/*
* Function used to register a table for sync on a group of peers
*
*/
void peers_register_table(struct peers *peers, struct stktable *table)
{
struct shared_table *st;
struct peer * curpeer;
struct peer_session *ps;
st = (struct shared_table *)calloc(1,sizeof(struct shared_table));
st->table = table;
st->next = peers->tables;
st->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
peers->tables = st;
for (curpeer = peers->remote; curpeer; curpeer = curpeer->next) {
ps = (struct peer_session *)calloc(1,sizeof(struct peer_session));
ps->table = st;
ps->peer = curpeer;
if (curpeer->local)
st->local_session = ps;
ps->next = st->sessions;
ps->reconnect = now_ms;
st->sessions = ps;
peers->peers_fe->maxconn += 3;
}
peers->peers_fe->listen->maxconn = peers->peers_fe->maxconn;
st->sync_task = task_new();
st->sync_task->process = process_peer_sync;
st->sync_task->expire = TICK_ETERNITY;
st->sync_task->context = (void *)st;
table->sync_task =st->sync_task;
signal_register_task(0, table->sync_task, 0);
task_wakeup(st->sync_task, TASK_WOKEN_INIT);
}
/*
* Callback to release a session with a peer
*/
static void peer_session_release(struct stream_interface *si)
{
struct task *t = (struct task *)si->owner;
struct session *s = (struct session *)t->context;
struct peer_session *ps = (struct peer_session *)si->conn.data_ctx;
/* si->conn.data_ctx is not a peer session */
if (si->applet.st0 < PEER_SESSION_SENDSUCCESS)
return;
/* peer session identified */
if (ps) {
if (ps->session == s) {
ps->session = NULL;
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer for learning */
ps->flags &= ~(PEER_F_LEARN_ASSIGN);
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* reschedule a resync */
ps->table->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
}
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
}
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
}
}
/*
* Callback to release a session with a peer
*/
static void peer_session_release(struct appctx *appctx)
{
struct stream_interface *si = appctx->owner;
struct stream *s = si_strm(si);
struct peer_session *ps = (struct peer_session *)appctx->ctx.peers.ptr;
/* appctx->ctx.peers.ptr is not a peer session */
if (appctx->st0 < PEER_SESS_ST_SENDSUCCESS)
return;
/* peer session identified */
if (ps) {
if (ps->stream == s) {
ps->stream = NULL;
ps->appctx = NULL;
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer for learning */
ps->flags &= ~(PEER_F_LEARN_ASSIGN);
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* reschedule a resync */
ps->table->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
}
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
}
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
}
}
/*
* Use this function to force a close of a peer session
*/
static void peer_session_forceshutdown(struct stream * stream)
{
struct appctx *appctx = NULL;
struct peer_session *ps;
int i;
for (i = 0; i <= 1; i++) {
appctx = objt_appctx(stream->si[i].end);
if (!appctx)
continue;
if (appctx->applet != &peer_applet)
continue;
break;
}
if (!appctx)
return;
ps = (struct peer_session *)appctx->ctx.peers.ptr;
/* we're killing a connection, we must apply a random delay before
* retrying otherwise the other end will do the same and we can loop
* for a while.
*/
if (ps)
ps->reconnect = tick_add(now_ms, MS_TO_TICKS(50 + random() % 2000));
/* call release to reinit resync states if needed */
peer_session_release(appctx);
appctx->st0 = PEER_SESS_ST_END;
appctx->ctx.peers.ptr = NULL;
task_wakeup(stream->task, TASK_WOKEN_MSG);
}
/*
* this function disables health-check servers so that the process will quickly be ignored
* by load balancers. Note that if a proxy was already in the PAUSED state, then its grace
* time will not be used since it would already not listen anymore to the socket.
*/
void soft_stop(void)
{
struct proxy *p;
struct peers *prs;
stopping = 1;
p = proxy;
tv_update_date(0,1); /* else, the old time before select will be used */
while (p) {
if (p->state != PR_STSTOPPED) {
Warning("Stopping %s %s in %d ms.\n", proxy_cap_str(p->cap), p->id, p->grace);
send_log(p, LOG_WARNING, "Stopping %s %s in %d ms.\n", proxy_cap_str(p->cap), p->id, p->grace);
p->stop_time = tick_add(now_ms, p->grace);
}
if (p->table.size && p->table.sync_task)
task_wakeup(p->table.sync_task, TASK_WOKEN_MSG);
/* wake every proxy task up so that they can handle the stopping */
task_wakeup(p->task, TASK_WOKEN_MSG);
p = p->next;
}
prs = peers;
while (prs) {
stop_proxy((struct proxy *)prs->peers_fe);
prs = prs->next;
}
/* signal zero is used to broadcast the "stopping" event */
signal_handler(0);
}
/*
* this function disables health-check servers so that the process will quickly be ignored
* by load balancers. Note that if a proxy was already in the PAUSED state, then its grace
* time will not be used since it would already not listen anymore to the socket.
*/
void soft_stop(void)
{
struct proxy *p;
stopping = 1;
p = proxy;
tv_update_date(0,1); /* else, the old time before select will be used */
while (p) {
if (p->state != PR_STSTOPPED) {
Warning("Stopping %s %s in %d ms.\n", proxy_cap_str(p->cap), p->id, p->grace);
send_log(p, LOG_WARNING, "Stopping %s %s in %d ms.\n", proxy_cap_str(p->cap), p->id, p->grace);
p->stop_time = tick_add(now_ms, p->grace);
}
p = p->next;
}
}
/*
* IO Handler to handle message exchance with a peer
*/
static void peer_io_handler(struct stream_interface *si)
{
struct task *t= (struct task *)si->owner;
struct session *s = (struct session *)t->context;
struct peers *curpeers = (struct peers *)s->fe->parent;
int reql = 0;
int repl = 0;
while (1) {
switchstate:
switch(si->applet.st0) {
case PEER_SESSION_ACCEPT:
si->conn.data_ctx = NULL;
si->applet.st0 = PEER_SESSION_GETVERSION;
/* fall through */
case PEER_SESSION_GETVERSION:
reql = bo_getline(si->ob, trash, trashlen);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
if (trash[reql-1] != '\n') {
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
else if (reql > 1 && (trash[reql-2] == '\r'))
trash[reql-2] = 0;
else
trash[reql-1] = 0;
bo_skip(si->ob, reql);
/* test version */
if (strcmp(PEER_SESSION_PROTO_NAME " 1.0", trash) != 0) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRVERSION;
/* test protocol */
if (strncmp(PEER_SESSION_PROTO_NAME " ", trash, strlen(PEER_SESSION_PROTO_NAME)+1) != 0)
si->applet.st1 = PEER_SESSION_ERRPROTO;
goto switchstate;
}
si->applet.st0 = PEER_SESSION_GETHOST;
/* fall through */
case PEER_SESSION_GETHOST:
reql = bo_getline(si->ob, trash, trashlen);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
if (trash[reql-1] != '\n') {
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
else if (reql > 1 && (trash[reql-2] == '\r'))
trash[reql-2] = 0;
else
trash[reql-1] = 0;
bo_skip(si->ob, reql);
/* test hostname match */
if (strcmp(localpeer, trash) != 0) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRHOST;
goto switchstate;
}
si->applet.st0 = PEER_SESSION_GETPEER;
/* fall through */
case PEER_SESSION_GETPEER: {
struct peer *curpeer;
char *p;
reql = bo_getline(si->ob, trash, trashlen);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
if (trash[reql-1] != '\n') {
/* Incomplete line, we quit */
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
else if (reql > 1 && (trash[reql-2] == '\r'))
trash[reql-2] = 0;
else
trash[reql-1] = 0;
bo_skip(si->ob, reql);
/* parse line "<peer name> <pid>" */
p = strchr(trash, ' ');
if (!p) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRPROTO;
goto switchstate;
}
*p = 0;
/* lookup known peer */
for (curpeer = curpeers->remote; curpeer; curpeer = curpeer->next) {
if (strcmp(curpeer->id, trash) == 0)
break;
}
/* if unknown peer */
if (!curpeer) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRPEER;
goto switchstate;
}
si->conn.data_ctx = curpeer;
si->applet.st0 = PEER_SESSION_GETTABLE;
/* fall through */
}
case PEER_SESSION_GETTABLE: {
struct peer *curpeer = (struct peer *)si->conn.data_ctx;
