Page Menu
Home
ClusterLabs Projects
Search
Configure Global Search
Log In
Files
F3151698
No One
Temporary
Actions
View File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Flag For Later
Award Token
Size
162 KB
Referenced Files
None
Subscribers
None
View Options
diff --git a/src/attr.c b/src/attr.c
index 919b18b..b4468d5 100644
--- a/src/attr.c
+++ b/src/attr.c
@@ -1,477 +1,479 @@
/*
* Copyright (C) 2015 Dejan Muhamedagic <dejan@hello-penguin.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <string.h>
#include "attr.h"
#include "booth.h"
#include "ticket.h"
#include "pacemaker.h"
void print_geostore_usage(void)
{
printf(
"Usage:\n"
" geostore {list|set|get|delete} [-t ticket] [options] attr [value]\n"
"\n"
" list: List all attributes\n"
" set: Set attribute to a value\n"
" get: Get attribute's value\n"
" delete: Delete attribute\n"
"\n"
" -t <ticket> Ticket where attribute resides\n"
" (required, if more than one ticket is configured)\n"
"\n"
"Options:\n"
" -c FILE Specify config file [default " BOOTH_DEFAULT_CONF "]\n"
" Can be a path or just a name without \".conf\" suffix\n"
" -s <site> Connect to a different site\n"
" -h Print this help\n"
"\n"
"Examples:\n"
"\n"
" # geostore list -t ticket-A -s 10.121.8.183\n"
" # geostore set -s 10.121.8.183 sr_status ACTIVE\n"
" # geostore get -t ticket-A -s 10.121.8.183 sr_status\n"
" # geostore delete -s 10.121.8.183 sr_status\n"
"\n"
"See the geostore(8) man page for more details.\n"
);
}
/*
* the client side
*/
/* cl has all the input parameters:
* ticket, attr name, attr value
*/
int test_attr_reply(cmd_result_t reply_code, cmd_request_t cmd)
{
int rv = 0;
const char *op_str = "";
switch (cmd) {
case ATTR_SET: op_str = "set"; break;
case ATTR_GET: op_str = "get"; break;
case ATTR_LIST: op_str = "list"; break;
case ATTR_DEL: op_str = "delete"; break;
default:
log_error("internal error reading reply result!");
return -1;
}
switch (reply_code) {
case RLT_ASYNC:
log_info("%s command sent, result will be returned "
"asynchronously.", op_str);
rv = 0;
break;
case RLT_SYNC_SUCC:
case RLT_SUCCESS:
if (cmd == ATTR_SET)
log_info("%s succeeded!", op_str);
rv = 0;
break;
case RLT_SYNC_FAIL:
log_info("%s failed!", op_str);
rv = -1;
break;
case RLT_INVALID_ARG:
log_error("ticket \"%s\" does not exist",
cl.attr_msg.attr.tkt_id);
rv = 1;
break;
case RLT_NO_SUCH_ATTR:
log_error("attribute \"%s\" not set",
cl.attr_msg.attr.name);
rv = 1;
break;
case RLT_AUTH:
log_error("authentication error");
rv = -1;
break;
default:
log_error("got an error code: %x", rv);
rv = -1;
}
return rv;
}
/* read the server's reply
* need to first get the header which contains the length of the
* reply
* return codes:
* -2: header not received
* -1: header received, but message too short
* >=0: success
*/
static int read_server_reply(
struct booth_transport const *tpt, struct booth_site *site,
char *msg)
{
struct boothc_header *header;
int rv;
int len;
header = (struct boothc_header *)msg;
rv = tpt->recv(site, header, sizeof(*header));
if (rv < 0) {
return -2;
}
len = ntohl(header->length);
rv = tpt->recv(site, msg+len, len-sizeof(*header));
if (rv < 0) {
return -1;
}
return rv;
}
int do_attr_command(struct booth_config *conf_ptr, cmd_request_t cmd)
{
struct booth_site *site = NULL;
struct boothc_header *header;
struct booth_transport const *tpt = NULL;
int len, rv = -1;
char *msg = NULL;
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
if (!*cl.site)
site = local;
else {
if (!find_site_by_name(conf_ptr, cl.site, &site, 1)) {
log_error("Site \"%s\" not configured.", cl.site);
goto out_close;
}
}
if (site->type == ARBITRATOR) {
if (site == local) {
log_error("We're just an arbitrator, no attributes here.");
} else {
log_error("%s is just an arbitrator, no attributes there.", cl.site);
}
goto out_close;
}
- tpt = booth_transport + TCP;
+ tpt = *conf_ptr->transport + TCP;
init_header(conf_ptr, &cl.attr_msg.header, cmd, 0, cl.options, 0, 0,
sizeof(cl.attr_msg));
rv = tpt->open(site);
if (rv < 0)
goto out_close;
rv = tpt->send(conf_ptr, site, &cl.attr_msg, sendmsglen(&cl.attr_msg));
if (rv < 0)
goto out_close;
msg = malloc(MAX_MSG_LEN);
if (!msg) {
log_error("out of memory");
rv = -1;
goto out_close;
}
rv = read_server_reply(tpt, site, msg);
header = (struct boothc_header *)msg;
if (rv < 0) {
if (rv == -1)
(void)test_attr_reply(ntohl(header->result), cmd);
goto out_close;
}
len = ntohl(header->length);
if (check_boothc_header(header, len) < 0) {
log_error("message from %s receive error", site_string(site));
rv = -1;
goto out_close;
}
if (check_auth(conf_ptr, site, msg, len)) {
log_error("%s failed to authenticate", site_string(site));
rv = -1;
goto out_close;
}
rv = test_attr_reply(ntohl(header->result), cmd);
out_close:
if (tpt && site)
tpt->close(site);
if (msg)
free(msg);
return rv;
}
/*
* the server side
*/
/* need to invert gboolean, our success is 0
*/
#define gbool2rlt(i) (i ? RLT_SUCCESS : RLT_SYNC_FAIL)
static void free_geo_attr(gpointer data)
{
struct geo_attr *a = (struct geo_attr *)data;
if (!a)
return;
g_free(a->val);
g_free(a);
}
int store_geo_attr(struct ticket_config *tk, const char *name,
const char *val, int notime)
{
struct geo_attr *a;
GDestroyNotify free_geo_attr_notify = free_geo_attr;
if (!tk)
return -1;
/*
* allocate new, if attr doesn't already exist
* copy the attribute value
* send status
*/
if (!tk->attr)
tk->attr = g_hash_table_new_full(g_str_hash, g_str_equal,
g_free, free_geo_attr_notify);
if (!tk->attr) {
log_error("out of memory");
return -1;
}
if (strnlen(name, BOOTH_NAME_LEN) == BOOTH_NAME_LEN)
tk_log_warn("name of the attribute too long (%d+ bytes), skipped",
BOOTH_NAME_LEN);
else if (strnlen(val, BOOTH_ATTRVAL_LEN) == BOOTH_ATTRVAL_LEN)
tk_log_warn("value of the attribute too long (%d+ bytes), skipped",
BOOTH_ATTRVAL_LEN);
else {
a = (struct geo_attr *)calloc(1, sizeof(struct geo_attr));
if (!a) {
log_error("out of memory");
return -1;
}
a->val = g_strdup(val);
if (!notime)
get_time(&a->update_ts);
g_hash_table_insert(tk->attr,
g_strdup(name), a);
}
return 0;
}
static cmd_result_t attr_set(struct ticket_config *tk, struct boothc_attr_msg *msg)
{
int rc;
rc = store_geo_attr(tk, msg->attr.name, msg->attr.val, 0);
if (rc) {
return RLT_SYNC_FAIL;
}
(void)pcmk_handler.set_attr(tk, msg->attr.name, msg->attr.val);
return RLT_SUCCESS;
}
static cmd_result_t attr_del(struct ticket_config *tk, struct boothc_attr_msg *msg)
{
gboolean rv;
gpointer orig_key, value;
/*
* lookup attr
* deallocate, if found
* send status
*/
if (!tk->attr)
return RLT_NO_SUCH_ATTR;
rv = g_hash_table_lookup_extended(tk->attr, msg->attr.name,
&orig_key, &value);
if (!rv)
return RLT_NO_SUCH_ATTR;
rv = g_hash_table_remove(tk->attr, msg->attr.name);
(void)pcmk_handler.del_attr(tk, msg->attr.name);
return gbool2rlt(rv);
}
static void
append_attr(gpointer key, gpointer value, gpointer user_data)
{
char *attr_name = (char *)key;
struct geo_attr *a = (struct geo_attr *)value;
GString *data = (GString *)user_data;
char time_str[64];
time_t ts;
if (is_time_set(&a->update_ts)) {
ts = wall_ts(&a->update_ts);
strftime(time_str, sizeof(time_str), "%F %T",
localtime(&ts));
} else {
time_str[0] = '\0';
}
g_string_append_printf(data, "%s %s %s\n",
attr_name, a->val, time_str);
}
static cmd_result_t attr_get(struct booth_config *conf_ptr,
struct ticket_config *tk, int fd,
struct boothc_attr_msg *msg)
{
cmd_result_t rv = RLT_SUCCESS;
struct boothc_hdr_msg hdr;
struct geo_attr *a;
GString *attr_val;
/*
* lookup attr
* send value
*/
a = (struct geo_attr *)g_hash_table_lookup(tk->attr, msg->attr.name);
if (!a)
return RLT_NO_SUCH_ATTR;
attr_val = g_string_new(NULL);
if (!attr_val) {
log_error("out of memory");
return RLT_SYNC_FAIL;
}
g_string_printf(attr_val, "%s\n", a->val);
init_header(conf_ptr, &hdr.header, ATTR_GET, 0, 0, RLT_SUCCESS, 0,
sizeof(hdr) + attr_val->len);
if (send_header_plus(conf_ptr, fd, &hdr, attr_val->str, attr_val->len))
rv = RLT_SYNC_FAIL;
if (attr_val)
g_string_free(attr_val, FALSE);
return rv;
}
static cmd_result_t attr_list(struct booth_config *conf_ptr,
struct ticket_config *tk, int fd,
struct boothc_attr_msg *msg)
{
GString *data;
cmd_result_t rv;
struct boothc_hdr_msg hdr;
/*
* list all attributes for the ticket
* send the list
*/
data = g_string_sized_new(512);
if (!data) {
log_error("out of memory");
return RLT_SYNC_FAIL;
}
g_hash_table_foreach(tk->attr, append_attr, data);
init_header(conf_ptr, &hdr.header, ATTR_LIST, 0, 0, RLT_SUCCESS, 0,
sizeof(hdr) + data->len);
rv = send_header_plus(conf_ptr, fd, &hdr, data->str, data->len);
if (data)
g_string_free(data, FALSE);
return rv;
}
int process_attr_request(struct booth_config *conf_ptr,
struct client *req_client, void *buf)
{
cmd_result_t rv = RLT_SYNC_FAIL;
struct ticket_config *tk;
int cmd;
struct boothc_attr_msg *msg;
struct boothc_hdr_msg hdr;
msg = (struct boothc_attr_msg *)buf;
cmd = ntohl(msg->header.cmd);
if (!check_ticket(conf_ptr, msg->attr.tkt_id, &tk)) {
log_warn("client referenced unknown ticket %s",
msg->attr.tkt_id);
rv = RLT_INVALID_ARG;
goto reply_now;
}
switch (cmd) {
case ATTR_LIST:
rv = attr_list(conf_ptr, tk, req_client->fd, msg);
if (rv)
goto reply_now;
return 1;
case ATTR_GET:
rv = attr_get(conf_ptr, tk, req_client->fd, msg);
if (rv)
goto reply_now;
return 1;
case ATTR_SET:
rv = attr_set(tk, msg);
break;
case ATTR_DEL:
rv = attr_del(tk, msg);
break;
}
reply_now:
init_header(conf_ptr, &hdr.header, CL_RESULT, 0, 0, rv, 0, sizeof(hdr));
send_header_plus(conf_ptr, req_client->fd, &hdr, NULL, 0);
return 1;
}
/* read attr message from another site */
/* this is a NOOP and it should never be invoked
* only clients retrieve/manage attributes and they connect
* directly to the target site
*/
int attr_recv(struct booth_config *conf_ptr, void *buf,
struct booth_site *source)
{
struct boothc_attr_msg *msg;
struct ticket_config *tk;
msg = (struct boothc_attr_msg *)buf;
log_warn("unexpected attribute message from %s",
site_string(source));
if (!check_ticket(conf_ptr, msg->attr.tkt_id, &tk)) {
log_warn("got invalid ticket name %s from %s",
msg->attr.tkt_id, site_string(source));
source->invalid_cnt++;
return -1;
}
return 0;
}
diff --git a/src/config.c b/src/config.c
index a1be053..3e8b8f7 100644
--- a/src/config.c
+++ b/src/config.c
@@ -1,1026 +1,1027 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <ctype.h>
#include <stdlib.h>
#include <assert.h>
#include <zlib.h>
#include <sys/types.h>
#include <pwd.h>
#include <grp.h>
#include <errno.h>
#include <string.h>
#include <netdb.h>
#include "b_config.h"
#include "booth.h"
#include "config.h"
#include "raft.h"
#include "ticket.h"
#include "log.h"
static int ticket_size = 0;
static int ticket_realloc(struct booth_config *conf_ptr)
{
const int added = 5;
int had, want;
void *p;
assert(conf_ptr != NULL);
had = conf_ptr->ticket_allocated;
want = had + added;
p = realloc(conf_ptr->ticket, sizeof(struct ticket_config) * want);
if (!p) {
log_error("can't alloc more tickets");
return -ENOMEM;
}
conf_ptr->ticket = p;
memset(conf_ptr->ticket + had, 0,
sizeof(struct ticket_config) * added);
conf_ptr->ticket_allocated = want;
return 0;
}
static void hostname_to_ip(char * hostname)
{
struct hostent *he;
struct in_addr **addr_list;
if ((he = gethostbyname(hostname)) == NULL) {
log_error("can't find IP for the host \"%s\"", hostname);
return;
}
addr_list = (struct in_addr **) he->h_addr_list;
/* Return the first found address */
if (addr_list[0] != NULL) {
strncpy(hostname, inet_ntoa(*addr_list[0]), BOOTH_NAME_LEN - 1);
/* buffer overflow will not happen (IPv6 notation < 63 chars),
but suppress the warnings */
hostname[BOOTH_NAME_LEN - 1] = '\0';
}
else {
log_error("no IP addresses found for the host \"%s\"", hostname);
}
}
static int add_site(struct booth_config *conf_ptr, char *addr_string, int type)
{
int rv;
struct booth_site *site;
uLong nid;
uint32_t mask;
int i;
assert(conf_ptr != NULL);
rv = 1;
if (conf_ptr->site_count == MAX_NODES) {
log_error("too many nodes");
goto out;
}
if (strnlen(addr_string, sizeof(conf_ptr->site[0].addr_string))
>= sizeof(conf_ptr->site[0].addr_string)) {
log_error("site address \"%s\" too long", addr_string);
goto out;
}
site = conf_ptr->site + conf_ptr->site_count;
site->family = AF_INET;
site->type = type;
/* buffer overflow will not hapen (we've already checked that
addr_string will fit incl. terminating '\0' above), but
suppress the warnings with copying everything but the boundary
byte, which is valid as-is, since this last byte will be safely
pre-zeroed from the struct booth_config initialization */
strncpy(site->addr_string, addr_string, sizeof(site->addr_string) - 1);
if (!(inet_pton(AF_INET, site->addr_string, &site->sa4.sin_addr) > 0) &&
!(inet_pton(AF_INET6, site->addr_string, &site->sa6.sin6_addr) > 0)) {
/* Not a valid address, so let us try to convert it into an IP address */
hostname_to_ip(site->addr_string);
}
site->index = conf_ptr->site_count;
site->bitmask = 1 << conf_ptr->site_count;
/* Catch site overflow */
assert(site->bitmask);
conf_ptr->all_bits |= site->bitmask;
if (type == SITE)
conf_ptr->sites_bits |= site->bitmask;
site->tcp_fd = -1;
conf_ptr->site_count++;
rv = 0;
memset(&site->sa6, 0, sizeof(site->sa6));
nid = crc32(0L, NULL, 0);
/* Using the ASCII representation in site->addr_string (both sizeof()
* and strlen()) gives quite a lot of collisions; a brute-force run
* from 0.0.0.0 to 24.0.0.0 gives ~4% collisions, and this tends to
* increase even more.
* Whether there'll be a collision in real-life, with 3 or 5 nodes, is
* another question ... but for now get the ID from the binary
* representation - that had *no* collisions up to 32.0.0.0.
* Note that POSIX mandates inet_pton to arange the address pointed
* to by "dst" in network byte order, assuring little/big-endianess
* mutual compatibility. */
if (inet_pton(AF_INET,
site->addr_string,
&site->sa4.sin_addr) > 0) {
site->family = AF_INET;
site->sa4.sin_family = site->family;
site->sa4.sin_port = htons(conf_ptr->port);
site->saddrlen = sizeof(site->sa4);
site->addrlen = sizeof(site->sa4.sin_addr);
site->site_id = crc32(nid, (void*)&site->sa4.sin_addr, site->addrlen);
} else if (inet_pton(AF_INET6,
site->addr_string,
&site->sa6.sin6_addr) > 0) {
site->family = AF_INET6;
site->sa6.sin6_family = site->family;
site->sa6.sin6_flowinfo = 0;
site->sa6.sin6_port = htons(conf_ptr->port);
site->saddrlen = sizeof(site->sa6);
site->addrlen = sizeof(site->sa6.sin6_addr);
site->site_id = crc32(nid, (void*)&site->sa6.sin6_addr, site->addrlen);
} else {
log_error("Address string \"%s\" is bad", site->addr_string);
rv = EINVAL;
}
/* Make sure we will never collide with NO_ONE,
* or be negative (to get "get_local_id() < 0" working). */
mask = 1 << (sizeof(site->site_id)*8 -1);
assert(NO_ONE & mask);
site->site_id &= ~mask;
/* Test for collisions with other sites */
for(i=0; i<site->index; i++)
if (conf_ptr->site[i].site_id == site->site_id) {
log_error("Got a site-ID collision. Please file a bug on https://github.com/ClusterLabs/booth/issues/new, attaching the configuration file.");
exit(1);
}
out:
return rv;
}
inline static char *skip_while_in(const char *cp, int (*fn)(int), const char *allowed)
{
/* strchr() returns a pointer to the terminator if *cp == 0. */
while (*cp &&
(fn(*cp) ||
strchr(allowed, *cp)))
cp++;
/* discard "const" qualifier */
return (char*)cp;
}
inline static char *skip_while(char *cp, int (*fn)(int))
{
while (fn(*cp))
cp++;
return cp;
}
inline static char *skip_until(char *cp, char expected)
{
while (*cp && *cp != expected)
cp++;
return cp;
}
static inline int is_end_of_line(char *cp)
{
char c = *cp;
return c == '\n' || c == 0 || c == '#';
}
static int add_ticket(struct booth_config *conf_ptr, const char *name,
struct ticket_config **tkp, const struct ticket_config *def)
{
int rv;
struct ticket_config *tk;
assert(conf_ptr != NULL);
if (conf_ptr->ticket_count == conf_ptr->ticket_allocated) {
rv = ticket_realloc(conf_ptr);
if (rv < 0)
return rv;
}
tk = conf_ptr->ticket + conf_ptr->ticket_count;
conf_ptr->ticket_count++;
if (!check_max_len_valid(name, sizeof(tk->name))) {
log_error("ticket name \"%s\" too long.", name);
return -EINVAL;
}
if (find_ticket_by_name(conf_ptr, name, NULL)) {
log_error("ticket name \"%s\" used again.", name);
return -EINVAL;
}
if (* skip_while_in(name, isalnum, "-/")) {
log_error("ticket name \"%s\" invalid; only alphanumeric names.", name);
return -EINVAL;
}
strcpy(tk->name, name);
tk->timeout = def->timeout;
tk->term_duration = def->term_duration;
tk->retries = def->retries;
memcpy(tk->weight, def->weight, sizeof(tk->weight));
tk->mode = def->mode;
if (tkp)
*tkp = tk;
return 0;
}
static int postproc_ticket(struct ticket_config *tk)
{
if (!tk)
return 1;
if (!tk->renewal_freq) {
tk->renewal_freq = tk->term_duration/2;
}
if (tk->timeout*(tk->retries+1) >= tk->renewal_freq) {
log_error("%s: total amount of time to "
"retry sending packets cannot exceed "
"renewal frequency "
"(%d*(%d+1) >= %d)",
tk->name, tk->timeout, tk->retries, tk->renewal_freq);
return 0;
}
return 1;
}
/* returns number of weights, or -1 on bad input. */
static int parse_weights(const char *input, int weights[MAX_NODES])
{
int i, v;
char *cp;
for(i=0; i<MAX_NODES; i++) {
/* End of input? */
if (*input == 0)
break;
v = strtol(input, &cp, 0);
if (input == cp) {
log_error("No integer weight value at \"%s\"", input);
return -1;
}
weights[i] = v;
while (*cp) {
/* Separator characters */
if (isspace(*cp) ||
strchr(",;:-+", *cp))
cp++;
/* Next weight */
else if (isdigit(*cp))
break;
/* Rest */
else {
log_error("Invalid character at \"%s\"", cp);
return -1;
}
}
input = cp;
}
/* Fill rest of vector. */
for(v=i; v<MAX_NODES; v++) {
weights[v] = 0;
}
return i;
}
/* returns TICKET_MODE_AUTO if failed to parse the ticket mode. */
static ticket_mode_e retrieve_ticket_mode(const char *input)
{
if (strcasecmp(input, "manual") == 0) {
return TICKET_MODE_MANUAL;
}
return TICKET_MODE_AUTO;
}
/* scan val for time; time is [0-9]+(ms)?, i.e. either in seconds
* or milliseconds
* returns -1 on failure, otherwise time in ms
*/
static long read_time(char *val)
{
long t;
char *ep;
t = strtol(val, &ep, 10);
if (ep == val) { /* matched none */
t = -1L;
} else if (*ep == '\0') { /* matched all */
t = t*1000L; /* in seconds, convert to ms */
} else if (strcmp(ep, "ms")) { /* ms not exactly matched */
t = -1L;
} /* otherwise, time in ms */
/* if second fractions configured, send finer resolution
* times (i.e. term_valid_for) */
if (t % 1000L) {
TIME_MULT = 1000;
}
return t;
}
/* make arguments for execv(2)
* tk_test.path points to the path
* tk_test.argv is argument vector (starts with the prog)
* (strtok pokes holes in the configuration parameter value, i.e.
