diff --git a/lib/services/services.c b/lib/services/services.c
index db6a4a06d2..6279221c3c 100644
--- a/lib/services/services.c
+++ b/lib/services/services.c
@@ -1,1150 +1,1152 @@
/*
- * Copyright 2010-2019 Andrew Beekhof <andrew@beekhof.net>
+ * Copyright 2010-2019 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 <crm_internal.h>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <sys/types.h>
#include <stdio.h>
#include <errno.h>
#include <unistd.h>
#include <dirent.h>
#include <fcntl.h>
#include <crm/crm.h>
#include <crm/common/mainloop.h>
#include <crm/services.h>
#include <crm/stonith-ng.h>
#include <crm/msg_xml.h>
#include "services_private.h"
#include "services_lsb.h"
#if SUPPORT_UPSTART
# include <upstart.h>
#endif
#if SUPPORT_SYSTEMD
# include <systemd.h>
#endif
/* TODO: Develop a rollover strategy */
static int operations = 0;
static GHashTable *recurring_actions = NULL;
/* ops waiting to run async because of conflicting active
* pending ops */
static GList *blocked_ops = NULL;
/* ops currently active (in-flight) */
static GList *inflight_ops = NULL;
static void handle_blocked_ops(void);
/*!
* \brief Find first service class that can provide a specified agent
*
* \param[in] agent Name of agent to search for
*
* \return Service class if found, NULL otherwise
*
* \note The priority is LSB, then systemd, then upstart. It would be preferable
* to put systemd first, but LSB merely requires a file existence check,
* while systemd requires contacting D-Bus.
*/
const char *
resources_find_service_class(const char *agent)
{
if (services__lsb_agent_exists(agent)) {
return PCMK_RESOURCE_CLASS_LSB;
}
#if SUPPORT_SYSTEMD
if (systemd_unit_exists(agent)) {
return PCMK_RESOURCE_CLASS_SYSTEMD;
}
#endif
#if SUPPORT_UPSTART
if (upstart_job_exists(agent)) {
return PCMK_RESOURCE_CLASS_UPSTART;
}
#endif
return NULL;
}
static inline void
init_recurring_actions(void)
{
if (recurring_actions == NULL) {
recurring_actions = g_hash_table_new_full(g_str_hash, g_str_equal, NULL,
NULL);
}
}
/*!
* \internal
* \brief Check whether op is in-flight systemd or upstart op
*
* \param[in] op Operation to check
*
* \return TRUE if op is in-flight systemd or upstart op
*/
static inline gboolean
inflight_systemd_or_upstart(svc_action_t *op)
{
return (safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD)
|| safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_UPSTART))
&& (g_list_find(inflight_ops, op) != NULL);
}
/*!
* \internal
* \brief Expand "service" alias to an actual resource class
*
* \param[in] rsc Resource name (for logging only)
* \param[in] standard Resource class as configured
* \param[in] agent Agent name to look for
*
* \return Newly allocated string with actual resource class
*
* \note The caller is responsible for calling free() on the result.
*/
static char *
expand_resource_class(const char *rsc, const char *standard, const char *agent)
{
char *expanded_class = NULL;
if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0) {
const char *found_class = resources_find_service_class(agent);
if (found_class) {
crm_debug("Found %s agent %s for %s", found_class, agent, rsc);
expanded_class = strdup(found_class);
} else {
crm_info("Assuming resource class lsb for agent %s for %s",
agent, rsc);
expanded_class = strdup(PCMK_RESOURCE_CLASS_LSB);
}
} else {
expanded_class = strdup(standard);
}
CRM_ASSERT(expanded_class);
return expanded_class;
}
/*!
* \brief Duplicate a file path, inserting a prefix if not absolute
*
* \param[in] filename File path to duplicate
* \param[in] dirname If filename is not absolute, prefix to add
*
* \return Newly allocated memory with full path
*/
static char *
dup_file_path(const char *filename, const char *dirname)
{
return (*filename == '/')? strdup(filename)
: crm_strdup_printf("%s/%s", dirname, filename);
}
svc_action_t *
resources_action_create(const char *name, const char *standard,
const char *provider, const char *agent,
const char *action, guint interval_ms, int timeout,
GHashTable *params, enum svc_action_flags flags)
{
svc_action_t *op = NULL;
uint32_t ra_caps = 0;
/*
* Do some up front sanity checks before we go off and
* build the svc_action_t instance.
*/
if (crm_strlen_zero(name)) {
crm_err("Cannot create operation without resource name");
goto return_error;
}
if (crm_strlen_zero(standard)) {
crm_err("Cannot create operation for %s without resource class", name);
goto return_error;
}
ra_caps = pcmk_get_ra_caps(standard);
if (is_set(ra_caps, pcmk_ra_cap_provider) && crm_strlen_zero(provider)) {
crm_err("Cannot create operation for %s without provider", name);
goto return_error;
}
if (crm_strlen_zero(agent)) {
crm_err("Cannot create operation for %s without agent name", name);
goto return_error;
}
if (crm_strlen_zero(action)) {
crm_err("Cannot create operation for %s without operation name", name);
goto return_error;
}
/*
* Sanity checks passed, proceed!
*/
op = calloc(1, sizeof(svc_action_t));
op->opaque = calloc(1, sizeof(svc_action_private_t));
op->rsc = strdup(name);
op->interval_ms = interval_ms;
op->timeout = timeout;
op->standard = expand_resource_class(name, standard, agent);
op->agent = strdup(agent);
op->sequence = ++operations;
op->flags = flags;
op->id = generate_op_key(name, action, interval_ms);
if (is_set(ra_caps, pcmk_ra_cap_status) && safe_str_eq(action, "monitor")) {
op->action = strdup("status");
} else {
op->action = strdup(action);
}
if (is_set(ra_caps, pcmk_ra_cap_provider)) {
op->provider = strdup(provider);
}
if (is_set(ra_caps, pcmk_ra_cap_params)) {
op->params = params;
params = NULL; // so we don't free them in this function
}
if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
op->opaque->exec = crm_strdup_printf("%s/resource.d/%s/%s",
OCF_ROOT_DIR, provider, agent);
op->opaque->args[0] = strdup(op->opaque->exec);
op->opaque->args[1] = strdup(op->action);
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
op->opaque->exec = services__lsb_agent_path(op->agent);
op->opaque->args[0] = strdup(op->opaque->exec);
op->opaque->args[1] = strdup(op->action);
#if SUPPORT_SYSTEMD
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
op->opaque->exec = strdup("systemd-dbus");
#endif
#if SUPPORT_UPSTART
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
op->opaque->exec = strdup("upstart-dbus");
#endif
#if SUPPORT_NAGIOS
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
op->opaque->exec = dup_file_path(op->agent, NAGIOS_PLUGIN_DIR);
op->opaque->args[0] = strdup(op->opaque->exec);
if (safe_str_eq(op->action, "monitor") && (op->interval_ms == 0)) {
/* Invoke --version for a nagios probe */
op->opaque->args[1] = strdup("--version");
} else if (op->params) {
GHashTableIter iter;
char *key = NULL;
char *value = NULL;
int index = 1;
static int args_size = sizeof(op->opaque->args) / sizeof(char *);
g_hash_table_iter_init(&iter, op->params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value) &&
index <= args_size - 3) {
if (safe_str_eq(key, XML_ATTR_CRM_VERSION) || strstr(key, CRM_META "_")) {
continue;
}
op->opaque->args[index++] = crm_strdup_printf("--%s", key);
op->opaque->args[index++] = strdup(value);
}
}
// Nagios actions don't need to keep the parameters
if (op->params != NULL) {
g_hash_table_destroy(op->params);
op->params = NULL;
}
#endif
} else {
crm_err("Unknown resource standard: %s", op->standard);
goto return_error;
}
if(params) {
g_hash_table_destroy(params);
}
return op;
return_error:
if(params) {
g_hash_table_destroy(params);
}
services_action_free(op);
return NULL;
}
svc_action_t *
services_action_create_generic(const char *exec, const char *args[])
{
svc_action_t *op;
unsigned int cur_arg;
op = calloc(1, sizeof(*op));
op->opaque = calloc(1, sizeof(svc_action_private_t));
op->opaque->exec = strdup(exec);
op->opaque->args[0] = strdup(exec);
for (cur_arg = 1; args && args[cur_arg - 1]; cur_arg++) {
op->opaque->args[cur_arg] = strdup(args[cur_arg - 1]);
if (cur_arg == DIMOF(op->opaque->args) - 1) {
crm_err("svc_action_t args list not long enough for '%s' execution request.", exec);
break;
}
}
return op;
}
/*!