struct shared_table *st;
struct peer_session *ps = NULL;
unsigned long key_type;
size_t key_size;
char *p;
reql = bo_getline(si->ob, trash, trashlen);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
si->conn.data_ctx = NULL;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* Re init si->conn.data_ctx to null, to handle correctly a release case */
si->conn.data_ctx = NULL;
if (trash[reql-1] != '\n') {
/* Incomplete line, we quit */
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
else if (reql > 1 && (trash[reql-2] == '\r'))
trash[reql-2] = 0;
else
trash[reql-1] = 0;
bo_skip(si->ob, reql);
/* Parse line "<table name> <type> <size>" */
p = strchr(trash, ' ');
if (!p) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRPROTO;
goto switchstate;
}
*p = 0;
key_type = (unsigned long)atol(p+1);
p = strchr(p+1, ' ');
if (!p) {
si->conn.data_ctx = NULL;
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRPROTO;
goto switchstate;
}
key_size = (size_t)atoi(p);
for (st = curpeers->tables; st; st = st->next) {
/* If table name matches */
if (strcmp(st->table->id, trash) == 0) {
/* If key size mismatches */
if (key_size != st->table->key_size) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRSIZE;
goto switchstate;
}
/* If key type mismatches */
if (key_type != st->table->type) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRTYPE;
goto switchstate;
}
/* lookup peer session of current peer */
for (ps = st->sessions; ps; ps = ps->next) {
if (ps->peer == curpeer) {
/* If session already active, replaced by new one */
if (ps->session && ps->session != s) {
if (ps->peer->local) {
/* Local connection, reply a retry */
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_TRYAGAIN;
goto switchstate;
}
peer_session_forceshutdown(ps->session);
}
ps->session = s;
break;
}
}
break;
}
}
/* If table not found */
if (!st){
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRTABLE;
goto switchstate;
}
/* If no peer session for current peer */
if (!ps) {
si->applet.st0 = PEER_SESSION_EXIT;
si->applet.st1 = PEER_SESSION_ERRPEER;
goto switchstate;
}
si->conn.data_ctx = ps;
si->applet.st0 = PEER_SESSION_SENDSUCCESS;
/* fall through */
}
case PEER_SESSION_SENDSUCCESS:{
struct peer_session *ps = (struct peer_session *)si->conn.data_ctx;
repl = snprintf(trash, trashlen, "%d\n", PEER_SESSION_SUCCESSCODE);
repl = bi_putblk(si->ib, trash, repl);
if (repl <= 0) {
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* Register status code */
ps->statuscode = PEER_SESSION_SUCCESSCODE;
/* Awake main task */
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
/* Init cursors */
ps->teaching_origin =ps->lastpush = ps->lastack = ps->pushack = 0;
ps->pushed = ps->update;
/* Init confirm counter */
ps->confirm = 0;
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
/* if current peer is local */
if (ps->peer->local) {
/* if table need resyncfrom local and no process assined */
if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMLOCAL &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* assign local peer for a lesson, consider lesson already requested */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= (SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
}
}
else if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* assign peer for a lesson */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= SHTABLE_F_RESYNC_ASSIGN;
}
/* switch to waiting message state */
si->applet.st0 = PEER_SESSION_WAITMSG;
goto switchstate;
}
case PEER_SESSION_CONNECT: {
struct peer_session *ps = (struct peer_session *)si->conn.data_ctx;
/* Send headers */
repl = snprintf(trash, trashlen,
PEER_SESSION_PROTO_NAME " 1.0\n%s\n%s %d\n%s %lu %d\n",
ps->peer->id,
localpeer,
(int)getpid(),
ps->table->table->id,
ps->table->table->type,
(int)ps->table->table->key_size);
if (repl >= trashlen) {
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
repl = bi_putblk(si->ib, trash, repl);
if (repl <= 0) {
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* switch to the waiting statuscode state */
si->applet.st0 = PEER_SESSION_GETSTATUS;
/* fall through */
}
case PEER_SESSION_GETSTATUS: {
struct peer_session *ps = (struct peer_session *)si->conn.data_ctx;
if (si->ib->flags & CF_WRITE_PARTIAL)
ps->statuscode = PEER_SESSION_CONNECTEDCODE;
reql = bo_getline(si->ob, trash, trashlen);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
if (trash[reql-1] != '\n') {
/* Incomplete line, we quit */
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
else if (reql > 1 && (trash[reql-2] == '\r'))
trash[reql-2] = 0;
else
trash[reql-1] = 0;
bo_skip(si->ob, reql);
/* Register status code */
ps->statuscode = atoi(trash);
/* Awake main task */
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
/* If status code is success */
if (ps->statuscode == PEER_SESSION_SUCCESSCODE) {
/* Init cursors */
ps->teaching_origin = ps->lastpush = ps->lastack = ps->pushack = 0;
ps->pushed = ps->update;
/* Init confirm counter */
ps->confirm = 0;
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
/* If current peer is local */
if (ps->peer->local) {
/* Init cursors to push a resync */
ps->teaching_origin = ps->pushed = ps->table->table->update;
/* flag to start to teach lesson */
ps->flags |= PEER_F_TEACH_PROCESS;
}
else if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* If peer is remote and resync from remote is needed,
and no peer currently assigned */
/* assign peer for a lesson */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= SHTABLE_F_RESYNC_ASSIGN;
}
}
else {
/* Status code is not success, abort */
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
si->applet.st0 = PEER_SESSION_WAITMSG;
/* fall through */
}
case PEER_SESSION_WAITMSG: {
struct peer_session *ps = (struct peer_session *)si->conn.data_ctx;
char c;
int totl = 0;
reql = bo_getblk(si->ob, (char *)&c, sizeof(c), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
/* nothing to read */
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
if ((c & 0x80) || (c == 'D')) {
/* Here we have data message */
unsigned int pushack;
struct stksess *ts;
struct stksess *newts;
struct stktable_key stkey;
int srvid;
uint32_t netinteger;
/* Compute update remote version */
if (c & 0x80) {
pushack = ps->pushack + (unsigned int)(c & 0x7F);
}
else {
reql = bo_getblk(si->ob, (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
pushack = ntohl(netinteger);
}
/* read key */
if (ps->table->table->type == STKTABLE_TYPE_STRING) {
/* type string */
stkey.key = stkey.data.buf;
reql = bo_getblk(si->ob, (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
stkey.key_len = ntohl(netinteger);
reql = bo_getblk(si->ob, stkey.key, stkey.key_len, totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
}
else if (ps->table->table->type == STKTABLE_TYPE_INTEGER) {
/* type integer */
stkey.key_len = (size_t)-1;
stkey.key = &stkey.data.integer;
reql = bo_getblk(si->ob, (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
stkey.data.integer = ntohl(netinteger);
}
else {
/* type ip */
stkey.key_len = (size_t)-1;
stkey.key = stkey.data.buf;
reql = bo_getblk(si->ob, (char *)&stkey.data.buf, ps->table->table->key_size, totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
}
/* read server id */
reql = bo_getblk(si->ob, (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
srvid = ntohl(netinteger);
/* update entry */
newts = stksess_new(ps->table->table, &stkey);
if (newts) {
/* lookup for existing entry */
ts = stktable_lookup(ps->table->table, newts);
if (ts) {
/* the entry already exist, we can free ours */
stktable_touch(ps->table->table, ts, 0);
stksess_free(ps->table->table, newts);
}
else {
struct eb32_node *eb;
/* create new entry */
ts = stktable_store(ps->table->table, newts, 0);
ts->upd.key= (++ps->table->table->update)+(2^31);
eb = eb32_insert(&ps->table->table->updates, &ts->upd);
if (eb != &ts->upd) {
eb32_delete(eb);
eb32_insert(&ps->table->table->updates, &ts->upd);