* we don't need to allocate memory for arguments)
*/
static int parse_extprog(char *val, struct ticket_config *tk)
{
char *p;
int i = 0;
if (tk_test.path) {
free(tk_test.path);
}
if (!(tk_test.path = strdup(val))) {
log_error("out of memory");
return -1;
}
p = strtok(tk_test.path, " \t");
tk_test.argv[i++] = p;
do {
p = strtok(NULL, " \t");
if (i >= MAX_ARGS) {
log_error("too many arguments for the acquire-handler");
free(tk_test.path);
return -1;
}
tk_test.argv[i++] = p;
} while (p);
return 0;
}
struct toktab grant_type[] = {
{ "auto", GRANT_AUTO},
{ "manual", GRANT_MANUAL},
{ NULL, 0},
};
struct toktab attr_op[] = {
{"eq", ATTR_OP_EQ},
{"ne", ATTR_OP_NE},
{NULL, 0},
};
static int lookup_tokval(char *key, struct toktab *tab)
{
struct toktab *tp;
for (tp = tab; tp->str; tp++) {
if (!strcmp(tp->str, key))
return tp->val;
}
return 0;
}
/* attribute prerequisite
*/
static int parse_attr_prereq(char *val, struct ticket_config *tk)
{
struct attr_prereq *ap = NULL;
char *p;
ap = (struct attr_prereq *)calloc(1, sizeof(struct attr_prereq));
if (!ap) {
log_error("out of memory");
return -1;
}
p = strtok(val, " \t");
if (!p) {
log_error("not enough arguments to attr-prereq");
goto err_out;
}
ap->grant_type = lookup_tokval(p, grant_type);
if (!ap->grant_type) {
log_error("%s is not a grant type", p);
goto err_out;
}
p = strtok(NULL, " \t");
if (!p) {
log_error("not enough arguments to attr-prereq");
goto err_out;
}
if (!(ap->attr_name = strdup(p))) {
log_error("out of memory");
goto err_out;
}
p = strtok(NULL, " \t");
if (!p) {
log_error("not enough arguments to attr-prereq");
goto err_out;
}
ap->op = lookup_tokval(p, attr_op);
if (!ap->op) {
log_error("%s is not an attribute operation", p);
goto err_out;
}
p = strtok(NULL, " \t");
if (!p) {
log_error("not enough arguments to attr-prereq");
goto err_out;
}
if (!(ap->attr_val = strdup(p))) {
log_error("out of memory");
goto err_out;
}
tk->attr_prereqs = g_list_append(tk->attr_prereqs, ap);
if (!tk->attr_prereqs) {
log_error("out of memory");
goto err_out;
}
return 0;
err_out:
if (ap) {
if (ap->attr_val)
free(ap->attr_val);
if (ap->attr_name)
free(ap->attr_name);
free(ap);
}
return -1;
}
extern int poll_timeout;
-int read_config(struct booth_config **conf_pptr, const char *path, int type)
+int read_config(struct booth_config **conf_pptr,
+ const booth_transport_table_t *transport, const char *path,
+ int type)
{
char line[1024];
FILE *fp;
char *s, *key, *val, *end_of_key;
const char *error;
char *cp, *cp2;
int i;
int lineno = 0;
int got_transport = 0;
int min_timeout = 0;
struct ticket_config defaults = { { 0 } };
struct ticket_config *current_tk = NULL;
assert(conf_pptr != NULL);
free(*conf_pptr);
fp = fopen(path, "r");
if (!fp) {
log_error("failed to open %s: %s", path, strerror(errno));
*conf_pptr = NULL;
return -1;
}
*conf_pptr = malloc(sizeof(struct booth_config)
+ TICKET_ALLOC * sizeof(struct ticket_config));
if (*conf_pptr == NULL) {
fclose(fp);
log_error("failed to alloc memory for booth config");
return -ENOMEM;
}
memset(*conf_pptr, 0, sizeof(struct booth_config)
+ TICKET_ALLOC * sizeof(struct ticket_config));
ticket_size = TICKET_ALLOC;
-
+ (*conf_pptr)->transport = transport;
(*conf_pptr)->proto = UDP;
(*conf_pptr)->port = BOOTH_DEFAULT_PORT;
(*conf_pptr)->maxtimeskew = BOOTH_DEFAULT_MAX_TIME_SKEW;
(*conf_pptr)->authkey[0] = '\0';
-
/* Provide safe defaults. -1 is reserved, though. */
(*conf_pptr)->uid = -2;
(*conf_pptr)->gid = -2;
strcpy((*conf_pptr)->site_user, "hacluster");
strcpy((*conf_pptr)->site_group, "haclient");
strcpy((*conf_pptr)->arb_user, "nobody");
strcpy((*conf_pptr)->arb_group, "nobody");
parse_weights("", defaults.weight);
defaults.clu_test.path = NULL;
defaults.clu_test.pid = 0;
defaults.clu_test.status = 0;
defaults.clu_test.progstate = EXTPROG_IDLE;
defaults.term_duration = DEFAULT_TICKET_EXPIRY;
defaults.timeout = DEFAULT_TICKET_TIMEOUT;
defaults.retries = DEFAULT_RETRIES;
defaults.acquire_after = 0;
defaults.mode = TICKET_MODE_AUTO;
error = "";
log_debug("reading config file %s", path);
while (fgets(line, sizeof(line), fp)) {
lineno++;
s = skip_while(line, isspace);
if (is_end_of_line(s) || *s == '#')
continue;
key = s;
/* Key */
end_of_key = skip_while_in(key, isalnum, "-_");
if (end_of_key == key) {
error = "No key";
goto err;
}
if (!*end_of_key)
goto exp_equal;
/* whitespace, and something else but nothing more? */
s = skip_while(end_of_key, isspace);
if (*s != '=') {
exp_equal:
error = "Expected '=' after key";
goto err;
}
s++;
/* It's my buffer, and I terminate if I want to. */
/* But not earlier than that, because we had to check for = */
*end_of_key = 0;
/* Value tokenizing */
s = skip_while(s, isspace);
switch (*s) {
case '"':
case '\'':
val = s+1;
s = skip_until(val, *s);
/* Terminate value */
if (!*s) {
error = "Unterminated quoted string";
goto err;
}
/* Remove and skip quote */
*s = 0;
s++;
if (*(s = skip_while(s, isspace)) && *s != '#') {
error = "Surplus data after value";
goto err;
}
*s = 0;
break;
case 0:
no_value:
error = "No value";
goto err;
break;
default:
val = s;
/* Rest of line. */
i = strlen(s);
/* i > 0 because of "case 0" above. */
while (i > 0 && isspace(s[i-1]))
i--;
s += i;
*s = 0;
}
if (val == s)
goto no_value;
if (strlen(key) > BOOTH_NAME_LEN
|| strlen(val) > BOOTH_NAME_LEN) {
error = "key/value too long";
goto err;
}
if (strcmp(key, "transport") == 0) {
if (got_transport) {
error = "config file has multiple transport lines";
goto err;
}
if (strcasecmp(val, "UDP") == 0)
(*conf_pptr)->proto = UDP;
else if (strcasecmp(val, "SCTP") == 0)
(*conf_pptr)->proto = SCTP;
else {
error = "invalid transport protocol";
goto err;
}
got_transport = 1;
continue;
}
if (strcmp(key, "port") == 0) {
(*conf_pptr)->port = atoi(val);
continue;
}
if (strcmp(key, "name") == 0) {
safe_copy((*conf_pptr)->name,
val, BOOTH_NAME_LEN,
"name");
continue;
}
#if HAVE_LIBGCRYPT || HAVE_LIBMHASH
if (strcmp(key, "authfile") == 0) {
safe_copy((*conf_pptr)->authfile,
val, BOOTH_PATH_LEN,
"authfile");
continue;
}
if (strcmp(key, "maxtimeskew") == 0) {
(*conf_pptr)->maxtimeskew = atoi(val);
continue;
}
#endif
if (strcmp(key, "site") == 0) {
if (add_site(*conf_pptr, val, SITE))
goto err;
continue;
}
if (strcmp(key, "arbitrator") == 0) {
if (add_site(*conf_pptr, val, ARBITRATOR))
goto err;
continue;
}
if (strcmp(key, "site-user") == 0) {
safe_copy((*conf_pptr)->site_user, optarg, BOOTH_NAME_LEN,
"site-user");
continue;
}
if (strcmp(key, "site-group") == 0) {
safe_copy((*conf_pptr)->site_group, optarg, BOOTH_NAME_LEN,
"site-group");
continue;
}
if (strcmp(key, "arbitrator-user") == 0) {
safe_copy((*conf_pptr)->arb_user, optarg, BOOTH_NAME_LEN,
"arbitrator-user");
continue;
}
if (strcmp(key, "arbitrator-group") == 0) {
safe_copy((*conf_pptr)->arb_group, optarg, BOOTH_NAME_LEN,
"arbitrator-group");
continue;
}
if (strcmp(key, "debug") == 0) {
if (type != CLIENT && type != GEOSTORE)
debug_level = max(debug_level, atoi(val));
continue;
}
if (strcmp(key, "ticket") == 0) {
if (current_tk && strcmp(current_tk->name, "__defaults__")) {
if (!postproc_ticket(current_tk)) {
goto err;
}
}
if (!strcmp(val, "__defaults__")) {
current_tk = &defaults;
} else if (add_ticket(*conf_pptr, val, ¤t_tk,
&defaults)) {
goto err;
}
continue;
}
/* current_tk must be allocated at this point, otherwise
* we don't know to which ticket the key refers
*/
if (!current_tk) {
error = "Unexpected keyword";
goto err;
}
if (strcmp(key, "expire") == 0) {
current_tk->term_duration = read_time(val);
if (current_tk->term_duration <= 0) {
error = "Expected time >0 for expire";
goto err;
}
continue;
}
if (strcmp(key, "timeout") == 0) {
current_tk->timeout = read_time(val);
if (current_tk->timeout <= 0) {
error = "Expected time >0 for timeout";
goto err;
}
if (!min_timeout) {
min_timeout = current_tk->timeout;
} else {
min_timeout = min(min_timeout, current_tk->timeout);
}
continue;
}
if (strcmp(key, "retries") == 0) {
current_tk->retries = strtol(val, &s, 0);
if (*s || s == val ||
current_tk->retries<3 || current_tk->retries > 100) {
error = "Expected plain integer value in the range [3, 100] for retries";
goto err;
}
continue;
}
if (strcmp(key, "renewal-freq") == 0) {
current_tk->renewal_freq = read_time(val);
if (current_tk->renewal_freq <= 0) {
error = "Expected time >0 for renewal-freq";
goto err;
}
continue;
}
if (strcmp(key, "acquire-after") == 0) {
current_tk->acquire_after = read_time(val);
if (current_tk->acquire_after < 0) {
error = "Expected time >=0 for acquire-after";
goto err;
}
continue;
}
if (strcmp(key, "before-acquire-handler") == 0) {
if (parse_extprog(val, current_tk)) {
goto err;
}
continue;
}
if (strcmp(key, "attr-prereq") == 0) {
if (parse_attr_prereq(val, current_tk)) {
goto err;
}
continue;
}
if (strcmp(key, "mode") == 0) {
current_tk->mode = retrieve_ticket_mode(val);
continue;
}
if (strcmp(key, "weights") == 0) {
if (parse_weights(val, current_tk->weight) < 0)
goto err;
continue;
}
error = "Unknown keyword";
goto err;
}
fclose(fp);
if (((*conf_pptr)->site_count % 2) == 0) {
log_warn("Odd number of nodes is strongly recommended!");
}
/* Default: make config name match config filename. */
if (!(*conf_pptr)->name[0]) {
cp = strrchr(path, '/');
cp = cp ? cp+1 : (char *)path;
cp2 = strrchr(cp, '.');
if (!cp2)
cp2 = cp + strlen(cp);
if (cp2-cp >= BOOTH_NAME_LEN) {
log_error("booth config file name too long");
goto out;
}
strncpy((*conf_pptr)->name, cp, cp2-cp);
*((*conf_pptr)->name+(cp2-cp)) = '\0';
}
if (!postproc_ticket(current_tk)) {
goto out;
}
poll_timeout = min(POLL_TIMEOUT, min_timeout/10);
if (!poll_timeout)
poll_timeout = POLL_TIMEOUT;
return 0;
err:
fclose(fp);
out:
log_error("%s in config file line %d",
error, lineno);
free(*conf_pptr);
*conf_pptr = NULL;
return -1;
}
int check_config(struct booth_config *conf_ptr, int type)
{
struct passwd *pw;
struct group *gr;
char *cp, *input;
if (conf_ptr == NULL)
return -1;
input = (type == ARBITRATOR)
? conf_ptr->arb_user
: conf_ptr->site_user;
if (!*input)
goto u_inval;
if (isdigit(input[0])) {
conf_ptr->uid = strtol(input, &cp, 0);
if (*cp != 0) {
u_inval:
log_error("User \"%s\" cannot be resolved into a UID.", input);
return ENOENT;
}
} else {
pw = getpwnam(input);
if (!pw)
goto u_inval;
conf_ptr->uid = pw->pw_uid;
}
input = (type == ARBITRATOR)
? conf_ptr->arb_group
: conf_ptr->site_group;
if (!*input)
goto g_inval;
if (isdigit(input[0])) {
conf_ptr->gid = strtol(input, &cp, 0);
if (*cp != 0) {
g_inval:
log_error("Group \"%s\" cannot be resolved into a UID.", input);
return ENOENT;
}
} else {
gr = getgrnam(input);
if (!gr)
goto g_inval;
conf_ptr->gid = gr->gr_gid;
}
return 0;
}
static int get_other_site(struct booth_config *conf_ptr,
struct booth_site **node)
{
struct booth_site *n;
int i;
*node = NULL;
if (conf_ptr == NULL)
return 0;
FOREACH_NODE(conf_ptr, i, n) {
if (n != local && n->type == SITE) {
if (!*node) {
*node = n;
} else {
return 0;
}
}
}
return !*node ? 0 : 1;
}
int find_site_by_name(struct booth_config *conf_ptr, const char *site,
struct booth_site **node, int any_type)
{
struct booth_site *n;
int i;
if (conf_ptr == NULL)
return 0;
if (!strcmp(site, OTHER_SITE))
return get_other_site(conf_ptr, node);
FOREACH_NODE(conf_ptr, i, n) {
if ((n->type == SITE || any_type) &&
strncmp(n->addr_string, site, sizeof(n->addr_string)) == 0) {
*node = n;
return 1;
}
}
return 0;
}
int find_site_by_id(struct booth_config *conf_ptr, uint32_t site_id,
struct booth_site **node)
{
struct booth_site *n;
int i;
if (site_id == NO_ONE) {
*node = no_leader;
return 1;
}
if (conf_ptr == NULL)
return 0;
FOREACH_NODE(conf_ptr, i, n) {
if (n->site_id == site_id) {
*node = n;
return 1;
}
}
return 0;
}
const char *type_to_string(int type)
{
switch (type)
{
case ARBITRATOR: return "arbitrator";
case SITE: return "site";
case CLIENT: return "client";
case GEOSTORE: return "attr";
}
return "??invalid-type??";
}
diff --git a/src/config.h b/src/config.h
index 68e2175..d325350 100644
--- a/src/config.h
+++ b/src/config.h
@@ -1,388 +1,391 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#ifndef _CONFIG_H
#define _CONFIG_H
#include <stdint.h>
#include <sys/stat.h>
struct booth_config;
#include "booth.h"
#include "timer.h"
#include "raft.h"
#include "transport.h"
/** @{ */
/** Definitions for in-RAM data. */
#define MAX_NODES 16
#define MAX_ARGS 16
#define TICKET_ALLOC 16
#define OTHER_SITE "other"
typedef enum {
EXTPROG_IDLE,
EXTPROG_RUNNING,
EXTPROG_EXITED,
EXTPROG_IGNORE,
} extprog_state_e;
#define tk_test tk->clu_test
typedef enum {
ATTR_OP_EQ = 1,
ATTR_OP_NE,
} attr_op_e;
typedef enum {
GRANT_AUTO = 1,
GRANT_MANUAL,
} grant_type_e;
typedef enum {
TICKET_MODE_AUTO = 1,
TICKET_MODE_MANUAL,
} ticket_mode_e;
struct toktab {
const char *str;
int val;
};
struct attr_prereq {
grant_type_e grant_type; /* grant type */
attr_op_e op; /* attribute operation */
char *attr_name;
char *attr_val;
};
struct ticket_config {
/** \name Configuration items.
* @{ */
/** Name of ticket. */
boothc_ticket name;
/** How long a term lasts if not refreshed (in ms) */
int term_duration;
/** Network related timeouts (in ms) */
int timeout;
/** Retries before giving up. */
int retries;
/** If >0, time to wait for a site to get fenced.
* The ticket may be acquired after that timespan by
* another site. */
int acquire_after;
/* How often to renew the ticket (in ms)
*/
int renewal_freq;
/* Program to ask whether it makes sense to
* acquire the ticket */
struct clu_test {
char *path;
int is_dir;
char *argv[MAX_ARGS];
pid_t pid;
int status; /* child exit status */
extprog_state_e progstate; /* program running/idle/waited on */
} clu_test;
/** Node weights. */
int weight[MAX_NODES];
/* Mode operation of the ticket.
* Set to MANUAL to make sure that the ticket will be manipulated
* only by manual commands of the administrator. In such a case
* automatic elections will be disabled.
* Manual tickets do not have to be renewed every some time.
* The leader will continue to send heartbeat messages to other sites.
*/
ticket_mode_e mode;
/** @} */
/** \name Runtime values.
* @{ */
/** Current state. */
server_state_e state;
/** Next state. Used at startup. */
server_state_e next_state;
/** When something has to be done */
timetype next_cron;
/** Current leader. This is effectively the log[] in Raft. */
struct booth_site *leader;
/** Leader that got lost. */
struct booth_site *lost_leader;
/** Is the ticket granted? */
int is_granted;
/** Which site considered itself a leader.
* For manual tickets it is possible, that
* more than one site will act as a leader.
* This array is used for tracking that situation
* and notifying the user about the issue.
*
* Possible values for every site:
* 0: the site does not claim to be the leader
* 1: the site considers itself a leader and
* is sending or used to send heartbeat messages
*
* The site will be marked as '1' until this site
* receives revoke confirmation.
*
* If more than one site has '1', the geo cluster is
* considered to have multiple leadership and proper
* warning are generated.
*/
int sites_where_granted[MAX_NODES];
/** Timestamp of leadership expiration */
timetype term_expires;
/** End of election period */
timetype election_end;
struct booth_site *voted_for;
/** Who the various sites vote for.
* NO_OWNER = no vote yet. */
struct booth_site *votes_for[MAX_NODES];
/* bitmap */
uint64_t votes_received;
/** Last voting round that was seen. */
uint32_t current_term;
/** Do ticket updates whenever we get enough heartbeats.
* But do that only once.
* This is reset to 0 whenever we broadcast heartbeat and set
* to 1 once enough acks are received.
* Increased to 2 when the ticket is commited to the CIB (see
* delay_commit).
*/
uint32_t ticket_updated;
/** Outcome of whatever ticket request was processed.
* Can also be an intermediate stage.
*/
uint32_t outcome;
/** @} */
/** */
uint32_t last_applied;
uint32_t next_index[MAX_NODES];
uint32_t match_index[MAX_NODES];
/* Why did we start the elections?
*/
cmd_reason_t election_reason;
/* if it is potentially dangerous to grant the ticket
* immediately, then this is set to some point in time,
* usually (now + term_duration + acquire_after)
*/
timetype delay_commit;
/* the last request RPC we sent
*/
uint32_t last_request;
/* if we expect some acks, then set this to the id of
* the RPC which others will send us; it is cleared once all
* replies were received
*/
uint32_t acks_expected;
/* bitmask of servers which sent acks
*/
uint64_t acks_received;
/* timestamp of the request */
timetype req_sent_at;
/* we need to wait for MY_INDEX from other servers,
* hold the ticket processing for a while until they reply
*/
int start_postpone;
/** Last renewal time */
timetype last_renewal;
/* Do we need to update the copy in the CIB?