* \brief Create an alert agent action
*
* \param[in] id Alert ID
* \param[in] exec Path to alert agent executable
* \param[in] timeout Action timeout
* \param[in] params Parameters to use with action
* \param[in] sequence Action sequence number
* \param[in] cb_data Data to pass to callback function
*
* \return New action on success, NULL on error
* \note It is the caller's responsibility to free cb_data.
* The caller should not free params explicitly.
*/
svc_action_t *
services_alert_create(const char *id, const char *exec, int timeout,
GHashTable *params, int sequence, void *cb_data)
{
svc_action_t *action = services_action_create_generic(exec, NULL);
CRM_ASSERT(action);
action->timeout = timeout;
action->id = strdup(id);
action->params = params;
action->sequence = sequence;
action->cb_data = cb_data;
return action;
}
/*!
* \brief Set the user and group that an action will execute as
*
* \param[in,out] action Action to modify
* \param[in] user Name of user to execute action as
* \param[in] group Name of group to execute action as
*
* \return pcmk_ok on success, -errno otherwise
*
* \note This will have no effect unless the process executing the action runs
* as root, and the action is not a systemd or upstart action.
* We could implement this for systemd by adding User= and Group= to
* [Service] in the override file, but that seems more likely to cause
* problems than be useful.
*/
int
services_action_user(svc_action_t *op, const char *user)
{
CRM_CHECK((op != NULL) && (user != NULL), return -EINVAL);
return crm_user_lookup(user, &(op->opaque->uid), &(op->opaque->gid));
}
/*!
* \brief Execute an alert agent action
*
* \param[in] action Action to execute
* \param[in] cb Function to call when action completes
*
* \return TRUE if the library will free action, FALSE otherwise
*
* \note If this function returns FALSE, it is the caller's responsibility to
* free the action with services_action_free().
*/
gboolean
services_alert_async(svc_action_t *action, void (*cb)(svc_action_t *op))
{
action->synchronous = false;
action->opaque->callback = cb;
return services_os_action_execute(action);
}
#if SUPPORT_DBUS
/*!
* \internal
* \brief Update operation's pending DBus call, unreferencing old one if needed
*
* \param[in,out] op Operation to modify
* \param[in] pending Pending call to set
*/
void
services_set_op_pending(svc_action_t *op, DBusPendingCall *pending)
{
if (op->opaque->pending && (op->opaque->pending != pending)) {
if (pending) {
crm_info("Lost pending %s DBus call (%p)", op->id, op->opaque->pending);
} else {
crm_trace("Done with pending %s DBus call (%p)", op->id, op->opaque->pending);
}
dbus_pending_call_unref(op->opaque->pending);
}
op->opaque->pending = pending;
if (pending) {
crm_trace("Updated pending %s DBus call (%p)", op->id, pending);
} else {
crm_trace("Cleared pending %s DBus call", op->id);
}
}
#endif
void
services_action_cleanup(svc_action_t * op)
{
if ((op == NULL) || (op->opaque == NULL)) {
return;
}
#if SUPPORT_DBUS
if(op->opaque->timerid != 0) {
crm_trace("Removing timer for call %s to %s", op->action, op->rsc);
g_source_remove(op->opaque->timerid);
op->opaque->timerid = 0;
}
if(op->opaque->pending) {
if (dbus_pending_call_get_completed(op->opaque->pending)) {
// This should never be the case
crm_warn("Result of %s op %s was unhandled",
op->standard, op->id);
} else {
crm_debug("Will ignore any result of canceled %s op %s",
op->standard, op->id);
}
dbus_pending_call_cancel(op->opaque->pending);
services_set_op_pending(op, NULL);
}
#endif
if (op->opaque->stderr_gsource) {
mainloop_del_fd(op->opaque->stderr_gsource);
op->opaque->stderr_gsource = NULL;
}
if (op->opaque->stdout_gsource) {
mainloop_del_fd(op->opaque->stdout_gsource);
op->opaque->stdout_gsource = NULL;
}
}
void
services_action_free(svc_action_t * op)
{
unsigned int i;
if (op == NULL) {
return;
}
/* The operation should be removed from all tracking lists by this point.
* If it's not, we have a bug somewhere, so bail. That may lead to a
* memory leak, but it's better than a use-after-free segmentation fault.
*/
CRM_CHECK(g_list_find(inflight_ops, op) == NULL, return);
CRM_CHECK(g_list_find(blocked_ops, op) == NULL, return);
CRM_CHECK((recurring_actions == NULL)
|| (g_hash_table_lookup(recurring_actions, op->id) == NULL),
return);
services_action_cleanup(op);
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
free(op->id);
free(op->opaque->exec);
for (i = 0; i < DIMOF(op->opaque->args); i++) {
free(op->opaque->args[i]);
}
free(op->opaque);
free(op->rsc);
free(op->action);
free(op->standard);
free(op->agent);
free(op->provider);
free(op->stdout_data);
free(op->stderr_data);
if (op->params) {
g_hash_table_destroy(op->params);
op->params = NULL;
}
free(op);
}
gboolean
cancel_recurring_action(svc_action_t * op)
{
crm_info("Cancelling %s operation %s", op->standard, op->id);
if (recurring_actions) {
g_hash_table_remove(recurring_actions, op->id);
}
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
return TRUE;
}
/*!
* \brief Cancel a recurring action
*
* \param[in] name Name of resource that operation is for
* \param[in] action Name of operation to cancel
* \param[in] interval_ms Interval of operation to cancel
*
* \return TRUE if action was successfully cancelled, FALSE otherwise
*/
gboolean
services_action_cancel(const char *name, const char *action, guint interval_ms)
{
gboolean cancelled = FALSE;
char *id = generate_op_key(name, action, interval_ms);
svc_action_t *op = NULL;
/* We can only cancel a recurring action */
init_recurring_actions();
op = g_hash_table_lookup(recurring_actions, id);
if (op == NULL) {
goto done;
}
/* Tell operation_finalize() not to reschedule the operation */
op->cancel = TRUE;
/* Stop tracking it as a recurring operation, and stop its repeat timer */
cancel_recurring_action(op);
/* If the op has a PID, it's an in-flight child process, so kill it.
*
* Whether the kill succeeds or fails, the main loop will send the op to
* operation_finished() (and thus operation_finalize()) when the process
* goes away.
*/
if (op->pid != 0) {
- crm_info("Terminating in-flight op %s (pid %d) early because it was cancelled",
+ crm_info("Terminating in-flight op %s[%d] early because it was cancelled",
id, op->pid);
cancelled = mainloop_child_kill(op->pid);
if (cancelled == FALSE) {
- crm_err("Termination of %s (pid %d) failed", id, op->pid);
+ crm_err("Termination of %s[%d] failed", id, op->pid);
}
goto done;
}
#if SUPPORT_DBUS
// In-flight systemd and upstart ops don't have a pid
if (inflight_systemd_or_upstart(op)) {
inflight_ops = g_list_remove(inflight_ops, op);
/* This will cause any result that comes in later to be discarded, so we
* don't call the callback and free the operation twice.
*/
services_action_cleanup(op);
}
#endif
// The rest of this is essentially equivalent to operation_finalize(),
// except without calling handle_blocked_ops()
// Report operation as cancelled
op->status = PCMK_LRM_OP_CANCELLED;
if (op->opaque->callback) {
op->opaque->callback(op);
}
blocked_ops = g_list_remove(blocked_ops, op);
services_action_free(op);
cancelled = TRUE;
// @TODO Initiate handle_blocked_ops() asynchronously
done:
free(id);
return cancelled;
}
gboolean
services_action_kick(const char *name, const char *action, guint interval_ms)
{
svc_action_t * op = NULL;
char *id = generate_op_key(name, action, interval_ms);
init_recurring_actions();
op = g_hash_table_lookup(recurring_actions, id);
free(id);
if (op == NULL) {
return FALSE;
}
if (op->pid || inflight_systemd_or_upstart(op)) {
return TRUE;
} else {
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
recurring_action_timer(op);
return TRUE;
}
}
/*!
