probes_allowed); } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in,out] node Cluster node whose attributes are being unpacked * \param[in,out] scheduler Scheduler data */ static void unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, node, TRUE, scheduler); if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(node)); pcmk__set_node_flags(node, pcmk__node_standby); } if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current))) { crm_info("%s is in maintenance mode", pcmk__node_name(node)); node->details->maintenance = TRUE; } discovery = pcmk__node_attr(node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute for %s because disabling resource" " discovery is not allowed for cluster nodes", pcmk__node_name(node)); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not * the resource history inside it. Multiple passes through the status are needed * to fully unpack everything. * * \param[in] state CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *uname = NULL; pcmk_node_t *this_node = NULL; id = pcmk__xe_get(state, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without " PCMK_XA_ID); crm_log_xml_info(state, "missing-id"); return; } uname = pcmk__xe_get(state, PCMK_XA_UNAME); if (uname == NULL) { /* If a joining peer makes the cluster acquire the quorum from Corosync * but has not joined the controller CPG membership yet, it's possible * that the created PCMK__XE_NODE_STATE entry doesn't have a * PCMK_XA_UNAME yet. Recognize the node as pending and wait for it to * join CPG. */ crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" " "without " PCMK_XA_UNAME, id); } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { crm_notice("Ignoring recorded state for removed node with name %s and " PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id); return; } if (pcmk__is_pacemaker_remote_node(this_node)) { int remote_fenced = 0; /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(pcmk__xe_get(state, PCMK__XA_NODE_FENCED), &remote_fenced, 0); if (remote_fenced) { pcmk__set_node_flags(this_node, pcmk__node_remote_fenced); } else { pcmk__clear_node_flags(this_node, pcmk__node_remote_fenced); } return; } unpack_transient_attributes(state, this_node, scheduler); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; pcmk__set_node_flags(this_node, pcmk__node_seen); crm_trace("Determining online status of cluster node %s (id %s)", pcmk__node_name(this_node), id); determine_online_status(state, this_node, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk__sched_quorate) && this_node->details->online && (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(scheduler, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(const xmlNode *status, bool fence, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; // Loop through all PCMK__XE_NODE_STATE entries in CIB status for (const xmlNode *state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL, NULL); state != NULL; state = pcmk__xe_next(state, PCMK__XE_NODE_STATE)) { const char *id = pcmk__xe_id(state); const char *uname = pcmk__xe_get(state, PCMK_XA_UNAME); pcmk_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " PCMK__XE_NODE_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pcmk__is_guest_or_bundle_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ const pcmk_resource_t *remote = this_node->priv->remote; const pcmk_resource_t *launcher = remote->priv->launcher; if ((remote->priv->orig_role != pcmk_role_started) || (launcher->priv->orig_role != pcmk_role_started)) { crm_trace("Not unpacking resource history for guest node %s " "because launcher and connection are not known to " "be up", id); continue; } } else if (pcmk__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pcmk_resource_t *rsc = this_node->priv->remote; if ((rsc == NULL) || (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock) && (rsc->priv->orig_role != pcmk_role_started))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(scheduler->flags, pcmk__sched_fencing_enabled |pcmk__sched_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); unpack_handle_remote_attrs(this_node, state, scheduler); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); pcmk__set_node_flags(this_node, pcmk__node_unpacked); unpack_node_lrm(this_node, state, scheduler); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (scheduler->priv->ticket_constraints == NULL) { scheduler->priv->ticket_constraints = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL; state = pcmk__xe_next(state, NULL)) { if (pcmk__xe_is(state, PCMK_XE_TICKETS)) { pcmk__xe_foreach_child(state, PCMK__XE_TICKET_STATE, unpack_ticket_state, scheduler); } else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { unpack_node_state(state, scheduler); } } while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled), scheduler); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (scheduler->priv->stop_needed != NULL) { for (GList *item = scheduler->priv->stop_needed; item != NULL; item = item->next) { pcmk_resource_t *container = item->data; pcmk_node_t *node = pcmk__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(scheduler->priv->stop_needed); scheduler->priv->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *this_node = gIter->data; if (!pcmk__is_pacemaker_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->priv->remote != NULL)) { pe__set_next_role(this_node->priv->remote, pcmk_role_stopped, "remote shutdown"); } if (!pcmk_is_set(this_node->priv->flags, pcmk__node_unpacked)) { determine_remote_online_status(scheduler, this_node); } } return TRUE; } /*! * \internal * \brief Unpack node's time when it became a member at the cluster layer * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * \param[in,out] scheduler Scheduler data * * \return Epoch time when node became a cluster member * (or scheduler effective time for legacy entries) if a member, * 0 if not a member, or -1 if no valid information available */ static long long unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler) { const char *member_time = pcmk__xe_get(node_state, PCMK__XA_IN_CCM); int member = 0; if (member_time == NULL) { return -1LL; } else if (crm_str_to_boolean(member_time, &member) == 1) { /* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was * recorded as a boolean for a DC < 2.1.7, or the node is pending * shutdown and has left the CPG, in which case it was set to 1 to avoid * fencing for PCMK_OPT_NODE_PENDING_TIMEOUT. * * We return the effective time for in_ccm=1 because what's important to * avoid fencing is that effective time minus this value is less than * the pending node timeout. */ return member? (long long) pcmk__scheduler_epoch_time(scheduler) : 0LL; } else { long long when_member = 0LL; if ((pcmk__scan_ll(member_time, &when_member, 0LL) != pcmk_rc_ok) || (when_member < 0LL)) { crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM " in " PCMK__XE_NODE_STATE " entry", member_time); return -1LL; } return when_member; } } /*! * \internal * \brief Unpack node's time when it became online in process group * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * * \return Epoch time when node became online in process group (or 0 if not * online, or 1 for legacy online entries) */ static long long unpack_node_online(const xmlNode *node_state) { const char *peer_time = pcmk__xe_get(node_state, PCMK_XA_CRMD); // @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline" if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE, pcmk__str_casei|pcmk__str_null_matches)) { return 0LL; } else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) { return 1LL; } else { long long when_online = 0LL; if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok) || (when_online < 0)) { crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in " PCMK__XE_NODE_STATE " entry, assuming offline", peer_time); return 0LL; } return when_online; } } /*! * \internal * \brief Unpack node attribute for user-requested fencing * * \param[in] node Node to check * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status * * \return \c true if fencing has been requested for \p node, otherwise \c false */ static bool unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state) { long long value = 0LL; int value_i = 0; int rc = pcmk_rc_ok; const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE, NULL, pcmk__rsc_node_current); // Value may be boolean or an epoch time if (crm_str_to_boolean(value_s, &value_i) == 1) { return (value_i != 0); } rc = pcmk__scan_ll(value_s, &value, 0LL); if (rc == pcmk_rc_ok) { return (value > 0); } crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE "node attribute for %s: %s", value_s, pcmk__node_name(node), pcmk_rc_str(rc)); return false; } static gboolean determine_online_status_no_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { gboolean online = FALSE; const char *join = pcmk__xe_get(node_state, PCMK__XA_JOIN); const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); if (when_member <= 0) { crm_trace("Node %s is %sdown", pcmk__node_name(this_node), ((when_member < 0)? "presumed " : "")); } else if (when_online > 0) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node %s is not ready to run resources: %s", pcmk__node_name(this_node), join); } } else if (!pcmk_is_set(this_node->priv->flags, pcmk__node_expected_up)) { crm_trace("Node %s controller is down: " "member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } else { /* mark it unclean */ pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE); crm_info("Node %s member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } return online; } /*! * \internal * \brief Check whether a node has taken too long to join controller group * * \param[in,out] scheduler Scheduler data * \param[in] node Node to check * \param[in] when_member Epoch time when node became a cluster member * \param[in] when_online Epoch time when node joined controller group * * \return true if node has been pending (on the way up) longer than * \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false * \note This will also update the cluster's recheck time if appropriate. */ static inline bool pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, long long when_member, long long when_online) { if ((scheduler->priv->node_pending_ms > 0U) && (when_member > 0) && (when_online <= 0)) { // There is a timeout on pending nodes, and node is pending time_t timeout = when_member + pcmk__timeout_ms2s(scheduler->priv->node_pending_ms); if (pcmk__scheduler_epoch_time(node->priv->scheduler) >= timeout) { return true; // Node has timed out } // Node is pending, but still has time pcmk__update_recheck_time(timeout, scheduler, "pending node timeout"); } return false; } static bool determine_online_status_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { bool termination_requested = unpack_node_terminate(this_node, node_state); const char *join = pcmk__xe_get(node_state, PCMK__XA_JOIN); const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); /* - PCMK__XA_JOIN ::= member|down|pending|banned - PCMK_XA_EXPECTED ::= member|down @COMPAT with entries recorded for DCs < 2.1.7 - PCMK__XA_IN_CCM ::= true|false - PCMK_XA_CRMD ::= online|offline Since crm_feature_set 3.18.0 (pacemaker-2.1.7): - PCMK__XA_IN_CCM ::= |0 Since when node has been a cluster member. A value 0 of means the node is not a cluster member. - PCMK_XA_CRMD ::= |0 Since when peer has been online in CPG. A value 0 means the peer is offline in CPG. */ crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, ""), (termination_requested? " (termination requested)" : "")); if (this_node->details->shutdown) { crm_debug("%s is shutting down", pcmk__node_name(this_node)); /* Slightly different criteria since we can't shut down a dead peer */ return (when_online > 0); } if (when_member < 0) { pe_fence_node(scheduler, this_node, "peer has not been seen by the cluster", FALSE); return false; } if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) { pe_fence_node(scheduler, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (termination_requested) { if ((when_member <= 0) && (when_online <= 0) && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) { crm_info("%s was fenced as requested", pcmk__node_name(this_node)); return false; } pe_fence_node(scheduler, this_node, "fencing was requested", false); } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_null_matches)) { if (pending_too_long(scheduler, this_node, when_member, when_online)) { pe_fence_node(scheduler, this_node, "peer pending timed out on joining the process group", FALSE); } else if ((when_member > 0) || (when_online > 0)) { crm_info("- %s is not ready to run resources", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", pcmk__node_name(this_node)); } } else if (when_member <= 0) { // Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes pe_fence_node(scheduler, this_node, "peer is no longer part of the cluster", TRUE); } else if (when_online <= 0) { pe_fence_node(scheduler, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) { crm_info("%s is active", pcmk__node_name(this_node)); } else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("%s is not ready to run resources", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); this_node->details->pending = TRUE; } else { pe_fence_node(scheduler, this_node, "peer was in an unknown state", FALSE); } return (when_member > 0); } static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node) { pcmk_resource_t *rsc = this_node->priv->remote; pcmk_resource_t *launcher = NULL; pcmk_node_t *host = NULL; const char *node_type = "Remote"; if (rsc == NULL) { /* This is a leftover node state entry for a former Pacemaker Remote * node whose connection resource was removed. Consider it offline. */ crm_trace("Pacemaker Remote node %s is considered OFFLINE because " "its connection resource has been removed from the CIB", this_node->priv->id); this_node->details->online = FALSE; return; } launcher = rsc->priv->launcher; if (launcher != NULL) { node_type = "Guest"; if (pcmk__list_of_1(rsc->priv->active_nodes)) { host = rsc->priv->active_nodes->data; } } /* If the resource is currently started, mark it online. */ if (rsc->priv->orig_role == pcmk_role_started) { this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if ((rsc->priv->orig_role == pcmk_role_started) && (rsc->priv->next_role == pcmk_role_stopped)) { crm_trace("%s node %s shutting down because connection resource is stopping", node_type, this_node->priv->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if ((launcher != NULL) && pcmk_is_set(launcher->flags, pcmk__rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->priv->id); this_node->details->online = FALSE; pcmk__set_node_flags(this_node, pcmk__node_remote_reset); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", node_type, this_node->priv->id); this_node->details->online = FALSE; } else if ((rsc->priv->orig_role == pcmk_role_stopped) || ((launcher != NULL) && (launcher->priv->orig_role == pcmk_role_stopped))) { crm_trace("%s node %s OFFLINE because its resource is stopped", node_type, this_node->priv->id); this_node->details->online = FALSE; pcmk__clear_node_flags(this_node, pcmk__node_remote_reset); } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->priv->id); this_node->details->online = FALSE; pcmk__set_node_flags(this_node, pcmk__node_remote_reset); } else { crm_trace("%s node %s is %s", node_type, this_node->priv->id, this_node->details->online? "ONLINE" : "OFFLINE"); } } static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler) { gboolean online = FALSE; const char *exp_state = pcmk__xe_get(node_state, PCMK_XA_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { pcmk__set_node_flags(this_node, pcmk__node_expected_up); } if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { online = determine_online_status_no_fencing(scheduler, node_state, this_node); } else { online = determine_online_status_fencing(scheduler, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->assign->score = -PCMK_SCORE_INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->assign->score = -PCMK_SCORE_INFINITY; } if (this_node->details->unclean) { pcmk__sched_warn(scheduler, "%s is unclean", pcmk__node_name(this_node)); } else if (!this_node->details->online) { crm_trace("%s is offline", pcmk__node_name(this_node)); } else if (this_node->details->shutdown) { crm_info("%s is shutting down", pcmk__node_name(this_node)); } else if (this_node->details->pending) { crm_info("%s is pending", pcmk__node_name(this_node)); } else if (pcmk_is_set(this_node->priv->flags, pcmk__node_standby)) { crm_info("%s is in standby", pcmk__node_name(this_node)); } else if (this_node->details->maintenance) { crm_info("%s is in maintenance", pcmk__node_name(this_node)); } else { crm_info("%s is online", pcmk__node_name(this_node)); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; pcmk__assert(end != NULL); basename = strndup(last_rsc_id, end - last_rsc_id + 1); pcmk__assert(basename != NULL); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; pcmk__assert(end != NULL); zero = pcmk__assert_alloc(base_name_len + 3, sizeof(char)); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pcmk_resource_t * create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE); pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pcmk_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pcmk_find_node(scheduler, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, 0, scheduler); } link_rsc2remotenode(scheduler, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (pcmk__xe_get(rsc_entry, PCMK__META_CONTAINER)) { // This removed resource needs to be mapped to a launcher crm_trace("Launched resource %s was removed from the configuration", rsc_id); pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_launched); } pcmk__set_rsc_flags(rsc, pcmk__rsc_removed); scheduler->priv->resources = g_list_append(scheduler->priv->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history * * \param[in,out] parent Clone resource that orphan will be added to * \param[in] rsc_id Orphan's resource ID * \param[in] node Where orphan is active (for logging only) * \param[in,out] scheduler Scheduler data * * \return Newly added orphaned instance of \p parent */ static pcmk_resource_t * create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *top = pe__create_clone_child(parent, scheduler); pcmk_resource_t *orphan = NULL; // find_rsc() because we might be a cloned group orphan = top->priv->fns->find_rsc(top, rsc_id, NULL, pcmk_rsc_match_clone_only); pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, pcmk__node_name(node)); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX * instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX * instances are already active). * * \param[in,out] scheduler Scheduler data * \param[in] node Node on which to check for instance * \param[in,out] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (no instance) */ static pcmk_resource_t * find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, pcmk_resource_t *parent, const char *rsc_id) { GList *rIter = NULL; pcmk_resource_t *rsc = NULL; pcmk_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; pcmk__assert(pcmk__is_anonymous_clone(parent)); // Check for active (or partially active, for cloned groups) instance pcmk__rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, pcmk__node_name(node), parent->id); for (rIter = parent->priv->children; (rIter != NULL) && (rsc == NULL); rIter = rIter->next) { GList *locations = NULL; pcmk_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if * PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and * (3) when we re-run calculations on the same scheduler data as part of * a simulation. */ child->priv->fns->location(child, &locations, pcmk__rsc_node_current |pcmk__rsc_node_pending); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (pcmk__same_node((pcmk_node_t *) locations->data, node)) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->priv->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if PCMK_META_GLOBALLY_UNIQUE is switched from true * to false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->priv->active_nodes != NULL) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, pcmk__node_name(node)); skip_inactive = TRUE; rsc = NULL; } else { pcmk__rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pcmk__rsc_blocked)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->priv->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); /* ... but don't use it if it was already associated with a * pending action on another node */ if (inactive_instance != NULL) { const pcmk_node_t *pending_node = NULL; pending_node = inactive_instance->priv->pending_node; if ((pending_node != NULL) && !pcmk__same_node(pending_node, node)) { inactive_instance = NULL; } } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pcmk__rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or * PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we * don't want to consume a valid instance number for unclean nodes. Such * instances may appear to be active according to the history, but should be * considered inactive, so we can start an instance elsewhere. Treat such * instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pcmk__is_guest_or_bundle_node(node) && !pe__is_universal_clone(parent, scheduler)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler); pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pcmk_resource_t * unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, const char *rsc_id) { pcmk_resource_t *rsc = NULL; pcmk_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(scheduler->priv->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0, * we create a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pcmk_resource_t *clone0 = pe_find_resource(scheduler->priv->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->priv->variant > pcmk__rsc_variant_primitive) { crm_trace("Resource history for %s is orphaned " "because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pcmk__is_anonymous_clone(parent)) { if (pcmk__is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->priv->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(scheduler, node, parent, base); free(base); pcmk__assert(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none) && !pcmk__str_eq(rsc_id, rsc->priv->history_id, pcmk__str_none)) { pcmk__str_update(&(rsc->priv->history_id), rsc_id); pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, pcmk__node_name(node), rsc->id, pcmk_is_set(rsc->flags, pcmk__rsc_removed)? " (ORPHAN)" : ""); } return rsc; } static pcmk_resource_t * process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; const char *rsc_id = pcmk__xe_get(rsc_entry, PCMK_XA_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, pcmk__node_name(node)); rsc = create_fake_resource(rsc_id, rsc_entry, scheduler); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(scheduler->flags, pcmk__sched_stop_removed_resources)) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__orphan_do_not_run__", scheduler); } return rsc; } static void process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node, enum pcmk__on_fail on_fail) { pcmk_node_t *tmpnode = NULL; char *reason = NULL; enum pcmk__on_fail save_on_fail = pcmk__on_fail_ignore; pcmk_scheduler_t *scheduler = NULL; bool known_active = false; pcmk__assert(rsc != NULL); scheduler = rsc->priv->scheduler; known_active = (rsc->priv->orig_role > pcmk_role_stopped); pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, pcmk_role_text(rsc->priv->orig_role), pcmk__node_name(node), pcmk__on_fail_text(on_fail)); /* process current state */ if (rsc->priv->orig_role != pcmk_role_unknown) { pcmk_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->priv->probed_nodes, node->priv->id) == NULL) { pcmk_node_t *n = pe__copy_node(node); pcmk__rsc_trace(rsc, "%s (%s in history) known on %s", rsc->id, pcmk__s(rsc->priv->history_id, "the same"), pcmk__node_name(n)); g_hash_table_insert(iter->priv->probed_nodes, (gpointer) n->priv->id, n); } if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) { break; } iter = iter->priv->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (known_active && !node->details->online && !node->details->maintenance && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); should_fence = TRUE; } else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { if (pcmk__is_remote_node(node) && (node->priv->remote != NULL) && !pcmk_is_set(node->priv->remote->flags, pcmk__rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ pcmk__clear_node_flags(node, pcmk__node_seen); reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(scheduler, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = pcmk__on_fail_ignore; } switch (on_fail) { case pcmk__on_fail_ignore: /* nothing to do */ break; case pcmk__on_fail_demote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); demote_action(rsc, node, FALSE); break; case pcmk__on_fail_fence_node: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(scheduler, node, reason, FALSE); free(reason); break; case pcmk__on_fail_standby_node: pcmk__set_node_flags(node, pcmk__node_standby|pcmk__node_fail_standby); break; case pcmk__on_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); break; case pcmk__on_fail_ban: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -PCMK_SCORE_INFINITY, "__action_migration_auto__", scheduler); break; case pcmk__on_fail_stop: pe__set_next_role(rsc, pcmk_role_stopped, PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP); break; case pcmk__on_fail_restart: if (known_active) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); stop_action(rsc, node, FALSE); } break; case pcmk__on_fail_restart_container: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if ((rsc->priv->launcher != NULL) && pcmk__is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ scheduler->priv->stop_needed = g_list_prepend(scheduler->priv->stop_needed, rsc->priv->launcher); } else if (rsc->priv->launcher != NULL) { stop_action(rsc->priv->launcher, node, FALSE); } else if (known_active) { stop_action(rsc, node, FALSE); } break; case pcmk__on_fail_reset_remote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { tmpnode = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); } if (pcmk__is_remote_node(tmpnode) && !pcmk_is_set(tmpnode->priv->flags, pcmk__node_remote_fenced)) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(scheduler, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (known_active) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->priv->remote_reconnect_ms > 0U) { pe__set_next_role(rsc, pcmk_role_stopped, "remote reset"); } break; } /* Ensure a remote connection failure forces an unclean Pacemaker Remote * node to be fenced. By marking the node as seen, the failure will result * in a fencing operation regardless if we're going to attempt to reconnect * in this transition. */ if (pcmk_all_flags_set(rsc->flags, pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); if (tmpnode && tmpnode->details->unclean) { pcmk__set_node_flags(tmpnode, pcmk__node_seen); } } if (known_active) { if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Removed resource %s is active on %s and will be " "stopped when possible", rsc->id, pcmk__node_name(node)); } else { crm_notice("Removed resource %s must be stopped manually on %s " "because " PCMK_OPT_STOP_ORPHAN_RESOURCES " is set to false", rsc->id, pcmk__node_name(node)); } } native_add_running(rsc, node, scheduler, (save_on_fail != pcmk__on_fail_ignore)); switch (on_fail) { case pcmk__on_fail_ignore: break; case pcmk__on_fail_demote: case pcmk__on_fail_block: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); break; default: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); break; } } else if ((rsc->priv->history_id != NULL) && (strchr(rsc->priv->history_id, ':') != NULL)) { /* @COMPAT This is for older (<1.1.8) status sections that included * instance numbers, otherwise stopped instances are considered orphans. * * @TODO We should be able to drop this, but some old regression tests * will need to be updated. Double-check that this is not still needed * for unique clones (which may have been later converted to anonymous). */ pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)", rsc->priv->history_id, rsc->id); free(rsc->priv->history_id); rsc->priv->history_id = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *stop = (pcmk_action_t *) gIter->data; pcmk__set_action_flags(stop, pcmk__action_optional); } g_list_free(possible_matches); } /* A successful stop after migrate_to on the migration source doesn't make * the partially migrated resource stopped on the migration target. */ if ((rsc->priv->orig_role == pcmk_role_stopped) && (rsc->priv->active_nodes != NULL) && (rsc->priv->partial_migration_target != NULL) && pcmk__same_node(rsc->priv->partial_migration_source, node)) { rsc->priv->orig_role = pcmk_role_started; } } /* create active recurring operations as optional */ static void process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc, int start_index, int stop_index, GList *sorted_op_list, pcmk_scheduler_t *scheduler) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; pcmk__assert(rsc != NULL); pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = pcmk__xe_id(rsc_op); counter++; if (node->details->online == FALSE) { pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline", rsc->id, pcmk__node_name(node)); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active", id, pcmk__node_name(node)); continue; } else if (counter < start_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d", id, pcmk__node_name(node), counter); continue; } pcmk__xe_get_guint(rsc_op, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring", id, pcmk__node_name(node)); continue; } status = pcmk__xe_get(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pcmk__rsc_trace(rsc, "Skipping %s on %s: status", id, pcmk__node_name(node)); continue; } task = pcmk__xe_get(rsc_op, PCMK_XA_OPERATION); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node)); custom_action(rsc, key, task, node, TRUE, scheduler); } } void calculate_active_ops(const GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; *stop_index = -1; *start_index = -1; for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { const xmlNode *rsc_op = (const xmlNode *) iter->data; counter++; task = pcmk__xe_get(rsc_op, PCMK_XA_OPERATION); status = pcmk__xe_get(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { const char *rc = pcmk__xe_get(rsc_op, PCMK__XA_RC_CODE); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) time_t sched_time = 0; guint shutdown_lock_ms = scheduler->priv->shutdown_lock_ms; pcmk__xe_get_time(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time); if (lock_time == 0) { return; } sched_time = pcmk__scheduler_epoch_time(scheduler); if ((shutdown_lock_ms > 0U) && (sched_time > (lock_time + pcmk__timeout_ms2s(shutdown_lock_ms)))) { pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, pcmk__node_name(node)); pe__clear_resource_history(rsc, node); } else { rsc->priv->lock_node = node; rsc->priv->lock_time = lock_time; } } /*! * \internal * \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status * * \param[in,out] node Node whose status is being unpacked * \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked * \param[in,out] scheduler Scheduler data * * \return Resource corresponding to the entry, or NULL if no operation history */ static pcmk_resource_t * unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource, pcmk_scheduler_t *scheduler) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = pcmk_role_unknown; const char *rsc_id = pcmk__xe_id(lrm_resource); pcmk_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum pcmk__on_fail on_fail = pcmk__on_fail_ignore; enum rsc_role_e saved_role = pcmk_role_unknown; if (rsc_id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE " entry: No " PCMK_XA_ID); crm_log_xml_info(lrm_resource, "missing-id"); return NULL; } crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s", rsc_id, pcmk__node_name(node)); /* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort * them */ for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(scheduler, node, rsc_id); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, scheduler); } } pcmk__assert(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(scheduler->flags, pcmk__sched_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, scheduler); } /* process