}
}
/* update entry */
if (srvid && stktable_data_ptr(ps->table->table, ts, STKTABLE_DT_SERVER_ID))
stktable_data_cast(stktable_data_ptr(ps->table->table, ts, STKTABLE_DT_SERVER_ID), server_id) = srvid;
ps->pushack = pushack;
}
}
else if (c == 'R') {
/* Reset message: remote need resync */
/* reinit counters for a resync */
ps->lastpush = 0;
ps->teaching_origin = ps->pushed = ps->table->table->update;
/* reset teaching flags to 0 */
ps->flags &= PEER_TEACH_RESET;
/* flag to start to teach lesson */
ps->flags |= PEER_F_TEACH_PROCESS;
}
else if (c == 'F') {
/* Finish message, all known updates have been pushed by remote */
/* and remote is up to date */
/* If resync is in progress with remote peer */
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer for learning */
ps->flags &= ~PEER_F_LEARN_ASSIGN;
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* Consider table is now up2date, resync resync no more needed from local neither remote */
ps->table->flags |= (SHTABLE_F_RESYNC_LOCAL|SHTABLE_F_RESYNC_REMOTE);
}
/* Increase confirm counter to launch a confirm message */
ps->confirm++;
}
else if (c == 'c') {
/* confirm message, remote peer is now up to date with us */
/* If stopping state */
if (stopping) {
/* Close session, push resync no more needed */
ps->flags |= PEER_F_TEACH_COMPLETE;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* reset teaching flags to 0 */
ps->flags &= PEER_TEACH_RESET;
}
else if (c == 'C') {
/* Continue message, all known updates have been pushed by remote */
/* but remote is not up to date */
/* If resync is in progress with current peer */
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer */
ps->flags &= ~PEER_F_LEARN_ASSIGN;
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* flag current peer is not up 2 date to try from an other */
ps->flags |= PEER_F_LEARN_NOTUP2DATE;
/* reschedule a resync */
ps->table->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
}
ps->confirm++;
}
else if (c == 'A') {
/* ack message */
uint32_t netinteger;
reql = bo_getblk(si->ob, (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0) {
goto incomplete;
}
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
totl += reql;
/* Consider remote is up to date with "acked" version */
ps->update = ntohl(netinteger);
}
else {
/* Unknown message */
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* skip consumed message */
bo_skip(si->ob, totl);
/* loop on that state to peek next message */
continue;
incomplete:
/* Nothing to read, now we start to write */
/* Confirm finished or partial messages */
while (ps->confirm) {
/* There is a confirm messages to send */
repl = bi_putchr(si->ib, 'c');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->confirm--;
}
/* Need to request a resync */
if ((ps->flags & PEER_F_LEARN_ASSIGN) &&
(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN) &&
!(ps->table->flags & SHTABLE_F_RESYNC_PROCESS)) {
/* Current peer was elected to request a resync */
repl = bi_putchr(si->ib, 'R');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->table->flags |= SHTABLE_F_RESYNC_PROCESS;
}
/* It remains some updates to ack */
if (ps->pushack != ps->lastack) {
uint32_t netinteger;
trash[0] = 'A';
netinteger = htonl(ps->pushack);
memcpy(&trash[1], &netinteger, sizeof(netinteger));
repl = bi_putblk(si->ib, trash, 1+sizeof(netinteger));
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->lastack = ps->pushack;
}
if (ps->flags & PEER_F_TEACH_PROCESS) {
/* current peer was requested for a lesson */
if (!(ps->flags & PEER_F_TEACH_STAGE1)) {
/* lesson stage 1 not complete */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
if (!eb) {
/* flag lesson stage1 complete */
ps->flags |= PEER_F_TEACH_STAGE1;
eb = eb32_first(&ps->table->table->updates);
if (eb)
ps->pushed = eb->key - 1;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash, trashlen);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si->ib, trash, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* !TEACH_STAGE1 */
if (!(ps->flags & PEER_F_TEACH_STAGE2)) {
/* lesson stage 2 not complete */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
if (!eb || eb->key > ps->teaching_origin) {
/* flag lesson stage1 complete */
ps->flags |= PEER_F_TEACH_STAGE2;
ps->pushed = ps->teaching_origin;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash, trashlen);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si->ib, trash, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* !TEACH_STAGE2 */
if (!(ps->flags & PEER_F_TEACH_FINISHED)) {
/* process final lesson message */
repl = bi_putchr(si->ib, ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FINISHED) ? 'F' : 'C');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
/* flag finished message sent */
ps->flags |= PEER_F_TEACH_FINISHED;
} /* !TEACH_FINISHED */
} /* TEACH_PROCESS */
if (!(ps->flags & PEER_F_LEARN_ASSIGN) &&
(int)(ps->pushed - ps->table->table->localupdate) < 0) {
/* Push local updates, only if no learning in progress (to avoid ping-pong effects) */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
/* push local updates */
if (!eb) {
eb = eb32_first(&ps->table->table->updates);
if (!eb || ((int)(eb->key - ps->pushed) <= 0)) {
ps->pushed = ps->table->table->localupdate;
break;
}
}
if ((int)(eb->key - ps->table->table->localupdate) > 0) {
ps->pushed = ps->table->table->localupdate;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash, trashlen);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si->ib, trash, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* ! LEARN_ASSIGN */
/* noting more to do */
goto out;
}
case PEER_SESSION_EXIT:
repl = snprintf(trash, trashlen, "%d\n", si->applet.st1);
if (bi_putblk(si->ib, trash, repl) == -1)
goto out;
si->applet.st0 = PEER_SESSION_END;
/* fall through */
case PEER_SESSION_END: {
si_shutw(si);
si_shutr(si);
si->ib->flags |= CF_READ_NULL;
goto quit;
}
}
}
out:
si_update(si);
si->ob->flags |= CF_READ_DONTWAIT;
/* we don't want to expire timeouts while we're processing requests */
si->ib->rex = TICK_ETERNITY;
si->ob->wex = TICK_ETERNITY;
quit:
return;
}
/*
* Task processing function to manage re-connect and peer session
* tasks wakeup on local update.
*/
static struct task *process_peer_sync(struct task * task)
{
struct shared_table *st = (struct shared_table *)task->context;
struct peer_session *ps;
task->expire = TICK_ETERNITY;
if (!stopping) {
/* Normal case (not soft stop)*/
if (((st->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMLOCAL) &&
(!nb_oldpids || tick_is_expired(st->resync_timeout, now_ms)) &&
!(st->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* Resync from local peer needed
no peer was assigned for the lesson
and no old local peer found
or resync timeout expire */
/* flag no more resync from local, to try resync from remotes */
st->flags |= SHTABLE_F_RESYNC_LOCAL;
/* reschedule a resync */
st->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
}
/* For each session */
for (ps = st->sessions; ps; ps = ps->next) {
/* For each remote peers */
if (!ps->peer->local) {
if (!ps->session) {
/* no active session */
if (ps->statuscode == 0 ||
ps->statuscode == PEER_SESSION_SUCCESSCODE ||
((ps->statuscode == PEER_SESSION_CONNECTCODE ||
ps->statuscode == PEER_SESSION_CONNECTEDCODE) &&
tick_is_expired(ps->reconnect, now_ms))) {
/* connection never tried
* or previous session established with success
* or previous session failed during connection
* and reconnection timer is expired */
/* retry a connect */
ps->session = peer_session_create(ps->peer, ps);
}
else if (ps->statuscode == PEER_SESSION_CONNECTCODE ||
ps->statuscode == PEER_SESSION_CONNECTEDCODE) {
/* If previous session failed during connection
* but reconnection timer is not expired */
/* reschedule task for reconnect */
task->expire = tick_first(task->expire, ps->reconnect);
}
/* else do nothing */
} /* !ps->session */
else if (ps->statuscode == PEER_SESSION_SUCCESSCODE) {