* \internal
* \brief Add a new recurring operation, checking for duplicates
*
* \param[in] op Operation to add
*
* \return TRUE if duplicate found (and reschedule), FALSE otherwise
*/
static gboolean
handle_duplicate_recurring(svc_action_t * op)
{
svc_action_t * dup = NULL;
/* check for duplicates */
dup = g_hash_table_lookup(recurring_actions, op->id);
if (dup && (dup != op)) {
/* update user data */
if (op->opaque->callback) {
dup->opaque->callback = op->opaque->callback;
dup->cb_data = op->cb_data;
op->cb_data = NULL;
}
/* immediately execute the next interval */
if (dup->pid != 0) {
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
recurring_action_timer(dup);
}
/* free the duplicate */
services_action_free(op);
return TRUE;
}
return FALSE;
}
inline static gboolean
action_exec_helper(svc_action_t * op)
{
/* Whether a/synchronous must be decided (op->synchronous) beforehand. */
if (op->standard
&& (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_UPSTART) == 0)) {
#if SUPPORT_UPSTART
return upstart_job_exec(op);
#endif
} else if (op->standard && strcasecmp(op->standard,
PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
#if SUPPORT_SYSTEMD
return systemd_unit_exec(op);
#endif
} else {
return services_os_action_execute(op);
}
/* The 'op' has probably been freed if the execution functions return TRUE
for the asynchronous 'op'. */
/* Avoid using the 'op' in here. */
return FALSE;
}
void
services_add_inflight_op(svc_action_t * op)
{
if (op == NULL) {
return;
}
CRM_ASSERT(op->synchronous == FALSE);
/* keep track of ops that are in-flight to avoid collisions in the same namespace */
if (op->rsc) {
inflight_ops = g_list_append(inflight_ops, op);
}
}
/*!
* \internal
* \brief Stop tracking an operation that completed
*
* \param[in] op Operation to stop tracking
*/
void
services_untrack_op(svc_action_t *op)
{
/* Op is no longer in-flight or blocked */
inflight_ops = g_list_remove(inflight_ops, op);
blocked_ops = g_list_remove(blocked_ops, op);
/* Op is no longer blocking other ops, so check if any need to run */
handle_blocked_ops();
}
gboolean
services_action_async_fork_notify(svc_action_t * op,
void (*action_callback) (svc_action_t *),
void (*action_fork_callback) (svc_action_t *))
{
op->synchronous = false;
if (action_callback) {
op->opaque->callback = action_callback;
}
if (action_fork_callback) {
op->opaque->fork_callback = action_fork_callback;
}
if (op->interval_ms > 0) {
init_recurring_actions();
if (handle_duplicate_recurring(op) == TRUE) {
/* entry rescheduled, dup freed */
/* exit early */
return TRUE;
}
g_hash_table_replace(recurring_actions, op->id, op);
}
if (is_not_set(op->flags, SVC_ACTION_NON_BLOCKED)
&& op->rsc && is_op_blocked(op->rsc)) {
blocked_ops = g_list_append(blocked_ops, op);
return TRUE;
}
return action_exec_helper(op);
}
gboolean
services_action_async(svc_action_t * op,
void (*action_callback) (svc_action_t *))
{
return services_action_async_fork_notify(op, action_callback, NULL);
}
static gboolean processing_blocked_ops = FALSE;
gboolean
is_op_blocked(const char *rsc)
{
GList *gIter = NULL;
svc_action_t *op = NULL;
for (gIter = inflight_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
if (safe_str_eq(op->rsc, rsc)) {
return TRUE;
}
}
return FALSE;
}
static void
handle_blocked_ops(void)
{
GList *executed_ops = NULL;
GList *gIter = NULL;
svc_action_t *op = NULL;
gboolean res = FALSE;
if (processing_blocked_ops) {
/* avoid nested calling of this function */
return;
}
processing_blocked_ops = TRUE;
/* n^2 operation here, but blocked ops are incredibly rare. this list
* will be empty 99% of the time. */
for (gIter = blocked_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
if (is_op_blocked(op->rsc)) {
continue;
}
executed_ops = g_list_append(executed_ops, op);
res = action_exec_helper(op);
if (res == FALSE) {
op->status = PCMK_LRM_OP_ERROR;
/* this can cause this function to be called recursively
* which is why we have processing_blocked_ops static variable */
operation_finalize(op);
}
}
for (gIter = executed_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
blocked_ops = g_list_remove(blocked_ops, op);
}
g_list_free(executed_ops);
processing_blocked_ops = FALSE;
}
#if SUPPORT_NAGIOS
static int
nagios_get_metadata(const char *type, char **output)
{
int rc = pcmk_ok;
FILE *file_strm = NULL;
int start = 0, length = 0, read_len = 0;
char *metadata_file = crm_strdup_printf("%s/%s.xml",
NAGIOS_METADATA_DIR, type);
file_strm = fopen(metadata_file, "r");
if (file_strm == NULL) {
crm_err("Metadata file %s does not exist", metadata_file);
free(metadata_file);
return -EIO;
}
/* see how big the file is */
start = ftell(file_strm);
fseek(file_strm, 0L, SEEK_END);
length = ftell(file_strm);
fseek(file_strm, 0L, start);
CRM_ASSERT(length >= 0);
CRM_ASSERT(start == ftell(file_strm));
if (length <= 0) {
crm_info("%s was not valid", metadata_file);
free(*output);
*output = NULL;
rc = -EIO;
} else {
crm_trace("Reading %d bytes from file", length);
*output = calloc(1, (length + 1));
read_len = fread(*output, 1, length, file_strm);
if (read_len != length) {
crm_err("Calculated and read bytes differ: %d vs. %d",
length, read_len);
free(*output);
*output = NULL;
rc = -EIO;
}
}
fclose(file_strm);
free(metadata_file);
return rc;
}
#endif
static gboolean
action_get_metadata(svc_action_t *op)
{
const char *class = op->standard;
if (op->agent == NULL) {
crm_err("meta-data requested without specifying agent");
return FALSE;
}
if (class == NULL) {
crm_err("meta-data requested for agent %s without specifying class",
op->agent);
return FALSE;
}
if (!strcmp(class, PCMK_RESOURCE_CLASS_SERVICE)) {
class = resources_find_service_class(op->agent);
}
if (class == NULL) {
crm_err("meta-data requested for %s, but could not determine class",
op->agent);
return FALSE;
}
if (safe_str_eq(class, PCMK_RESOURCE_CLASS_LSB)) {
return (services__get_lsb_metadata(op->agent, &op->stdout_data) >= 0);
}
#if SUPPORT_NAGIOS
if (safe_str_eq(class, PCMK_RESOURCE_CLASS_NAGIOS)) {
return (nagios_get_metadata(op->agent, &op->stdout_data) >= 0);
}
#endif
return action_exec_helper(op);
}
gboolean
services_action_sync(svc_action_t * op)
{
gboolean rc = TRUE;
if (op == NULL) {
crm_trace("No operation to execute");
return FALSE;
}
op->synchronous = true;
if (safe_str_eq(op->action, "meta-data")) {
/* Synchronous meta-data operations are handled specially. Since most
* resource classes don't provide any meta-data, it has to be
* synthesized from available information about the agent.
*
* services_action_async() doesn't treat meta-data actions specially, so
* it will result in an error for classes that don't support the action.