* Normally, the ticket is written only when it changes via
* the UPDATE RPC (for followers) and on expiration update
* (for leaders)
*/
int update_cib;
/* Is this ticket in election?
*/
int in_election;
/* don't log warnings unnecessarily
*/
int expect_more_rejects;
/** \name Needed while proposals are being done.
* @{ */
/* Need to keep the previous valid ticket in case we moved to
* start new elections and another server asks for the ticket
* status. It would be wrong to send our candidate ticket.
*/
struct ticket_config *last_valid_tk;
/** Attributes, user defined
*/
GHashTable *attr;
/** Attribute prerequisites
*/
GList *attr_prereqs;
/** Whom to vote for the next time.
* Needed to push a ticket to someone else. */
#if 0
/** Bitmap of sites that acknowledge that state. */
uint64_t proposal_acknowledges;
/** When an incompletely acknowledged proposal gets done.
* If all peers agree, that happens sooner.
* See switch_state_to(). */
struct timeval proposal_switch;
/** Timestamp of proposal expiration. */
time_t proposal_expires;
#endif
/** Number of send retries left.
* Used on the new owner.
* Starts at 0, counts up. */
int retry_number;
/** @} */
};
struct booth_config {
char name[BOOTH_NAME_LEN];
/** File containing the authentication file. */
char authfile[BOOTH_PATH_LEN];
struct stat authstat;
char authkey[BOOTH_MAX_KEY_LEN];
int authkey_len;
/** Maximum time skew between peers allowed */
int maxtimeskew;
transport_layer_t proto;
uint16_t port;
/** Stores the OR of sites bitmasks. */
uint64_t sites_bits;
/** Stores the OR of all members' bitmasks. */
uint64_t all_bits;
char site_user[BOOTH_NAME_LEN];
char site_group[BOOTH_NAME_LEN];
char arb_user[BOOTH_NAME_LEN];
char arb_group[BOOTH_NAME_LEN];
uid_t uid;
gid_t gid;
int site_count;
struct booth_site site[MAX_NODES];
int ticket_count;
int ticket_allocated;
struct ticket_config *ticket;
-};
-extern struct booth_config *booth_conf;
+ const booth_transport_table_t *transport;
+};
#define is_auth_req(b_) ((b_)->authkey[0] != '\0')
/**
* @internal
* Parse booth configuration file and store as structured data
*
* @param[inout] conf_pptr config object to free-alloc cycle & fill accordingly
+ * @param[in] transport transport handlers table
* @param[in] path where the configuration file is expected
* @param[in] type role currently being acted as
*
* @return 0 or negative value (-1 or -errno) on error
*/
-int read_config(struct booth_config **conf_pptr, const char *path, int type);
+int read_config(struct booth_config **conf_pptr,
+ const booth_transport_table_t *transport, const char *path,
+ int type);
/**
* @internal
* Check booth configuration
*
* Currently it means checking that login user/group indeed exists,
* while converting it to respective numeric values for further use.
*
* @param[inout] conf_ptr config object to check
* @param[in] type role currently being acted as
*
* @return 0 or negative value (-1 or -errno) on error
*/
int check_config(struct booth_config *conf_ptr, int type);
/**
* @internal
* Find site in booth configuration by resolved host name
*
* @param[inout] conf_ptr config object to refer to
* @param[in] site name to match against previously resolved host names
* @param[out] node relevant tracked data when found
* @param[in] any_type whether or not to consider also non-site members
*
* @return 0 if nothing found, or 1 when found (node assigned accordingly)
*/
int find_site_by_name(struct booth_config *conf_ptr, const char *site,
struct booth_site **node, int any_type);
/**
* @internal
* Find site in booth configuration by a hash (id)
*
* @param[inout] conf_ptr config object to refer to
* @param[in] site_id hash (id) to match against previously resolved ones
* @param[out] node relevant tracked data when found
*
* @return 0 if nothing found, or 1 when found (node assigned accordingly)
*/
int find_site_by_id(struct booth_config *conf_ptr, uint32_t site_id,
struct booth_site **node);
const char *type_to_string(int type);
#endif /* _CONFIG_H */
diff --git a/src/inline-fn.h b/src/inline-fn.h
index 4958038..b3a56d5 100644
--- a/src/inline-fn.h
+++ b/src/inline-fn.h
@@ -1,325 +1,326 @@
/*
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#ifndef _INLINE_FN_H
#define _INLINE_FN_H
#include <time.h>
#include <sys/time.h>
#include <assert.h>
#include <string.h>
#include "timer.h"
#include "config.h"
#include "transport.h"
inline static int get_local_id(void)
{
return local ? local->site_id : -1;
}
inline static uint32_t get_node_id(struct booth_site *node)
{
return node ? node->site_id : 0;
}
/** Returns number of seconds left, if any. */
inline static int term_time_left(struct ticket_config *tk)
{
int left = 0;
if (is_time_set(&tk->term_expires)) {
left = time_left(&tk->term_expires);
}
return (left < 0) ? 0 : left;
}
inline static int leader_and_valid(struct ticket_config *tk)
{
if (tk->leader != local)
return 0;
return term_time_left(tk);
}
/** Is this some leader? */
inline static int is_owned(const struct ticket_config *tk)
{
return (tk->leader && tk->leader != no_leader);
}
inline static int is_resend(struct ticket_config *tk)
{
timetype now;
get_time(&now);
return time_sub_int(&now, &tk->req_sent_at) >= tk->timeout;
}
static inline void init_header_bare(struct booth_config *conf_ptr,
struct boothc_header *h) {
timetype now;
assert(local && local->site_id);
h->magic = htonl(BOOTHC_MAGIC);
h->version = htonl(BOOTHC_VERSION);
h->from = htonl(local->site_id);
if (is_auth_req(conf_ptr)) {
get_time(&now);
h->opts = htonl(BOOTH_OPT_AUTH);
h->secs = htonl(secs_since_epoch(&now));
h->usecs = htonl(get_usecs(&now));
} else {
h->opts = htonl(0);
h->secs = htonl(0);
h->usecs = htonl(0);
}
}
/* get the _real_ message length out of the header
*/
#define sendmsglen(msg) ntohl((msg)->header.length)
static inline void init_header(struct booth_config *conf_ptr,
struct boothc_header *h, int cmd, int request,
int options, int result, int reason,
int data_len)
{
init_header_bare(conf_ptr, h);
h->length = htonl(data_len -
(is_auth_req(conf_ptr) ? 0 : sizeof(struct hmac)));
h->cmd = htonl(cmd);
h->request = htonl(request);
h->options = htonl(options);
h->result = htonl(result);
h->reason = htonl(reason);
}
#define my_last_term(tk) \
(((tk)->state == ST_CANDIDATE && (tk)->last_valid_tk) ? \
(tk)->last_valid_tk->current_term : (tk)->current_term)
extern int TIME_RES, TIME_MULT;
#define msg_term_time(msg) \
ntohl((msg)->ticket.term_valid_for)*TIME_RES/TIME_MULT
#define set_msg_term_time(msg, tk) \
(msg)->ticket.term_valid_for = htonl(term_time_left(tk)*TIME_MULT/TIME_RES)
static inline void init_ticket_msg(struct booth_config *conf_ptr,
struct boothc_ticket_msg *msg, int cmd,
int request, int rv, int reason,
struct ticket_config *tk)
{
assert(sizeof(msg->ticket.id) == sizeof(tk->name));
init_header(conf_ptr, &msg->header, cmd, request, 0, rv, reason,
sizeof(*msg));
if (!tk) {
memset(&msg->ticket, 0, sizeof(msg->ticket));
} else {
memcpy(msg->ticket.id, tk->name, sizeof(msg->ticket.id));
msg->ticket.leader = htonl(get_node_id(
(tk->leader && tk->leader != no_leader) ? tk->leader :
(tk->voted_for ? tk->voted_for : no_leader)));
msg->ticket.term = htonl(tk->current_term);
set_msg_term_time(msg, tk);
}
}
-/* XXX uses globals: booth_transport, booth_conf */
-static inline struct booth_transport const *transport(void)
+static inline struct booth_transport const *transport(struct booth_config *conf_ptr)
{
- return booth_transport + booth_conf->proto;
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
+ return *conf_ptr->transport + conf_ptr->proto;
}
static inline const char *site_string(const struct booth_site *site)
{
return site ? site->addr_string : "NONE";
}
/**
* @internal
* Parse booth configuration file and store as structured data
*
* @param[in] site subject of TCP/UDP port extraction
*
* @return 0 for "undefined", actual port number otherwise
*/
static inline uint16_t site_port(const struct booth_site *site)
{
assert(site != NULL);
return site
? site->family == AF_INET
? ntohs(site->sa4.sin_port)
: site->family == AF_INET6
? ntohs(site->sa6.sin6_port)
: 0
: 0;
}
static inline const char *ticket_leader_string(struct ticket_config *tk)
{
return site_string(tk->leader);
}
/* We allow half of the uint32_t to be used;
* half of that below, half of that above the current known "good" value.
* 0 UINT32_MAX
* |--------------------------+----------------+------------|
* | | |
* |--------+-------| allowed range
* |
* current commit index
*
* So, on overflow it looks like that:
* UINT32_MAX 0
* |--------------------------+-----------||---+------------|
* | | |
* |--------+-------| allowed range
* |
* current commit index
*
* This should be possible by using the same datatype and relying
* on the under/overflow semantics.
*
*
* Having 30 bits available, and assuming an expire time of
* one minute and a (high) commit index step of 64 == 2^6 (because
* of weights), we get 2^24 minutes of range - which is ~750
* years. "Should be enough for everybody."
*/
static inline int index_is_higher_than(uint32_t c_high, uint32_t c_low)
{
uint32_t diff;
if (c_high == c_low)
return 0;
diff = c_high - c_low;
if (diff < UINT32_MAX/4)
return 1;
diff = c_low - c_high;
if (diff < UINT32_MAX/4)
return 0;
assert(!"commit index out of range - invalid");
}
static inline uint32_t index_max2(uint32_t a, uint32_t b)
{
return index_is_higher_than(a, b) ? a : b;
}
static inline uint32_t index_max3(uint32_t a, uint32_t b, uint32_t c)
{
return index_max2( index_max2(a, b), c);
}
/* only invoked when ticket leader */
static inline void get_next_election_time(struct ticket_config *tk, timetype *next)
{
assert(tk->leader == local);
/* if last_renewal is not set, which is unusual, it may mean
* that the ticket never got updated, i.e. nobody acked
* ticket updates (say, due to a temporary connection
* problem)
* we may try a bit later again */
if (!is_time_set(&tk->last_renewal)) {
time_reset(next);
} else {
interval_add(&tk->last_renewal, tk->renewal_freq, next);
}
/* if delay_commit is earlier than next, then set next to
* delay_commit */
if (is_time_set(&tk->delay_commit) &&
time_cmp(next, &tk->delay_commit, >)) {
copy_time(&tk->delay_commit, next);
}
}
static inline void expect_replies(struct ticket_config *tk,
int reply_type)
{
tk->retry_number = 0;
tk->acks_expected = reply_type;
tk->acks_received = local->bitmask;
get_time(&tk->req_sent_at);
}
static inline void no_resends(struct ticket_config *tk)
{
tk->retry_number = 0;
tk->acks_expected = 0;
}
static inline struct booth_site *my_vote(struct ticket_config *tk)
{
return tk->votes_for[ local->index ];
}
static inline int count_bits(uint64_t val) {
return __builtin_popcount(val);
}
static inline int majority_of_bits(struct booth_config *conf_ptr,
struct ticket_config *tk, uint64_t val)
{
assert(conf_ptr != NULL);
/* Use ">" to get majority decision, even for an even number
* of participants. */
return count_bits(val) * 2 > conf_ptr->site_count;
}
static inline int all_replied(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
assert(conf_ptr != NULL);
return !(tk->acks_received ^ conf_ptr->all_bits);
}
static inline int all_sites_replied(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
assert(conf_ptr != NULL);
return !((tk->acks_received & conf_ptr->sites_bits) \
^ conf_ptr->sites_bits);
}
#endif
diff --git a/src/main.c b/src/main.c
index 995e2b7..ebbf5b5 100644
--- a/src/main.c
+++ b/src/main.c
@@ -1,1653 +1,1659 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <sched.h>
#include <errno.h>
#include <limits.h>
#include <sys/file.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/poll.h>
#include <sys/wait.h>
#include <fcntl.h>
#include <string.h>
#include <ctype.h>
#include <assert.h>
#include <signal.h>
#include <netdb.h>
#include <arpa/inet.h>
#include <sys/types.h>
#include <pacemaker/crm/services.h>
#include "b_config.h"
#if HAVE_LIBGCRYPT
#include <gcrypt.h>
#endif
#ifndef NAMETAG_LIBSYSTEMD
#include <clplumbing/setproctitle.h>
#else
#include "alt/nametag_libsystemd.h"
#endif
#ifdef COREDUMP_NURSING
#include <sys/prctl.h>
#include <clplumbing/coredumps.h>
#endif
#include "log.h"
#include "booth.h"
#include "config.h"
#include "transport.h"
#include "inline-fn.h"
#include "pacemaker.h"
#include "ticket.h"
#include "request.h"
#include "attr.h"
#include "handler.h"
#define RELEASE_VERSION "1.0"
#define RELEASE_STR RELEASE_VERSION " (build " BOOTH_BUILD_VERSION ")"
#define CLIENT_NALLOC 32
+extern const booth_transport_table_t booth__transport;
+
static int daemonize = 1;
int enable_stderr = 0;
timetype start_time;
+static struct booth_config *booth_conf;
/** Structure for "clients".
* Filehandles with incoming data get registered here (and in pollfds),
* along with their callbacks.
* Because these can be reallocated with every new fd, addressing
* happens _only_ by their numeric index. */
struct client *clients = NULL;
struct pollfd *pollfds = NULL;
static int client_maxi;
static int client_size = 0;
static const struct booth_site _no_leader = {
.addr_string = "none",
.site_id = NO_ONE,
.index = -1,
};
struct booth_site *const no_leader = (struct booth_site*) &_no_leader;
typedef enum
{
BOOTHD_STARTED=0,
BOOTHD_STARTING
} BOOTH_DAEMON_STATE;
int poll_timeout;
-
-
-
-struct booth_config *booth_conf;
struct command_line cl;
static void client_alloc(void)
{
int i;
if (!(clients = realloc(
clients, (client_size + CLIENT_NALLOC) * sizeof(*clients))
) || !(pollfds = realloc(
pollfds, (client_size + CLIENT_NALLOC) * sizeof(*pollfds))
)) {
log_error("can't alloc for client array");
exit(1);
}
for (i = client_size; i < client_size + CLIENT_NALLOC; i++) {
clients[i].workfn = NULL;
clients[i].deadfn = NULL;
clients[i].fd = -1;
pollfds[i].fd = -1;
pollfds[i].revents = 0;
}
client_size += CLIENT_NALLOC;
}
static void client_dead(int ci)
{
struct client *c = clients + ci;
if (c->fd != -1) {
log_debug("removing client %d", c->fd);
close(c->fd);
}
c->fd = -1;
c->workfn = NULL;
if (c->msg) {
free(c->msg);
c->msg = NULL;
c->offset = 0;
}
pollfds[ci].fd = -1;
}
int client_add(int fd, const struct booth_transport *tpt,
void (*workfn)(struct booth_config *conf_ptr, int ci),
void (*deadfn)(int ci))
{
int i;
struct client *c;
if (client_size - 1 <= client_maxi ) {
client_alloc();
}
for (i = 0; i < client_size; i++) {
c = clients + i;
if (c->fd != -1)
continue;
c->workfn = workfn;
if (deadfn)
c->deadfn = deadfn;
else
c->deadfn = client_dead;
c->transport = tpt;
c->fd = fd;
c->msg = NULL;
c->offset = 0;
pollfds[i].fd = fd;
pollfds[i].events = POLLIN;
if (i > client_maxi)
client_maxi = i;
return i;
}
assert(!"no client");
}
int find_client_by_fd(int fd)
{
int i;
if (fd < 0)
return -1;
for (i = 0; i <= client_maxi; i++) {
if (clients[i].fd == fd)
return i;
}
return -1;
}
static int format_peers(struct booth_config *conf_ptr,
char **pdata, unsigned int *len)
{
struct booth_site *s;
char *data, *cp;
char time_str[64];
int i, alloc;
assert(conf_ptr != NULL);
*pdata = NULL;
*len = 0;
alloc = conf_ptr->site_count * (BOOTH_NAME_LEN + 256);
data = malloc(alloc);
if (!data)
return -ENOMEM;
cp = data;
FOREACH_NODE(conf_ptr, i, s) {
if (s == local)
continue;
strftime(time_str, sizeof(time_str), "%F %T",
localtime(&s->last_recv));
cp += snprintf(cp,
alloc - (cp - data),
"%-12s %s, last recv: %s\n",
type_to_string(s->type),
s->addr_string,
time_str);
cp += snprintf(cp,
alloc - (cp - data),
"\tSent pkts:%u error:%u resends:%u\n",
s->sent_cnt,
s->sent_err_cnt,
s->resend_cnt);
cp += snprintf(cp,
alloc - (cp - data),
"\tRecv pkts:%u error:%u authfail:%u invalid:%u\n\n",
s->recv_cnt,
s->recv_err_cnt,
s->sec_cnt,
s->invalid_cnt);
if (alloc - (cp - data) <= 0) {
free(data);
return -ENOMEM;
}
}
*pdata = data;
*len = cp - data;
return 0;
}
void list_peers(struct booth_config *conf_ptr, int fd)
{
char *data;
unsigned int olen;
struct boothc_hdr_msg hdr;
if (format_peers(conf_ptr, &data, &olen) < 0)
goto out;
init_header(conf_ptr, &hdr.header, CL_LIST, 0, 0, RLT_SUCCESS,
0, sizeof(hdr) + olen);
(void) send_header_plus(conf_ptr, fd, &hdr, data, olen);
out:
if (data)
free(data);
}
/* trim trailing spaces if the key is ascii
*/
static void trim_key(struct booth_config *conf_ptr)
{
char *p;
int i;
assert(conf_ptr != NULL);
for (i = 0, p = conf_ptr->authkey; i < conf_ptr->authkey_len; i++, p++)
if (!isascii(*p))
return;
p = conf_ptr->authkey;
while (conf_ptr->authkey_len > 0 && isspace(*p)) {
p++;
conf_ptr->authkey_len--;
}
memmove(conf_ptr->authkey, p, conf_ptr->authkey_len);
p = conf_ptr->authkey + conf_ptr->authkey_len - 1;
while (conf_ptr->authkey_len > 0 && isspace(*p)) {
conf_ptr->authkey_len--;
p--;
}
}
static int read_authkey(struct booth_config *conf_ptr)
{
int fd;
assert(conf_ptr != NULL);
conf_ptr->authkey[0] = '\0';
fd = open(conf_ptr->authfile, O_RDONLY);
if (fd < 0) {
log_error("cannot open %s: %s",
conf_ptr->authfile, strerror(errno));
return -1;
}
if (fstat(fd, &conf_ptr->authstat) < 0) {
log_error("cannot stat authentication file %s (%d): %s",
conf_ptr->authfile, fd, strerror(errno));
close(fd);
return -1;
}
if (conf_ptr->authstat.st_mode & (S_IRGRP | S_IROTH)) {
log_error("%s: file shall not be readable for anyone but the owner",
conf_ptr->authfile);
close(fd);
return -1;
}
conf_ptr->authkey_len = read(fd, conf_ptr->authkey, BOOTH_MAX_KEY_LEN);
close(fd);
trim_key(conf_ptr);
log_debug("read key of size %d in authfile %s",