/* current session is active and established */
if (((st->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE) &&
!(st->flags & SHTABLE_F_RESYNC_ASSIGN) &&
!(ps->flags & PEER_F_LEARN_NOTUP2DATE)) {
/* Resync from a remote is needed
* and no peer was assigned for lesson
* and current peer may be up2date */
/* assign peer for the lesson */
ps->flags |= PEER_F_LEARN_ASSIGN;
st->flags |= SHTABLE_F_RESYNC_ASSIGN;
/* awake peer session task to handle a request of resync */
task_wakeup(ps->session->task, TASK_WOKEN_MSG);
}
else if ((int)(ps->pushed - ps->table->table->localupdate) < 0) {
/* awake peer session task to push local updates */
task_wakeup(ps->session->task, TASK_WOKEN_MSG);
}
/* else do nothing */
} /* SUCCESSCODE */
} /* !ps->peer->local */
} /* for */
/* Resync from remotes expired: consider resync is finished */
if (((st->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE) &&
!(st->flags & SHTABLE_F_RESYNC_ASSIGN) &&
tick_is_expired(st->resync_timeout, now_ms)) {
/* Resync from remote peer needed
* no peer was assigned for the lesson
* and resync timeout expire */
/* flag no more resync from remote, consider resync is finished */
st->flags |= SHTABLE_F_RESYNC_REMOTE;
}
if ((st->flags & SHTABLE_RESYNC_STATEMASK) != SHTABLE_RESYNC_FINISHED) {
/* Resync not finished*/
/* reschedule task to resync timeout, to ended resync if needed */
task->expire = tick_first(task->expire, st->resync_timeout);
}
} /* !stopping */
else {
/* soft stop case */
if (task->state & TASK_WOKEN_SIGNAL) {
/* We've just recieved the signal */
if (!(st->flags & SHTABLE_F_DONOTSTOP)) {
/* add DO NOT STOP flag if not present */
jobs++;
st->flags |= SHTABLE_F_DONOTSTOP;
}
/* disconnect all connected peers */
for (ps = st->sessions; ps; ps = ps->next) {
if (ps->session) {
peer_session_forceshutdown(ps->session);
ps->session = NULL;
}
}
}
ps = st->local_session;
if (ps->flags & PEER_F_TEACH_COMPLETE) {
if (st->flags & SHTABLE_F_DONOTSTOP) {
/* resync of new process was complete, current process can die now */
jobs--;
st->flags &= ~SHTABLE_F_DONOTSTOP;
}
}
else if (!ps->session) {
/* If session is not active */
if (ps->statuscode == 0 ||
ps->statuscode == PEER_SESSION_SUCCESSCODE ||
ps->statuscode == PEER_SESSION_CONNECTEDCODE ||
ps->statuscode == PEER_SESSION_TRYAGAIN) {
/* connection never tried
* or previous session was successfully established
* or previous session tcp connect success but init state incomplete
* or during previous connect, peer replies a try again statuscode */
/* connect to the peer */
ps->session = peer_session_create(ps->peer, ps);
}
else {
/* Other error cases */
if (st->flags & SHTABLE_F_DONOTSTOP) {
/* unable to resync new process, current process can die now */
jobs--;
st->flags &= ~SHTABLE_F_DONOTSTOP;
}
}
}
else if (ps->statuscode == PEER_SESSION_SUCCESSCODE &&
(int)(ps->pushed - ps->table->table->localupdate) < 0) {
/* current session active and established
awake session to push remaining local updates */
task_wakeup(ps->session->task, TASK_WOKEN_MSG);
}
} /* stopping */
/* Wakeup for re-connect */
return task;
}
/*
* Create a new peer session in assigned state (connect will start automatically)
*/
static struct session *peer_session_create(struct peer *peer, struct peer_session *ps)
{
struct listener *l = ((struct proxy *)peer->peers->peers_fe)->listen;
struct proxy *p = (struct proxy *)l->frontend; /* attached frontend */
struct session *s;
struct http_txn *txn;
struct task *t;
if ((s = pool_alloc2(pool2_session)) == NULL) { /* disable this proxy for a while */
Alert("out of memory in event_accept().\n");
goto out_close;
}
LIST_ADDQ(&sessions, &s->list);
LIST_INIT(&s->back_refs);
s->flags = SN_ASSIGNED|SN_ADDR_SET;
s->term_trace = 0;
/* if this session comes from a known monitoring system, we want to ignore
* it as soon as possible, which means closing it immediately for TCP.
*/
if ((t = task_new()) == NULL) { /* disable this proxy for a while */
Alert("out of memory in event_accept().\n");
goto out_free_session;
}
ps->reconnect = tick_add(now_ms, MS_TO_TICKS(5000));
ps->statuscode = PEER_SESSION_CONNECTCODE;
t->process = l->handler;
t->context = s;
t->nice = l->nice;
memcpy(&s->si[1].conn.addr.to, &peer->addr, sizeof(s->si[1].conn.addr.to));
s->task = t;
s->listener = l;
/* Note: initially, the session's backend points to the frontend.
* This changes later when switching rules are executed or
* when the default backend is assigned.
*/
s->be = s->fe = p;
s->req = s->rep = NULL; /* will be allocated later */
s->si[0].conn.t.sock.fd = -1;
s->si[0].conn.flags = CO_FL_NONE;
s->si[0].owner = t;
s->si[0].state = s->si[0].prev_state = SI_ST_EST;
s->si[0].err_type = SI_ET_NONE;
s->si[0].err_loc = NULL;
s->si[0].release = NULL;
s->si[0].send_proxy_ofs = 0;
set_target_client(&s->si[0].conn.target, l);
s->si[0].exp = TICK_ETERNITY;
s->si[0].flags = SI_FL_NONE;
if (s->fe->options2 & PR_O2_INDEPSTR)
s->si[0].flags |= SI_FL_INDEP_STR;
stream_int_register_handler(&s->si[0], &peer_applet);
s->si[0].applet.st0 = PEER_SESSION_CONNECT;
s->si[0].conn.data_ctx = (void *)ps;
s->si[1].conn.t.sock.fd = -1; /* just to help with debugging */
s->si[1].conn.flags = CO_FL_NONE;
s->si[1].owner = t;
s->si[1].state = s->si[1].prev_state = SI_ST_ASS;
s->si[1].conn_retries = p->conn_retries;
s->si[1].err_type = SI_ET_NONE;
s->si[1].err_loc = NULL;
s->si[1].release = NULL;
s->si[1].send_proxy_ofs = 0;
set_target_proxy(&s->si[1].conn.target, s->be);
si_prepare_conn(&s->si[1], peer->proto, peer->data);
s->si[1].exp = TICK_ETERNITY;
s->si[1].flags = SI_FL_NONE;
if (s->be->options2 & PR_O2_INDEPSTR)
s->si[1].flags |= SI_FL_INDEP_STR;
session_init_srv_conn(s);
set_target_proxy(&s->target, s->be);
s->pend_pos = NULL;
/* init store persistence */
s->store_count = 0;
s->stkctr1_entry = NULL;
s->stkctr2_entry = NULL;
/* FIXME: the logs are horribly complicated now, because they are
* defined in <p>, <p>, and later <be> and <be>.
*/
s->logs.logwait = 0;
s->do_log = NULL;
/* default error reporting function, may be changed by analysers */
s->srv_error = default_srv_error;
s->uniq_id = 0;
s->unique_id = NULL;
txn = &s->txn;
/* Those variables will be checked and freed if non-NULL in
* session.c:session_free(). It is important that they are
* properly initialized.
*/
txn->sessid = NULL;
txn->srv_cookie = NULL;
txn->cli_cookie = NULL;
txn->uri = NULL;
txn->req.cap = NULL;
txn->rsp.cap = NULL;
txn->hdr_idx.v = NULL;
txn->hdr_idx.size = txn->hdr_idx.used = 0;
if ((s->req = pool_alloc2(pool2_channel)) == NULL)
goto out_fail_req; /* no memory */
s->req->buf.size = global.tune.bufsize;
channel_init(s->req);
s->req->prod = &s->si[0];
s->req->cons = &s->si[1];
s->si[0].ib = s->si[1].ob = s->req;
s->req->flags |= CF_READ_ATTACHED; /* the producer is already connected */
/* activate default analysers enabled for this listener */
s->req->analysers = l->analysers;
/* note: this should not happen anymore since there's always at least the switching rules */
if (!s->req->analysers) {
channel_auto_connect(s->req);/* don't wait to establish connection */
channel_auto_close(s->req);/* let the producer forward close requests */
}
s->req->rto = s->fe->timeout.client;
s->req->wto = s->be->timeout.server;
if ((s->rep = pool_alloc2(pool2_channel)) == NULL)
goto out_fail_rep; /* no memory */
s->rep->buf.size = global.tune.bufsize;
channel_init(s->rep);
s->rep->prod = &s->si[1];
s->rep->cons = &s->si[0];
s->si[0].ob = s->si[1].ib = s->rep;
s->rep->rto = s->be->timeout.server;
s->rep->wto = s->fe->timeout.client;
s->req->rex = TICK_ETERNITY;
s->req->wex = TICK_ETERNITY;
s->req->analyse_exp = TICK_ETERNITY;
s->rep->rex = TICK_ETERNITY;
s->rep->wex = TICK_ETERNITY;
s->rep->analyse_exp = TICK_ETERNITY;
t->expire = TICK_ETERNITY;
s->rep->flags |= CF_READ_DONTWAIT;
/* it is important not to call the wakeup function directly but to
* pass through task_wakeup(), because this one knows how to apply
* priorities to tasks.