*/
rc = action_get_metadata(op);
} else {
rc = action_exec_helper(op);
}
crm_trace(" > " CRM_OP_FMT ": %s = %d",
op->rsc, op->action, op->interval_ms, op->opaque->exec, op->rc);
if (op->stdout_data) {
crm_trace(" > stdout: %s", op->stdout_data);
}
if (op->stderr_data) {
crm_trace(" > stderr: %s", op->stderr_data);
}
return rc;
}
GList *
get_directory_list(const char *root, gboolean files, gboolean executable)
{
return services_os_get_directory_list(root, files, executable);
}
GList *
resources_list_standards(void)
{
GList *standards = NULL;
GList *agents = NULL;
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_OCF));
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_LSB));
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_SERVICE));
#if SUPPORT_SYSTEMD
agents = systemd_unit_listall();
if (agents) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_SYSTEMD));
g_list_free_full(agents, free);
}
#endif
#if SUPPORT_UPSTART
agents = upstart_job_listall();
if (agents) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_UPSTART));
g_list_free_full(agents, free);
}
#endif
#if SUPPORT_NAGIOS
agents = resources_os_list_nagios_agents();
if (agents) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_NAGIOS));
g_list_free_full(agents, free);
}
#endif
return standards;
}
GList *
resources_list_providers(const char *standard)
{
if (is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider)) {
return resources_os_list_ocf_providers();
}
return NULL;
}
GList *
resources_list_agents(const char *standard, const char *provider)
{
if ((standard == NULL)
|| (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0)) {
GList *tmp1;
GList *tmp2;
GList *result = services__list_lsb_agents();
if (standard == NULL) {
tmp1 = result;
tmp2 = resources_os_list_ocf_agents(NULL);
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
}
#if SUPPORT_SYSTEMD
tmp1 = result;
tmp2 = systemd_unit_listall();
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
#endif
#if SUPPORT_UPSTART
tmp1 = result;
tmp2 = upstart_job_listall();
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
#endif
return result;
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
return resources_os_list_ocf_agents(provider);
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
return services__list_lsb_agents();
#if SUPPORT_SYSTEMD
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
return systemd_unit_listall();
#endif
#if SUPPORT_UPSTART
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
return upstart_job_listall();
#endif
#if SUPPORT_NAGIOS
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
return resources_os_list_nagios_agents();
#endif
}
return NULL;
}
gboolean
resources_agent_exists(const char *standard, const char *provider, const char *agent)
{
GList *standards = NULL;
GList *providers = NULL;
GListPtr iter = NULL;
gboolean rc = FALSE;
gboolean has_providers = FALSE;
standards = resources_list_standards();
for (iter = standards; iter != NULL; iter = iter->next) {
if (crm_str_eq(iter->data, standard, TRUE)) {
rc = TRUE;
break;
}
}
if (rc == FALSE) {
goto done;
}
rc = FALSE;
has_providers = is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider);
if (has_providers == TRUE && provider != NULL) {
providers = resources_list_providers(standard);
for (iter = providers; iter != NULL; iter = iter->next) {
if (crm_str_eq(iter->data, provider, TRUE)) {
rc = TRUE;
break;
}
}
} else if (has_providers == FALSE && provider == NULL) {
rc = TRUE;
}
if (rc == FALSE) {
goto done;
}
if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_SERVICE)) {
if (services__lsb_agent_exists(agent)) {
rc = TRUE;
#if SUPPORT_SYSTEMD
} else if (systemd_unit_exists(agent)) {
rc = TRUE;
#endif
#if SUPPORT_UPSTART
} else if (upstart_job_exists(agent)) {
rc = TRUE;
#endif
} else {
rc = FALSE;
}
} else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_OCF)) {
rc = services__ocf_agent_exists(provider, agent);
} else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_LSB)) {
rc = services__lsb_agent_exists(agent);
#if SUPPORT_SYSTEMD
} else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD)) {
rc = systemd_unit_exists(agent);
#endif
#if SUPPORT_UPSTART
} else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART)) {
rc = upstart_job_exists(agent);
#endif
#if SUPPORT_NAGIOS
} else if (safe_str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS)) {
rc = services__nagios_agent_exists(agent);
#endif
} else {
rc = FALSE;
}
done:
g_list_free(standards);
g_list_free(providers);
return rc;
}
diff --git a/lib/services/services_linux.c b/lib/services/services_linux.c
index 97821dfa63..c3154fefd3 100644
--- a/lib/services/services_linux.c
+++ b/lib/services/services_linux.c
@@ -1,1187 +1,1218 @@
/*
- * Copyright 2010-2019 Andrew Beekhof <andrew@beekhof.net>
+ * Copyright 2010-2019 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 <crm_internal.h>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <errno.h>
#include <unistd.h>
#include <dirent.h>
#include <grp.h>
#include <string.h>
#include <sys/time.h>
#include <sys/resource.h>
#include "crm/crm.h"
#include "crm/common/mainloop.h"
#include "crm/services.h"
#include "services_private.h"
#if SUPPORT_CIBSECRETS
# include "crm/common/cib_secrets.h"
#endif
/* We have two alternative ways of handling SIGCHLD when synchronously waiting
* for spawned processes to complete. Both rely on polling a file descriptor to
* discover SIGCHLD events.
*
* If sys/signalfd.h is available (e.g. on Linux), we call signalfd() to
* generate the file descriptor. Otherwise, we use the "self-pipe trick"
* (opening a pipe and writing a byte to it when SIGCHLD is received).
*/
#ifdef HAVE_SYS_SIGNALFD_H
// signalfd() implementation
#include <sys/signalfd.h>
// Everything needed to manage SIGCHLD handling
struct sigchld_data_s {
sigset_t mask; // Signals to block now (including SIGCHLD)
sigset_t old_mask; // Previous set of blocked signals
};
// Initialize SIGCHLD data and prepare for use
static void
sigchld_setup(struct sigchld_data_s *data)
{
sigemptyset(&(data->mask));
sigaddset(&(data->mask), SIGCHLD);
sigemptyset(&(data->old_mask));
// Block SIGCHLD (saving previous set of blocked signals to restore later)
if (sigprocmask(SIG_BLOCK, &(data->mask), &(data->old_mask)) < 0) {
- crm_perror(LOG_ERR, "sigprocmask() failed to block sigchld");
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=sigprocmask", pcmk_strerror(errno));
}
}
// Get a file descriptor suitable for polling for SIGCHLD events
static int
sigchld_open(struct sigchld_data_s *data)
{
int fd;
CRM_CHECK(data != NULL, return -1);
fd = signalfd(-1, &(data->mask), SFD_NONBLOCK);
if (fd < 0) {
- crm_perror(LOG_ERR, "signalfd() failed");
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=signalfd", pcmk_strerror(errno));
}
return fd;
}
// Close a file descriptor returned by sigchld_open()
static void
sigchld_close(int fd)
{
if (fd > 0) {
close(fd);
}
}
// Return true if SIGCHLD was received from polled fd
static bool
sigchld_received(int fd)
{
struct signalfd_siginfo fdsi;
ssize_t s;
s = read(fd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
- crm_perror(LOG_ERR, "Read from signal fd %d failed", fd);
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=read", pcmk_strerror(errno));
} else if (fdsi.ssi_signo == SIGCHLD) {
return true;
}
return false;
}
// Do anything needed after done waiting for SIGCHLD
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the original set of blocked signals
if ((sigismember(&(data->old_mask), SIGCHLD) == 0)
&& (sigprocmask(SIG_UNBLOCK, &(data->mask), NULL) < 0)) {
- crm_perror(LOG_ERR, "sigprocmask() failed to unblock sigchld");
+ crm_warn("Could not clean up after child process completion: %s",
+ pcmk_strerror(errno));
}
}
#else // HAVE_SYS_SIGNALFD_H not defined
// Self-pipe implementation (see above for function descriptions)
struct sigchld_data_s {