conf_ptr->authkey_len, conf_ptr->authfile);
/* make sure that the key is of minimum length */
return (conf_ptr->authkey_len >= BOOTH_MIN_KEY_LEN) ? 0 : -1;
}
/* XXX UNUSED */
int update_authkey(struct booth_config *conf_ptr)
{
struct stat statbuf;
assert(conf_ptr != NULL);
if (stat(conf_ptr->authfile, &statbuf) < 0) {
log_error("cannot stat authentication file %s: %s",
conf_ptr->authfile, strerror(errno));
return -1;
}
if (statbuf.st_mtime > conf_ptr->authstat.st_mtime) {
return read_authkey(conf_ptr);
}
return 0;
}
static int setup_config(struct booth_config **conf_pptr, int type)
{
int rv;
assert(conf_pptr != NULL);
- rv = read_config(conf_pptr, cl.configfile, type);
+ rv = read_config(conf_pptr, &booth__transport, cl.configfile, type);
if (rv < 0)
goto out;
if (is_auth_req(*conf_pptr)) {
rv = read_authkey(*conf_pptr);
if (rv < 0)
goto out;
#if HAVE_LIBGCRYPT
if (!gcry_check_version(NULL)) {
log_error("gcry_check_version");
rv = -ENOENT;
goto out;
}
gcry_control(GCRYCTL_DISABLE_SECMEM, 0);
gcry_control(GCRYCTL_INITIALIZATION_FINISHED, 0);
#endif
}
/* Set "local" pointer, ignoring errors. */
if (cl.type == DAEMON && cl.site[0]) {
if (!find_site_by_name(*conf_pptr, cl.site, &local, 1)) {
log_error("Cannot find \"%s\" in the configuration.",
cl.site);
return -EINVAL;
}
local->local = 1;
} else
find_myself(*conf_pptr, NULL,
type == CLIENT || type == GEOSTORE);
rv = check_config(*conf_pptr, type);
if (rv < 0)
goto out;
/* Per default the PID file name is derived from the
* configuration name. */
if (!cl.lockfile[0]) {
snprintf(cl.lockfile, sizeof(cl.lockfile) - 1,
"%s/%s.pid", BOOTH_RUN_DIR, (*conf_pptr)->name);
}
out:
return rv;
}
-static int setup_transport(void)
+static int setup_transport(struct booth_config *conf_ptr)
{
int rv;
- rv = transport()->init(message_recv);
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
+ rv = transport(conf_ptr)->init(conf_ptr, message_recv);
if (rv < 0) {
- log_error("failed to init booth_transport %s", transport()->name);
+ log_error("failed to init booth_transport %s",
+ transport(conf_ptr)->name);
goto out;
}
- rv = booth_transport[TCP].init(NULL);
+ rv = (*conf_ptr->transport)[TCP].init(conf_ptr, NULL);
if (rv < 0) {
log_error("failed to init booth_transport[TCP]");
goto out;
}
out:
return rv;
}
static int write_daemon_state(struct booth_config *conf_ptr, int fd, int state)
{
char buffer[1024];
int rv, size;
size = sizeof(buffer) - 1;
rv = snprintf(buffer, size,
"booth_pid=%d "
"booth_state=%s "
"booth_type=%s "
"booth_cfg_name='%s' "
"booth_id=%d "
"booth_addr_string='%s' "
"booth_port=%d\n",
getpid(),
( state == BOOTHD_STARTED ? "started" :
state == BOOTHD_STARTING ? "starting" : "invalid"),
type_to_string(local->type), conf_ptr->name,
get_local_id(), site_string(local),
site_port(local));
if (rv < 0 || rv == size) {
log_error("Buffer filled up in write_daemon_state().");
return -1;
}
size = rv;
rv = ftruncate(fd, 0);
if (rv < 0) {
log_error("lockfile %s truncate error %d: %s",
cl.lockfile, errno, strerror(errno));
return rv;
}
rv = lseek(fd, 0, SEEK_SET);
if (rv < 0) {
log_error("lseek set fd(%d) offset to 0 error, return(%d), message(%s)",
fd, rv, strerror(errno));
rv = -1;
return rv;
}
rv = write(fd, buffer, size);
if (rv != size) {
log_error("write to fd(%d, %d) returned %d, errno %d, message(%s)",
fd, size,
rv, errno, strerror(errno));
return -1;
}
return 0;
}
static int loop(struct booth_config *conf_ptr, int fd)
{
void (*workfn) (struct booth_config *conf_ptr, int ci);
void (*deadfn) (int ci);
int rv, i;
- rv = setup_transport();
+ rv = setup_transport(conf_ptr);
if (rv < 0)
goto fail;
rv = setup_ticket(conf_ptr);
if (rv < 0)
goto fail;
rv = write_daemon_state(conf_ptr, fd, BOOTHD_STARTED);
if (rv != 0) {
log_error("write daemon state %d to lockfile error %s: %s",
BOOTHD_STARTED, cl.lockfile, strerror(errno));
goto fail;
}
log_info("BOOTH %s daemon started, node id is 0x%08X (%d).",
type_to_string(local->type),
local->site_id, local->site_id);
while (1) {
rv = poll(pollfds, client_maxi + 1, poll_timeout);
if (rv == -1 && errno == EINTR)
continue;
if (rv < 0) {
log_error("poll failed: %s (%d)", strerror(errno), errno);
goto fail;
}
for (i = 0; i <= client_maxi; i++) {
if (clients[i].fd < 0)
continue;
if (pollfds[i].revents & POLLIN) {
workfn = clients[i].workfn;
if (workfn)
workfn(conf_ptr, i);
}
if (pollfds[i].revents &
(POLLERR | POLLHUP | POLLNVAL)) {
deadfn = clients[i].deadfn;
if (deadfn)
deadfn(i);
}
}
process_tickets(conf_ptr);
}
return 0;
fail:
return -1;
}
static int test_reply(cmd_result_t reply_code, cmd_request_t cmd)
{
int rv = 0;
const char *op_str = "";
if (cmd == CMD_GRANT)
op_str = "grant";
else if (cmd == CMD_REVOKE)
op_str = "revoke";
else if (cmd == CMD_LIST)
op_str = "list";
else if (cmd == CMD_PEERS)
op_str = "peers";
else {
log_error("internal error reading reply result!");
return -1;
}
switch (reply_code) {
case RLT_OVERGRANT:
log_info("You're granting a granted ticket. "
"If you wanted to migrate a ticket, "
"use revoke first, then use grant.");
rv = -1;
break;
case RLT_TICKET_IDLE:
log_info("ticket is not owned");
rv = 0;
break;
case RLT_ASYNC:
log_info("%s command sent, result will be returned "
"asynchronously. Please use \"booth list\" to "
"see the outcome.", op_str);
rv = 0;
break;
case RLT_CIB_PENDING:
log_info("%s succeeded (CIB commit pending)", op_str);
/* wait for the CIB commit? */
rv = (cl.options & OPT_WAIT_COMMIT) ? 3 : 0;
break;
case RLT_MORE:
rv = 2;
break;
case RLT_SYNC_SUCC:
case RLT_SUCCESS:
if (cmd != CMD_LIST && cmd != CMD_PEERS)
log_info("%s succeeded!", op_str);
rv = 0;
break;
case RLT_SYNC_FAIL:
log_info("%s failed!", op_str);
rv = -1;
break;
case RLT_INVALID_ARG:
log_error("ticket \"%s\" does not exist",
cl.msg.ticket.id);
rv = -1;
break;
case RLT_AUTH:
log_error("authentication error");
rv = -1;
break;
case RLT_EXT_FAILED:
log_error("before-acquire-handler for ticket \"%s\" failed, grant denied",
cl.msg.ticket.id);
rv = -1;
break;
case RLT_ATTR_PREREQ:
log_error("attr-prereq for ticket \"%s\" failed, grant denied",
cl.msg.ticket.id);
rv = -1;
break;
case RLT_REDIRECT:
/* talk to another site */
rv = 1;
break;
default:
log_error("got an error code: %x", rv);
rv = -1;
}
return rv;
}
static int query_get_string_answer(struct booth_config *conf_ptr,
cmd_request_t cmd)
{
struct booth_site *site;
struct boothc_hdr_msg reply;
struct boothc_header *header;
char *data;
int data_len;
int rv;
struct booth_transport const *tpt;
int (*test_reply_f) (cmd_result_t reply_code, cmd_request_t cmd);
size_t msg_size;
void *request;
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
if (cl.type == GEOSTORE) {
test_reply_f = test_attr_reply;
msg_size = sizeof(cl.attr_msg);
request = &cl.attr_msg;
} else {
test_reply_f = test_reply;
msg_size = sizeof(cl.msg);
request = &cl.msg;
}
header = (struct boothc_header *)request;
data = NULL;
init_header(conf_ptr, header, cmd, 0, cl.options, 0, 0, msg_size);
if (!*cl.site)
site = local;
else if (!find_site_by_name(conf_ptr, cl.site, &site, 1)) {
log_error("cannot find site \"%s\"", cl.site);
rv = ENOENT;
goto out;
}
- tpt = booth_transport + TCP;
+ tpt = *conf_ptr->transport + TCP;
rv = tpt->open(site);
if (rv < 0)
goto out_close;
rv = tpt->send(conf_ptr, site, request, msg_size);
if (rv < 0)
goto out_close;
rv = tpt->recv_auth(conf_ptr, site, &reply, sizeof(reply));
if (rv < 0)
goto out_close;
data_len = ntohl(reply.header.length) - rv;
/* no attribute, or no ticket found */
if (!data_len) {
goto out_test_reply;
}
data = malloc(data_len+1);
if (!data) {
rv = -ENOMEM;
goto out_close;
}
rv = tpt->recv(site, data, data_len);
if (rv < 0)
goto out_close;
*(data+data_len) = '\0';
*(data + data_len) = '\0';
(void)fputs(data, stdout);
fflush(stdout);
rv = 0;
out_test_reply:
rv = test_reply_f(ntohl(reply.header.result), cmd);
out_close:
tpt->close(site);
out:
if (data)
free(data);
return rv;
}
static int do_command(struct booth_config *conf_ptr,
cmd_request_t cmd)
{
struct booth_site *site;
struct boothc_ticket_msg reply;
struct booth_transport const *tpt;
uint32_t leader_id;
int rv;
int reply_cnt = 0, msg_logged = 0;
const char *op_str = "";
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
if (cmd == CMD_GRANT)
op_str = "grant";
else if (cmd == CMD_REVOKE)
op_str = "revoke";
rv = 0;
site = NULL;
/* Always use TCP for client - at least for now. */
- tpt = booth_transport + TCP;
+ tpt = *conf_ptr->transport + TCP;
if (!*cl.site)
site = local;
else {
if (!find_site_by_name(conf_ptr, cl.site, &site, 1)) {
log_error("Site \"%s\" not configured.", cl.site);
goto out_close;
}
}
if (site->type == ARBITRATOR) {
if (site == local) {
log_error("We're just an arbitrator, cannot grant/revoke tickets here.");
} else {
log_error("%s is just an arbitrator, cannot grant/revoke tickets there.", cl.site);
}
goto out_close;
}
assert(site->type == SITE);
/* We don't check for existence of ticket, so that asking can be
* done without local configuration, too.
* Although, that means that the UDP port has to be specified, too. */
if (!cl.msg.ticket.id[0]) {
/* If the loaded configuration has only a single ticket defined, use that. */
if (conf_ptr->ticket_count == 1) {
strncpy(cl.msg.ticket.id, conf_ptr->ticket[0].name,
sizeof(cl.msg.ticket.id));
} else {
log_error("No ticket given.");
goto out_close;
}
}
redirect:
init_header(conf_ptr, &cl.msg.header, cmd, 0, cl.options, 0, 0, sizeof(cl.msg));
rv = tpt->open(site);
if (rv < 0)
goto out_close;
rv = tpt->send(conf_ptr, site, &cl.msg, sendmsglen(&cl.msg));
if (rv < 0)
goto out_close;
read_more:
rv = tpt->recv_auth(conf_ptr, site, &reply, sizeof(reply));
if (rv < 0) {
/* print any errors depending on the code sent by the
* server */
(void)test_reply(ntohl(reply.header.result), cmd);
goto out_close;
}
rv = test_reply(ntohl(reply.header.result), cmd);
if (rv == 1) {
tpt->close(site);
leader_id = ntohl(reply.ticket.leader);
if (!find_site_by_id(conf_ptr, leader_id, &site)) {
log_error("Message with unknown redirect site %x received", leader_id);
rv = -1;
goto out_close;
}
goto redirect;
} else if (rv == 2 || rv == 3) {
/* the server has more to say */
/* don't wait too long */
if (reply_cnt > 1 && !(cl.options & OPT_WAIT)) {
rv = 0;
log_info("Giving up on waiting for the definite result. "
"Please use \"booth list\" later to "
"see the outcome.");
goto out_close;
}
if (reply_cnt == 0) {
log_info("%s request sent, "
"waiting for the result ...", op_str);
msg_logged++;
} else if (rv == 3 && msg_logged < 2) {
log_info("waiting for the CIB commit ...");
msg_logged++;
}
reply_cnt++;
goto read_more;
}
out_close:
if (site)
tpt->close(site);
return rv;
}
static int _lockfile(int mode, int *fdp, pid_t *locked_by)
{
struct flock lock;
int fd, rv;
/* After reboot the directory may not yet exist.
* Try to create it, but ignore errors. */
if (strncmp(cl.lockfile, BOOTH_RUN_DIR,
strlen(BOOTH_RUN_DIR)) == 0)
mkdir(BOOTH_RUN_DIR, 0775);
if (locked_by)
*locked_by = 0;
*fdp = -1;
fd = open(cl.lockfile, mode, 0664);
if (fd < 0)
return errno;
*fdp = fd;
lock.l_type = F_WRLCK;
lock.l_start = 0;
lock.l_whence = SEEK_SET;
lock.l_len = 0;
lock.l_pid = 0;
if (fcntl(fd, F_SETLK, &lock) == 0)
return 0;
rv = errno;
if (locked_by)
if (fcntl(fd, F_GETLK, &lock) == 0)
*locked_by = lock.l_pid;
return rv;
}
static inline int is_root(void)
{
return geteuid() == 0;
}
static int create_lockfile(struct booth_config *conf_ptr)
{
int rv, fd;
fd = -1;
rv = _lockfile(O_CREAT | O_WRONLY, &fd, NULL);
if (fd == -1) {
log_error("lockfile %s open error %d: %s",
cl.lockfile, rv, strerror(rv));
return -1;
}
if (rv < 0) {
log_error("lockfile %s setlk error %d: %s",
cl.lockfile, rv, strerror(rv));
goto fail;
}
rv = write_daemon_state(conf_ptr, fd, BOOTHD_STARTING);
if (rv != 0) {
log_error("write daemon state %d to lockfile error %s: %s",
BOOTHD_STARTING, cl.lockfile, strerror(errno));
goto fail;
}
if (is_root()) {
if (fchown(fd, conf_ptr->uid, conf_ptr->gid) < 0)
log_error("fchown() on lockfile said %d: %s",
errno, strerror(errno));
}
return fd;
fail:
close(fd);
return -1;
}
static void unlink_lockfile(int fd)
{
unlink(cl.lockfile);
close(fd);
}
static void print_usage(void)
{
printf(
"Usage:\n"
" booth list [options]\n"
" booth {grant|revoke} [options] <ticket>\n"
" booth status [options]\n"
"\n"
" list: List all tickets\n"
" grant: Grant ticket to site\n"
" revoke: Revoke ticket\n"
"\n"
"Options:\n"
" -c FILE Specify config file [default " BOOTH_DEFAULT_CONF "]\n"
" Can be a path or just a name without \".conf\" suffix\n"
" -s <site> Connect/grant to a different site\n"
" -F Try to grant the ticket immediately\n"
" even if not all sites are reachable\n"
" For manual tickets:\n"
" grant a manual ticket even if it has been already granted\n"
" -w Wait forever for the outcome of the request\n"
" -C Wait until the ticket is committed to the CIB (grant only)\n"
" -h Print this help\n"
"\n"
"Examples:\n"
"\n"
" # booth list (list tickets)\n"
" # booth grant ticket-A (grant ticket here)\n"
" # booth grant -s 10.121.8.183 ticket-A (grant ticket to site 10.121.8.183)\n"
" # booth revoke ticket-A (revoke ticket)\n"
"\n"
"See the booth(8) man page for more details.\n"
);
}
#define OPTION_STRING "c:Dl:t:s:FhSwC"
#define ATTR_OPTION_STRING "c:Dt:s:h"
void safe_copy(char *dest, char *value, size_t buflen, const char *description) {
int content_len = buflen - 1;
if (strlen(value) >= content_len) {
fprintf(stderr, "'%s' exceeds maximum %s length of %d\n",
value, description, content_len);
exit(EXIT_FAILURE);
}
strncpy(dest, value, content_len);
dest[content_len] = 0;
}
static int host_convert(char *hostname, char *ip_str, size_t ip_size)
{
struct addrinfo *result = NULL, hints = {0};
int re = -1;
memset(&hints, 0, sizeof(hints));
hints.ai_family = AF_INET;
hints.ai_socktype = SOCK_DGRAM;
re = getaddrinfo(hostname, NULL, &hints, &result);
if (re == 0) {
struct in_addr addr = ((struct sockaddr_in *)result->ai_addr)->sin_addr;
const char *re_ntop = inet_ntop(AF_INET, &addr, ip_str, ip_size);
if (re_ntop == NULL) {
re = -1;
}
}
freeaddrinfo(result);
return re;
}
#define cparg(dest, descr) do { \
if (optind >= argc) \
goto missingarg; \
safe_copy(dest, argv[optind], sizeof(dest), descr); \
optind++; \
} while(0)
static int read_arguments(int argc, char **argv)
{
int optchar;
char *arg1 = argv[1];
char *op = NULL;
char *cp;
const char *opt_string = OPTION_STRING;
char site_arg[INET_ADDRSTRLEN] = {0};
int left;
cl.type = 0;
if ((cp = strstr(argv[0], ATTR_PROG)) &&
!strcmp(cp, ATTR_PROG)) {
cl.type = GEOSTORE;
op = argv[1];
optind = 2;
opt_string = ATTR_OPTION_STRING;
} else if (argc > 1 && (strcmp(arg1, "arbitrator") == 0 ||
strcmp(arg1, "site") == 0 ||
strcmp(arg1, "start") == 0 ||
strcmp(arg1, "daemon") == 0)) {
cl.type = DAEMON;
optind = 2;
} else if (argc > 1 && (strcmp(arg1, "status") == 0)) {
cl.type = STATUS;
optind = 2;
} else if (argc > 1 && (strcmp(arg1, "client") == 0)) {
cl.type = CLIENT;
if (argc < 3) {
print_usage();
exit(EXIT_FAILURE);
}
op = argv[2];
optind = 3;
}
if (!cl.type) {
cl.type = CLIENT;
op = argv[1];
optind = 2;
}
if (argc < 2 || !strcmp(arg1, "help") || !strcmp(arg1, "--help") ||
!strcmp(arg1, "-h")) {
if (cl.type == GEOSTORE)
print_geostore_usage();
else
print_usage();
exit(EXIT_SUCCESS);
}
if (!strcmp(arg1, "version") || !strcmp(arg1, "--version") ||
!strcmp(arg1, "-V")) {
printf("%s %s\n", argv[0], RELEASE_STR);
exit(EXIT_SUCCESS);
}
if (cl.type == CLIENT) {
if (!strcmp(op, "list"))
cl.op = CMD_LIST;
else if (!strcmp(op, "grant"))
cl.op = CMD_GRANT;
else if (!strcmp(op, "revoke"))
cl.op = CMD_REVOKE;
else if (!strcmp(op, "peers"))
cl.op = CMD_PEERS;
else {
fprintf(stderr, "client operation \"%s\" is unknown\n",
op);
exit(EXIT_FAILURE);
}
} else if (cl.type == GEOSTORE) {
if (!strcmp(op, "list"))
cl.op = ATTR_LIST;
else if (!strcmp(op, "set"))
cl.op = ATTR_SET;
else if (!strcmp(op, "get"))
cl.op = ATTR_GET;
else if (!strcmp(op, "delete"))
cl.op = ATTR_DEL;
else {
fprintf(stderr, "attribute operation \"%s\" is unknown\n",
op);
exit(EXIT_FAILURE);
}
}
while (optind < argc) {
optchar = getopt(argc, argv, opt_string);
switch (optchar) {
case 'c':
if (strchr(optarg, '/')) {
safe_copy(cl.configfile, optarg,
sizeof(cl.configfile), "config file");
} else {
/* If no "/" in there, use with default directory. */
strcpy(cl.configfile, BOOTH_DEFAULT_CONF_DIR);
cp = cl.configfile + strlen(BOOTH_DEFAULT_CONF_DIR);
assert(cp > cl.configfile);
assert(*(cp-1) == '/');
/* Write at the \0, ie. after the "/" */
safe_copy(cp, optarg,
(sizeof(cl.configfile) -
(cp - cl.configfile) -
strlen(BOOTH_DEFAULT_CONF_EXT)),
"config name");
/* If no extension, append ".conf".
* Space is available, see -strlen() above. */
if (!strchr(cp, '.'))
strcat(cp, BOOTH_DEFAULT_CONF_EXT);
}
break;
case 'D':
debug_level++;
break;
case 'S':
daemonize = 0;
enable_stderr = 1;
break;
case 'l':
safe_copy(cl.lockfile, optarg, sizeof(cl.lockfile), "lock file");
break;
case 't':
if (cl.op == CMD_GRANT || cl.op == CMD_REVOKE) {
safe_copy(cl.msg.ticket.id, optarg,
sizeof(cl.msg.ticket.id), "ticket name");
} else if (cl.type == GEOSTORE) {
safe_copy(cl.attr_msg.attr.tkt_id, optarg,
sizeof(cl.attr_msg.attr.tkt_id), "ticket name");
} else {
print_usage();
exit(EXIT_FAILURE);
}
break;
case 's':
/* For testing and debugging: allow "-s site" also for
* daemon start, so that the address that should be used
* can be set manually.