*/
task_wakeup(t, TASK_WOKEN_INIT);
l->nbconn++; /* warning! right now, it's up to the handler to decrease this */
p->feconn++;/* beconn will be increased later */
jobs++;
if (!(s->listener->options & LI_O_UNLIMITED))
actconn++;
totalconn++;
return s;
/* Error unrolling */
out_fail_rep:
pool_free2(pool2_channel, s->req);
out_fail_req:
task_free(t);
out_free_session:
LIST_DEL(&s->list);
pool_free2(pool2_session, s);
out_close:
return s;
}
/* This function is called on a read event from a listening socket, corresponding
* to an accept. It tries to accept as many connections as possible, and for each
* calls the listener's accept handler (generally the frontend's accept handler).
*/
int stream_sock_accept(int fd)
{
struct listener *l = fdtab[fd].owner;
struct proxy *p = l->frontend;
int max_accept = global.tune.maxaccept;
int cfd;
int ret;
if (unlikely(l->nbconn >= l->maxconn)) {
listener_full(l);
return 0;
}
if (global.cps_lim && !(l->options & LI_O_UNLIMITED)) {
int max = freq_ctr_remain(&global.conn_per_sec, global.cps_lim, 0);
if (unlikely(!max)) {
/* frontend accept rate limit was reached */
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, next_event_delay(&global.conn_per_sec, global.cps_lim, 0)));
return 0;
}
if (max_accept > max)
max_accept = max;
}
if (p && p->fe_sps_lim) {
int max = freq_ctr_remain(&p->fe_sess_per_sec, p->fe_sps_lim, 0);
if (unlikely(!max)) {
/* frontend accept rate limit was reached */
limit_listener(l, &p->listener_queue);
task_schedule(p->task, tick_add(now_ms, next_event_delay(&p->fe_sess_per_sec, p->fe_sps_lim, 0)));
return 0;
}
if (max_accept > max)
max_accept = max;
}
/* Note: if we fail to allocate a connection because of configured
* limits, we'll schedule a new attempt worst 1 second later in the
* worst case. If we fail due to system limits or temporary resource
* shortage, we try again 100ms later in the worst case.
*/
while (max_accept--) {
struct sockaddr_storage addr;
socklen_t laddr = sizeof(addr);
if (unlikely(actconn >= global.maxconn) && !(l->options & LI_O_UNLIMITED)) {
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 1000)); /* try again in 1 second */
return 0;
}
if (unlikely(p && p->feconn >= p->maxconn)) {
limit_listener(l, &p->listener_queue);
return 0;
}
cfd = accept(fd, (struct sockaddr *)&addr, &laddr);
if (unlikely(cfd == -1)) {
switch (errno) {
case EAGAIN:
case EINTR:
case ECONNABORTED:
return 0; /* nothing more to accept */
case ENFILE:
if (p)
send_log(p, LOG_EMERG,
"Proxy %s reached system FD limit at %d. Please check system tunables.\n",
p->id, maxfd);
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 100)); /* try again in 100 ms */
return 0;
case EMFILE:
if (p)
send_log(p, LOG_EMERG,
"Proxy %s reached process FD limit at %d. Please check 'ulimit-n' and restart.\n",
p->id, maxfd);
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 100)); /* try again in 100 ms */
return 0;
case ENOBUFS:
case ENOMEM:
if (p)
send_log(p, LOG_EMERG,
"Proxy %s reached system memory limit at %d sockets. Please check system tunables.\n",
p->id, maxfd);
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 100)); /* try again in 100 ms */
return 0;
default:
return 0;
}
}
if (unlikely(cfd >= global.maxsock)) {
send_log(p, LOG_EMERG,
"Proxy %s reached the configured maximum connection limit. Please check the global 'maxconn' value.\n",
p->id);
close(cfd);
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 1000)); /* try again in 1 second */
return 0;
}
/* increase the per-process number of cumulated connections */
if (!(l->options & LI_O_UNLIMITED)) {
update_freq_ctr(&global.conn_per_sec, 1);
if (global.conn_per_sec.curr_ctr > global.cps_max)
global.cps_max = global.conn_per_sec.curr_ctr;
actconn++;
}
jobs++;
totalconn++;
l->nbconn++;
if (l->counters) {
if (l->nbconn > l->counters->conn_max)
l->counters->conn_max = l->nbconn;
}
ret = l->accept(l, cfd, &addr);
if (unlikely(ret <= 0)) {
/* The connection was closed by session_accept(). Either
* we just have to ignore it (ret == 0) or it's a critical
* error due to a resource shortage, and we must stop the
* listener (ret < 0).
*/
if (!(l->options & LI_O_UNLIMITED))
actconn--;
jobs--;
l->nbconn--;
if (ret == 0) /* successful termination */
continue;
limit_listener(l, &global_listener_queue);
task_schedule(global_listener_queue_task, tick_add(now_ms, 100)); /* try again in 100 ms */
return 0;
}
if (l->nbconn >= l->maxconn) {
listener_full(l);
return 0;
}
} /* end of while (p->feconn < p->maxconn) */
return 0;
}
/*
* this function enables proxies when there are enough free sessions,
* or stops them when the table is full. It is designed to be called from the
* select_loop(). It adjusts the date of next expiration event during stop
* time if appropriate.
*/
void maintain_proxies(int *next)
{
struct proxy *p;
struct listener *l;
unsigned int wait;
p = proxy;
/* if there are enough free sessions, we'll activate proxies */
if (actconn < global.maxconn) {
for (; p; p = p->next) {
/* check the various reasons we may find to block the frontend */
if (p->feconn >= p->maxconn)
goto do_block;
if (p->fe_sps_lim &&
(wait = next_event_delay(&p->fe_sess_per_sec, p->fe_sps_lim, 1))) {
/* we're blocking because a limit was reached on the number of
* requests/s on the frontend. We want to re-check ASAP, which
* means in 1 ms before estimated expiration date, because the
* timer will have settled down. Note that we may already be in
* IDLE state here.