operations */ saved_role = rsc->priv->orig_role; rsc->priv->orig_role = pcmk_role_unknown; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, scheduler); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail); if (get_target_role(rsc, &req_role)) { if ((rsc->priv->next_role == pcmk_role_unknown) || (req_role < rsc->priv->next_role)) { pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE); } else if (req_role > rsc->priv->next_role) { pcmk__rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, pcmk_role_text(rsc->priv->next_role), pcmk_role_text(req_role)); } } if (saved_role > rsc->priv->orig_role) { rsc->priv->orig_role = saved_role; } return rsc; } static void handle_removed_launched_resources(const xmlNode *lrm_rsc_list, pcmk_scheduler_t *scheduler) { for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) { pcmk_resource_t *rsc; pcmk_resource_t *launcher = NULL; const char *rsc_id; const char *launcher_id = NULL; launcher_id = pcmk__xe_get(rsc_entry, PCMK__META_CONTAINER); rsc_id = pcmk__xe_get(rsc_entry, PCMK_XA_ID); if ((launcher_id == NULL) || (rsc_id == NULL)) { continue; } launcher = pe_find_resource(scheduler->priv->resources, launcher_id); if (launcher == NULL) { continue; } rsc = pe_find_resource(scheduler->priv->resources, rsc_id); if ((rsc == NULL) || (rsc->priv->launcher != NULL) || !pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) { continue; } pcmk__rsc_trace(rsc, "Mapped launcher of removed resource %s to %s", rsc->id, launcher_id); rsc->priv->launcher = launcher; launcher->priv->launched = g_list_append(launcher->priv->launched, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in,out] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler) { bool found_removed_launched_resource = false; // Drill down to PCMK__XE_LRM_RESOURCES section xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL); if (xml == NULL) { return; } xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL); if (xml == NULL) { return; } // Unpack each PCMK__XE_LRM_RESOURCE entry for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry, PCMK__XE_LRM_RESOURCE)) { pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) { found_removed_launched_resource = true; } } /* Now that all resource state has been unpacked for this node, map any * removed launched resources to their launchers. */ if (found_removed_launched_resource) { handle_removed_launched_resources(xml, scheduler); } } static void set_active(pcmk_resource_t *rsc) { const pcmk_resource_t *top = pe__const_top_resource(rsc, false); if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) { rsc->priv->orig_role = pcmk_role_unpromoted; } else { rsc->priv->orig_role = pcmk_role_started; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pcmk_node_t *node = value; int *score = user_data; node->assign->score = *score; } #define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE #define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \ "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE #define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, int target_rc, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']" SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'", NULL); /* Need to check against transition_magic too? */ if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_TARGET "='", source, "']", NULL); } else if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_SOURCE "='", source, "']", NULL); } else { g_string_append_c(xpath, ']'); } xml = pcmk__xpath_find_one(scheduler->input->doc, xpath->str, LOG_DEBUG); g_string_free(xpath, TRUE); if (xml && target_rc >= 0) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; pcmk__xe_get_int(xml, PCMK__XA_RC_CODE, &rc); pcmk__xe_get_int(xml, PCMK__XA_OP_STATUS, &status); if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static xmlNode * find_lrm_resource(const char *rsc_id, const char *node_name, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']", NULL); xml = pcmk__xpath_find_one(scheduler->input->doc, xpath->str, LOG_DEBUG); g_string_free(xpath, TRUE); return xml; } /*! * \internal * \brief Check whether a resource has no completed action history on a node * * \param[in,out] rsc Resource to check * \param[in] node_name Node to check * * \return true if \p rsc_id is unknown on \p node_name, otherwise false */ static bool unknown_on_node(pcmk_resource_t *rsc, const char *node_name) { bool result = false; xmlXPathObject *search; char *xpath = NULL; xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" SUB_XPATH_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "!='%d']", node_name, rsc->id, PCMK_OCF_UNKNOWN); search = pcmk__xpath_search(rsc->priv->scheduler->input->doc, xpath); result = (pcmk__xpath_num_results(search) == 0); xmlXPathFreeObject(search); free(xpath); return result; } /*! * \internal * \brief Check whether a probe/monitor indicating the resource was not running * on a node happened after some event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that monitor is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a monitor happened after event, false otherwise */ static bool monitor_not_running_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, pcmk_scheduler_t *scheduler) { /* Any probe/monitor operation on the node indicating it was not running * there */ xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name, NULL, PCMK_OCF_NOT_RUNNING, scheduler); return (monitor != NULL) && (pe__is_newer_op(monitor, xml_op) > 0); } /*! * \internal * \brief Check whether any non-monitor operation on a node happened after some * event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that non-monitor is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool non_monitor_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, pcmk_scheduler_t *scheduler) { xmlNode *lrm_resource = NULL; lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler); if (lrm_resource == NULL) { return false; } for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); op != NULL; op = pcmk__xe_next(op, PCMK__XE_LRM_RSC_OP)) { const char * task = NULL; if (op == xml_op) { continue; } task = pcmk__xe_get(op, PCMK_XA_OPERATION); if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL) && pe__is_newer_op(op, xml_op) > 0) { return true; } } return false; } /*! * \internal * \brief Check whether the resource has newer state on a node after a migration * attempt * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] migrate_to Any migrate_to event that is being compared to * \param[in] migrate_from Any migrate_from event that is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool newer_state_after_migrate(const char *rsc_id, const char *node_name, const xmlNode *migrate_to, const xmlNode *migrate_from, pcmk_scheduler_t *scheduler) { const xmlNode *xml_op = (migrate_from != NULL)? migrate_from : migrate_to; const char *source = pcmk__xe_get(xml_op, PCMK__META_MIGRATE_SOURCE); /* It's preferred to compare to the migrate event on the same node if * existing, since call ids are more reliable. */ if ((xml_op != migrate_to) && (migrate_to != NULL) && pcmk__str_eq(node_name, source, pcmk__str_casei)) { xml_op = migrate_to; } /* If there's any newer non-monitor operation on the node, or any newer * probe/monitor operation on the node indicating it was not running there, * the migration events potentially no longer matter for the node. */ return non_monitor_after(rsc_id, node_name, xml_op, scheduler) || monitor_not_running_after(rsc_id, node_name, xml_op, scheduler); } /*! * \internal * \brief Parse migration source and target node names from history entry * * \param[in] entry Resource history entry for a migration action * \param[in] source_node If not NULL, source must match this node * \param[in] target_node If not NULL, target must match this node * \param[out] source_name Where to store migration source node name * \param[out] target_name Where to store migration target node name * * \return Standard Pacemaker return code */ static int get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node, const pcmk_node_t *target_node, const char **source_name, const char **target_name) { *source_name = pcmk__xe_get(entry, PCMK__META_MIGRATE_SOURCE); *target_name = pcmk__xe_get(entry, PCMK__META_MIGRATE_TARGET); if ((*source_name == NULL) || (*target_name == NULL)) { pcmk__config_err("Ignoring resource history entry %s without " PCMK__META_MIGRATE_SOURCE " and " PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry)); return pcmk_rc_unpack_error; } if ((source_node != NULL) && !pcmk__str_eq(*source_name, source_node->priv->name, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_SOURCE "='%s' does not match %s", pcmk__xe_id(entry), *source_name, pcmk__node_name(source_node)); return pcmk_rc_unpack_error; } if ((target_node != NULL) && !pcmk__str_eq(*target_name, target_node->priv->name, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_TARGET "='%s' does not match %s", pcmk__xe_id(entry), *target_name, pcmk__node_name(target_node)); return pcmk_rc_unpack_error; } return pcmk_rc_ok; } /* * \internal * \brief Add a migration source to a resource's list of dangling migrations * * If the migrate_to and migrate_from actions in a live migration both * succeeded, but there is no stop on the source, the migration is considered * "dangling." Add the source to the resource's dangling migration list, which * will be used to schedule a stop on the source without affecting the target. * * \param[in,out] rsc Resource involved in migration * \param[in] node Migration source */ static void add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s", rsc->id, pcmk__node_name(node)); rsc->priv->orig_role = pcmk_role_stopped; rsc->priv->dangling_migration_sources = g_list_prepend(rsc->priv->dangling_migration_sources, (gpointer) node); } /*! * \internal * \brief Update resource role etc. after a successful migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_success(struct action_history *history) { /* A complete migration sequence is: * 1. migrate_to on source node (which succeeded if we get to this function) * 2. migrate_from on target node * 3. stop on source node * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from succeeded but no stop has happened, the * migration is considered to be "dangling". * * If a successful migrate_to and stop have happened on the source node, we * still need to check for a partial migration, due to scenarios (easier to * produce with batch-limit=1) like: * * - A resource is migrating from node1 to node2, and a migrate_to is * initiated for it on node1. * * - node2 goes into standby mode while the migrate_to is pending, which * aborts the