* This makes it easier to start multiple processes
* on one machine. */
if (cl.type == CLIENT || cl.type == GEOSTORE ||
(cl.type == DAEMON && debug_level)) {
if (strcmp(optarg, OTHER_SITE) &&
host_convert(optarg, site_arg, INET_ADDRSTRLEN) == 0) {
safe_copy(cl.site, site_arg, sizeof(cl.site), "site name");
} else {
safe_copy(cl.site, optarg, sizeof(cl.site), "site name");
}
} else {
log_error("\"-s\" not allowed in daemon mode.");
exit(EXIT_FAILURE);
}
break;
case 'F':
if (cl.type != CLIENT || cl.op != CMD_GRANT) {
log_error("use \"-F\" only for client grant");
exit(EXIT_FAILURE);
}
cl.options |= OPT_IMMEDIATE;
break;
case 'w':
if (cl.type != CLIENT ||
(cl.op != CMD_GRANT && cl.op != CMD_REVOKE)) {
log_error("use \"-w\" only for grant and revoke");
exit(EXIT_FAILURE);
}
cl.options |= OPT_WAIT;
break;
case 'C':
if (cl.type != CLIENT || cl.op != CMD_GRANT) {
log_error("use \"-C\" only for grant");
exit(EXIT_FAILURE);
}
cl.options |= OPT_WAIT | OPT_WAIT_COMMIT;
break;
case 'h':
if (cl.type == GEOSTORE)
print_geostore_usage();
else
print_usage();
exit(EXIT_SUCCESS);
break;
case ':':
case '?':
fprintf(stderr, "Please use '-h' for usage.\n");
exit(EXIT_FAILURE);
break;
case -1:
/* No more parameters on cmdline, only arguments. */
goto extra_args;
default:
goto unknown;
};
}
return 0;
extra_args:
if (cl.type == CLIENT && !cl.msg.ticket.id[0]) {
cparg(cl.msg.ticket.id, "ticket name");
} else if (cl.type == GEOSTORE) {
if (cl.op != ATTR_LIST) {
cparg(cl.attr_msg.attr.name, "attribute name");
}
if (cl.op == ATTR_SET) {
cparg(cl.attr_msg.attr.val, "attribute value");
}
}
if (optind == argc)
return 0;
left = argc - optind;
fprintf(stderr, "Superfluous argument%s: %s%s\n",
left == 1 ? "" : "s",
argv[optind],
left == 1 ? "" : "...");
exit(EXIT_FAILURE);
unknown:
fprintf(stderr, "unknown option: %s\n", argv[optind]);
exit(EXIT_FAILURE);
missingarg:
fprintf(stderr, "not enough arguments\n");
exit(EXIT_FAILURE);
}
static void set_scheduler(void)
{
struct sched_param sched_param;
struct rlimit rlimit;
int rv;
rlimit.rlim_cur = RLIM_INFINITY;
rlimit.rlim_max = RLIM_INFINITY;
setrlimit(RLIMIT_MEMLOCK, &rlimit);
rv = mlockall(MCL_CURRENT | MCL_FUTURE);
if (rv < 0) {
log_error("mlockall failed");
}
rv = sched_get_priority_max(SCHED_RR);
if (rv != -1) {
sched_param.sched_priority = rv;
rv = sched_setscheduler(0, SCHED_RR, &sched_param);
if (rv == -1)
log_error("could not set SCHED_RR priority %d: %s (%d)",
sched_param.sched_priority,
strerror(errno), errno);
} else {
log_error("could not get maximum scheduler priority err %d",
errno);
}
}
static int set_procfs_val(const char *path, const char *val)
{
int rc = -1;
FILE *fp = fopen(path, "w");
if (fp) {
if (fprintf(fp, "%s", val) > 0)
rc = 0;
fclose(fp);
}
return rc;
}
static int do_status(struct booth_config **conf_pptr, int type)
{
pid_t pid;
int rv, status_lock_fd, ret;
const char *reason = NULL;
char lockfile_data[1024], *cp;
assert(conf_pptr != NULL);
ret = PCMK_OCF_NOT_RUNNING;
rv = setup_config(conf_pptr, type);
if (rv) {
reason = "Error reading configuration.";
ret = PCMK_OCF_UNKNOWN_ERROR;
goto quit;
}
if (!local) {
reason = "No Service IP active here.";
goto quit;
}
rv = _lockfile(O_RDWR, &status_lock_fd, &pid);
if (status_lock_fd == -1) {
reason = "No PID file.";
goto quit;
}
if (rv == 0) {
close(status_lock_fd);
reason = "PID file not locked.";
goto quit;
}
if (pid) {
fprintf(stdout, "booth_lockpid=%d ", pid);
fflush(stdout);
}
rv = read(status_lock_fd, lockfile_data, sizeof(lockfile_data) - 1);
if (rv < 4) {
close(status_lock_fd);
reason = "Cannot read lockfile data.";
ret = PCMK_LSB_UNKNOWN_ERROR;
goto quit;
}
lockfile_data[rv] = 0;
close(status_lock_fd);
/* Make sure it's only a single line */
cp = strchr(lockfile_data, '\r');
if (cp)
*cp = 0;
cp = strchr(lockfile_data, '\n');
if (cp)
*cp = 0;
rv = setup_tcp_listener(1);
if (rv == 0) {
reason = "TCP port not in use.";
goto quit;
}
fprintf(stdout, "booth_lockfile='%s' %s\n",
cl.lockfile, lockfile_data);
if (!daemonize)
fprintf(stderr, "Booth at %s port %d seems to be running.\n",
site_string(local), site_port(local));
return 0;
quit:
log_debug("not running: %s", reason);
/* Ie. "DEBUG" */
if (!daemonize)
fprintf(stderr, "not running: %s\n", reason);
return ret;
}
static int limit_this_process(struct booth_config *conf_ptr)
{
int rv;
if (!is_root())
return 0;
if (setregid(conf_ptr->gid, conf_ptr->gid) < 0) {
rv = errno;
log_error("setregid() didn't work: %s", strerror(rv));
return rv;
}
if (setreuid(conf_ptr->uid, conf_ptr->uid) < 0) {
rv = errno;
log_error("setreuid() didn't work: %s", strerror(rv));
return rv;
}
return 0;
}
static int lock_fd = -1;
static void server_exit(void)
{
int rv;
if (lock_fd >= 0) {
/* We might not be able to delete it, but at least
* make it empty. */
rv = ftruncate(lock_fd, 0);
(void)rv;
unlink_lockfile(lock_fd);
}
log_info("exiting");
}
static void sig_exit_handler(int sig)
{
log_info("caught signal %d", sig);
exit(0);
}
static void wait_child_adaptor(int sig)
{
wait_child(booth_conf);
}
static int do_server(struct booth_config **conf_pptr, int type)
{
int rv = -1;
static char log_ent[128] = DAEMON_NAME "-";
assert(conf_pptr != NULL);
rv = setup_config(conf_pptr, type);
if (rv < 0)
return rv;
if (!local) {
log_error("Cannot find myself in the configuration.");
exit(EXIT_FAILURE);
}
if (daemonize) {
if (daemon(0, 0) < 0) {
perror("daemon error");
exit(EXIT_FAILURE);
}
}
/* The lockfile must be written to _after_ the call to daemon(), so
* that the lockfile contains the pid of the daemon, not the parent. */
lock_fd = create_lockfile(*conf_pptr);
if (lock_fd < 0)
return lock_fd;
atexit(server_exit);
strcat(log_ent, type_to_string(local->type));
cl_log_set_entity(log_ent);
cl_log_enable_stderr(enable_stderr ? TRUE : FALSE);
cl_log_set_facility(HA_LOG_FACILITY);
cl_inherit_logging_environment(0);
log_info("BOOTH %s %s daemon is starting",
type_to_string(local->type), RELEASE_STR);
signal(SIGUSR1, (__sighandler_t)tickets_log_info);
signal(SIGTERM, (__sighandler_t)sig_exit_handler);
signal(SIGINT, (__sighandler_t)sig_exit_handler);
/* we'll handle errors there and then */
signal(SIGPIPE, SIG_IGN);
set_scheduler();
/* we don't want to be killed by the OOM-killer */
if (set_procfs_val("/proc/self/oom_score_adj", "-999"))
(void)set_procfs_val("/proc/self/oom_adj", "-16");
set_proc_title("%s %s %s for [%s]:%d",
DAEMON_NAME, cl.configfile, type_to_string(local->type),
site_string(local), site_port(local));
rv = limit_this_process(*conf_pptr);
if (rv)
return rv;
#ifdef COREDUMP_NURSING
if (cl_enable_coredumps(TRUE) < 0){
log_error("enabling core dump failed");
}
cl_cdtocoredir();
prctl(PR_SET_DUMPABLE, (unsigned long)TRUE, 0UL, 0UL, 0UL);
#else
if (chdir(BOOTH_CORE_DIR) < 0) {
log_error("cannot change working directory to %s", BOOTH_CORE_DIR);
}
#endif
signal(SIGCHLD, (__sighandler_t) wait_child_adaptor);
rv = loop(*conf_pptr, lock_fd);
return rv;
}
static int do_client(struct booth_config **conf_pptr)
{
int rv;
rv = setup_config(conf_pptr, CLIENT);
if (rv < 0) {
log_error("cannot read config");
goto out;
}
switch (cl.op) {
case CMD_LIST:
case CMD_PEERS:
rv = query_get_string_answer(*conf_pptr, cl.op);
break;
case CMD_GRANT:
case CMD_REVOKE:
rv = do_command(*conf_pptr, cl.op);
break;
}
out:
return rv;
}
static int do_attr(struct booth_config **conf_pptr)
{
int rv = -1;
assert(conf_pptr != NULL);
rv = setup_config(conf_pptr, GEOSTORE);
if (rv < 0) {
log_error("cannot read config");
goto out;
}
/* We don't check for existence of ticket, so that asking can be
* done without local configuration, too.
* Although, that means that the UDP port has to be specified, too. */
if (!cl.attr_msg.attr.tkt_id[0]) {
/* If the loaded configuration has only a single ticket defined, use that. */
if ((*conf_pptr)->ticket_count == 1) {
strncpy(cl.attr_msg.attr.tkt_id,
(*conf_pptr)->ticket[0].name,
sizeof(cl.attr_msg.attr.tkt_id));
} else {
rv = 1;
log_error("No ticket given.");
goto out;
}
}
switch (cl.op) {
case ATTR_LIST:
case ATTR_GET:
rv = query_get_string_answer(*conf_pptr, cl.op);
break;
case ATTR_SET:
case ATTR_DEL:
rv = do_attr_command((*conf_pptr), cl.op);
break;
}
out:
return rv;
}
int main(int argc, char *argv[], char *envp[])
{
int rv;
const char *cp;
#ifdef LOGGING_LIBQB
enum qb_log_target_slot i;
#endif
init_set_proc_title(argc, argv, envp);
get_time(&start_time);
memset(&cl, 0, sizeof(cl));
strncpy(cl.configfile,
BOOTH_DEFAULT_CONF, BOOTH_PATH_LEN - 1);
cl.lockfile[0] = 0;
debug_level = 0;
cp = ((cp = strstr(argv[0], ATTR_PROG)) && !strcmp(cp, ATTR_PROG)
? ATTR_PROG
: "booth");
#ifndef LOGGING_LIBQB
cl_log_set_entity(cp);
#else
qb_log_init(cp, LOG_USER, LOG_DEBUG); /* prio driven by debug_level */
for (i = QB_LOG_TARGET_START; i < QB_LOG_TARGET_MAX; i++) {
if (i == QB_LOG_SYSLOG || i == QB_LOG_BLACKBOX)
continue;
qb_log_format_set(i, "%t %H %N: [%P]: %p: %b");
}
(void) qb_log_filter_ctl(QB_LOG_STDERR, QB_LOG_FILTER_ADD,
QB_LOG_FILTER_FILE, "*", LOG_DEBUG);
#endif
cl_log_enable_stderr(TRUE);
cl_log_set_facility(0);
rv = read_arguments(argc, argv);
if (rv < 0)
goto out;
switch (cl.type) {
case STATUS:
rv = do_status(&booth_conf, cl.type);
break;
case ARBITRATOR:
case DAEMON:
case SITE:
rv = do_server(&booth_conf, cl.type);
break;
case CLIENT:
rv = do_client(&booth_conf);
break;
case GEOSTORE:
rv = do_attr(&booth_conf);
break;
}
out:
#ifdef LOGGING_LIBQB
qb_log_fini();
#endif
/* Normalize values. 0x100 would be seen as "OK" by waitpid(). */
return (rv >= 0 && rv < 0x70) ? rv : 1;
}
diff --git a/src/ticket.c b/src/ticket.c
index 91133d5..396b10b 100644
--- a/src/ticket.c
+++ b/src/ticket.c
@@ -1,1480 +1,1481 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#include <arpa/inet.h>
#include <inttypes.h>
#include <stdio.h>
#include <assert.h>
#include <time.h>
#include "b_config.h"
#ifndef RANGE2RANDOM_GLIB
#include <clplumbing/cl_random.h>
#else
#include "alt/range2random_glib.h"
#endif
#include "ticket.h"
#include "config.h"
#include "pacemaker.h"
#include "inline-fn.h"
#include "log.h"
#include "booth.h"
#include "raft.h"
#include "handler.h"
#include "request.h"
#include "manual.h"
#define TK_LINE 256
extern int TIME_RES;
/* Untrusted input, must fit (incl. \0) in a buffer of max chars. */
int check_max_len_valid(const char *s, int max)
{
int i;
for(i=0; i<max; i++)
if (s[i] == 0)
return 1;
return 0;
}
int find_ticket_by_name(struct booth_config *conf_ptr,
const char *ticket, struct ticket_config **found)
{
struct ticket_config *tk;
int i;
assert(conf_ptr != NULL);
if (found)
*found = NULL;
FOREACH_TICKET(conf_ptr, i, tk) {
if (!strncmp(tk->name, ticket, sizeof(tk->name))) {
if (found)
*found = tk;
return 1;
}
}
return 0;
}
int check_ticket(struct booth_config *conf_ptr, const char *ticket,
struct ticket_config **found)
{
if (found)
*found = NULL;
if (conf_ptr == NULL)
return 0;
if (!check_max_len_valid(ticket, sizeof(conf_ptr->ticket[0].name)))
return 0;
return find_ticket_by_name(conf_ptr, ticket, found);
}
/* XXX UNUSED */
int check_site(struct booth_config *conf_ptr, const char *site,
int *is_local)
{
struct booth_site *node;
if (!check_max_len_valid(site, sizeof(node->addr_string)))
return 0;
if (find_site_by_name(conf_ptr, site, &node, 0)) {
*is_local = node->local;
return 1;
}
return 0;
}
/* is it safe to commit the grant?
* if we didn't hear from all sites on the initial grant, we may
* need to delay the commit
*
* TODO: investigate possibility to devise from history whether a
* missing site could be holding a ticket or not
*/
static int ticket_dangerous(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int tdiff;
/* we may be invoked often, don't spam the log unnecessarily
*/
static int no_log_delay_msg;
if (!is_time_set(&tk->delay_commit))
return 0;
if (is_past(&tk->delay_commit) || all_sites_replied(conf_ptr, tk)) {
if (tk->leader == local) {
tk_log_info("%s, committing to CIB",
is_past(&tk->delay_commit) ?
"ticket delay expired" : "all sites replied");
}
time_reset(&tk->delay_commit);
no_log_delay_msg = 0;
return 0;
}
tdiff = time_left(&tk->delay_commit);
tk_log_debug("delay ticket commit for another " intfmt(tdiff));
if (!no_log_delay_msg) {
tk_log_info("delaying ticket commit to CIB for " intfmt(tdiff));
tk_log_info("(or all sites are reached)");
no_log_delay_msg = 1;
}
return 1;
}
int ticket_write(struct booth_config *conf_ptr, struct ticket_config *tk)
{
assert(conf_ptr != NULL);
if (local->type != SITE)
return -EINVAL;
if (ticket_dangerous(conf_ptr, tk))
return 1;
if (tk->leader == local) {
if (tk->state != ST_LEADER) {
tk_log_info("ticket state not yet consistent, "
"delaying ticket grant to CIB");
return 1;
}
pcmk_handler.grant_ticket(tk);
} else {
pcmk_handler.revoke_ticket(tk);
}
tk->update_cib = 0;
return 0;
}
void save_committed_tkt(struct ticket_config *tk)
{
if (!tk->last_valid_tk) {
tk->last_valid_tk = malloc(sizeof(struct ticket_config));
if (!tk->last_valid_tk) {
log_error("out of memory");
return;
}
}
memcpy(tk->last_valid_tk, tk, sizeof(struct ticket_config));
}
static void ext_prog_failed(struct booth_config *conf_ptr,
struct ticket_config *tk, int start_election)
{
if (!is_manual(tk)) {
/* Give it to somebody else.
* Just send a VOTE_FOR message, so the
* others can start elections. */
if (leader_and_valid(tk)) {
save_committed_tkt(tk);
reset_ticket(tk);
ticket_write(conf_ptr, tk);
if (start_election) {
ticket_broadcast(conf_ptr, tk, OP_VOTE_FOR,
OP_REQ_VOTE, RLT_SUCCESS,
OR_LOCAL_FAIL);
}
}
} else {
/* There is not much we can do now because
* the manual ticket cannot be relocated.
* Just warn the user. */
if (tk->leader == local) {
save_committed_tkt(tk);
reset_ticket(tk);
ticket_write(conf_ptr, tk);
log_error("external test failed on the specified machine, cannot acquire a manual ticket");
}
}
}
#define attr_found(geo_ap, ap) \
((geo_ap) && !strcmp((geo_ap)->val, (ap)->attr_val))
int check_attr_prereq(struct ticket_config *tk, grant_type_e grant_type)
{
GList *el;
struct attr_prereq *ap;
struct geo_attr *geo_ap;
for (el = g_list_first(tk->attr_prereqs); el; el = g_list_next(el))
{
ap = (struct attr_prereq *)el->data;
if (ap->grant_type != grant_type)
continue;
geo_ap = (struct geo_attr *)g_hash_table_lookup(tk->attr, ap->attr_name);
switch(ap->op) {
case ATTR_OP_EQ:
if (!attr_found(geo_ap, ap))
goto fail;
break;
case ATTR_OP_NE:
if (attr_found(geo_ap, ap))
goto fail;
break;
default:
break;
}
}
return 0;
fail:
tk_log_warn("'%s' attr-prereq failed", ap->attr_name);
return 1;
}
/* do we need to run the external program?
* or we already done that and waiting for the outcome
* or program exited and we can collect the status
* return codes
* 0: no program defined
* RUNCMD_MORE: program forked, results later
* != 0: executing program failed (or some other failure)
*/
static int do_ext_prog(struct booth_config *conf_ptr,
struct ticket_config *tk, int start_election)
{
int rv = 0;
if (!tk_test.path)
return 0;
switch(tk_test.progstate) {
case EXTPROG_IDLE:
rv = run_handler(conf_ptr, tk);
if (rv == RUNCMD_ERR) {
tk_log_warn("couldn't run external test, not allowed to acquire ticket");
ext_prog_failed(conf_ptr, tk, start_election);
}
break;
case EXTPROG_RUNNING:
/* should never get here, but just in case */
rv = RUNCMD_MORE;
break;
case EXTPROG_EXITED:
rv = tk_test_exit_status(tk);
if (rv) {
ext_prog_failed(conf_ptr, tk, start_election);
}
break;
case EXTPROG_IGNORE:
/* nothing to do here */
break;
}
return rv;
}
/* Try to acquire a ticket
* Could be manual grant or after start (if the ticket is granted
* and still valid in the CIB)
* If the external program needs to run, this is run twice, once
* to start the program, and then to get the result and start
* elections.
*/
static int acquire_ticket(struct booth_config *conf_ptr,
struct ticket_config *tk, cmd_reason_t reason)
{
int rv;
if (reason == OR_ADMIN && check_attr_prereq(tk, GRANT_MANUAL))
return RLT_ATTR_PREREQ;
switch(do_ext_prog(conf_ptr, tk, 0)) {
case 0:
/* everything fine */
break;
case RUNCMD_MORE:
/* need to wait for the outcome before starting elections */
return 0;
default:
return RLT_EXT_FAILED;
}
if (is_manual(tk)) {
rv = manual_selection(conf_ptr, tk, local, 1, reason);
} else {
rv = new_election(conf_ptr, tk, local, 1, reason);
}
return rv ? RLT_SYNC_FAIL : 0;
}
/** Try to get the ticket for the local site.
* */
static int do_grant_ticket(struct booth_config *conf_ptr,
struct ticket_config *tk, int options)
{
int rv;
tk_log_info("granting ticket");
if (tk->leader == local)
return RLT_SUCCESS;
if (is_owned(tk)) {
if (is_manual(tk) && (options & OPT_IMMEDIATE)) {
/* -F flag has been used while granting a manual ticket.
* The ticket will be granted and may end up being granted
* on multiple sites */
tk_log_warn("manual ticket forced to be granted! be aware that "
"you may end up having two sites holding the same manual "
"ticket! revoke the ticket from the unnecessary site!");
} else {
return RLT_OVERGRANT;
}
}
set_future_time(&tk->delay_commit, tk->term_duration + tk->acquire_after);
if (options & OPT_IMMEDIATE) {
tk_log_warn("granting ticket immediately! If there are "
"unreachable sites, _hope_ you are sure that they don't "
"have the ticket!");
time_reset(&tk->delay_commit);
}
rv = acquire_ticket(conf_ptr, tk, OR_ADMIN);
if (rv) {
time_reset(&tk->delay_commit);
return rv;
} else {
return RLT_MORE;
}
}
static void start_revoke_ticket(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
tk_log_info("revoking ticket");
save_committed_tkt(tk);
reset_ticket_and_set_no_leader(tk);
ticket_write(conf_ptr, tk);
ticket_broadcast(conf_ptr, tk, OP_REVOKE, OP_ACK, RLT_SUCCESS, OR_ADMIN);
}
/** Ticket revoke.