*/
*next = tick_first(*next, tick_add(now_ms, wait));
goto do_block;
}
/* OK we have no reason to block, so let's unblock if we were blocking */
if (p->state == PR_STIDLE) {
for (l = p->listen; l != NULL; l = l->next)
enable_listener(l);
p->state = PR_STRUN;
}
continue;
do_block:
if (p->state == PR_STRUN) {
for (l = p->listen; l != NULL; l = l->next)
disable_listener(l);
p->state = PR_STIDLE;
}
}
}
else { /* block all proxies */
while (p) {
if (p->state == PR_STRUN) {
for (l = p->listen; l != NULL; l = l->next)
disable_listener(l);
p->state = PR_STIDLE;
}
p = p->next;
}
}
if (stopping) {
p = proxy;
while (p) {
if (p->state != PR_STSTOPPED) {
int t;
t = tick_remain(now_ms, p->stop_time);
if (t == 0) {
Warning("Proxy %s stopped (FE: %lld conns, BE: %lld conns).\n",
p->id, p->cum_feconn, p->cum_beconn);
send_log(p, LOG_WARNING, "Proxy %s stopped (FE: %lld conns, BE: %lld conns).\n",
p->id, p->cum_feconn, p->cum_beconn);
stop_proxy(p);
/* try to free more memory */
pool_gc2();
}
else {
*next = tick_first(*next, p->stop_time);
}
}
p = p->next;
}
}
return;
}
/*
* IO Handler to handle message exchance with a peer
*/
static void peer_io_handler(struct appctx *appctx)
{
struct stream_interface *si = appctx->owner;
struct stream *s = si_strm(si);
struct peers *curpeers = (struct peers *)strm_fe(s)->parent;
int reql = 0;
int repl = 0;
while (1) {
switchstate:
switch(appctx->st0) {
case PEER_SESS_ST_ACCEPT:
appctx->ctx.peers.ptr = NULL;
appctx->st0 = PEER_SESS_ST_GETVERSION;
/* fall through */
case PEER_SESS_ST_GETVERSION:
reql = bo_getline(si_oc(si), trash.str, trash.size);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
if (trash.str[reql-1] != '\n') {
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
else if (reql > 1 && (trash.str[reql-2] == '\r'))
trash.str[reql-2] = 0;
else
trash.str[reql-1] = 0;
bo_skip(si_oc(si), reql);
/* test version */
if (strcmp(PEER_SESSION_PROTO_NAME " 1.0", trash.str) != 0) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRVERSION;
/* test protocol */
if (strncmp(PEER_SESSION_PROTO_NAME " ", trash.str, strlen(PEER_SESSION_PROTO_NAME)+1) != 0)
appctx->st1 = PEER_SESS_SC_ERRPROTO;
goto switchstate;
}
appctx->st0 = PEER_SESS_ST_GETHOST;
/* fall through */
case PEER_SESS_ST_GETHOST:
reql = bo_getline(si_oc(si), trash.str, trash.size);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
if (trash.str[reql-1] != '\n') {
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
else if (reql > 1 && (trash.str[reql-2] == '\r'))
trash.str[reql-2] = 0;
else
trash.str[reql-1] = 0;
bo_skip(si_oc(si), reql);
/* test hostname match */
if (strcmp(localpeer, trash.str) != 0) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRHOST;
goto switchstate;
}
appctx->st0 = PEER_SESS_ST_GETPEER;
/* fall through */
case PEER_SESS_ST_GETPEER: {
struct peer *curpeer;
char *p;
reql = bo_getline(si_oc(si), trash.str, trash.size);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
if (trash.str[reql-1] != '\n') {
/* Incomplete line, we quit */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
else if (reql > 1 && (trash.str[reql-2] == '\r'))
trash.str[reql-2] = 0;
else
trash.str[reql-1] = 0;
bo_skip(si_oc(si), reql);
/* parse line "<peer name> <pid>" */
p = strchr(trash.str, ' ');
if (!p) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRPROTO;
goto switchstate;
}
*p = 0;
/* lookup known peer */
for (curpeer = curpeers->remote; curpeer; curpeer = curpeer->next) {
if (strcmp(curpeer->id, trash.str) == 0)
break;
}
/* if unknown peer */
if (!curpeer) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRPEER;
goto switchstate;
}
appctx->ctx.peers.ptr = curpeer;
appctx->st0 = PEER_SESS_ST_GETTABLE;
/* fall through */
}
case PEER_SESS_ST_GETTABLE: {
struct peer *curpeer = (struct peer *)appctx->ctx.peers.ptr;
struct shared_table *st;
struct peer_session *ps = NULL;
unsigned long key_type;
size_t key_size;
char *p;
reql = bo_getline(si_oc(si), trash.str, trash.size);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
appctx->ctx.peers.ptr = NULL;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* Re init appctx->ctx.peers.ptr to null, to handle correctly a release case */
appctx->ctx.peers.ptr = NULL;
if (trash.str[reql-1] != '\n') {
/* Incomplete line, we quit */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
else if (reql > 1 && (trash.str[reql-2] == '\r'))
trash.str[reql-2] = 0;
else
trash.str[reql-1] = 0;
bo_skip(si_oc(si), reql);
/* Parse line "<table name> <type> <size>" */
p = strchr(trash.str, ' ');
if (!p) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRPROTO;
goto switchstate;
}
*p = 0;
key_type = (unsigned long)atol(p+1);
p = strchr(p+1, ' ');
if (!p) {
appctx->ctx.peers.ptr = NULL;
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRPROTO;
goto switchstate;
}
key_size = (size_t)atoi(p);
for (st = curpeers->tables; st; st = st->next) {
/* If table name matches */
if (strcmp(st->table->id, trash.str) == 0) {
/* Check key size mismatches, except for strings
* which may be truncated as long as they fit in
* a buffer.
*/
if (key_size != st->table->key_size &&
(key_type != STKTABLE_TYPE_STRING ||
1 + 4 + 4 + key_size - 1 >= trash.size)) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRSIZE;
goto switchstate;
}
/* If key type mismatches */
if (key_type != st->table->type) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRTYPE;
goto switchstate;
}
/* lookup peer stream of current peer */
for (ps = st->sessions; ps; ps = ps->next) {
if (ps->peer == curpeer) {
/* If stream already active, replaced by new one */
if (ps->stream && ps->stream != s) {
if (ps->peer->local) {
/* Local connection, reply a retry */
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_TRYAGAIN;
goto switchstate;
}
peer_session_forceshutdown(ps->stream);
}
ps->stream = s;
ps->appctx = appctx;
break;
}
}
break;
}
}
/* If table not found */
if (!st){
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRTABLE;
goto switchstate;
}
/* If no peer session for current peer */
if (!ps) {
appctx->st0 = PEER_SESS_ST_EXIT;
appctx->st1 = PEER_SESS_SC_ERRPEER;
goto switchstate;
}
appctx->ctx.peers.ptr = ps;
appctx->st0 = PEER_SESS_ST_SENDSUCCESS;
/* fall through */
}
case PEER_SESS_ST_SENDSUCCESS: {
struct peer_session *ps = (struct peer_session *)appctx->ctx.peers.ptr;
repl = snprintf(trash.str, trash.size, "%d\n", PEER_SESS_SC_SUCCESSCODE);
repl = bi_putblk(si_ic(si), trash.str, repl);
if (repl <= 0) {
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* Register status code */
ps->statuscode = PEER_SESS_SC_SUCCESSCODE;
/* Awake main task */
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
/* Init cursors */
ps->teaching_origin =ps->lastpush = ps->lastack = ps->pushack = 0;
ps->pushed = ps->update;
/* Init confirm counter */
ps->confirm = 0;
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
/* if current peer is local */
if (ps->peer->local) {
/* if table need resyncfrom local and no process assined */
if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMLOCAL &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* assign local peer for a lesson, consider lesson already requested */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= (SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
}
}
else if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* assign peer for a lesson */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= SHTABLE_F_RESYNC_ASSIGN;
}
/* switch to waiting message state */
appctx->st0 = PEER_SESS_ST_WAITMSG;
goto switchstate;
}
case PEER_SESS_ST_CONNECT: {
struct peer_session *ps = (struct peer_session *)appctx->ctx.peers.ptr;
/* Send headers */
repl = snprintf(trash.str, trash.size,
PEER_SESSION_PROTO_NAME " 1.0\n%s\n%s %d\n%s %lu %d\n",
ps->peer->id,
localpeer,
(int)getpid(),
ps->table->table->id,
ps->table->table->type,
(int)ps->table->table->key_size);
if (repl >= trash.size) {
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
repl = bi_putblk(si_ic(si), trash.str, repl);
if (repl <= 0) {
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* switch to the waiting statuscode state */
appctx->st0 = PEER_SESS_ST_GETSTATUS;
/* fall through */
}
case PEER_SESS_ST_GETSTATUS: {
struct peer_session *ps = (struct peer_session *)appctx->ctx.peers.ptr;
if (si_ic(si)->flags & CF_WRITE_PARTIAL)
ps->statuscode = PEER_SESS_SC_CONNECTEDCODE;
reql = bo_getline(si_oc(si), trash.str, trash.size);
if (reql <= 0) { /* closed or EOL not found */
if (reql == 0)
goto out;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
if (trash.str[reql-1] != '\n') {
/* Incomplete line, we quit */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
else if (reql > 1 && (trash.str[reql-2] == '\r'))
trash.str[reql-2] = 0;
else
trash.str[reql-1] = 0;
bo_skip(si_oc(si), reql);
/* Register status code */
ps->statuscode = atoi(trash.str);
/* Awake main task */
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
/* If status code is success */
if (ps->statuscode == PEER_SESS_SC_SUCCESSCODE) {
/* Init cursors */
ps->teaching_origin = ps->lastpush = ps->lastack = ps->pushack = 0;
ps->pushed = ps->update;
/* Init confirm counter */
ps->confirm = 0;
/* reset teaching and learning flags to 0 */
ps->flags &= PEER_TEACH_RESET;
ps->flags &= PEER_LEARN_RESET;
/* If current peer is local */
if (ps->peer->local) {
/* Init cursors to push a resync */
ps->teaching_origin = ps->pushed = ps->table->table->update;
/* flag to start to teach lesson */
ps->flags |= PEER_F_TEACH_PROCESS;
}
else if ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FROMREMOTE &&
!(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN)) {
/* If peer is remote and resync from remote is needed,
and no peer currently assigned */
/* assign peer for a lesson */
ps->flags |= PEER_F_LEARN_ASSIGN;
ps->table->flags |= SHTABLE_F_RESYNC_ASSIGN;
}
}
else {
/* Status code is not success, abort */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
appctx->st0 = PEER_SESS_ST_WAITMSG;
/* fall through */
}
case PEER_SESS_ST_WAITMSG: {
struct peer_session *ps = (struct peer_session *)appctx->ctx.peers.ptr;
struct stksess *ts, *newts = NULL;
char c;
int totl = 0;
reql = bo_getblk(si_oc(si), (char *)&c, sizeof(c), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
if ((c & 0x80) || (c == 'D')) {
/* Here we have data message */
unsigned int pushack;
int srvid;
uint32_t netinteger;
/* Compute update remote version */
if (c & 0x80) {
pushack = ps->pushack + (unsigned int)(c & 0x7F);
}
else {
reql = bo_getblk(si_oc(si), (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
pushack = ntohl(netinteger);
}
/* Read key. The string keys are read in two steps, the first step
* consists in reading whatever fits into the table directly into
* the pre-allocated key. The second step consists in simply
* draining all exceeding data. This can happen for example after a
* config reload with a smaller key size for the stick table than
* what was previously set, or when facing the impossibility to
* allocate a new stksess (for example when the table is full with
* "nopurge").