int pipe_fd[2]; // Pipe file descriptors
struct sigaction sa; // Signal handling info (with SIGCHLD)
struct sigaction old_sa; // Previous signal handling info
};
// We need a global to use in the signal handler
volatile struct sigchld_data_s *last_sigchld_data = NULL;
static void
sigchld_handler()
{
// We received a SIGCHLD, so trigger pipe polling
if ((last_sigchld_data != NULL)
&& (last_sigchld_data->pipe_fd[1] >= 0)
&& (write(last_sigchld_data->pipe_fd[1], "", 1) == -1)) {
- crm_perror(LOG_TRACE, "Could not poke SIGCHLD self-pipe");
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=write", pcmk_strerror(errno));
}
}
static void
sigchld_setup(struct sigchld_data_s *data)
{
int rc;
data->pipe_fd[0] = data->pipe_fd[1] = -1;
if (pipe(data->pipe_fd) == -1) {
- crm_perror(LOG_ERR, "pipe() failed");
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=pipe", pcmk_strerror(errno));
}
rc = crm_set_nonblocking(data->pipe_fd[0]);
if (rc < 0) {
crm_warn("Could not set pipe input non-blocking: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
}
rc = crm_set_nonblocking(data->pipe_fd[1]);
if (rc < 0) {
crm_warn("Could not set pipe output non-blocking: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
}
// Set SIGCHLD handler
data->sa.sa_handler = sigchld_handler;
data->sa.sa_flags = 0;
sigemptyset(&(data->sa.sa_mask));
if (sigaction(SIGCHLD, &(data->sa), &(data->old_sa)) < 0) {
- crm_perror(LOG_ERR, "sigaction() failed to set sigchld handler");
+ crm_err("Wait for child process completion failed: %s "
+ CRM_XS " source=sigaction", pcmk_strerror(errno));
}
// Remember data for use in signal handler
last_sigchld_data = data;
}
static int
sigchld_open(struct sigchld_data_s *data)
{
CRM_CHECK(data != NULL, return -1);
return data->pipe_fd[0];
}
static void
sigchld_close(int fd)
{
// Pipe will be closed in sigchld_cleanup()
return;
}
static bool
sigchld_received(int fd)
{
char ch;
// Clear out the self-pipe
while (read(fd, &ch, 1) == 1) /*omit*/;
return true;
}
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the previous SIGCHLD handler
if (sigaction(SIGCHLD, &(data->old_sa), NULL) < 0) {
- crm_perror(LOG_ERR, "sigaction() failed to remove sigchld handler");
+ crm_warn("Could not clean up after child process completion: %s",
+ pcmk_strerror(errno));
}
// Close the pipe
if (data->pipe_fd[0] >= 0) {
close(data->pipe_fd[0]);
data->pipe_fd[0] = -1;
}
if (data->pipe_fd[1] >= 0) {
close(data->pipe_fd[1]);
data->pipe_fd[1] = -1;
}
}
#endif
static gboolean
svc_read_output(int fd, svc_action_t * op, bool is_stderr)
{
char *data = NULL;
int rc = 0, len = 0;
char buf[500];
static const size_t buf_read_len = sizeof(buf) - 1;
if (fd < 0) {
crm_trace("No fd for %s", op->id);
return FALSE;
}
if (is_stderr && op->stderr_data) {
len = strlen(op->stderr_data);
data = op->stderr_data;
crm_trace("Reading %s stderr into offset %d", op->id, len);
} else if (is_stderr == FALSE && op->stdout_data) {
len = strlen(op->stdout_data);
data = op->stdout_data;
crm_trace("Reading %s stdout into offset %d", op->id, len);
} else {
crm_trace("Reading %s %s into offset %d", op->id, is_stderr?"stderr":"stdout", len);
}
do {
rc = read(fd, buf, buf_read_len);
if (rc > 0) {
buf[rc] = 0;
crm_trace("Got %d chars: %.80s", rc, buf);
data = realloc_safe(data, len + rc + 1);
len += sprintf(data + len, "%s", buf);
} else if (errno != EINTR) {
/* error or EOF
* Cleanup happens in pipe_done()
*/
rc = FALSE;
break;
}
} while (rc == buf_read_len || rc < 0);
if (is_stderr) {
op->stderr_data = data;
} else {
op->stdout_data = data;
}
return rc;
}
static int
dispatch_stdout(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
static int
dispatch_stderr(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
static void
pipe_out_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
crm_trace("%p", op);
op->opaque->stdout_gsource = NULL;
if (op->opaque->stdout_fd > STDOUT_FILENO) {
close(op->opaque->stdout_fd);
}
op->opaque->stdout_fd = -1;
}
static void
pipe_err_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
op->opaque->stderr_gsource = NULL;
if (op->opaque->stderr_fd > STDERR_FILENO) {
close(op->opaque->stderr_fd);
}
op->opaque->stderr_fd = -1;
}
static struct mainloop_fd_callbacks stdout_callbacks = {
.dispatch = dispatch_stdout,
.destroy = pipe_out_done,
};
static struct mainloop_fd_callbacks stderr_callbacks = {
.dispatch = dispatch_stderr,
.destroy = pipe_err_done,
};
static void
set_ocf_env(const char *key, const char *value, gpointer user_data)
{
if (setenv(key, value, 1) != 0) {
crm_perror(LOG_ERR, "setenv failed for key:%s and value:%s", key, value);
}
}
static void
set_ocf_env_with_prefix(gpointer key, gpointer value, gpointer user_data)
{
char buffer[500];
snprintf(buffer, sizeof(buffer), "OCF_RESKEY_%s", (char *)key);
set_ocf_env(buffer, value, user_data);
}
static void
set_alert_env(gpointer key, gpointer value, gpointer user_data)
{
int rc;
if (value != NULL) {
rc = setenv(key, value, 1);
} else {
rc = unsetenv(key);
}
if (rc < 0) {
crm_perror(LOG_ERR, "setenv %s=%s",
(char*)key, (value? (char*)value : ""));
} else {
crm_trace("setenv %s=%s", (char*)key, (value? (char*)value : ""));
}
}
/*!
* \internal
* \brief Add environment variables suitable for an action
*
* \param[in] op Action to use
*/
static void
add_action_env_vars(const svc_action_t *op)
{
void (*env_setter)(gpointer, gpointer, gpointer) = NULL;
if (op->agent == NULL) {
env_setter = set_alert_env; /* we deal with alert handler */
} else if (safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF)) {
env_setter = set_ocf_env_with_prefix;
}
if (env_setter != NULL && op->params != NULL) {
g_hash_table_foreach(op->params, env_setter, NULL);
}
if (env_setter == NULL || env_setter == set_alert_env) {
return;
}
set_ocf_env("OCF_RA_VERSION_MAJOR", "1", NULL);
set_ocf_env("OCF_RA_VERSION_MINOR", "0", NULL);
set_ocf_env("OCF_ROOT", OCF_ROOT_DIR, NULL);
set_ocf_env("OCF_EXIT_REASON_PREFIX", PCMK_OCF_REASON_PREFIX, NULL);
if (op->rsc) {
set_ocf_env("OCF_RESOURCE_INSTANCE", op->rsc, NULL);
}
if (op->agent != NULL) {
set_ocf_env("OCF_RESOURCE_TYPE", op->agent, NULL);
}
/* Notes: this is not added to specification yet. Sept 10,2004 */
if (op->provider != NULL) {
set_ocf_env("OCF_RESOURCE_PROVIDER", op->provider, NULL);
}
}
static void
pipe_in_single_parameter(gpointer key, gpointer value, gpointer user_data)
{
svc_action_t *op = user_data;
char *buffer = crm_strdup_printf("%s=%s\n", (char *)key, (char *) value);
int ret, total = 0, len = strlen(buffer);
do {
errno = 0;
ret = write(op->opaque->stdin_fd, buffer + total, len - total);
if (ret > 0) {
total += ret;
}
} while ((errno == EINTR) && (total < len));
free(buffer);
}
/*!
* \internal
* \brief Pipe parameters in via stdin for action
*
* \param[in] op Action to use
*/
static void
pipe_in_action_stdin_parameters(const svc_action_t *op)
{
crm_debug("sending args");
if (op->params) {
g_hash_table_foreach(op->params, pipe_in_single_parameter, (gpointer) op);
}
}
gboolean
recurring_action_timer(gpointer data)
{
svc_action_t *op = data;
crm_debug("Scheduling another invocation of %s", op->id);
/* Clean out the old result */
free(op->stdout_data);
op->stdout_data = NULL;
free(op->stderr_data);
op->stderr_data = NULL;
op->opaque->repeat_timer = 0;
services_action_async(op, NULL);
return FALSE;
}
/* Returns FALSE if 'op' should be free'd by the caller */
gboolean
operation_finalize(svc_action_t * op)
{
int recurring = 0;
if (op->interval_ms) {
if (op->cancel) {
op->status = PCMK_LRM_OP_CANCELLED;
cancel_recurring_action(op);
} else {
recurring = 1;
op->opaque->repeat_timer = g_timeout_add(op->interval_ms,
recurring_action_timer, (void *)op);
}
}
if (op->opaque->callback) {
op->opaque->callback(op);
}
op->pid = 0;
services_untrack_op(op);
if (!recurring && op->synchronous == FALSE) {
/*
* If this is a recurring action, do not free explicitly.
* It will get freed whenever the action gets cancelled.