* Only to be started from the leader. */
static int do_revoke_ticket(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
if (tk->acks_expected) {
tk_log_info("delay ticket revoke until the current operation finishes");
set_next_state(tk, ST_INIT);
return RLT_MORE;
} else {
start_revoke_ticket(conf_ptr, tk);
return RLT_SUCCESS;
}
}
static int number_sites_marked_as_granted(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
struct booth_site *ignored __attribute__((unused));
int i, result = 0;
assert(conf_ptr != NULL);
FOREACH_NODE(conf_ptr, i, ignored) {
result += tk->sites_where_granted[i];
}
return result;
}
static int list_ticket(struct booth_config *conf_ptr, char **pdata,
unsigned int *len)
{
struct ticket_config *tk;
struct booth_site *site;
char timeout_str[64];
char pending_str[64];
char *data, *cp;
int i, alloc, site_index;
time_t ts;
int multiple_grant_warning_length = 0;
assert(conf_ptr != NULL);
*pdata = NULL;
*len = 0;
alloc = conf_ptr->ticket_count * (BOOTH_NAME_LEN * 2 + 128 + 16);
FOREACH_TICKET(conf_ptr, i, tk) {
multiple_grant_warning_length = \
number_sites_marked_as_granted(conf_ptr, tk);
if (multiple_grant_warning_length > 1) {
// 164: 55 + 45 + 2*number_of_multiple_sites + some margin
alloc += 164 + BOOTH_NAME_LEN * (1+multiple_grant_warning_length);
}
}
data = malloc(alloc);
if (!data)
return -ENOMEM;
cp = data;
FOREACH_TICKET(conf_ptr, i, tk) {
if ((!is_manual(tk)) && is_time_set(&tk->term_expires)) {
/* Manual tickets doesn't have term_expires defined */
ts = wall_ts(&tk->term_expires);
strftime(timeout_str, sizeof(timeout_str), "%F %T",
localtime(&ts));
} else
strcpy(timeout_str, "INF");
if (tk->leader == local && is_time_set(&tk->delay_commit)
&& !is_past(&tk->delay_commit)) {
ts = wall_ts(&tk->delay_commit);
strcpy(pending_str, " (commit pending until ");
strftime(pending_str + strlen(" (commit pending until "),
sizeof(pending_str) - strlen(" (commit pending until ") - 1,
"%F %T", localtime(&ts));
strcat(pending_str, ")");
} else
*pending_str = '\0';
cp += snprintf(cp,
alloc - (cp - data),
"ticket: %s, leader: %s",
tk->name,
ticket_leader_string(tk));
if (is_owned(tk)) {
cp += snprintf(cp,
alloc - (cp - data),
", expires: %s%s",
timeout_str,
pending_str);
}
if (is_manual(tk)) {
cp += snprintf(cp,
alloc - (cp - data),
" [manual mode]");
}
cp += snprintf(cp, alloc - (cp - data), "\n");
if (alloc - (cp - data) <= 0) {
free(data);
return -ENOMEM;
}
}
FOREACH_TICKET(conf_ptr, i, tk) {
multiple_grant_warning_length = \
number_sites_marked_as_granted(conf_ptr, tk);
if (multiple_grant_warning_length > 1) {
cp += snprintf(cp,
alloc - (cp - data),
"\nWARNING: The ticket %s is granted to multiple sites: ", // ~55 characters
tk->name);
FOREACH_NODE(conf_ptr, site_index, site) {
if (tk->sites_where_granted[site_index] > 0) {
cp += snprintf(cp,
alloc - (cp - data),
"%s",
site_string(site));
if (--multiple_grant_warning_length > 0) {
cp += snprintf(cp,
alloc - (cp - data),
", ");
}
}
}
cp += snprintf(cp,
alloc - (cp - data),
". Revoke the ticket from the faulty sites.\n"); // ~45 characters
}
}
*pdata = data;
*len = cp - data;
return 0;
}
void disown_ticket(struct ticket_config *tk)
{
set_leader(tk, NULL);
tk->is_granted = 0;
get_time(&tk->term_expires);
}
int disown_if_expired(struct ticket_config *tk)
{
if (is_past(&tk->term_expires) ||
!tk->leader) {
disown_ticket(tk);
return 1;
}
return 0;
}
void reset_ticket(struct ticket_config *tk)
{
ignore_ext_test(tk);
disown_ticket(tk);
no_resends(tk);
set_state(tk, ST_INIT);
set_next_state(tk, 0);
tk->voted_for = NULL;
}
void reset_ticket_and_set_no_leader(struct ticket_config *tk)
{
mark_ticket_as_revoked_from_leader(tk);
reset_ticket(tk);
tk->leader = no_leader;
tk_log_debug("ticket leader set to no_leader");
}
static void log_reacquire_reason(struct ticket_config *tk)
{
int valid;
const char *where_granted = "\0";
char buff[75];
valid = is_time_set(&tk->term_expires) && !is_past(&tk->term_expires);
if (tk->leader == local) {
where_granted = "granted here";
} else {
snprintf(buff, sizeof(buff), "granted to %s",
site_string(tk->leader));
where_granted = buff;
}
if (!valid) {
tk_log_warn("%s, but not valid "
"anymore (will try to reacquire)", where_granted);
}
if (tk->is_granted && tk->leader != local) {
if (tk->leader && tk->leader != no_leader) {
tk_log_error("granted here, but also %s, "
"that's really too bad (will try to reacquire)",
where_granted);
} else {
tk_log_warn("granted here, but we're "
"not recorded as the grantee (will try to reacquire)");
}
}
}
void update_ticket_state(struct booth_config *conf_ptr,
struct ticket_config *tk, struct booth_site *sender)
{
if (tk->state == ST_CANDIDATE) {
tk_log_info("learned from %s about "
"newer ticket, stopping elections",
site_string(sender));
/* there could be rejects coming from others; don't log
* warnings unnecessarily */
tk->expect_more_rejects = 1;
}
if (tk->leader == local || tk->is_granted) {
/* message from a live leader with valid ticket? */
if (sender == tk->leader && term_time_left(tk)) {
if (tk->is_granted) {
tk_log_warn("ticket was granted here, "
"but it's live at %s (revoking here)",
site_string(sender));
} else {
tk_log_info("ticket live at %s",
site_string(sender));
}
disown_ticket(tk);
ticket_write(conf_ptr, tk);
set_state(tk, ST_FOLLOWER);
set_next_state(tk, ST_FOLLOWER);
} else {
if (tk->state == ST_CANDIDATE) {
set_state(tk, ST_FOLLOWER);
}
set_next_state(tk, ST_LEADER);
}
} else {
if (!tk->leader || tk->leader == no_leader) {
if (sender)
tk_log_info("ticket is not granted");
else
tk_log_info("ticket is not granted (from CIB)");
set_state(tk, ST_INIT);
} else {
if (sender)
tk_log_info("ticket granted to %s (says %s)",
site_string(tk->leader),
tk->leader == sender ? "they" : site_string(sender));
else
tk_log_info("ticket granted to %s (from CIB)",
site_string(tk->leader));
set_state(tk, ST_FOLLOWER);
/* just make sure that we check the ticket soon */
set_next_state(tk, ST_FOLLOWER);
}
}
}
int setup_ticket(struct booth_config *conf_ptr)
{
struct ticket_config *tk;
int i;
assert(conf_ptr != NULL);
FOREACH_TICKET(conf_ptr, i, tk) {
reset_ticket(tk);
if (local->type == SITE) {
if (!pcmk_handler.load_ticket(conf_ptr, tk)) {
update_ticket_state(conf_ptr, tk, NULL);
}
tk->update_cib = 1;
}
tk_log_info("broadcasting state query");
/* wait until all send their status (or the first
* timeout) */
tk->start_postpone = 1;
ticket_broadcast(conf_ptr, tk, OP_STATUS, OP_MY_INDEX,
RLT_SUCCESS, 0);
}
return 0;
}
int ticket_answer_list(struct booth_config *conf_ptr, int fd)
{
char *data;
int rv;
unsigned int olen;
struct boothc_hdr_msg hdr;
rv = list_ticket(conf_ptr, &data, &olen);
if (rv < 0)
goto out;
init_header(conf_ptr, &hdr.header, CL_LIST, 0, 0, RLT_SUCCESS, 0,
sizeof(hdr) + olen);
rv = send_header_plus(conf_ptr, fd, &hdr, data, olen);
out:
if (data)
free(data);
return rv;
}
int process_client_request(struct booth_config *conf_ptr,
struct client *req_client, void *buf)
{
int rv, rc = 1;
struct ticket_config *tk;
int cmd;
struct boothc_ticket_msg omsg;
struct boothc_ticket_msg *msg;
msg = (struct boothc_ticket_msg *)buf;
cmd = ntohl(msg->header.cmd);
if (!check_ticket(conf_ptr, msg->ticket.id, &tk)) {
log_warn("client referenced unknown ticket %s",
msg->ticket.id);
rv = RLT_INVALID_ARG;
goto reply_now;
}
/* Perform the initial check before granting
* an already granted non-manual ticket */
if ((!is_manual(tk) && (cmd == CMD_GRANT) && is_owned(tk))) {
log_warn("client wants to grant an (already granted!) ticket %s",
msg->ticket.id);
rv = RLT_OVERGRANT;
goto reply_now;
}
if ((cmd == CMD_REVOKE) && !is_owned(tk)) {
log_info("client wants to revoke a free ticket %s",
msg->ticket.id);
rv = RLT_TICKET_IDLE;
goto reply_now;
}
if ((cmd == CMD_REVOKE) && tk->leader != local) {
tk_log_info("not granted here, redirect to %s",
ticket_leader_string(tk));
rv = RLT_REDIRECT;
goto reply_now;
}
if (cmd == CMD_REVOKE)
rv = do_revoke_ticket(conf_ptr, tk);
else
rv = do_grant_ticket(conf_ptr, tk, ntohl(msg->header.options));
if (rv == RLT_MORE) {
/* client may receive further notifications, save the
* request for further processing */
add_req(tk, req_client, msg);
tk_log_debug("queue request %s for client %d",
state_to_string(cmd), req_client->fd);
rc = 0; /* we're not yet done with the message */
}
reply_now:
init_ticket_msg(conf_ptr, &omsg, CL_RESULT, 0, rv, 0, tk);
send_client_msg(conf_ptr, req_client->fd, &omsg);
return rc;
}
int notify_client(struct booth_config *conf_ptr, struct ticket_config *tk,
int client_fd, struct boothc_ticket_msg *msg)
{
struct boothc_ticket_msg omsg;
void (*deadfn) (int ci);
int rv, rc, ci;
int cmd, options;
struct client *req_client;
cmd = ntohl(msg->header.cmd);
options = ntohl(msg->header.options);
rv = tk->outcome;
ci = find_client_by_fd(client_fd);
if (ci < 0) {
tk_log_info("client %d (request %s) left before being notified",
client_fd, state_to_string(cmd));
return 0;
}
tk_log_debug("notifying client %d (request %s)",
client_fd, state_to_string(cmd));
init_ticket_msg(conf_ptr, &omsg, CL_RESULT, 0, rv, 0, tk);
rc = send_client_msg(conf_ptr, client_fd, &omsg);
if (rc == 0 && ((rv == RLT_MORE) ||
(rv == RLT_CIB_PENDING && (options & OPT_WAIT_COMMIT)))) {
/* more to do here, keep the request */
return 1;
} else {
/* we sent a definite answer or there was a write error, drop
* the client */
if (rc) {
tk_log_debug("failed to notify client %d (request %s)",
client_fd, state_to_string(cmd));
} else {
tk_log_debug("client %d (request %s) got final notification",
client_fd, state_to_string(cmd));
}
req_client = clients + ci;
deadfn = req_client->deadfn;
if(deadfn) {
deadfn(ci);
}
return 0; /* we're done with this request */
}
}
int ticket_broadcast(struct booth_config *conf_ptr,
struct ticket_config *tk, cmd_request_t cmd,
cmd_request_t expected_reply, cmd_result_t res,
cmd_reason_t reason)
{
struct boothc_ticket_msg msg;
init_ticket_msg(conf_ptr, &msg, cmd, 0, res, reason, tk);
tk_log_debug("broadcasting '%s' (term=%d, valid=%d)",
state_to_string(cmd),
ntohl(msg.ticket.term),
msg_term_time(&msg));
tk->last_request = cmd;
if (expected_reply) {
expect_replies(tk, expected_reply);
}
ticket_activate_timeout(tk);
- return transport()->broadcast_auth(conf_ptr, &msg, sendmsglen(&msg));
+ return transport(conf_ptr)->broadcast_auth(conf_ptr, &msg,
+ sendmsglen(&msg));
}
/* update the ticket on the leader, write it to the CIB, and
send out the update message to others with the new expiry
time
*/
int leader_update_ticket(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int rv = 0, rv2;
timetype now;
if (tk->ticket_updated >= 2)
return 0;
/* for manual tickets, we don't set time expiration */
if (!is_manual(tk)) {
if (tk->ticket_updated < 1) {
tk->ticket_updated = 1;
get_time(&now);
copy_time(&now, &tk->last_renewal);
set_future_time(&tk->term_expires, tk->term_duration);
rv = ticket_broadcast(conf_ptr, tk, OP_UPDATE, OP_ACK,
RLT_SUCCESS, 0);
}
}
if (tk->ticket_updated < 2) {
rv2 = ticket_write(conf_ptr, tk);
switch(rv2) {
case 0:
tk->ticket_updated = 2;
tk->outcome = RLT_SUCCESS;
foreach_tkt_req(conf_ptr, tk, notify_client);
break;
case 1:
if (tk->outcome != RLT_CIB_PENDING) {
tk->outcome = RLT_CIB_PENDING;
foreach_tkt_req(conf_ptr, tk, notify_client);
}
break;
default:
break;
}
}
return rv;
}
static void log_lost_servers(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
struct booth_site *n;
int i;
assert(conf_ptr != NULL);
if (tk->retry_number > 1)
/* log those that we couldn't reach, but do
* that only on the first retry
*/
return;
FOREACH_NODE(conf_ptr, i, n) {
if (!(tk->acks_received & n->bitmask)) {
tk_log_warn("%s %s didn't acknowledge our %s, "
"will retry %d times",
(n->type == ARBITRATOR ? "arbitrator" : "site"),
site_string(n),
state_to_string(tk->last_request),
tk->retries);
}
}
}
static void resend_msg(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
struct booth_site *n;
int i;
assert(conf_ptr != NULL);
if (!(tk->acks_received ^ local->bitmask)) {
ticket_broadcast(conf_ptr, tk, tk->last_request, 0,
RLT_SUCCESS, 0);
} else {
FOREACH_NODE(conf_ptr, i, n) {
if (!(tk->acks_received & n->bitmask)) {
n->resend_cnt++;
tk_log_debug("resending %s to %s",
state_to_string(tk->last_request),
site_string(n)
);
send_msg(conf_ptr, tk->last_request, tk, n,
NULL);
}
}
ticket_activate_timeout(tk);
}
}
static void handle_resends(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int ack_cnt;
if (++tk->retry_number > tk->retries) {
tk_log_info("giving up on sending retries");
no_resends(tk);
set_ticket_wakeup(tk);
return;
}
/* try to reach some sites again if we just stepped down */
if (tk->last_request == OP_VOTE_FOR) {
tk_log_warn("no answers to our VtFr request to step down (try #%d), "
"we are alone",
tk->retry_number);
goto just_resend;
}
if (!majority_of_bits(conf_ptr, tk, tk->acks_received)) {
ack_cnt = count_bits(tk->acks_received) - 1;
if (!ack_cnt) {
tk_log_warn("no answers to our request (try #%d), "
"we are alone",
tk->retry_number);
} else {
tk_log_warn("not enough answers to our request (try #%d): "
"only got %d answers",
tk->retry_number,
ack_cnt);
}
} else {
log_lost_servers(conf_ptr, tk);
}
just_resend:
resend_msg(conf_ptr, tk);
}
int postpone_ticket_processing(struct ticket_config *tk)
{
extern timetype start_time;
return tk->start_postpone &&
(-time_left(&start_time) < tk->timeout);
}
#define has_extprog_exited(tk) ((tk)->clu_test.progstate == EXTPROG_EXITED)
static void process_next_state(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int rv;
switch(tk->next_state) {
case ST_LEADER:
if (has_extprog_exited(tk)) {
if (tk->state != ST_LEADER) {
rv = acquire_ticket(conf_ptr, tk, OR_ADMIN);
if (rv != 0) { /* external program failed */
tk->outcome = rv;
foreach_tkt_req(conf_ptr, tk, notify_client);
}
}
} else {
log_reacquire_reason(tk);
acquire_ticket(conf_ptr, tk, OR_REACQUIRE);
}
break;
case ST_INIT:
no_resends(tk);
start_revoke_ticket(conf_ptr, tk);
tk->outcome = RLT_SUCCESS;
foreach_tkt_req(conf_ptr, tk, notify_client);
break;
/* wanting to be follower is not much of an ambition; no
* processing, just return; don't reset start_postpone until
* we got some replies to status */
case ST_FOLLOWER:
return;
default:
break;
}
tk->start_postpone = 0;
}
static void ticket_lost(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int reason = OR_TKT_LOST;
if (tk->leader != local) {
tk_log_warn("lost at %s", site_string(tk->leader));
} else {
if (is_ext_prog_running(tk)) {
ext_prog_timeout(tk);
reason = OR_LOCAL_FAIL;
} else {
tk_log_warn("lost majority (revoking locally)");
reason = tk->election_reason ? tk->election_reason : OR_REACQUIRE;
}
}
tk->lost_leader = tk->leader;
save_committed_tkt(tk);
mark_ticket_as_revoked_from_leader(tk);
reset_ticket(tk);
set_state(tk, ST_FOLLOWER);
if (local->type == SITE) {
ticket_write(conf_ptr, tk);
schedule_election(tk, reason);
}
}
static void next_action(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
int rv;
switch(tk->state) {
case ST_INIT:
/* init state, handle resends for ticket revoke */
/* and rebroadcast if stepping down */
/* try to acquire ticket on grant */
if (has_extprog_exited(tk)) {
rv = acquire_ticket(conf_ptr, tk, OR_ADMIN);
if (rv != 0) { /* external program failed */
tk->outcome = rv;
foreach_tkt_req(conf_ptr, tk, notify_client);
}
} else {
if (tk->acks_expected) {
handle_resends(conf_ptr, tk);
}
}
break;
case ST_FOLLOWER:
if (!is_manual(tk)) {
/* leader/ticket lost? and we didn't vote yet */
tk_log_debug("leader: %s, voted_for: %s",
site_string(tk->leader),
site_string(tk->voted_for));
if (!tk->leader) {
if (!tk->voted_for || !tk->in_election) {
disown_ticket(tk);
if (!new_election(conf_ptr, tk, NULL,
1, OR_AGAIN)) {
ticket_activate_timeout(tk);
}
} else {
/* we should restart elections in case nothing
* happens in the meantime */
tk->in_election = 0;
ticket_activate_timeout(tk);
}
}
} else {
/* for manual tickets, also try to acquire ticket on grant
* in the Follower state (because we may end up having
* two Leaders) */
if (has_extprog_exited(tk)) {
rv = acquire_ticket(conf_ptr, tk, OR_ADMIN);
if (rv != 0) { /* external program failed */
tk->outcome = rv;
foreach_tkt_req(conf_ptr, tk,
notify_client);
}
} else {
/* Otherwise, just send ACKs if needed */
if (tk->acks_expected) {
handle_resends(conf_ptr, tk);
}
}
}
break;
case ST_CANDIDATE:
/* elections timed out? */
elections_end(conf_ptr, tk);
break;
case ST_LEADER:
/* timeout or ticket renewal? */
if (tk->acks_expected) {
handle_resends(conf_ptr, tk);
if (majority_of_bits(conf_ptr, tk, tk->acks_received)) {
leader_update_ticket(conf_ptr, tk);
}
} else {
/* this is ticket renewal, run local test */
if (!do_ext_prog(conf_ptr, tk, 1)) {
ticket_broadcast(conf_ptr, tk, OP_HEARTBEAT,
OP_ACK, RLT_SUCCESS, 0);
tk->ticket_updated = 0;
}
}
break;
default:
break;
}
}
static void ticket_cron(struct booth_config *conf_ptr,
struct ticket_config *tk)
{
/* don't process the tickets too early after start */
if (postpone_ticket_processing(tk)) {
tk_log_debug("ticket processing postponed (start_postpone=%d)",
tk->start_postpone);
/* but run again soon */
ticket_activate_timeout(tk);
return;
}
/* no need for status resends, we hope we got at least one
* my_index back */
if (tk->acks_expected == OP_MY_INDEX) {
no_resends(tk);
}
/* after startup, we need to decide what to do based on the
* current ticket state; tk->next_state has a hint
* also used for revokes which had to be delayed
*/
if (tk->next_state) {
process_next_state(conf_ptr, tk);
goto out;
}
/* Has an owner, has an expiry date, and expiry date in the past?