transition. * * - Upon completion of the migrate_to, a new transition schedules a stop * on both nodes and a start on node1. * * - If the new transition is aborted for any reason while the resource is * stopping on node1, the transition after that stop completes will see * the migrate_to and stop on the source, but it's still a partial * migration, and the resource must be stopped on node2 because it is * potentially active there due to the migrate_to. * * We also need to take into account that either node's history may be * cleared at any point in the migration process. */ int from_rc = PCMK_OCF_OK; int from_status = PCMK_EXEC_PENDING; pcmk_node_t *target_node = NULL; xmlNode *migrate_from = NULL; const char *source = NULL; const char *target = NULL; bool source_newer_op = false; bool target_newer_state = false; bool active_on_target = false; pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } // Check for newer state on the source source_newer_op = non_monitor_after(history->rsc->id, source, history->xml, scheduler); // Check for a migrate_from action from this source on the target migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, -1, scheduler); if (migrate_from != NULL) { if (source_newer_op) { /* There's a newer non-monitor operation on the source and a * migrate_from on the target, so this migrate_to is irrelevant to * the resource's state. */ return; } pcmk__xe_get_int(migrate_from, PCMK__XA_RC_CODE, &from_rc); pcmk__xe_get_int(migrate_from, PCMK__XA_OP_STATUS, &from_status); } /* If the resource has newer state on both the source and target after the * migration events, this migrate_to is irrelevant to the resource's state. */ target_newer_state = newer_state_after_migrate(history->rsc->id, target, history->xml, migrate_from, scheduler); if (source_newer_op && target_newer_state) { return; } /* Check for dangling migration (migrate_from succeeded but stop not done). * We know there's no stop because we already returned if the target has a * migrate_from and the source has any newer non-monitor operation. */ if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) { add_dangling_migration(history->rsc, history->node); return; } /* Without newer state, this migrate_to implies the resource is active. * (Clones are not allowed to migrate, so role can't be promoted.) */ history->rsc->priv->orig_role = pcmk_role_started; target_node = pcmk_find_node(scheduler, target); active_on_target = !target_newer_state && (target_node != NULL) && target_node->details->online; if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target if (active_on_target) { native_add_running(history->rsc, target_node, scheduler, TRUE); } else { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } return; } // The migrate_from is pending, complete but erased, or to be scheduled /* If there is no history at all for the resource on an online target, then * it was likely cleaned. Just return, and we'll schedule a probe. Once we * have the probe result, it will be reflected in target_newer_state. */ if ((target_node != NULL) && target_node->details->online && unknown_on_node(history->rsc, target)) { return; } if (active_on_target) { pcmk_node_t *source_node = pcmk_find_node(scheduler, source); native_add_running(history->rsc, target_node, scheduler, FALSE); if ((source_node != NULL) && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ history->rsc->priv->partial_migration_target = target_node; history->rsc->priv->partial_migration_source = source_node; } } else if (!source_newer_op) { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_failure(struct action_history *history) { xmlNode *target_migrate_from = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->priv->orig_role = pcmk_role_started; // Check for migrate_from on the target target_migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the target, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, target) /* If the resource has newer state on the target after the migration * events, this migrate_to no longer matters for the target. */ && !newer_state_after_migrate(history->rsc->id, target, history->xml, target_migrate_from, scheduler)) { /* The resource has no newer state on the target, so assume it's still * active there. * (if it is up). */ pcmk_node_t *target_node = pcmk_find_node(scheduler, target); if (target_node && target_node->details->online) { native_add_running(history->rsc, target_node, scheduler, FALSE); } } else if (!non_monitor_after(history->rsc->id, source, history->xml, scheduler)) { /* We know the resource has newer state on the target, but this * migrate_to still matters for the source as long as there's no newer * non-monitor operation there. */ // Mark node as having dangling migration so we can force a stop later history->rsc->priv->dangling_migration_sources = g_list_prepend(history->rsc->priv->dangling_migration_sources, (gpointer) history->node); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_from action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_from_failure(struct action_history *history) { xmlNode *source_migrate_to = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, NULL, history->node, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->priv->orig_role = pcmk_role_started; // Check for a migrate_to on the source source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO, source, target, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the source, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, source) /* If the resource has newer state on the source after the migration * events, this migrate_from no longer matters for the source. */ && !newer_state_after_migrate(history->rsc->id, source, source_migrate_to, history->xml, scheduler)) { /* The resource has no newer state on the source, so assume it's still * active there (if it is up). */ pcmk_node_t *source_node = pcmk_find_node(scheduler, source); if (source_node && source_node->details->online) { native_add_running(history->rsc, source_node, scheduler, TRUE); } } } /*! * \internal * \brief Add an action to cluster's list of failed actions * * \param[in,out] history Parsed action result history */ static void record_failed_op(struct action_history *history) { const pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; if (!(history->node->details->online)) { return; } for (const xmlNode *xIter = scheduler->priv->failed->children; xIter != NULL; xIter = xIter->next) { const char *key = pcmk__xe_history_key(xIter); const char *uname = pcmk__xe_get(xIter, PCMK_XA_UNAME); if (pcmk__str_eq(history->key, key, pcmk__str_none) && pcmk__str_eq(uname, history->node->priv->name, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", history->key, pcmk__node_name(history->node)); return; } } crm_trace("Adding entry for %s on %s to failed action list", history->key, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->priv->name); crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id); pcmk__xml_copy(scheduler->priv->failed, history->xml); } static char * last_change_str(const xmlNode *xml_op) { time_t when; char *result = NULL; if (pcmk__xe_get_time(xml_op, PCMK_XA_LAST_RC_CHANGE, &when) == pcmk_rc_ok) { char *when_s = pcmk__epoch2str(&when, 0); const char *p = strchr(when_s, ' '); // Skip day of week to make message shorter if ((p != NULL) && (*(++p) != '\0')) { result = pcmk__str_copy(p); } free(when_s); } if (result == NULL) { result = pcmk__str_copy("unknown_time"); } return result; } /*! * \internal * \brief Ban a resource (or its clone if an anonymous instance) from all nodes * * \param[in,out] rsc Resource to ban */ static void ban_from_all_nodes(pcmk_resource_t *rsc) { int score = -PCMK_SCORE_INFINITY; const pcmk_scheduler_t *scheduler = rsc->priv->scheduler; if (rsc->priv->parent != NULL) { pcmk_resource_t *parent = uber_parent(rsc); if (pcmk__is_anonymous_clone(parent)) { /* For anonymous clones, if an operation with * PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the * entire clone must stop. */ rsc = parent; } } // Ban the resource from all nodes crm_notice("%s will not be started under current conditions", rsc->id); if (rsc->priv->allowed_nodes != NULL) { g_hash_table_destroy(rsc->priv->allowed_nodes); } rsc->priv->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_foreach(rsc->priv->allowed_nodes, set_node_score, &score); } /*! * \internal * \brief Get configured failure handling and role after failure for an action * * \param[in,out] history Unpacked action history entry * \param[out] on_fail Where to set configured failure handling * \param[out] fail_role Where to set to role after failure */ static void unpack_failure_handling(struct action_history *history, enum pcmk__on_fail *on_fail, enum rsc_role_e *fail_role) { xmlNode *config = pcmk__find_action_config(history->rsc, history->task, history->interval_ms, true); GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node, history->task, history->interval_ms, config); const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); *on_fail = pcmk__parse_on_fail(history->rsc, history->task, history->interval_ms, on_fail_str); *fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail, meta); g_hash_table_destroy(meta); } /*! * \internal * \brief Update resource role, failure handling, etc., after a failed action * * \param[in,out] history Parsed action result history * \param[in] config_on_fail Action failure handling from configuration * \param[in] fail_role Resource's role after failure of this action * \param[out] last_failure This will be set to the history XML * \param[in,out] on_fail Actual handling of action result */ static void unpack_rsc_op_failure(struct action_history *history, enum pcmk__on_fail config_on_fail, enum rsc_role_e fail_role, xmlNode **last_failure, enum pcmk__on_fail *on_fail) { bool is_probe = false; char *last_change_s = NULL; pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; *last_failure = history->xml; is_probe = pcmk_xe_is_probe(history->xml); last_change_s = last_change_str(history->xml); if (!pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster) && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " QB_XS " exit-status=%d id=%s", crm_exit_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); } else { pcmk__sched_warn(scheduler, "Unexpected result (%s%s%s) was recorded for %s of " "%s on %s at %s " QB_XS " exit-status=%d id=%s", crm_exit_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); if (is_probe && (history->exit_status != PCMK_OCF_OK) && (history->exit_status != PCMK_OCF_NOT_RUNNING) && (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the " PCMK_XA_RESOURCE_DISCOVERY " option for location " "constraints", history->rsc->id, pcmk__node_name(history->node)); } record_failed_op(history); } free(last_change_s); if (*on_fail < config_on_fail) { pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s", pcmk__on_fail_text(*on_fail), pcmk__on_fail_text(config_on_fail), history->key); *on_fail = config_on_fail; } if (strcmp(history->task, PCMK_ACTION_STOP) == 0) { resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY, "__stop_fail__", scheduler); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) { unpack_migrate_to_failure(history); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) { unpack_migrate_from_failure(history); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->priv->orig_role = pcmk_role_promoted; } else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) { if (config_on_fail == pcmk__on_fail_block) { history->rsc->priv->orig_role = pcmk_role_promoted; pe__set_next_role(history->rsc, pcmk_role_stopped, "demote with " PCMK_META_ON_FAIL "=block"); } else if (history->exit_status == PCMK_OCF_NOT_RUNNING) { history->rsc->priv->orig_role = pcmk_role_stopped; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ history->rsc->priv->orig_role = pcmk_role_unpromoted; } } if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { /* leave stopped */ pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id); history->rsc->priv->orig_role = pcmk_role_stopped; } else if (history->rsc->priv->orig_role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id); set_active(history->rsc); } pcmk__rsc_trace(history->rsc, "Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s", history->rsc->id, pcmk_role_text(history->rsc->priv->orig_role), pcmk__btoa(history->node->details->unclean), pcmk__on_fail_text(config_on_fail), pcmk_role_text(fail_role)); if ((fail_role != pcmk_role_started) && (history->rsc->priv->next_role < fail_role)) { pe__set_next_role(history->rsc, fail_role, "failure"); } if (fail_role == pcmk_role_stopped) { ban_from_all_nodes(history->rsc); } } /*! * \internal * \brief Block a resource with a failed action if it cannot be recovered * * If resource action is a failed stop and fencing is not possible, mark the * resource as unmanaged and blocked, since recovery cannot be done. * * \param[in,out] history Parsed action history entry */ static void block_if_unrecoverable(struct action_history *history) { char *last_change_s = NULL; if (strcmp(history->task, PCMK_ACTION_STOP) != 0) { return; // All actions besides stop are always recoverable } if (pe_can_fence(history->node->priv->scheduler, history->node)) { return; // Failed stops are recoverable via fencing } last_change_s = last_change_str(history->xml); pcmk__sched_err(history->node->priv->scheduler, "No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " QB_XS " rc=%d id=%s", history->rsc->id, history->task, pcmk__node_name(history->node), crm_exit_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), last_change_s, history->exit_status, history->id); free(last_change_s); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked); } /*! * \internal * \brief Update action history's execution status and why * * \param[in,out] history Parsed action history entry * \param[out] why Where to store reason for update * \param[in] value New value * \param[in] reason Description of why value was changed */ static inline void remap_because(struct action_history *history, const char **why, int value, const char *reason) { if (history->execution_status != value) { history->execution_status = value; *why = reason; } } /*! * \internal * \brief Remap informational monitor results and operation status * * For the monitor results, certain OCF codes are for providing extended information * to the user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by Pacemaker. * * For operation status, the action result can be used to determine an appropriate * status for the purposes of responding to the action. The status provided by the * executor is not directly usable since the executor does not know what was expected. * * \param[in,out] history Parsed action history entry * \param[in,out] on_fail What should be done about the result * \param[in] expired Whether result is expired * * \note If the result is remapped and the node is not shutting down or failed, * the operation will be recorded in the scheduler data's list of failed * operations to highlight it for the user. * * \note This may update the resource's current and next role. */ static void remap_operation(struct action_history *history, enum pcmk__on_fail *on_fail, bool expired) { /* @TODO It would probably also be a good idea to map an exit status of * CRM_EX_PROMOTED or CRM_EX_DEGRADED_PROMOTED to CRM_EX_OK for promote * actions */ bool is_probe = false; int orig_exit_status = history->exit_status; int orig_exec_status = history->execution_status; const char *why = NULL; const char *task = history->task; // Remap degraded results to their successful counterparts history->exit_status = pcmk__effective_rc(history->exit_status); if (history->exit_status != orig_exit_status) { why = "degraded result"; if (!expired && (!history->node->details->shutdown || history->node->details->online)) { record_failed_op(history); } } if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && ((history->execution_status != PCMK_EXEC_DONE) || (history->exit_status != PCMK_OCF_NOT_RUNNING))) { history->execution_status = PCMK_EXEC_DONE; history->exit_status = PCMK_OCF_NOT_RUNNING; why = "equivalent probe result"; } /* If the executor reported an execution status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ switch (history->execution_status) { case PCMK_EXEC_DONE: case PCMK_EXEC_ERROR: break; // These should be treated as node-fatal case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "node-fatal error"); goto remap_done; default: goto remap_done; } is_probe = pcmk_xe_is_probe(history->xml); if (is_probe) { task = "probe"; } if (history->expected_exit_status < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * expected exit status in the transition key, which (along with the * similar case of a corrupted transition key in the CIB) will be * reported to this function as -1. Pacemaker 2.0+ does not support * rolling upgrades from those versions or processing of saved CIB files * from those versions, so we do not need to care much about this case. */ remap_because(history, &why, PCMK_EXEC_ERROR, "obsolete history format"); pcmk__config_warn("Expected result not found for %s on %s " "(corrupt or obsolete CIB?)", history->key, pcmk__node_name(history->node)); } else if (history->exit_status == history->expected_exit_status) { remap_because(history, &why, PCMK_EXEC_DONE, "expected result"); } else { remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result"); pcmk__rsc_debug(history->rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", history->key, pcmk__node_name(history->node), history->expected_exit_status, crm_exit_str(history->expected_exit_status), history->exit_status, crm_exit_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, "")); } switch (history->exit_status) { case PCMK_OCF_OK: if (is_probe && (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (history->expected_exit_status == history->exit_status) || !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) { /* For probes, recurring monitors for the Stopped role, and * unmanaged resources, "not running" is not considered a * failure. */ remap_because(history, &why, PCMK_EXEC_DONE, "exit status"); history->rsc->priv->orig_role = pcmk_role_stopped; *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (history->exit_status != history->expected_exit_status)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active and promoted on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } if (!expired || (history->exit_status == history->expected_exit_status)) { history->rsc->priv->orig_role = pcmk_role_promoted; } break; case PCMK_OCF_FAILED_PROMOTED: if (!expired) { history->rsc->priv->orig_role = pcmk_role_promoted; } remap_because(history, &why, PCMK_EXEC_ERROR, "exit status"); break; case PCMK_OCF_NOT_CONFIGURED: remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status"); break; case PCMK_OCF_UNIMPLEMENT_FEATURE: { guint interval_ms = 0; pcmk__xe_get_guint(history->xml, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); } else { remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED, "exit status"); } } break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); break; default: if (history->execution_status == PCMK_EXEC_DONE) { char *last_change_s = last_change_str(history->xml); crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", history->exit_status, task, history->rsc->id, pcmk__node_name(history->node), last_change_s); remap_because(history, &why, PCMK_EXEC_ERROR, "unknown exit status"); free(last_change_s); } break; } remap_done: if (why != NULL) { pcmk__rsc_trace(history->rsc, "Remapped %s result from [%s: %s] to [%s: %s] " "because of %s", history->key, pcmk_exec_status_str(orig_exec_status), crm_exit_str(orig_exit_status), pcmk_exec_status_str(history->execution_status), crm_exit_str(history->exit_status), why); } } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(const xmlNode *xml_op, const char *task, pcmk_resource_t *rsc, pcmk_node_t *node) { if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pcmk__add_param_check(xml_op, rsc, node, pcmk__check_last_failure); } else { pcmk__op_digest_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->priv->scheduler); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, pcmk__xe_history_key(xml_op), node->priv->id); break; case pcmk__digest_match: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn, pcmk_scheduler_t *scheduler) { pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id); if (remote_node) { pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, scheduler); order_actions(fence, action, pcmk__ar_first_implies_then); } } static bool should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task, guint interval_ms, bool is_last_failure) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if ((rsc->priv->remote_reconnect_ms > 0U) && pcmk_is_set(rsc->priv->scheduler->flags, pcmk__sched_fencing_enabled) && (interval_ms != 0) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pcmk_node_t *remote_node = pcmk_find_node(rsc->priv->scheduler, rsc->id); if (remote_node && !pcmk_is_set(remote_node->priv->flags, pcmk__node_remote_fenced)) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in,out] history Parsed action result history * * \return true if operation history entry is expired, otherwise false */ static bool check_operation_expiry(struct action_history *history) { bool expired = false; - bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0"); + bool is_last_failure = (history->id != NULL) + && g_str_has_suffix(history->id, "_last_failure_0"); time_t last_run = 0; int unexpired_fail_count = 0; const