*/
if (ps->table->table->type == STKTABLE_TYPE_STRING) {
unsigned int to_read, to_store;
/* read size first */
reql = bo_getblk(si_oc(si), (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
to_store = 0;
to_read = ntohl(netinteger);
if (to_read + totl > si_ob(si)->size) {
/* impossible to read a key this large, abort */
reql = -1;
goto incomplete;
}
newts = stksess_new(ps->table->table, NULL);
if (newts)
to_store = MIN(to_read, ps->table->table->key_size - 1);
/* we read up to two blocks, the first one goes into the key,
* the rest is drained into the trash.
*/
if (to_store) {
reql = bo_getblk(si_oc(si), (char *)newts->key.key, to_store, totl);
if (reql <= 0) /* closed or incomplete */
goto incomplete;
newts->key.key[reql] = 0;
totl += reql;
to_read -= reql;
}
if (to_read) {
reql = bo_getblk(si_oc(si), trash.str, to_read, totl);
if (reql <= 0) /* closed or incomplete */
goto incomplete;
totl += reql;
}
}
else if (ps->table->table->type == STKTABLE_TYPE_INTEGER) {
reql = bo_getblk(si_oc(si), (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
newts = stksess_new(ps->table->table, NULL);
if (newts) {
netinteger = ntohl(netinteger);
memcpy(newts->key.key, &netinteger, sizeof(netinteger));
}
totl += reql;
}
else {
/* type ip or binary */
newts = stksess_new(ps->table->table, NULL);
reql = bo_getblk(si_oc(si), newts ? (char *)newts->key.key : trash.str, ps->table->table->key_size, totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
}
/* read server id */
reql = bo_getblk(si_oc(si), (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
srvid = ntohl(netinteger);
/* update entry */
if (newts) {
/* lookup for existing entry */
ts = stktable_lookup(ps->table->table, newts);
if (ts) {
/* the entry already exist, we can free ours */
stktable_touch(ps->table->table, ts, 0);
stksess_free(ps->table->table, newts);
newts = NULL;
}
else {
struct eb32_node *eb;
/* create new entry */
ts = stktable_store(ps->table->table, newts, 0);
newts = NULL; /* don't reuse it */
ts->upd.key= (++ps->table->table->update)+(2^31);
eb = eb32_insert(&ps->table->table->updates, &ts->upd);
if (eb != &ts->upd) {
eb32_delete(eb);
eb32_insert(&ps->table->table->updates, &ts->upd);
}
}
/* update entry */
if (srvid && stktable_data_ptr(ps->table->table, ts, STKTABLE_DT_SERVER_ID))
stktable_data_cast(stktable_data_ptr(ps->table->table, ts, STKTABLE_DT_SERVER_ID), server_id) = srvid;
ps->pushack = pushack;
}
}
else if (c == 'R') {
/* Reset message: remote need resync */
/* reinit counters for a resync */
ps->lastpush = 0;
ps->teaching_origin = ps->pushed = ps->table->table->update;
/* reset teaching flags to 0 */
ps->flags &= PEER_TEACH_RESET;
/* flag to start to teach lesson */
ps->flags |= PEER_F_TEACH_PROCESS;
}
else if (c == 'F') {
/* Finish message, all known updates have been pushed by remote */
/* and remote is up to date */
/* If resync is in progress with remote peer */
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer for learning */
ps->flags &= ~PEER_F_LEARN_ASSIGN;
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* Consider table is now up2date, resync resync no more needed from local neither remote */
ps->table->flags |= (SHTABLE_F_RESYNC_LOCAL|SHTABLE_F_RESYNC_REMOTE);
}
/* Increase confirm counter to launch a confirm message */
ps->confirm++;
}
else if (c == 'c') {
/* confirm message, remote peer is now up to date with us */
/* If stopping state */
if (stopping) {
/* Close session, push resync no more needed */
ps->flags |= PEER_F_TEACH_COMPLETE;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* reset teaching flags to 0 */
ps->flags &= PEER_TEACH_RESET;
}
else if (c == 'C') {
/* Continue message, all known updates have been pushed by remote */
/* but remote is not up to date */
/* If resync is in progress with current peer */
if (ps->flags & PEER_F_LEARN_ASSIGN) {
/* unassign current peer */
ps->flags &= ~PEER_F_LEARN_ASSIGN;
ps->table->flags &= ~(SHTABLE_F_RESYNC_ASSIGN|SHTABLE_F_RESYNC_PROCESS);
/* flag current peer is not up 2 date to try from an other */
ps->flags |= PEER_F_LEARN_NOTUP2DATE;
/* reschedule a resync */
ps->table->resync_timeout = tick_add(now_ms, MS_TO_TICKS(5000));
task_wakeup(ps->table->sync_task, TASK_WOKEN_MSG);
}
ps->confirm++;
}
else if (c == 'A') {
/* ack message */
uint32_t netinteger;
reql = bo_getblk(si_oc(si), (char *)&netinteger, sizeof(netinteger), totl);
if (reql <= 0) /* closed or EOL not found */
goto incomplete;
totl += reql;
/* Consider remote is up to date with "acked" version */
ps->update = ntohl(netinteger);
}
else {
/* Unknown message */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* skip consumed message */
bo_skip(si_oc(si), totl);
/* loop on that state to peek next message */
goto switchstate;
incomplete:
/* we get here when a bo_getblk() returns <= 0 in reql */
/* first, we may have to release newts */
if (newts) {
stksess_free(ps->table->table, newts);
newts = NULL;
}
if (reql < 0) {
/* there was an error */
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* Nothing to read, now we start to write */
/* Confirm finished or partial messages */
while (ps->confirm) {
/* There is a confirm messages to send */
repl = bi_putchr(si_ic(si), 'c');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->confirm--;
}
/* Need to request a resync */
if ((ps->flags & PEER_F_LEARN_ASSIGN) &&
(ps->table->flags & SHTABLE_F_RESYNC_ASSIGN) &&
!(ps->table->flags & SHTABLE_F_RESYNC_PROCESS)) {
/* Current peer was elected to request a resync */
repl = bi_putchr(si_ic(si), 'R');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->table->flags |= SHTABLE_F_RESYNC_PROCESS;
}
/* It remains some updates to ack */
if (ps->pushack != ps->lastack) {
uint32_t netinteger;
trash.str[0] = 'A';
netinteger = htonl(ps->pushack);
memcpy(&trash.str[1], &netinteger, sizeof(netinteger));
repl = bi_putblk(si_ic(si), trash.str, 1+sizeof(netinteger));
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->lastack = ps->pushack;
}
if (ps->flags & PEER_F_TEACH_PROCESS) {
/* current peer was requested for a lesson */
if (!(ps->flags & PEER_F_TEACH_STAGE1)) {
/* lesson stage 1 not complete */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
if (!eb) {
/* flag lesson stage1 complete */
ps->flags |= PEER_F_TEACH_STAGE1;
eb = eb32_first(&ps->table->table->updates);
if (eb)
ps->pushed = eb->key - 1;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash.str, trash.size);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si_ic(si), trash.str, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* !TEACH_STAGE1 */