* For automatic tickets, losing the ticket must happen
* in _every_ state.
*/
if ((!is_manual(tk)) &&
is_owned(tk) && is_time_set(&tk->term_expires)
&& is_past(&tk->term_expires)) {
ticket_lost(conf_ptr, tk);
goto out;
}
next_action(conf_ptr, tk);
out:
tk->next_state = 0;
if (!tk->in_election && tk->update_cib)
ticket_write(conf_ptr, tk);
}
void process_tickets(struct booth_config *conf_ptr)
{
struct ticket_config *tk;
int i;
timetype last_cron;
assert(conf_ptr != NULL);
FOREACH_TICKET(conf_ptr, i, tk) {
if (!has_extprog_exited(tk) &&
is_time_set(&tk->next_cron) && !is_past(&tk->next_cron))
continue;
tk_log_debug("ticket cron");
copy_time(&tk->next_cron, &last_cron);
ticket_cron(conf_ptr, tk);
if (time_cmp(&last_cron, &tk->next_cron, ==)) {
tk_log_debug("nobody set ticket wakeup");
set_ticket_wakeup(tk);
}
}
}
void tickets_log_info(struct booth_config *conf_ptr)
{
struct ticket_config *tk;
int i;
time_t ts;
assert(conf_ptr != NULL);
FOREACH_TICKET(conf_ptr, i, tk) {
ts = wall_ts(&tk->term_expires);
tk_log_info("state '%s' "
"term %d "
"leader %s "
"expires %-24.24s",
state_to_string(tk->state),
tk->current_term,
ticket_leader_string(tk),
ctime(&ts));
}
}
static void update_acks(struct booth_config *conf_ptr, struct ticket_config *tk,
struct booth_site *sender, struct booth_site *leader,
struct boothc_ticket_msg *msg)
{
uint32_t cmd;
uint32_t req;
cmd = ntohl(msg->header.cmd);
req = ntohl(msg->header.request);
if (req != tk->last_request ||
(tk->acks_expected != cmd &&
tk->acks_expected != OP_REJECTED))
return;
/* got an ack! */
tk->acks_received |= sender->bitmask;
if (all_replied(conf_ptr, tk) ||
/* we just stepped down, need only one site to start
* elections */
(cmd == OP_REQ_VOTE && tk->last_request == OP_VOTE_FOR)) {
no_resends(tk);
tk->start_postpone = 0;
set_ticket_wakeup(tk);
}
}
/* read ticket message */
int ticket_recv(struct booth_config *conf_ptr, void *buf,
struct booth_site *source)
{
struct boothc_ticket_msg *msg;
struct ticket_config *tk;
struct booth_site *leader;
uint32_t leader_u;
msg = (struct boothc_ticket_msg *)buf;
if (!check_ticket(conf_ptr, msg->ticket.id, &tk)) {
log_warn("got invalid ticket name %s from %s",
msg->ticket.id, site_string(source));
source->invalid_cnt++;
return -EINVAL;
}
leader_u = ntohl(msg->ticket.leader);
if (!find_site_by_id(conf_ptr, leader_u, &leader)) {
tk_log_error("message with unknown leader %u received", leader_u);
source->invalid_cnt++;
return -EINVAL;
}
update_acks(conf_ptr, tk, source, leader, msg);
return raft_answer(conf_ptr, tk, source, leader, msg);
}
static void log_next_wakeup(struct ticket_config *tk)
{
int left;
left = time_left(&tk->next_cron);
tk_log_debug("set ticket wakeup in " intfmt(left));
}
/* New vote round; §5.2 */
/* delay the next election start for some random time
* (up to 1 second)
*/
void add_random_delay(struct ticket_config *tk)
{
timetype tv;
interval_add(&tk->next_cron, rand_time(min(1000, tk->timeout)), &tv);
ticket_next_cron_at(tk, &tv);
if (ANYDEBUG) {
log_next_wakeup(tk);
}
}
void set_ticket_wakeup(struct ticket_config *tk)
{
timetype near_future, tv, next_vote;
set_future_time(&near_future, 10);
if (!is_manual(tk)) {
/* At least every hour, perhaps sooner (default) */
tk_log_debug("ticket will be woken up after up to one hour");
ticket_next_cron_in(tk, 3600*TIME_RES);
switch (tk->state) {
case ST_LEADER:
assert(tk->leader == local);
get_next_election_time(tk, &next_vote);
/* If timestamp is in the past, wakeup in
* near future */
if (!is_time_set(&next_vote)) {
tk_log_debug("next ts unset, wakeup soon");
ticket_next_cron_at(tk, &near_future);
} else if (is_past(&next_vote)) {
int tdiff = time_left(&next_vote);
tk_log_debug("next ts in the past " intfmt(tdiff));
ticket_next_cron_at(tk, &near_future);
} else {
ticket_next_cron_at(tk, &next_vote);
}
break;
case ST_CANDIDATE:
assert(is_time_set(&tk->election_end));
ticket_next_cron_at(tk, &tk->election_end);
break;
case ST_INIT:
case ST_FOLLOWER:
/* If there is (or should be) some owner, check on it later on.
* If no one is interested - don't care. */
if (is_owned(tk)) {
interval_add(&tk->term_expires, tk->acquire_after, &tv);
ticket_next_cron_at(tk, &tv);
}
break;
default:
tk_log_error("unknown ticket state: %d", tk->state);
}
if (tk->next_state) {
/* we need to do something soon here */
if (!tk->acks_expected) {
ticket_next_cron_at(tk, &near_future);
} else {
ticket_activate_timeout(tk);
}
}
} else {
/* At least six minutes, to make sure that multi-leader situations
* will be solved promptly.
*/
tk_log_debug("manual ticket will be woken up after up to six minutes");
ticket_next_cron_in(tk, 60*TIME_RES);
/* For manual tickets, no earlier timeout could be set in a similar
* way as it is done in a switch above for automatic tickets.
* The reason is that term's timeout is INF and no Raft-based elections
* are performed.
*/
}
if (ANYDEBUG) {
log_next_wakeup(tk);
}
}
void schedule_election(struct ticket_config *tk, cmd_reason_t reason)
{
if (local->type != SITE)
return;
tk->election_reason = reason;
get_time(&tk->next_cron);
/* introduce a short delay before starting election */
add_random_delay(tk);
}
int is_manual(struct ticket_config *tk)
{
return (tk->mode == TICKET_MODE_MANUAL) ? 1 : 0;
}
/* Given a state (in host byte order), return a human-readable (char*).
* An array is used so that multiple states can be printed in a single printf(). */
char *state_to_string(uint32_t state_ho)
{
union mu { cmd_request_t s; char c[5]; };
static union mu cache[6] = { { 0 } }, *cur;
static int current = 0;
current ++;
if (current >= sizeof(cache)/sizeof(cache[0]))
current = 0;
cur = cache + current;
cur->s = htonl(state_ho);
/* Shouldn't be necessary, union array is initialized with zeroes, and
* these bytes never get written. */
cur->c[4] = 0;
return cur->c;
}
int send_reject(struct booth_config *conf_ptr, struct booth_site *dest,
struct ticket_config *tk, cmd_result_t code,
struct boothc_ticket_msg *in_msg)
{
int req = ntohl(in_msg->header.cmd);
struct boothc_ticket_msg msg;
tk_log_debug("sending reject to %s",
site_string(dest));
init_ticket_msg(conf_ptr, &msg, OP_REJECTED, req, code, 0, tk);
return booth_udp_send_auth(conf_ptr, dest, &msg, sendmsglen(&msg));
}
int send_msg(struct booth_config *conf_ptr, int cmd, struct ticket_config *tk,
struct booth_site *dest, struct boothc_ticket_msg *in_msg)
{
int req = 0;
struct ticket_config *valid_tk = tk;
struct boothc_ticket_msg msg;
/* if we want to send the last valid ticket, then if we're in
* the ST_CANDIDATE state, the last valid ticket is in
* tk->last_valid_tk
*/
if (cmd == OP_MY_INDEX) {
if (tk->state == ST_CANDIDATE && tk->last_valid_tk) {
valid_tk = tk->last_valid_tk;
}
tk_log_info("sending status to %s",
site_string(dest));
}
if (in_msg)
req = ntohl(in_msg->header.cmd);
init_ticket_msg(conf_ptr, &msg, cmd, req, RLT_SUCCESS, 0, valid_tk);
return booth_udp_send_auth(conf_ptr, dest, &msg, sendmsglen(&msg));
}
diff --git a/src/transport.c b/src/transport.c
index 4e66890..c207fb0 100644
--- a/src/transport.c
+++ b/src/transport.c
@@ -1,1147 +1,1153 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013-2014 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <netdb.h> /* getnameinfo */
#include <poll.h>
#include <arpa/inet.h>
#include <asm/types.h>
#include <linux/rtnetlink.h>
#include <net/if.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <sys/socket.h> /* getnameinfo */
#include "b_config.h"
#include "config.h"
#include "transport.h"
#include "attr.h"
#include "auth.h"
#include "booth.h"
#include "inline-fn.h"
#include "log.h"
#include "ticket.h"
#define BOOTH_IPADDR_LEN (sizeof(struct in6_addr))
#define NETLINK_BUFSIZE 16384
#define SOCKET_BUFFER_SIZE 160000
#define FRAME_SIZE_MAX 10000
struct booth_site *local = NULL;
/* function to be called when handling booth-group-internal messages;
* it's expected to return 0 to indicate success, negative integer
* to indicate silent (or possibly already complained about) error,
* or positive integer to indicate sender's ID that will then be
* emitted in the error log message together with the real source
* address if this is available */
static int (*deliver_fn) (void *msg, int msglen);
static void parse_rtattr(struct rtattr *tb[],
int max, struct rtattr *rta, int len)
{
while (RTA_OK(rta, len)) {
if (rta->rta_type <= max)
tb[rta->rta_type] = rta;
rta = RTA_NEXT(rta,len);
}
}
enum match_type {
NO_MATCH = 0,
FUZZY_MATCH,
EXACT_MATCH,
};
static int find_address(struct booth_config *conf_ptr,
unsigned char ipaddr[BOOTH_IPADDR_LEN],
int family, int prefixlen, int fuzzy_allowed,
struct booth_site **me, int *address_bits_matched)
{
int i;
struct booth_site *node;
int bytes, bits_left, mask;
unsigned char node_bits, ip_bits;
uint8_t *n_a;
int matched;
enum match_type did_match = NO_MATCH;
assert(conf_ptr != NULL);
bytes = prefixlen / 8;
bits_left = prefixlen % 8;
/* One bit left to check means ignore 7 lowest bits. */
mask = ~( (1 << (8 - bits_left)) -1);
FOREACH_NODE(conf_ptr, i, node) {
if (family != node->family)
continue;
n_a = node_to_addr_pointer(node);
for(matched = 0; matched < node->addrlen; matched++)
if (ipaddr[matched] != n_a[matched])
break;
if (matched == node->addrlen) {
*address_bits_matched = matched * 8;
*me = node;
did_match = EXACT_MATCH;
break;
}
if (!fuzzy_allowed)
continue;
/* Check prefix, whole bytes */
if (matched < bytes)
continue;
if (matched * 8 < *address_bits_matched)
continue;
node_bits = n_a[bytes];
ip_bits = ipaddr[bytes];
if (((node_bits ^ ip_bits) & mask) == 0) {
/* _At_least_ prefixlen bits matched. */
if (did_match < EXACT_MATCH) {
*address_bits_matched = prefixlen;
*me = node;
did_match = FUZZY_MATCH;
}
}
}
return did_match;
}
static int _find_myself(struct booth_config *conf_ptr, int family,
struct booth_site **mep, int fuzzy_allowed)
{
int fd;
struct sockaddr_nl nladdr;
struct booth_site *me;
unsigned char ipaddr[BOOTH_IPADDR_LEN];
static char rcvbuf[NETLINK_BUFSIZE];
struct {
struct nlmsghdr nlh;
struct rtgenmsg g;
} req;
int address_bits_matched;
if (local)
goto found;
me = NULL;
address_bits_matched = 0;
if (mep)
*mep = NULL;
fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
if (fd < 0) {
log_error("failed to create netlink socket");
return 0;
}
setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, sizeof(rcvbuf));
memset(&nladdr, 0, sizeof(nladdr));
nladdr.nl_family = AF_NETLINK;
memset(&req, 0, sizeof(req));
req.nlh.nlmsg_len = sizeof(req);
req.nlh.nlmsg_type = RTM_GETADDR;
req.nlh.nlmsg_flags = NLM_F_ROOT|NLM_F_MATCH|NLM_F_REQUEST;
req.nlh.nlmsg_pid = 0;
req.nlh.nlmsg_seq = 1;
req.g.rtgen_family = family;
if (sendto(fd, (void *)&req, sizeof(req), 0,
(struct sockaddr*)&nladdr, sizeof(nladdr)) < 0) {
close(fd);
log_error("failed to send data to netlink socket");
return 0;
}
while (1) {
int status;
struct nlmsghdr *h;
struct iovec iov = { rcvbuf, sizeof(rcvbuf) };
struct msghdr msg = {
(void *)&nladdr, sizeof(nladdr),
&iov, 1,
NULL, 0,
0
};
status = recvmsg(fd, &msg, 0);
if (!status) {
close(fd);
log_error("failed to recvmsg from netlink socket");
return 0;
}
h = (struct nlmsghdr *)rcvbuf;
if (h->nlmsg_type == NLMSG_DONE)
break;
if (h->nlmsg_type == NLMSG_ERROR) {
close(fd);
log_error("netlink socket recvmsg error");
return 0;
}
while (NLMSG_OK(h, status)) {
if (h->nlmsg_type == RTM_NEWADDR) {
struct ifaddrmsg *ifa = NLMSG_DATA(h);
struct rtattr *tb[IFA_MAX+1];
int len = h->nlmsg_len
- NLMSG_LENGTH(sizeof(*ifa));
memset(tb, 0, sizeof(tb));
parse_rtattr(tb, IFA_MAX, IFA_RTA(ifa), len);
memset(ipaddr, 0, BOOTH_IPADDR_LEN);
/* prefer IFA_LOCAL if it exists, for p-t-p
* interfaces, otherwise use IFA_ADDRESS */
if (tb[IFA_LOCAL]) {
memcpy(ipaddr, RTA_DATA(tb[IFA_LOCAL]),
BOOTH_IPADDR_LEN);
} else {
memcpy(ipaddr, RTA_DATA(tb[IFA_ADDRESS]),
BOOTH_IPADDR_LEN);
}
/* Try to find the exact address or the address with subnet matching.
* The function find_address will be called for each address received
* from NLMSG_DATA above.
* The exact match will be preferred. If no exact match is found,
* the function find_address will try to return another, most similar
* address (with the longest possible number of same bytes). */
if (ifa->ifa_prefixlen > address_bits_matched) {
find_address(conf_ptr, ipaddr,
ifa->ifa_family, ifa->ifa_prefixlen,
fuzzy_allowed, &me, &address_bits_matched);
if (me) {
log_debug("found myself at %s (%d bits matched)",
site_string(me), address_bits_matched);
}
}
/* If the previous NLMSG_DATA calls have already allowed us
* to find an address with address_bits_matched matching bits,
* then no other better non-exact address can be found.
* But we can still try to find an exact match, so let us
* call the function find_address with disabled searching of
* similar addresses (fuzzy_allowed == 0) */
else if (ifa->ifa_prefixlen == address_bits_matched) {
find_address(conf_ptr, ipaddr,
ifa->ifa_family, ifa->ifa_prefixlen,
0 /* fuzzy_allowed */, &me, &address_bits_matched);
if (me) {
log_debug("found myself at %s (exact match)",
site_string(me));
}
}
}
h = NLMSG_NEXT(h, status);
}
}
close(fd);
if (!me)
return 0;
me->local = 1;
local = me;
found:
if (mep)
*mep = local;
return 1;
}
int find_myself(struct booth_config *conf_ptr, struct booth_site **mep,
int fuzzy_allowed)
{
return _find_myself(conf_ptr, AF_INET6, mep, fuzzy_allowed) ||
_find_myself(conf_ptr, AF_INET, mep, fuzzy_allowed);
}
/** Checks the header fields for validity.
* cf. init_header().
* For @len_incl_data < 0 the length is not checked.