*/
services_action_free(op);
return TRUE;
}
services_action_cleanup(op);
return FALSE;
}
+static void
+close_op_input(svc_action_t *op)
+{
+ if (op->opaque->stdin_fd >= 0) {
+ close(op->opaque->stdin_fd);
+ }
+}
+
+static void
+finish_op_output(svc_action_t *op, bool is_stderr)
+{
+ mainloop_io_t **source;
+ int fd;
+
+ if (is_stderr) {
+ source = &(op->opaque->stderr_gsource);
+ fd = op->opaque->stderr_fd;
+ } else {
+ source = &(op->opaque->stdout_gsource);
+ fd = op->opaque->stdout_fd;
+ }
+
+ if (op->synchronous || *source) {
+ crm_trace("Finish reading %s[%d] %s",
+ op->id, op->pid, (is_stderr? "stdout" : "stderr"));
+ svc_read_output(fd, op, is_stderr);
+ if (op->synchronous) {
+ close(fd);
+ } else {
+ mainloop_del_fd(*source);
+ *source = NULL;
+ }
+ }
+}
+
+// Log an operation's stdout and stderr
+static void
+log_op_output(svc_action_t *op)
+{
+ char *prefix = crm_strdup_printf("%s[%d] error output", op->id, op->pid);
+
+ crm_log_output(LOG_NOTICE, prefix, op->stderr_data);
+ strcpy(prefix + strlen(prefix) - strlen("error output"), "output");
+ crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
+ free(prefix);
+}
+
static void
operation_finished(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)
{
svc_action_t *op = mainloop_child_userdata(p);
- char *prefix = crm_strdup_printf("%s:%d", op->id, op->pid);
mainloop_clear_child_userdata(p);
- op->status = PCMK_LRM_OP_DONE;
CRM_ASSERT(op->pid == pid);
- crm_trace("%s %p %p", prefix, op->opaque->stderr_gsource, op->opaque->stdout_gsource);
- if (op->opaque->stderr_gsource) {
- /* Make sure we have read everything from the buffer.
- * Depending on the priority mainloop gives the fd, operation_finished
- * could occur before all the reads are done. Force the read now.*/
- crm_trace("%s dispatching stderr", prefix);
- dispatch_stderr(op);
- crm_trace("%s: %p", op->id, op->stderr_data);
- mainloop_del_fd(op->opaque->stderr_gsource);
- op->opaque->stderr_gsource = NULL;
- }
-
- if (op->opaque->stdout_gsource) {
- /* Make sure we have read everything from the buffer.
- * Depending on the priority mainloop gives the fd, operation_finished
- * could occur before all the reads are done. Force the read now.*/
- crm_trace("%s dispatching stdout", prefix);
- dispatch_stdout(op);
- crm_trace("%s: %p", op->id, op->stdout_data);
- mainloop_del_fd(op->opaque->stdout_gsource);
- op->opaque->stdout_gsource = NULL;
- }
+ /* Depending on the priority the mainloop gives the stdout and stderr
+ * file descriptors, this function could be called before everything has
+ * been read from them, so force a final read now.
+ */
+ finish_op_output(op, true);
+ finish_op_output(op, false);
- if (op->opaque->stdin_fd >= 0) {
- close(op->opaque->stdin_fd);
- }
+ close_op_input(op);
- if (signo) {
- if (mainloop_child_timeout(p)) {
- crm_warn("%s - timed out after %dms", prefix, op->timeout);
- op->status = PCMK_LRM_OP_TIMEOUT;
- op->rc = PCMK_OCF_TIMEOUT;
+ if (signo == 0) {
+ crm_debug("%s[%d] exited with status %d", op->id, op->pid, exitcode);
+ op->status = PCMK_LRM_OP_DONE;
+ op->rc = exitcode;
- } else if (op->cancel) {
- /* If an in-flight recurring operation was killed because it was
- * cancelled, don't treat that as a failure.
- */
- crm_info("%s - terminated with signal %d", prefix, signo);
- op->status = PCMK_LRM_OP_CANCELLED;
- op->rc = PCMK_OCF_OK;
+ } else if (mainloop_child_timeout(p)) {
+ crm_warn("%s[%d] timed out after %dms", op->id, op->pid, op->timeout);
+ op->status = PCMK_LRM_OP_TIMEOUT;
+ op->rc = PCMK_OCF_TIMEOUT;
- } else {
- crm_warn("%s - terminated with signal %d", prefix, signo);
- op->status = PCMK_LRM_OP_ERROR;
- op->rc = PCMK_OCF_SIGNAL;
- }
+ } else if (op->cancel) {
+ /* If an in-flight recurring operation was killed because it was
+ * cancelled, don't treat that as a failure.
+ */
+ crm_info("%s[%d] terminated with signal: %s " CRM_XS " (%d)",
+ op->id, op->pid, strsignal(signo), signo);
+ op->status = PCMK_LRM_OP_CANCELLED;
+ op->rc = PCMK_OCF_OK;
} else {
- op->rc = exitcode;
- crm_debug("%s - exited with rc=%d", prefix, exitcode);
+ crm_warn("%s[%d] terminated with signal: %s " CRM_XS " (%d)",
+ op->id, op->pid, strsignal(signo), signo);
+ op->status = PCMK_LRM_OP_ERROR;
+ op->rc = PCMK_OCF_SIGNAL;
}
- free(prefix);
- prefix = crm_strdup_printf("%s:%d:stderr", op->id, op->pid);
- crm_log_output(LOG_NOTICE, prefix, op->stderr_data);
-
- free(prefix);
- prefix = crm_strdup_printf("%s:%d:stdout", op->id, op->pid);
- crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
-
- free(prefix);
+ log_op_output(op);
operation_finalize(op);
}
/*!
* \internal
* \brief Set operation rc and status per errno from stat(), fork() or execvp()
*
* \param[in,out] op Operation to set rc and status for
* \param[in] error Value of errno after system call
*
* \return void
*/
static void
services_handle_exec_error(svc_action_t * op, int error)
{
int rc_not_installed, rc_insufficient_priv, rc_exec_error;
/* Mimic the return codes for each standard as that's what we'll convert back from in get_uniform_rc() */
if (safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB)
&& safe_str_eq(op->action, "status")) {
rc_not_installed = PCMK_LSB_STATUS_NOT_INSTALLED;
rc_insufficient_priv = PCMK_LSB_STATUS_INSUFFICIENT_PRIV;
rc_exec_error = PCMK_LSB_STATUS_UNKNOWN;
#if SUPPORT_NAGIOS
} else if (safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_NAGIOS)) {
rc_not_installed = NAGIOS_NOT_INSTALLED;
rc_insufficient_priv = NAGIOS_INSUFFICIENT_PRIV;
rc_exec_error = PCMK_OCF_EXEC_ERROR;
#endif
} else {
rc_not_installed = PCMK_OCF_NOT_INSTALLED;
rc_insufficient_priv = PCMK_OCF_INSUFFICIENT_PRIV;
rc_exec_error = PCMK_OCF_EXEC_ERROR;
}
switch (error) { /* see execve(2), stat(2) and fork(2) */
case ENOENT: /* No such file or directory */
case EISDIR: /* Is a directory */
case ENOTDIR: /* Path component is not a directory */
case EINVAL: /* Invalid executable format */
case ENOEXEC: /* Invalid executable format */
op->rc = rc_not_installed;
op->status = PCMK_LRM_OP_NOT_INSTALLED;
break;
case EACCES: /* permission denied (various errors) */
case EPERM: /* permission denied (various errors) */
op->rc = rc_insufficient_priv;
op->status = PCMK_LRM_OP_ERROR;
break;
default:
op->rc = rc_exec_error;
op->status = PCMK_LRM_OP_ERROR;
}
}
static void
action_launch_child(svc_action_t *op)
{
/* SIGPIPE is ignored (which is different from signal blocking) by the gnutls library.
* Depending on the libqb version in use, libqb may set SIGPIPE to be ignored as well.