char *clear_reason = NULL; const guint expiration_sec = pcmk__timeout_ms2s(history->rsc->priv->failure_expiration_ms); pcmk_scheduler_t *scheduler = history->rsc->priv->scheduler; if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) { pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not expired: " "Not Installed does not expire", history->id, pcmk__node_name(history->node)); return false; // "Not installed" must always be cleared manually } if ((expiration_sec > 0) && (pcmk__xe_get_time(history->xml, PCMK_XA_LAST_RC_CHANGE, &last_run) == pcmk_rc_ok)) { /* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a * timestamp */ time_t now = pcmk__scheduler_epoch_time(scheduler); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + expiration_sec)) && !should_ignore_failure_timeout(history->rsc, history->task, history->interval_ms, is_last_failure)) { expired = true; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(history->node, history->rsc, &last_failure, pcmk__fc_effective, history->xml); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d " "expiration=%s last-failure@%lld", history->id, (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, pcmk__readable_interval(expiration_sec * 1000), (long long) last_failure); last_failure += expiration_sec + 1; if (unexpired_fail_count && (now < last_failure)) { pcmk__update_recheck_time(last_failure, scheduler, "fail count expiration"); } } if (expired) { if (pe_get_failcount(history->node, history->rsc, NULL, pcmk__fc_default, history->xml)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Unexpired fail count", history->id, pcmk__node_name(history->node)); expired = false; } } else if (is_last_failure && (history->rsc->priv->remote_reconnect_ms > 0U)) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(history->xml, history->task, history->rsc, history->node)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { pcmk_action_t *clear_op = NULL; // Schedule clearing of the fail count clear_op = pe__clear_failcount(history->rsc, history->node, clear_reason, scheduler); if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled) && (history->rsc->priv->remote_reconnect_ms > 0)) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", history->task, history->rsc->id); order_after_remote_fencing(clear_op, history->rsc, scheduler); } } if (expired && (history->interval_ms == 0) && pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { switch (history->exit_status) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Probe result", history->id, pcmk__node_name(history->node)); expired = false; break; } } return expired; } int pe__target_rc_from_xml(const xmlNode *xml_op) { int target_rc = 0; const char *key = pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } /*! * \internal * \brief Update a resource's state for an action result * * \param[in,out] history Parsed action history entry * \param[in] exit_status Exit status to base new state on * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void update_resource_state(struct action_history *history, int exit_status, const xmlNode *last_failure, enum pcmk__on_fail *on_fail) { bool clear_past_failure = false; if ((exit_status == PCMK_OCF_NOT_INSTALLED) || (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml))) { history->rsc->priv->orig_role = pcmk_role_stopped; } else if (exit_status == PCMK_OCF_NOT_RUNNING) { clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { if ((last_failure != NULL) && pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure), pcmk__str_none)) { clear_past_failure = true; } if (history->rsc->priv->orig_role < pcmk_role_started) { set_active(history->rsc); } } else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) { history->rsc->priv->orig_role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) { history->rsc->priv->orig_role = pcmk_role_stopped; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { history->rsc->priv->orig_role = pcmk_role_promoted; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { if (*on_fail == pcmk__on_fail_demote) { /* Demote clears an error only if * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ clear_past_failure = true; } history->rsc->priv->orig_role = pcmk_role_unpromoted; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { history->rsc->priv->orig_role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { unpack_migrate_to_success(history); } else if (history->rsc->priv->orig_role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "%s active on %s", history->rsc->id, pcmk__node_name(history->node)); set_active(history->rsc); } if (!clear_past_failure) { return; } switch (*on_fail) { case pcmk__on_fail_stop: case pcmk__on_fail_ban: case pcmk__on_fail_standby_node: case pcmk__on_fail_fence_node: pcmk__rsc_trace(history->rsc, "%s (%s) is not cleared by a completed %s", history->rsc->id, pcmk__on_fail_text(*on_fail), history->task); break; case pcmk__on_fail_block: case pcmk__on_fail_ignore: case pcmk__on_fail_demote: case pcmk__on_fail_restart: case pcmk__on_fail_restart_container: *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures"); break; case pcmk__on_fail_reset_remote: if (history->rsc->priv->remote_reconnect_ms == 0U) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures and reset remote"); } break; } } /*! * \internal * \brief Check whether a given history entry matters for resource state * * \param[in] history Parsed action history entry * * \return true if action can affect resource state, otherwise false */ static inline bool can_affect_state(struct action_history *history) { return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, "asyncmon", NULL); } /*! * \internal * \brief Unpack execution/exit status and exit reason from a history entry * * \param[in,out] history Action history entry to unpack * * \return Standard Pacemaker return code */ static int unpack_action_result(struct action_history *history) { if ((pcmk__xe_get_int(history->xml, PCMK__XA_OP_STATUS, &(history->execution_status)) != pcmk_rc_ok) || (history->execution_status < PCMK_EXEC_PENDING) || (history->execution_status > PCMK_EXEC_MAX) || (history->execution_status == PCMK_EXEC_CANCELLED)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_OP_STATUS " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(pcmk__xe_get(history->xml, PCMK__XA_OP_STATUS), "")); return pcmk_rc_unpack_error; } if ((pcmk__xe_get_int(history->xml, PCMK__XA_RC_CODE, &(history->exit_status)) != pcmk_rc_ok) || (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_RC_CODE " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(pcmk__xe_get(history->xml, PCMK__XA_RC_CODE), "")); return pcmk_rc_unpack_error; } history->exit_reason = pcmk__xe_get(history->xml, PCMK_XA_EXIT_REASON); return pcmk_rc_ok; } /*! * \internal * \brief Process an action history entry whose result expired * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * * \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the * entry needs no further processing) */ static int process_expired_result(struct action_history *history, int orig_exit_status) { if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && (orig_exit_status != history->expected_exit_status)) { if (history->rsc->priv->orig_role <= pcmk_role_stopped) { history->rsc->priv->orig_role = pcmk_role_unknown; } crm_trace("Ignoring resource history entry %s for probe of %s on %s: " "Masked failure expired", history->id, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->exit_status == history->expected_exit_status) { return pcmk_rc_undetermined; // Only failures expire } if (history->interval_ms == 0) { crm_notice("Ignoring resource history entry %s for %s of %s on %s: " "Expired failure", history->id, history->task, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->node->details->online && !history->node->details->unclean) { /* Reschedule the recurring action. schedule_cancel() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so pcmk__check_action_config() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s-interval %s of %s on %s " "after failure expired", pcmk__readable_interval(history->interval_ms), history->task, history->rsc->id, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST, "calculated-failure-timeout"); return pcmk_rc_ok; } return pcmk_rc_undetermined; } /*! * \internal * \brief Process a masked probe failure * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void mask_probe_failure(struct action_history *history, int orig_exit_status, const xmlNode *last_failure, enum pcmk__on_fail *on_fail) { pcmk_resource_t *ban_rsc = history->rsc; if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) { ban_rsc = uber_parent(history->rsc); } crm_notice("Treating probe result '%s' for %s on %s as 'not running'", crm_exit_str(orig_exit_status), history->rsc->id, pcmk__node_name(history->node)); update_resource_state(history, history->expected_exit_status, last_failure, on_fail); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->priv->name); record_failed_op(history); resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY, "masked-probe-failure", ban_rsc->priv->scheduler); } /*! * \internal Check whether a given failure is for a given pending action * * \param[in] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known * * \return true if \p last_failure is failure of pending action in \p history, * otherwise false * \note Both \p history and \p last_failure must come from the same * \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be * the same. */ static bool failure_is_newer(const struct action_history *history, const xmlNode *last_failure) { guint failure_interval_ms = 0U; long long failure_change = 0LL; long long this_change = 0LL; if (last_failure == NULL) { return false; // Resource has no last_failure entry } if (!pcmk__str_eq(history->task, pcmk__xe_get(last_failure, PCMK_XA_OPERATION), pcmk__str_none)) { return false; // last_failure is for different action } if ((pcmk__xe_get_guint(last_failure, PCMK_META_INTERVAL, &failure_interval_ms) != pcmk_rc_ok) || (history->interval_ms != failure_interval_ms)) { return false; // last_failure is for action with different interval } if ((pcmk__scan_ll(pcmk__xe_get(history->xml, PCMK_XA_LAST_RC_CHANGE), &this_change, 0LL) != pcmk_rc_ok) || (pcmk__scan_ll(pcmk__xe_get(last_failure, PCMK_XA_LAST_RC_CHANGE), &failure_change, 0LL) != pcmk_rc_ok) || (failure_change < this_change)) { return false; // Failure is not known to be newer } return true; } /*! * \internal * \brief Update a resource's role etc. for a pending action * * \param[in,out] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known */ static void process_pending_action(struct action_history *history, const xmlNode *last_failure) { /* For recurring monitors, a failure is recorded only in RSC_last_failure_0, * and