* Return <0 if error, else bytes read. */
int check_boothc_header(struct boothc_header *h, int len_incl_data)
{
int l;
if (h->magic != htonl(BOOTHC_MAGIC)) {
log_error("magic error %x", ntohl(h->magic));
return -EINVAL;
}
if (h->version != htonl(BOOTHC_VERSION)) {
log_error("version error %x", ntohl(h->version));
return -EINVAL;
}
l = ntohl(h->length);
if (l < sizeof(*h)) {
log_error("length %d out of range", l);
return -EINVAL;
}
if (len_incl_data < 0)
return 0;
if (l != len_incl_data) {
log_error("length error - got %d, wanted %d",
len_incl_data, l);
return -EINVAL;
}
return len_incl_data;
}
static int do_read(int fd, void *buf, size_t count)
{
int rv, off = 0;
while (off < count) {
rv = read(fd, (char *)buf + off, count - off);
if (rv == 0)
return -1;
if (rv == -1 && errno == EINTR)
continue;
if (rv == -1 && errno == EWOULDBLOCK)
break;
if (rv == -1)
return -1;
off += rv;
}
return off;
}
static int do_write(int fd, void *buf, size_t count)
{
int rv, off = 0;
retry:
rv = send(fd, (char *)buf + off, count, MSG_NOSIGNAL);
if (rv == -1 && errno == EINTR)
goto retry;
/* If we cannot write _any_ data, we'd be in an (potential) loop. */
if (rv <= 0) {
log_error("send failed: %s (%d)", strerror(errno), errno);
return rv;
}
if (rv != count) {
count -= rv;
off += rv;
goto retry;
}
return 0;
}
/* Only used for client requests (tcp) */
int read_client(struct client *req_cl)
{
char *msg;
struct boothc_header *header;
int rv, fd;
int len = MAX_MSG_LEN;
if (!req_cl->msg) {
msg = malloc(MAX_MSG_LEN);
if (!msg) {
log_error("out of memory for client messages");
return -1;
}
req_cl->msg = (void *)msg;
} else {
msg = (char *)req_cl->msg;
}
header = (struct boothc_header *)msg;
/* update len if we read enough */
if (req_cl->offset >= sizeof(*header)) {
len = min(ntohl(header->length), MAX_MSG_LEN);
}
fd = req_cl->fd;
rv = do_read(fd, msg+req_cl->offset, len-req_cl->offset);
if (rv < 0) {
if (errno == ECONNRESET)
log_debug("client connection reset for fd %d", fd);
return -1;
}
req_cl->offset += rv;
/* update len if we read enough */
if (req_cl->offset >= sizeof(*header)) {
len = min(ntohl(header->length), MAX_MSG_LEN);
}
if (req_cl->offset < len) {
/* client promised to send more */
return 1;
}
if (check_boothc_header(header, len) < 0) {
return -1;
}
return 0;
}
/* Only used for client requests (tcp) */
static void process_connection(struct booth_config *conf_ptr, int ci)
{
struct client *req_cl;
void *msg = NULL;
struct boothc_header *header;
struct boothc_hdr_msg err_reply;
cmd_result_t errc;
void (*deadfn) (int ci);
req_cl = clients + ci;
switch (read_client(req_cl)) {
case -1: /* error */
goto kill;
case 1: /* more to read */
return;
case 0:
/* we can process the request now */
msg = req_cl->msg;
}
header = (struct boothc_header *)msg;
if (check_auth(conf_ptr, NULL, msg, ntohl(header->length))) {
errc = RLT_AUTH;
goto send_err;
}
/* For CMD_GRANT and CMD_REVOKE:
* Don't close connection immediately, but send
* result a second later? */
switch (ntohl(header->cmd)) {
case CMD_LIST:
ticket_answer_list(conf_ptr, req_cl->fd);
goto kill;
case CMD_PEERS:
list_peers(conf_ptr, req_cl->fd);
goto kill;
case CMD_GRANT:
case CMD_REVOKE:
if (process_client_request(conf_ptr, req_cl, msg) == 1)
goto kill; /* request processed definitely, close connection */
else
return;
case ATTR_LIST:
case ATTR_GET:
case ATTR_SET:
case ATTR_DEL:
if (process_attr_request(conf_ptr, req_cl, msg) == 1)
goto kill; /* request processed definitely, close connection */
else
return;
default:
log_error("connection %d cmd %x unknown",
ci, ntohl(header->cmd));
errc = RLT_INVALID_ARG;
goto send_err;
}
assert(0);
return;
send_err:
init_header(conf_ptr, &err_reply.header, CL_RESULT, 0, 0, errc, 0,
sizeof(err_reply));
send_client_msg(conf_ptr, req_cl->fd, &err_reply);
kill:
deadfn = req_cl->deadfn;
if(deadfn) {
deadfn(ci);
}
return;
}
static void process_tcp_listener(struct booth_config *conf_ptr, int ci)
{
int fd, i, flags, one = 1;
socklen_t addrlen = sizeof(struct sockaddr);
struct sockaddr addr;
fd = accept(clients[ci].fd, &addr, &addrlen);
if (fd < 0) {
log_error("process_tcp_listener: accept error %d %d",
fd, errno);
return;
}
setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char *)&one, sizeof(one));
flags = fcntl(fd, F_GETFL, 0);
fcntl(fd, F_SETFL, flags | O_NONBLOCK);
i = client_add(fd, clients[ci].transport,
process_connection, NULL);
log_debug("client connection %d fd %d", i, fd);
}
int setup_tcp_listener(int test_only)
{
int s, rv;
int one = 1;
s = socket(local->family, SOCK_STREAM, 0);
if (s == -1) {
log_error("failed to create tcp socket %s", strerror(errno));
return s;
}
rv = setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (char *)&one, sizeof(one));
if (rv == -1) {
close(s);
log_error("failed to set the SO_REUSEADDR option");
return rv;
}
rv = bind(s, &local->sa6, local->saddrlen);
if (test_only) {
rv = (rv == -1) ? errno : 0;
close(s);
return rv;
}
if (rv == -1) {
close(s);
log_error("failed to bind socket %s", strerror(errno));
return rv;
}
rv = listen(s, 5);
if (rv == -1) {
close(s);
log_error("failed to listen on socket %s", strerror(errno));
return rv;
}
return s;
}
-static int booth_tcp_init(void * unused __attribute__((unused)))
+static int booth_tcp_init(struct booth_config *conf_ptr,
+ void * unused __attribute__((unused)))
{
int rv;
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
if (get_local_id() < 0)
return -1;
rv = setup_tcp_listener(0);
if (rv < 0)
return rv;
- client_add(rv, booth_transport + TCP,
- process_tcp_listener, NULL);
+ client_add(rv, *conf_ptr->transport + TCP, process_tcp_listener, NULL);
return 0;
}
static int connect_nonb(int sockfd, const struct sockaddr *saptr,
socklen_t salen, int sec)
{
int flags, n, error;
socklen_t len;
fd_set rset, wset;
struct timeval tval;
flags = fcntl(sockfd, F_GETFL, 0);
fcntl(sockfd, F_SETFL, flags | O_NONBLOCK);
error = 0;
if ( (n = connect(sockfd, saptr, salen)) < 0)
if (errno != EINPROGRESS)
return -1;
if (n == 0)
goto done; /* connect completed immediately */
FD_ZERO(&rset);
FD_SET(sockfd, &rset);
wset = rset;
tval.tv_sec = sec;
tval.tv_usec = 0;
if ((n = select(sockfd + 1, &rset, &wset, NULL,
sec ? &tval : NULL)) == 0) {
/* leave outside function to close */
/* timeout */
/* close(sockfd); */
errno = ETIMEDOUT;
return -1;
}
if (FD_ISSET(sockfd, &rset) || FD_ISSET(sockfd, &wset)) {
len = sizeof(error);
if (getsockopt(sockfd, SOL_SOCKET, SO_ERROR, &error, &len) < 0)
return -1; /* Solaris pending error */
} else {
log_error("select error: sockfd not set");
return -1;
}
done:
fcntl(sockfd, F_SETFL, flags); /* restore file status flags */
if (error) {
/* leave outside function to close */
/* close(sockfd); */
errno = error;
return -1;
}
return 0;
}
static int booth_tcp_open(struct booth_site *to)
{
int s, rv;
if (to->tcp_fd >= STDERR_FILENO)
goto found;
s = socket(to->family, SOCK_STREAM, 0);
if (s == -1) {
log_error("cannot create socket of family %d", to->family);
return -1;
}
rv = connect_nonb(s, (struct sockaddr *)&to->sa6, to->saddrlen, 10);
if (rv == -1) {
if( errno == ETIMEDOUT)
log_error("connect to %s got a timeout", site_string(to));
else
log_error("connect to %s got an error: %s", site_string(to),
strerror(errno));
goto error;
}
to->tcp_fd = s;
found:
return 1;
error:
if (s >= 0)
close(s);
return -1;
}
/* data + (datalen-sizeof(struct hmac)) points to struct hmac
* i.e. struct hmac is always tacked on the payload
*/
static int add_hmac(struct booth_config *conf_ptr, void *data, int len)
{
int rv = 0;
#if HAVE_LIBGCRYPT || HAVE_LIBMHASH
int payload_len;
struct hmac *hp;
assert(conf_ptr != NULL);
if (!is_auth_req(conf_ptr))
return 0;
payload_len = len - sizeof(struct hmac);
hp = (struct hmac *)((unsigned char *)data + payload_len);
hp->hid = htonl(BOOTH_HASH);
memset(hp->hash, 0, BOOTH_MAC_SIZE);
rv = calc_hmac(data, payload_len, BOOTH_HASH, hp->hash,
conf_ptr->authkey, conf_ptr->authkey_len);
if (rv < 0) {
log_error("internal error: cannot calculate mac");
}
#endif
return rv;
}
static int booth_tcp_send(struct booth_config *conf_ptr,
struct booth_site *to, void *buf, int len)
{
int rv;
rv = add_hmac(conf_ptr, buf, len);
if (!rv)
rv = do_write(to->tcp_fd, buf, len);
return rv;
}
static int booth_tcp_recv(struct booth_site *from, void *buf, int len)
{
int got;
/* Needs timeouts! */
got = do_read(from->tcp_fd, buf, len);
if (got < 0) {
log_error("read failed (%d): %s", errno, strerror(errno));
return got;
}
return got;
}
static int booth_tcp_recv_auth(struct booth_config *conf_ptr,
struct booth_site *from, void *buf, int len)
{
int got, total;
int payload_len;
/* Needs timeouts! */
payload_len = len - sizeof(struct hmac);
got = booth_tcp_recv(from, buf, payload_len);
if (got < 0) {
return got;
}
total = got;
if (is_auth_req(conf_ptr)) {
got = booth_tcp_recv(from, (unsigned char *)buf+payload_len, sizeof(struct hmac));
if (got != sizeof(struct hmac)
|| check_auth(conf_ptr, from, buf, len)) {
return -1;
}
total += got;
}
return total;
}
static int booth_tcp_close(struct booth_site *to)
{
if (to) {
if (to->tcp_fd > STDERR_FILENO)
close(to->tcp_fd);
to->tcp_fd = -1;
}
return 0;
}
static int booth_tcp_exit(void)
{
return 0;
}
static int setup_udp_server(void)
{
int rv, fd;
int one = 1;
unsigned int recvbuf_size;
fd = socket(local->family, SOCK_DGRAM, 0);
if (fd == -1) {
log_error("failed to create UDP socket %s", strerror(errno));
goto ex;
}
rv = fcntl(fd, F_SETFL, O_NONBLOCK);
if (rv == -1) {
log_error("failed to set non-blocking operation "
"on UDP socket: %s", strerror(errno));
goto ex;
}
rv = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *)&one, sizeof(one));
if (rv == -1) {
log_error("failed to set the SO_REUSEADDR option");
goto ex;
}
rv = bind(fd, (struct sockaddr *)&local->sa6, local->saddrlen);
if (rv == -1) {
log_error("failed to bind UDP socket to [%s]:%d: %s",
site_string(local), site_port(local),
strerror(errno));
goto ex;
}
recvbuf_size = SOCKET_BUFFER_SIZE;
rv = setsockopt(fd, SOL_SOCKET, SO_RCVBUF,
&recvbuf_size, sizeof(recvbuf_size));
if (rv == -1) {
log_error("failed to set recvbuf size");
goto ex;
}
local->udp_fd = fd;
return 0;
ex:
if (fd >= 0)
close(fd);
return -1;
}
/* Receive/process callback for UDP */
static void process_recv(struct booth_config *conf_ptr, int ci)
{
struct sockaddr_storage sa;
int rv;
socklen_t sa_len;
/* beware, the buffer needs to be large enough to accept
* a packet */
char buffer[MAX_MSG_LEN];
/* Used for unit tests */
struct boothc_ticket_msg *msg;
sa_len = sizeof(sa);
msg = (void*)buffer;
rv = recvfrom(clients[ci].fd,
buffer, sizeof(buffer),
MSG_NOSIGNAL | MSG_DONTWAIT,
(struct sockaddr *)&sa, &sa_len);
if (rv == -1)
return;
rv = deliver_fn((void*)msg, rv);
if (rv > 0) {
if (getnameinfo((struct sockaddr *)&sa, sa_len,
buffer, sizeof(buffer), NULL, 0,
NI_NUMERICHOST) == 0)
log_error("unknown sender: %08x (real: %s)", rv, buffer);
else
log_error("unknown sender: %08x", rv);
}
}
-static int booth_udp_init(void *f)
+static int booth_udp_init(struct booth_config *conf_ptr, void *f)
{
int rv;
+ assert(conf_ptr != NULL && conf_ptr->transport != NULL);
+
rv = setup_udp_server();
if (rv < 0)
return rv;
deliver_fn = f;
- client_add(local->udp_fd,
- booth_transport + UDP,
- process_recv, NULL);
+ client_add(local->udp_fd, *conf_ptr->transport + UDP, process_recv,
+ NULL);
return 0;
}
static int booth_udp_send(struct booth_config *conf_ptr, struct booth_site *to,
void *buf, int len)
{
int rv;
to->sent_cnt++;
rv = sendto(local->udp_fd, buf, len, MSG_NOSIGNAL,
(struct sockaddr *)&to->sa6, to->saddrlen);
if (rv == len) {
rv = 0;
} else if (rv < 0) {
to->sent_err_cnt++;
log_error("Cannot send to %s: %d %s",
site_string(to),
errno,
strerror(errno));
} else {
rv = -1;
to->sent_err_cnt++;
log_error("Packet sent to %s got truncated",
site_string(to));
}
return rv;
}
int booth_udp_send_auth(struct booth_config *conf_ptr,
struct booth_site *to, void *buf, int len)
{
int rv;
rv = add_hmac(conf_ptr, buf, len);
if (rv < 0)
return rv;
return booth_udp_send(conf_ptr, to, buf, len);
}
static int booth_udp_broadcast_auth(struct booth_config *conf_ptr,
void *buf, int len)
{
int i, rv, rvs;
struct booth_site *site;
if (conf_ptr == NULL || !conf_ptr->site_count)
return -1;
rv = add_hmac(conf_ptr, buf, len);
if (rv < 0)
return rv;
rvs = 0;
FOREACH_NODE(conf_ptr, i, site) {
if (site != local) {
rv = booth_udp_send(conf_ptr, site, buf, len);
if (!rvs)
rvs = rv;
}
}
return rvs;
}
static int booth_udp_exit(void)
{
return 0;
}
/* SCTP transport layer has not been developed yet */
-static int booth_sctp_init(void *f __attribute__((unused)))
+static int booth_sctp_init(struct booth_config *conf_ptr __attribute__((unused)),
+ void *f __attribute__((unused)))
{
return 0;
}
static int booth_sctp_send(struct booth_config *conf_ptr __attribute__((unused)),
struct booth_site * to __attribute__((unused)),
void *buf __attribute__((unused)),
int len __attribute__((unused)))
{
return 0;
}
static int booth_sctp_broadcast(void *buf __attribute__((unused)),
int len __attribute__((unused)))
{
return 0;
}
static int return_0_booth_site(struct booth_site *v __attribute((unused)))
{
return 0;
}
static int return_0(void)
{
return 0;
}
-const struct booth_transport booth_transport[TRANSPORT_ENTRIES] = {
+
+/* semi-hidden, only main.c to have a knowledge about this */
+const booth_transport_table_t booth__transport = {
[TCP] = {
.name = "TCP",
.init = booth_tcp_init,
.open = booth_tcp_open,
.send = booth_tcp_send,
.recv = booth_tcp_recv,
.recv_auth = booth_tcp_recv_auth,
.close = booth_tcp_close,
.exit = booth_tcp_exit
},
[UDP] = {
.name = "UDP",
.init = booth_udp_init,
.open = return_0_booth_site,
.send = booth_udp_send,
.send_auth = booth_udp_send_auth,
.close = return_0_booth_site,
.broadcast_auth = booth_udp_broadcast_auth,
.exit = booth_udp_exit
},
[SCTP] = {
.name = "SCTP",
.init = booth_sctp_init,
.open = return_0_booth_site,
.send = booth_sctp_send,
.broadcast = booth_sctp_broadcast,
.exit = return_0,
}
};
#if HAVE_LIBGCRYPT || HAVE_LIBMHASH
/* TODO: we need some client identification for logging */
#define peer_string(p) (p ? site_string(p) : "client")
/* verify the validity of timestamp from the header
* the timestamp needs to be either greater than the one already
* recorded for the site or, and this is checked for clients,
* not to be older than conf_ptr->maxtimeskew
* update the timestamp for the site, if this packet is from a
* site
*/
static int verify_ts(struct booth_config *conf_ptr, struct booth_site *from,
void *buf, int len)
{
struct boothc_header *h;
struct timeval tv, curr_tv, now;
assert(conf_ptr != NULL);
if (len < sizeof(*h)) {
log_error("%s: packet too short", peer_string(from));
return -1;
}
h = (struct boothc_header *)buf;
tv.tv_sec = ntohl(h->secs);
tv.tv_usec = ntohl(h->usecs);
if (from) {
curr_tv.tv_sec = from->last_secs;
curr_tv.tv_usec = from->last_usecs;
if (timercmp(&tv, &curr_tv, >))
goto accept;
log_warn("%s: packet timestamp older than previous one",
site_string(from));
}
gettimeofday(&now, NULL);
now.tv_sec -= conf_ptr->maxtimeskew;
if (timercmp(&tv, &now, >))
goto accept;
log_error("%s: packet timestamp older than %d seconds",
peer_string(from), conf_ptr->maxtimeskew);
return -1;
accept:
if (from) {
from->last_secs = tv.tv_sec;
from->last_usecs = tv.tv_usec;
}
return 0;
}
#endif
int check_auth(struct booth_config *conf_ptr, struct booth_site *from,
void *buf, int len)
{
int rv = 0;
#if HAVE_LIBGCRYPT || HAVE_LIBMHASH
int payload_len;
struct hmac *hp;
assert(conf_ptr != NULL);
if (!is_auth_req(conf_ptr))
return 0;
payload_len = len - sizeof(struct hmac);
if (payload_len < 0) {
log_error("%s: failed to authenticate, packet too short (size:%d)",
peer_string(from), len);
return -1;
}
hp = (struct hmac *)((unsigned char *)buf + payload_len);
rv = verify_hmac(buf, payload_len, ntohl(hp->hid), hp->hash,
conf_ptr->authkey, conf_ptr->authkey_len);
if (!rv) {
rv = verify_ts(conf_ptr, from, buf, len);
}
if (rv != 0) {
log_error("%s: failed to authenticate", peer_string(from));
}
#endif
return rv;
}
int send_data(struct booth_config *conf_ptr, int fd, void *data, int datalen)
{
int rv = 0;
rv = add_hmac(conf_ptr, data, datalen);
if (!rv)
rv = do_write(fd, data, datalen);
return rv;
}
int send_header_plus(struct booth_config *conf_ptr, int fd,
struct boothc_hdr_msg *msg, void *data, int len)
{
int rv;
rv = send_data(conf_ptr, fd, msg, sendmsglen(msg)-len);
if (rv >= 0 && len)
rv = do_write(fd, data, len);
return rv;
}
/* UDP message receiver (see also deliver_fn declaration's comment) */
int message_recv(struct booth_config *conf_ptr, void *msg, int msglen)
{
uint32_t from;
struct boothc_header *header;
struct booth_site *source;
header = (struct boothc_header *)msg;
from = ntohl(header->from);
if (!find_site_by_id(conf_ptr, from, &source)) {
/* caller knows the actual source address, pass
the (assuredly) positive number and let it report */
from = from ? from : ~from; /* avoid 0 (success) */
return from & (~0U >> 1); /* avoid negative (error code} */
}
time(&source->last_recv);
source->recv_cnt++;
if (check_boothc_header(header, msglen) < 0) {
log_error("message from %s receive error", site_string(source));
source->recv_err_cnt++;
return -1;
}
if (check_auth(conf_ptr, source, msg, msglen)) {
log_error("%s failed to authenticate", site_string(source));
source->sec_cnt++;
return -1;
}
if (ntohl(header->opts) & BOOTH_OPT_ATTR) {
/* not used, clients send/retrieve attributes directly
* from sites
*/
return attr_recv(conf_ptr, msg, source);
} else {
return ticket_recv(conf_ptr, msg, source);
}
}
diff --git a/src/transport.h b/src/transport.h
index 47c1cce..99a5bea 100644
--- a/src/transport.h
+++ b/src/transport.h
@@ -1,158 +1,158 @@
/*
* Copyright (C) 2011 Jiaju Zhang <jjzhang@suse.de>
* Copyright (C) 2013 Philipp Marek <philipp.marek@linbit.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#ifndef _TRANSPORT_H
#define _TRANSPORT_H
#include "b_config.h"
#include "booth.h"
typedef enum {
TCP = 1,
UDP,
SCTP,
TRANSPORT_ENTRIES,
} transport_layer_t;
typedef enum {
ARBITRATOR = 0x50,
SITE,
CLIENT,
DAEMON,
STATUS,
GEOSTORE,
} action_t;
/* when allocating space for messages
*/
#define MAX_MSG_LEN 1024
struct booth_transport {
const char *name;
- int (*init) (void *);
+ int (*init) (struct booth_config *, void *);
int (*open) (struct booth_site *);
int (*send) (struct booth_config *, struct booth_site *, void *, int);
int (*send_auth) (struct booth_config *, struct booth_site *, void *, int);
int (*recv) (struct booth_site *, void *, int);
int (*recv_auth) (struct booth_config *, struct booth_site *, void *, int);
int (*broadcast) (void *, int);
int (*broadcast_auth) (struct booth_config *, void *, int);
int (*close) (struct booth_site *);
int (*exit) (void);
};
-extern const struct booth_transport booth_transport[TRANSPORT_ENTRIES];
+typedef struct booth_transport booth_transport_table_t[TRANSPORT_ENTRIES];
/**
* @internal
* Attempts to pick identity of self from config-tracked enumeration of sites
*
* @param[inout] conf_ptr config object to refer to
* @param[out] mep when self-discovery successful, site pointer is stored here
* @param[in] fuzzy_allowed whether it's OK to approximate the match
*
* @return 0 on success or negative value (-1 or -errno) on error
*/
int find_myself(struct booth_config *conf_ptr, struct booth_site **mep,
int fuzzy_allowed);
int read_client(struct client *req_cl);
int check_boothc_header(struct boothc_header *data, int len_incl_data);
int setup_tcp_listener(int test_only);
/**
* @internal
* Send data, with authentication added
*
* @param[inout] conf_ptr config object to refer to
* @param[in] to site structure of the recipient
* @param[in] buf message itself
* @param[in] len lenght of #buf
*
* @return see @add_hmac and @booth_udp_send
*/
int booth_udp_send_auth(struct booth_config *conf_ptr, struct booth_site *to,
void *buf, int len);
/**
* @internal
* First stage of incoming datagram handling (authentication)
*
* @param[inout] conf_ptr config object to refer to
* @param[in] msg raw message to act upon
* @param[in] msglen lenght of #msg
*
* @return 0 on success or negative value (-1 or -errno) on error
*/
int message_recv(struct booth_config *conf_ptr, void *msg, int msglen);
inline static void * node_to_addr_pointer(struct booth_site *node) {
switch (node->family) {
case AF_INET: return &node->sa4.sin_addr;
case AF_INET6: return &node->sa6.sin6_addr;
}
return NULL;
}
/**
* @internal
* Send data, with authentication added
*
* @param[inout] conf_ptr config object to refer to
* @param[in] fd descriptor of the socket to respond to
* @param[in] data message itself
* @param[in] datalen lenght of #data
*
* @return 0 on success or negative value (-1 or -errno) on error
*/
int send_data(struct booth_config *conf_ptr, int fd, void *data, int datalen);
/**
* @internal
* First stage of incoming datagram handling (authentication)
*
* @param[inout] conf_ptr config object to refer to
* @param[in] fd descriptor of the socket to respond to
* @param[in] hdr message header
* @param[in] data message itself
* @param[in] len lengh of @data
*
* @return see #send_data and #do_write
*/
int send_header_plus(struct booth_config *conf_ptr, int fd,
struct boothc_hdr_msg *hdr, void *data, int len);
#define send_client_msg(bc, fd, msg) send_data(bc, fd, msg, sendmsglen(msg))
/**
* @internal
* First stage of incoming datagram handling (authentication)
*
* @param[inout] conf_ptr config object to refer to
* @param[in] from site structure of the sender
* @param[in] buf message to check
* @param[in] len lengh of @buf
*
* @return see #send_data and #do_write
*/
int check_auth(struct booth_config *conf_ptr, struct booth_site *from,
void *buf, int len);
#endif /* _TRANSPORT_H */
File Metadata
Details
Attached
Mime Type
text/x-diff
Expires
Mon, Feb 24, 9:03 AM (21 h, 5 m ago)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1464095
Default Alt Text
(162 KB)
Attached To
Mode
rB Booth
Attached
Detach File
Event Timeline
Log In to Comment