* We do not want this to be inherited by the child process. By resetting this the signal
* to the default behavior, we avoid some potential odd problems that occur during OCF
* scripts when SIGPIPE is ignored by the environment. */
signal(SIGPIPE, SIG_DFL);
#if defined(HAVE_SCHED_SETSCHEDULER)
if (sched_getscheduler(0) != SCHED_OTHER) {
struct sched_param sp;
memset(&sp, 0, sizeof(sp));
sp.sched_priority = 0;
if (sched_setscheduler(0, SCHED_OTHER, &sp) == -1) {
crm_perror(LOG_ERR, "Could not reset scheduling policy to SCHED_OTHER for %s", op->id);
}
}
#endif
if (setpriority(PRIO_PROCESS, 0, 0) == -1) {
crm_perror(LOG_ERR, "Could not reset process priority to 0 for %s", op->id);
}
/* Man: The call setpgrp() is equivalent to setpgid(0,0)
* _and_ compiles on BSD variants too
* need to investigate if it works the same too.
*/
setpgid(0, 0);
pcmk__close_fds_in_child(false);
#if SUPPORT_CIBSECRETS
if (replace_secret_params(op->rsc, op->params) < 0) {
/* replacing secrets failed! */
if (safe_str_eq(op->action,"stop")) {
/* don't fail on stop! */
crm_info("proceeding with the stop operation for %s", op->rsc);
} else {
crm_err("failed to get secrets for %s, "
"considering resource not configured", op->rsc);
_exit(PCMK_OCF_NOT_CONFIGURED);
}
}
#endif
add_action_env_vars(op);
/* Become the desired user */
if (op->opaque->uid && (geteuid() == 0)) {
// If requested, set effective group
if (op->opaque->gid && (setgid(op->opaque->gid) < 0)) {
crm_perror(LOG_ERR, "Could not set child group to %d", op->opaque->gid);
_exit(PCMK_OCF_NOT_CONFIGURED);
}
// Erase supplementary group list
// (We could do initgroups() if we kept a copy of the username)
if (setgroups(0, NULL) < 0) {
crm_perror(LOG_ERR, "Could not set child groups");
_exit(PCMK_OCF_NOT_CONFIGURED);
}
// Set effective user
if (setuid(op->opaque->uid) < 0) {
crm_perror(LOG_ERR, "setting user to %d", op->opaque->uid);
_exit(PCMK_OCF_NOT_CONFIGURED);
}
}
/* execute the RA */
execvp(op->opaque->exec, op->opaque->args);
/* Most cases should have been already handled by stat() */
services_handle_exec_error(op, errno);
_exit(op->rc);
}
static void
action_synced_wait(svc_action_t *op, struct sigchld_data_s *data)
{
int status = 0;
int timeout = op->timeout;
time_t start = -1;
struct pollfd fds[3];
int wait_rc = 0;
fds[0].fd = op->opaque->stdout_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
fds[1].fd = op->opaque->stderr_fd;
fds[1].events = POLLIN;
fds[1].revents = 0;
fds[2].fd = sigchld_open(data);
fds[2].events = POLLIN;
fds[2].revents = 0;
- crm_trace("Waiting for %d", op->pid);
+ crm_trace("Waiting for %s[%d]", op->id, op->pid);
start = time(NULL);
do {
int poll_rc = poll(fds, 3, timeout);
if (poll_rc > 0) {
if (fds[0].revents & POLLIN) {
svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
if (fds[1].revents & POLLIN) {
svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
if ((fds[2].revents & POLLIN) && sigchld_received(fds[2].fd)) {
wait_rc = waitpid(op->pid, &status, WNOHANG);
- if (wait_rc > 0) {
+ if ((wait_rc > 0) || ((wait_rc < 0) && (errno == ECHILD))) {
+ // Child process exited or doesn't exist
break;
} else if (wait_rc < 0) {
- if (errno == ECHILD) {
- /* Here, don't dare to kill and bail out... */
- break;
-
- } else {
- /* ...otherwise pretend process still runs. */
- wait_rc = 0;
- }
- crm_perror(LOG_ERR, "waitpid() for %d failed", op->pid);
+ crm_warn("Wait for completion of %s[%d] failed: %s "
+ CRM_XS " source=waitpid",
+ op->id, op->pid, pcmk_strerror(errno));
+ wait_rc = 0; // Act as if process is still running
}
}
} else if (poll_rc == 0) {
+ // Poll timed out with no descriptors ready
timeout = 0;
break;
- } else if (poll_rc < 0) {
- if (errno != EINTR) {
- crm_perror(LOG_ERR, "poll() failed");
- break;
- }
+ } else if ((poll_rc < 0) && (errno != EINTR)) {
+ crm_err("Wait for completion of %s[%d] failed: %s "
+ CRM_XS " source=poll",
+ op->id, op->pid, pcmk_strerror(errno));
+ break;
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
- crm_trace("Child done: %d", op->pid);
+ crm_trace("Stopped waiting for %s[%d]", op->id, op->pid);
if (wait_rc <= 0) {
op->rc = PCMK_OCF_UNKNOWN_ERROR;
if (op->timeout > 0 && timeout <= 0) {
op->status = PCMK_LRM_OP_TIMEOUT;
- crm_warn("%s:%d - timed out after %dms", op->id, op->pid, op->timeout);
+ crm_warn("%s[%d] timed out after %dms",
+ op->id, op->pid, op->timeout);
} else {
op->status = PCMK_LRM_OP_ERROR;
}
/* If only child hasn't been successfully waited for, yet.
This is to limit killing wrong target a bit more. */
if (wait_rc == 0 && waitpid(op->pid, &status, WNOHANG) == 0) {
if (kill(op->pid, SIGKILL)) {
- crm_err("kill(%d, KILL) failed: %d", op->pid, errno);
+ crm_warn("Could not kill rogue child %s[%d]: %s",
+ op->id, op->pid, pcmk_strerror(errno));
}
/* Safe to skip WNOHANG here as we sent non-ignorable signal. */
while (waitpid(op->pid, &status, 0) == (pid_t) -1 && errno == EINTR) /*omit*/;
}
} else if (WIFEXITED(status)) {
op->status = PCMK_LRM_OP_DONE;
op->rc = WEXITSTATUS(status);
- crm_info("Managed %s process %d exited with rc=%d", op->id, op->pid, op->rc);
+ crm_info("%s[%d] exited with status %d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
op->status = PCMK_LRM_OP_ERROR;
- crm_err("Managed %s process %d exited with signal=%d", op->id, op->pid, signo);
+ crm_err("%s[%d] terminated with signal: %s " CRM_XS " (%d)",
+ op->id, op->pid, strsignal(signo), signo);
}
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
- crm_err("Managed %s process %d dumped core", op->id, op->pid);
+ crm_err("%s[%d] dumped core", op->id, op->pid);
}
#endif
- svc_read_output(op->opaque->stdout_fd, op, FALSE);
- svc_read_output(op->opaque->stderr_fd, op, TRUE);
-
- close(op->opaque->stdout_fd);
- close(op->opaque->stderr_fd);
- if (op->opaque->stdin_fd >= 0) {
- close(op->opaque->stdin_fd);
- }
+ finish_op_output(op, true);
+ finish_op_output(op, false);
+ close_op_input(op);
sigchld_close(fds[2].fd);
}
/* For an asynchronous 'op', returns FALSE if 'op' should be free'd by the caller */
/* For a synchronous 'op', returns FALSE if 'op' fails */
gboolean
services_os_action_execute(svc_action_t * op)
{
int stdout_fd[2];
int stderr_fd[2];
int stdin_fd[2] = {-1, -1};
int rc;
struct stat st;
struct sigchld_data_s data;
/* Fail fast */
if(stat(op->opaque->exec, &st) != 0) {
rc = errno;
- crm_warn("Cannot execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
+ crm_warn("Cannot execute '%s': %s " CRM_XS " stat rc=%d",
+ op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!op->synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (pipe(stdout_fd) < 0) {
rc = errno;
-
- crm_err("pipe(stdout_fd) failed. '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
-
+ crm_err("Cannot execute '%s': %s " CRM_XS " pipe(stdout) rc=%d",
+ op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!op->synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (pipe(stderr_fd) < 0) {
rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
- crm_err("pipe(stderr_fd) failed. '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
-
+ crm_err("Cannot execute '%s': %s " CRM_XS " pipe(stderr) rc=%d",