there might be a RSC_monitor_INTERVAL entry with the last successful * or pending result. * * If last_failure contains the failure of the pending recurring monitor * we're processing here, and is newer, the action is no longer pending. * (Pending results have call ID -1, which sorts last, so the last failure * if any should be known.) */ if (failure_is_newer(history, last_failure)) { return; } if (strcmp(history->task, PCMK_ACTION_START) == 0) { pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending); set_active(history->rsc); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->priv->orig_role = pcmk_role_promoted; } else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) && history->node->details->unclean) { /* A migrate_to action is pending on a unclean source, so force a stop * on the target. */ const char *migrate_target = NULL; pcmk_node_t *target = NULL; migrate_target = pcmk__xe_get(history->xml, PCMK__META_MIGRATE_TARGET); target = pcmk_find_node(history->rsc->priv->scheduler, migrate_target); if (target != NULL) { stop_action(history->rsc, target, FALSE); } } if (history->rsc->priv->pending_action != NULL) { /* There should never be multiple pending actions, but as a failsafe, * just remember the first one processed for display purposes. */ return; } if (pcmk_is_probe(history->task, history->interval_ms)) { /* Pending probes are currently never displayed, even if pending * operations are requested. If we ever want to change that, * enable the below and the corresponding part of * native.c:native_pending_action(). */ #if 0 history->rsc->private->pending_action = strdup("probe"); history->rsc->private->pending_node = history->node; #endif } else { history->rsc->priv->pending_action = strdup(history->task); history->rsc->priv->pending_node = history->node; } } static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum pcmk__on_fail *on_fail) { int old_rc = 0; bool expired = false; pcmk_resource_t *parent = rsc; enum rsc_role_e fail_role = pcmk_role_unknown; enum pcmk__on_fail failure_strategy = pcmk__on_fail_restart; struct action_history history = { .rsc = rsc, .node = node, .xml = xml_op, .execution_status = PCMK_EXEC_UNKNOWN, }; CRM_CHECK(rsc && node && xml_op, return); history.id = pcmk__xe_id(xml_op); if (history.id == NULL) { pcmk__config_err("Ignoring resource history entry for %s on %s " "without ID", rsc->id, pcmk__node_name(node)); return; } // Task and interval history.task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION); if (history.task == NULL) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "without " PCMK_XA_OPERATION, history.id, rsc->id, pcmk__node_name(node)); return; } pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &(history.interval_ms)); if (!can_affect_state(&history)) { pcmk__rsc_trace(rsc, "Ignoring resource history entry %s for %s on %s " "with irrelevant action '%s'", history.id, rsc->id, pcmk__node_name(node), history.task); return; } if (unpack_action_result(&history) != pcmk_rc_ok) { return; // Error already logged } history.expected_exit_status = pe__target_rc_from_xml(xml_op); history.key = pcmk__xe_history_key(xml_op); pcmk__xe_get_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id)); pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)", history.id, history.task, history.call_id, pcmk__node_name(node), pcmk_exec_status_str(history.execution_status), crm_exit_str(history.exit_status)); if (node->details->unclean) { pcmk__rsc_trace(rsc, "%s is running on %s, which is unclean (further action " "depends on value of stop's on-fail attribute)", rsc->id, pcmk__node_name(node)); } expired = check_operation_expiry(&history); old_rc = history.exit_status; remap_operation(&history, on_fail, expired); if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) { goto done; } if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) { mask_probe_failure(&history, old_rc, *last_failure, on_fail); goto done; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { parent = uber_parent(rsc); } switch (history.execution_status) { case PCMK_EXEC_PENDING: process_pending_action(&history, *last_failure); goto done; case PCMK_EXEC_DONE: update_resource_state(&history, history.exit_status, *last_failure, on_fail); goto done; case PCMK_EXEC_NOT_INSTALLED: unpack_failure_handling(&history, &failure_strategy, &fail_role); if (failure_strategy == pcmk__on_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " QB_XS " status=%d rc=%d id=%s", history.task, rsc->id, pcmk__node_name(node), history.execution_status, history.exit_status, history.id); /* Also for printing it as "FAILED" by marking it as * pcmk__rsc_failed later */ *on_fail = pcmk__on_fail_ban; } resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->priv->scheduler); unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pcmk__is_pacemaker_remote_node(node) && pcmk_is_set(node->priv->remote->flags, pcmk__rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pcmk__set_rsc_flags(node->priv->remote, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } break; // Not done, do error handling case PCMK_EXEC_ERROR: case PCMK_EXEC_ERROR_HARD: case PCMK_EXEC_ERROR_FATAL: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_INVALID: break; // Not done, do error handling default: // No other value should be possible at this point break; } unpack_failure_handling(&history, &failure_strategy, &fail_role); if ((failure_strategy == pcmk__on_fail_ignore) || ((failure_strategy == pcmk__on_fail_restart_container) && (strcmp(history.task, PCMK_ACTION_STOP) == 0))) { char *last_change_s = last_change_str(xml_op); crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded " QB_XS " %s", history.task, crm_exit_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), rsc->id, pcmk__node_name(node), last_change_s, history.id); free(last_change_s); update_resource_state(&history, history.expected_exit_status, *last_failure, on_fail); crm_xml_add(xml_op, PCMK_XA_UNAME, node->priv->name); pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure); record_failed_op(&history); if ((failure_strategy == pcmk__on_fail_restart_container) && (*on_fail <= pcmk__on_fail_restart)) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); if (history.execution_status == PCMK_EXEC_ERROR_HARD) { uint8_t log_level = LOG_ERR; if (history.exit_status == PCMK_OCF_NOT_INSTALLED) { log_level = LOG_NOTICE; } do_crm_log(log_level, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s) " QB_XS " %s", parent->id, pcmk__node_name(node), crm_exit_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->priv->scheduler); } else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) { pcmk__sched_err(rsc->priv->scheduler, "Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " QB_XS " %s", parent->id, crm_exit_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, NULL, -PCMK_SCORE_INFINITY, "fatal-error", rsc->priv->scheduler); } } done: pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)", rsc->id, pcmk__node_name(node), history.id, pcmk_role_text(rsc->priv->orig_role), pcmk_role_text(rsc->priv->next_role)); } /*! * \internal * \brief Insert a node attribute with value into a \c GHashTable * * \param[in,out] key Key to insert (either freed or owned by * \p user_data upon return) * \param[in] value Value to insert (owned by \p user_data upon return) * \param[in] user_data \c GHashTable to insert into */ static gboolean insert_attr(gpointer key, gpointer value, gpointer user_data) { GHashTable *table = user_data; g_hash_table_insert(table, key, value); return TRUE; } static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler) { const char *cluster_name = NULL; const char *dc_id = pcmk__xe_get(scheduler->input, PCMK_XA_DC_UUID); const pcmk_rule_input_t rule_input = { .now = scheduler->priv->now, }; pcmk__insert_dup(node->priv->attrs, CRM_ATTR_UNAME, node->priv->name); pcmk__insert_dup(node->priv->attrs, CRM_ATTR_ID, node->priv->id); if ((scheduler->dc_node == NULL) && pcmk__str_eq(node->priv->id, dc_id, pcmk__str_casei)) { scheduler->dc_node = node; pcmk__insert_dup(node->priv->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_TRUE); } else if (!pcmk__same_node(node, scheduler->dc_node)) { pcmk__insert_dup(node->priv->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_FALSE); } cluster_name = g_hash_table_lookup(scheduler->priv->options, PCMK_OPT_CLUSTER_NAME); if (cluster_name) { pcmk__insert_dup(node->priv->attrs, CRM_ATTR_CLUSTER_NAME, cluster_name); } if (overwrite) { /* @TODO Try to reorder some unpacking so that we don't need the * overwrite argument or to unpack into a temporary table */ GHashTable *unpacked = pcmk__strkey_table(free, free); pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_input, unpacked, NULL, scheduler); g_hash_table_foreach_steal(unpacked, insert_attr, node->priv->attrs); g_hash_table_destroy(unpacked); } else { pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_input, node->priv->attrs, NULL, scheduler); } pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_input, node->priv->utilization, NULL, scheduler); if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL, pcmk__rsc_node_current) == NULL) { const char *site_name = pcmk__node_attr(node, "site-name", NULL, pcmk__rsc_node_current); if (site_name) { pcmk__insert_dup(node->priv->attrs, CRM_ATTR_SITE_NAME, site_name); } else if (cluster_name) { /* Default to cluster-name if unset */ pcmk__insert_dup(node->priv->attrs, CRM_ATTR_SITE_NAME, cluster_name); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op, PCMK__XE_LRM_RSC_OP)) { crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc); crm_xml_add(rsc_op, PCMK_XA_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pcmk_scheduler_t *scheduler) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS, NULL, NULL); pcmk_node_t *this_node = NULL; xmlNode *node_state = NULL; CRM_CHECK(status != NULL, return NULL); for (node_state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL, NULL); node_state != NULL; node_state = pcmk__xe_next(node_state, PCMK__XE_NODE_STATE)) { const char *uname = pcmk__xe_get(node_state, PCMK_XA_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pcmk_find_node(scheduler, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); } else { determine_online_status(node_state, this_node, scheduler); } if (this_node->details->online || pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL, NULL); tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL, NULL); for (lrm_rsc = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCE, NULL, NULL); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc, PCMK__XE_LRM_RESOURCE)) { const char *rsc_id = pcmk__xe_get(lrm_rsc, PCMK_XA_ID); if ((rsc != NULL) && !pcmk__str_eq(rsc_id, rsc, pcmk__str_none)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } return output; }