if (!(ps->flags & PEER_F_TEACH_STAGE2)) {
/* lesson stage 2 not complete */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
if (!eb || eb->key > ps->teaching_origin) {
/* flag lesson stage1 complete */
ps->flags |= PEER_F_TEACH_STAGE2;
ps->pushed = ps->teaching_origin;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash.str, trash.size);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si_ic(si), trash.str, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* !TEACH_STAGE2 */
if (!(ps->flags & PEER_F_TEACH_FINISHED)) {
/* process final lesson message */
repl = bi_putchr(si_ic(si), ((ps->table->flags & SHTABLE_RESYNC_STATEMASK) == SHTABLE_RESYNC_FINISHED) ? 'F' : 'C');
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
/* flag finished message sent */
ps->flags |= PEER_F_TEACH_FINISHED;
} /* !TEACH_FINISHED */
} /* TEACH_PROCESS */
if (!(ps->flags & PEER_F_LEARN_ASSIGN) &&
(int)(ps->pushed - ps->table->table->localupdate) < 0) {
/* Push local updates, only if no learning in progress (to avoid ping-pong effects) */
struct eb32_node *eb;
eb = eb32_lookup_ge(&ps->table->table->updates, ps->pushed+1);
while (1) {
int msglen;
struct stksess *ts;
/* push local updates */
if (!eb) {
eb = eb32_first(&ps->table->table->updates);
if (!eb || ((int)(eb->key - ps->pushed) <= 0)) {
ps->pushed = ps->table->table->localupdate;
break;
}
}
if ((int)(eb->key - ps->table->table->localupdate) > 0) {
ps->pushed = ps->table->table->localupdate;
break;
}
ts = eb32_entry(eb, struct stksess, upd);
msglen = peer_prepare_datamsg(ts, ps, trash.str, trash.size);
if (msglen) {
/* message to buffer */
repl = bi_putblk(si_ic(si), trash.str, msglen);
if (repl <= 0) {
/* no more write possible */
if (repl == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
goto switchstate;
}
ps->lastpush = ps->pushed = ts->upd.key;
}
eb = eb32_next(eb);
}
} /* ! LEARN_ASSIGN */
/* noting more to do */
goto out;
}
case PEER_SESS_ST_EXIT:
repl = snprintf(trash.str, trash.size, "%d\n", appctx->st1);
if (bi_putblk(si_ic(si), trash.str, repl) == -1)
goto full;
appctx->st0 = PEER_SESS_ST_END;
/* fall through */
case PEER_SESS_ST_END: {
si_shutw(si);
si_shutr(si);
si_ic(si)->flags |= CF_READ_NULL;
goto out;
}
}
}
out:
si_oc(si)->flags |= CF_READ_DONTWAIT;
return;
full:
si_applet_cant_put(si);
goto out;
}
/*
* Create a new peer session in assigned state (connect will start automatically)
*/
static struct stream *peer_session_create(struct peer *peer, struct peer_session *ps)
{
struct listener *l = LIST_NEXT(&peer->peers->peers_fe->conf.listeners, struct listener *, by_fe);
struct proxy *p = (struct proxy *)l->frontend; /* attached frontend */
struct appctx *appctx;
struct session *sess;
struct stream *s;
struct task *t;
struct connection *conn;
ps->reconnect = tick_add(now_ms, MS_TO_TICKS(5000));
ps->statuscode = PEER_SESS_SC_CONNECTCODE;
s = NULL;
appctx = appctx_new(&peer_applet);
if (!appctx)
goto out_close;
appctx->st0 = PEER_SESS_ST_CONNECT;
appctx->ctx.peers.ptr = (void *)ps;
sess = session_new(p, l, &appctx->obj_type);
if (!sess) {
Alert("out of memory in peer_session_create().\n");
goto out_free_appctx;
}
if ((t = task_new()) == NULL) {
Alert("out of memory in peer_session_create().\n");
goto out_free_sess;
}
t->nice = l->nice;
if ((s = stream_new(sess, t, &appctx->obj_type)) == NULL) {
Alert("Failed to initialize stream in peer_session_create().\n");
goto out_free_task;
}
/* The tasks below are normally what is supposed to be done by
* fe->accept().
*/
s->flags = SF_ASSIGNED|SF_ADDR_SET;
/* applet is waiting for data */
si_applet_cant_get(&s->si[0]);
appctx_wakeup(appctx);
/* initiate an outgoing connection */
si_set_state(&s->si[1], SI_ST_ASS);
/* automatically prepare the stream interface to connect to the
* pre-initialized connection in si->conn.
*/
if (unlikely((conn = conn_new()) == NULL))
goto out_free_strm;
conn_prepare(conn, peer->proto, peer->xprt);
si_attach_conn(&s->si[1], conn);
conn->target = s->target = &s->be->obj_type;
memcpy(&conn->addr.to, &peer->addr, sizeof(conn->addr.to));
s->do_log = NULL;
s->uniq_id = 0;
s->res.flags |= CF_READ_DONTWAIT;
l->nbconn++; /* warning! right now, it's up to the handler to decrease this */
p->feconn++;/* beconn will be increased later */
jobs++;
if (!(s->sess->listener->options & LI_O_UNLIMITED))
actconn++;
totalconn++;
ps->appctx = appctx;
ps->stream = s;
return s;
/* Error unrolling */
out_free_strm:
LIST_DEL(&s->list);
pool_free2(pool2_stream, s);
out_free_task:
task_free(t);
out_free_sess:
session_free(sess);
out_free_appctx:
appctx_free(appctx);
out_close:
return s;
}
/*
* This is the proxy management task. It enables proxies when there are enough
* free sessions, or stops them when the table is full. It is designed to be
* called as a task which is woken up upon stopping or when rate limiting must
* be enforced.
*/
struct task *manage_proxy(struct task *t)
{
struct proxy *p = t->context;
int next = TICK_ETERNITY;
unsigned int wait;
/* We should periodically try to enable listeners waiting for a
* global resource here.
*/
/* first, let's check if we need to stop the proxy */
if (unlikely(stopping && p->state != PR_STSTOPPED)) {
int t;
t = tick_remain(now_ms, p->stop_time);
if (t == 0) {
Warning("Proxy %s stopped (FE: %lld conns, BE: %lld conns).\n",
p->id, p->fe_counters.cum_conn, p->be_counters.cum_conn);
send_log(p, LOG_WARNING, "Proxy %s stopped (FE: %lld conns, BE: %lld conns).\n",
p->id, p->fe_counters.cum_conn, p->be_counters.cum_conn);
stop_proxy(p);
/* try to free more memory */
pool_gc2();
}
else {
next = tick_first(next, p->stop_time);
}
}
/* the rest below is just for frontends */
if (!(p->cap & PR_CAP_FE))
goto out;
/* check the various reasons we may find to block the frontend */
if (unlikely(p->feconn >= p->maxconn)) {
if (p->state == PR_STREADY)
p->state = PR_STFULL;
goto out;
}
/* OK we have no reason to block, so let's unblock if we were blocking */
if (p->state == PR_STFULL)
p->state = PR_STREADY;
if (p->fe_sps_lim &&
(wait = next_event_delay(&p->fe_sess_per_sec, p->fe_sps_lim, 0))) {
/* we're blocking because a limit was reached on the number of
* requests/s on the frontend. We want to re-check ASAP, which
* means in 1 ms before estimated expiration date, because the
* timer will have settled down.
*/
next = tick_first(next, tick_add(now_ms, wait));
goto out;
}
/* The proxy is not limited so we can re-enable any waiting listener */
if (!LIST_ISEMPTY(&p->listener_queue))
dequeue_all_listeners(&p->listener_queue);
out:
t->expire = next;
task_queue(t);
return t;
}