+ op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!op->synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (safe_str_eq(op->standard, PCMK_RESOURCE_CLASS_STONITH)) {
if (pipe(stdin_fd) < 0) {
rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
close(stderr_fd[0]);
close(stderr_fd[1]);
- crm_err("pipe(stdin_fd) failed. '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
-
+ crm_err("Cannot execute '%s': %s " CRM_XS " pipe(stdin) rc=%d",
+ op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!op->synchronous) {
return operation_finalize(op);
}
return FALSE;
}
}
if (op->synchronous) {
sigchld_setup(&data);
}
op->pid = fork();
switch (op->pid) {
case -1:
rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
close(stderr_fd[0]);
close(stderr_fd[1]);
if (stdin_fd[0] >= 0) {
close(stdin_fd[0]);
close(stdin_fd[1]);
}
- crm_err("Could not execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
+ crm_err("Cannot execute '%s': %s " CRM_XS " fork rc=%d",
+ op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!op->synchronous) {
return operation_finalize(op);
}
sigchld_cleanup(&data);
return FALSE;
case 0: /* Child */
close(stdout_fd[0]);
close(stderr_fd[0]);
if (stdin_fd[1] >= 0) {
close(stdin_fd[1]);
}
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
- crm_err("dup2() failed (stdout)");
+ crm_warn("Can't redirect output from '%s': %s "
+ CRM_XS " errno=%d",
+ op->opaque->exec, pcmk_strerror(errno), errno);
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
- crm_err("dup2() failed (stderr)");
+ crm_warn("Can't redirect error output from '%s': %s "
+ CRM_XS " errno=%d",
+ op->opaque->exec, pcmk_strerror(errno), errno);
}
close(stderr_fd[1]);
}
if ((stdin_fd[0] >= 0) &&
(STDIN_FILENO != stdin_fd[0])) {
if (dup2(stdin_fd[0], STDIN_FILENO) != STDIN_FILENO) {
- crm_err("dup2() failed (stdin)");
+ crm_warn("Can't redirect input to '%s': %s "
+ CRM_XS " errno=%d",
+ op->opaque->exec, pcmk_strerror(errno), errno);
}
close(stdin_fd[0]);
}
if (op->synchronous) {
sigchld_cleanup(&data);
}
action_launch_child(op);
CRM_ASSERT(0); /* action_launch_child is effectively noreturn */
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
if (stdin_fd[0] >= 0) {
close(stdin_fd[0]);
}
op->opaque->stdout_fd = stdout_fd[0];
rc = crm_set_nonblocking(op->opaque->stdout_fd);
if (rc < 0) {
- crm_warn("Could not set child output non-blocking: %s "
+ crm_warn("Could not set '%s' output non-blocking: %s "
CRM_XS " rc=%d",
- pcmk_strerror(rc), rc);
+ op->opaque->exec, pcmk_strerror(rc), rc);
}
op->opaque->stderr_fd = stderr_fd[0];
rc = crm_set_nonblocking(op->opaque->stderr_fd);
if (rc < 0) {
- crm_warn("Could not set child error output non-blocking: %s "
+ crm_warn("Could not set '%s' error output non-blocking: %s "
CRM_XS " rc=%d",
- pcmk_strerror(rc), rc);
+ op->opaque->exec, pcmk_strerror(rc), rc);
}
op->opaque->stdin_fd = stdin_fd[1];
if (op->opaque->stdin_fd >= 0) {
// using buffer behind non-blocking-fd here - that could be improved
// as long as no other standard uses stdin_fd assume stonith
rc = crm_set_nonblocking(op->opaque->stdin_fd);
if (rc < 0) {
- crm_warn("Could not set child input non-blocking: %s "
- CRM_XS " fd=%d,rc=%d",
+ crm_warn("Could not set '%s' input non-blocking: %s "
+ CRM_XS " fd=%d,rc=%d", op->opaque->exec,
pcmk_strerror(rc), op->opaque->stdin_fd, rc);
}
pipe_in_action_stdin_parameters(op);
// as long as we are handling parameters directly in here just close
close(op->opaque->stdin_fd);
op->opaque->stdin_fd = -1;
}
// after fds are setup properly and before we plug anything into mainloop
if (op->opaque->fork_callback) {
op->opaque->fork_callback(op);
}
if (op->synchronous) {
action_synced_wait(op, &data);
sigchld_cleanup(&data);
} else {
-
- crm_trace("Async waiting for %d - %s", op->pid, op->opaque->exec);
+ crm_trace("Waiting async for '%s'[%d]", op->opaque->exec, op->pid);
mainloop_child_add_with_flags(op->pid,
op->timeout,
op->id,
op,
(op->flags & SVC_ACTION_LEAVE_GROUP) ? mainloop_leave_pid_group : 0,
operation_finished);
op->opaque->stdout_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stdout_fd, op, &stdout_callbacks);
op->opaque->stderr_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stderr_fd, op, &stderr_callbacks);
services_add_inflight_op(op);
}
return TRUE;
}
GList *
services_os_get_directory_list(const char *root, gboolean files, gboolean executable)
{
GList *list = NULL;
struct dirent **namelist;
int entries = 0, lpc = 0;
char buffer[PATH_MAX];
entries = scandir(root, &namelist, NULL, alphasort);
if (entries <= 0) {
return list;
}
for (lpc = 0; lpc < entries; lpc++) {
struct stat sb;
if ('.' == namelist[lpc]->d_name[0]) {
free(namelist[lpc]);
continue;
}
snprintf(buffer, sizeof(buffer), "%s/%s", root, namelist[lpc]->d_name);
if (stat(buffer, &sb)) {
continue;
}
if (S_ISDIR(sb.st_mode)) {
if (files) {
free(namelist[lpc]);
continue;
}
} else if (S_ISREG(sb.st_mode)) {
if (files == FALSE) {
free(namelist[lpc]);
continue;
} else if (executable
&& (sb.st_mode & S_IXUSR) == 0
&& (sb.st_mode & S_IXGRP) == 0 && (sb.st_mode & S_IXOTH) == 0) {
free(namelist[lpc]);
continue;
}
}
list = g_list_append(list, strdup(namelist[lpc]->d_name));
free(namelist[lpc]);
}
free(namelist);
return list;
}
GList *
resources_os_list_ocf_providers(void)
{
return get_directory_list(OCF_ROOT_DIR "/resource.d", FALSE, TRUE);
}
GList *
resources_os_list_ocf_agents(const char *provider)
{
GList *gIter = NULL;
GList *result = NULL;
GList *providers = NULL;
if (provider) {
char buffer[500];
snprintf(buffer, sizeof(buffer), "%s/resource.d/%s", OCF_ROOT_DIR, provider);
return get_directory_list(buffer, TRUE, TRUE);
}
providers = resources_os_list_ocf_providers();
for (gIter = providers; gIter != NULL; gIter = gIter->next) {
GList *tmp1 = result;
GList *tmp2 = resources_os_list_ocf_agents(gIter->data);
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
}
g_list_free_full(providers, free);
return result;
}
gboolean
services__ocf_agent_exists(const char *provider, const char *agent)
{
char *buf = NULL;
gboolean rc = FALSE;
struct stat st;
if (provider == NULL || agent == NULL) {
return rc;
}
buf = crm_strdup_printf(OCF_ROOT_DIR "/resource.d/%s/%s", provider, agent);
if (stat(buf, &st) == 0) {
rc = TRUE;
}
free(buf);
return rc;
}
#if SUPPORT_NAGIOS
GList *
resources_os_list_nagios_agents(void)
{
GList *plugin_list = NULL;
GList *result = NULL;
GList *gIter = NULL;
plugin_list = get_directory_list(NAGIOS_PLUGIN_DIR, TRUE, TRUE);
/* Make sure both the plugin and its metadata exist */
for (gIter = plugin_list; gIter != NULL; gIter = gIter->next) {
const char *plugin = gIter->data;
char *metadata = crm_strdup_printf(NAGIOS_METADATA_DIR "/%s.xml", plugin);
struct stat st;
if (stat(metadata, &st) == 0) {
result = g_list_append(result, strdup(plugin));
}
free(metadata);
}
g_list_free_full(plugin_list, free);
return result;
}
gboolean
services__nagios_agent_exists(const char *name)
{
char *buf = NULL;
gboolean rc = FALSE;
struct stat st;
if (name == NULL) {
return rc;
}
buf = crm_strdup_printf(NAGIOS_PLUGIN_DIR "/%s", name);
if (stat(buf, &st) == 0) {
rc = TRUE;
}
free(buf);
return rc;
}
#endif