/* return next server from the current tree, or a server from the init tree
* if appropriate. If both trees are empty, return NULL.
*/
struct srv *get_next_server() {
struct eb32_node *node;
struct srv *s;
node = eb32_first(&tree_0);
s = eb32_entry(node, struct srv, node);
if (!node || s->next > p) {
/* either we have no server left, or we have a hole */
struct eb32_node *node2;
node2 = eb32_first(init_tree);
if (node2) {
node = node2;
s = eb32_entry(node, struct srv, node);
get_srv_init(s);
if (s->w == 0)
node = NULL;
s->node.key = 0; // do not display random values
}
}
if (node)
return s;
else
return NULL;
}
/* return next server from the current tree in FWRR group <grp>, or a server
* from the "init" tree if appropriate. If both trees are empty, return NULL.
*/
static struct server *fwrr_get_server_from_group(struct fwrr_group *grp)
{
struct eb32_node *node;
struct server *s;
node = eb32_first(&grp->curr);
s = eb32_entry(node, struct server, lb_node);
if (!node || s->npos > grp->curr_pos) {
/* either we have no server left, or we have a hole */
struct eb32_node *node2;
node2 = eb32_first(grp->init);
if (node2) {
node = node2;
s = eb32_entry(node, struct server, lb_node);
fwrr_get_srv_init(s);
if (s->eweight == 0) /* FIXME: is it possible at all ? */
node = NULL;
}
}
if (node)
return s;
else
return NULL;
}
/* Return next server from the FS tree in backend <p>. If the tree is empty,
* return NULL. Saturated servers are skipped.
*/
struct server *fas_get_next_server(struct proxy *p, struct server *srvtoavoid)
{
struct server *srv, *avoided;
struct eb32_node *node;
srv = avoided = NULL;
if (p->srv_act)
node = eb32_first(&p->lbprm.fas.act);
else if (p->lbprm.fbck)
return p->lbprm.fbck;
else if (p->srv_bck)
node = eb32_first(&p->lbprm.fas.bck);
else
return NULL;
while (node) {
/* OK, we have a server. However, it may be saturated, in which
* case we don't want to reconsider it for now, so we'll simply
* skip it. Same if it's the server we try to avoid, in which
* case we simply remember it for later use if needed.
*/
struct server *s;
s = eb32_entry(node, struct server, lb_node);
if (!s->maxconn || (!s->nbpend && s->served < srv_dynamic_maxconn(s))) {
if (s != srvtoavoid) {
srv = s;
break;
}
avoided = s;
}
node = eb32_next(node);
}
if (!srv)
srv = avoided;
return srv;
}
/* Retrieve the first pendconn from tree <pendconns>. Classes are always
* considered first, then the time offset. The time does wrap, so the
* lookup is performed twice, one to retrieve the first class and a second
* time to retrieve the earliest time in this class.
*/
static struct pendconn *pendconn_first(struct eb_root *pendconns)
{
struct eb32_node *node, *node2 = NULL;
u32 key;
node = eb32_first(pendconns);
if (!node)
return NULL;
key = KEY_CLASS_OFFSET_BOUNDARY(node->key);
node2 = eb32_lookup_ge(pendconns, key);
if (!node2 ||
KEY_CLASS(node2->key) != KEY_CLASS(node->key)) {
/* no other key in the tree, or in this class */
return eb32_entry(node, struct pendconn, node);
}
/* found a better key */
return eb32_entry(node2, struct pendconn, node);
}
/* Redistribute pending connections when a server goes down. The number of
* connections redistributed is returned. It must be called with the server
* lock held.
*/
int pendconn_redistribute(struct server *s)
{
struct pendconn *p;
struct eb32_node *node;
int xferred = 0;
/* The REDISP option was specified. We will ignore cookie and force to
* balance or use the dispatcher. */
if ((s->proxy->options & (PR_O_REDISP|PR_O_PERSIST)) != PR_O_REDISP)
return 0;
for (node = eb32_first(&s->pendconns); node; node = eb32_next(node)) {
p = eb32_entry(&node, struct pendconn, node);
if (p->strm_flags & SF_FORCE_PRST)
continue;
/* it's left to the dispatcher to choose a server */
__pendconn_unlink(p);
p->strm_flags &= ~(SF_DIRECT | SF_ASSIGNED | SF_ADDR_SET);
task_wakeup(p->strm->task, TASK_WOKEN_RES);
}
return xferred;
}
/*
* 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;
}
int main(int argc, char **argv)
{
const char *b, *e, *p;
const char *output_file = NULL;
int f, tot, last, linenum, err, parse_err;
struct timer *t = NULL, *t2;
struct eb32_node *n;
int val, test;
int array[5];
int filter_acc_delay = 0, filter_acc_count = 0;
int filter_time_resp = 0;
int skip_fields = 1;
argc--; argv++;
while (argc > 0) {
if (*argv[0] != '-')
break;
if (strcmp(argv[0], "-ad") == 0) {
if (argc < 2) die("missing option for -ad");
argc--; argv++;
filter |= FILT_ACC_DELAY;
filter_acc_delay = atol(*argv);
}
else if (strcmp(argv[0], "-ac") == 0) {
if (argc < 2) die("missing option for -ac");
argc--; argv++;
filter |= FILT_ACC_COUNT;
filter_acc_count = atol(*argv);
}
else if (strcmp(argv[0], "-rt") == 0) {
if (argc < 2) die("missing option for -rt");
argc--; argv++;
filter |= FILT_TIME_RESP;
filter_time_resp = atol(*argv);
}
else if (strcmp(argv[0], "-RT") == 0) {
if (argc < 2) die("missing option for -RT");
argc--; argv++;
filter |= FILT_TIME_RESP | FILT_INVERT_TIME_RESP;
filter_time_resp = atol(*argv);
}
else if (strcmp(argv[0], "-s") == 0) {
if (argc < 2) die("missing option for -s");
argc--; argv++;
skip_fields = atol(*argv);
}
else if (strcmp(argv[0], "-e") == 0)
filter |= FILT_ERRORS_ONLY;
else if (strcmp(argv[0], "-E") == 0)
filter |= FILT_ERRORS_ONLY | FILT_INVERT_ERRORS;
else if (strcmp(argv[0], "-c") == 0)
filter |= FILT_COUNT_ONLY;
else if (strcmp(argv[0], "-q") == 0)
filter |= FILT_QUIET;
else if (strcmp(argv[0], "-v") == 0)
filter_invert = !filter_invert;
else if (strcmp(argv[0], "-gt") == 0)
filter |= FILT_GRAPH_TIMERS;
else if (strcmp(argv[0], "-pct") == 0)
filter |= FILT_PERCENTILE;
else if (strcmp(argv[0], "-o") == 0) {
if (output_file)
die("Fatal: output file name already specified.\n");
if (argc < 2)
die("Fatal: missing output file name.\n");
output_file = argv[1];
}
argc--;
argv++;
}
if (!filter)
die("No action specified.\n");
if (filter & FILT_ACC_COUNT && !filter_acc_count)
filter_acc_count=1;
if (filter & FILT_ACC_DELAY && !filter_acc_delay)
filter_acc_delay = 1;
linenum = 0;
tot = 0;
parse_err = 0;
while ((line = fgets2(stdin)) != NULL) {
linenum++;
test = 1;
if (filter & FILT_TIME_RESP) {
int tps;
/* only report lines with response times larger than filter_time_resp */
b = field_start(line, TIME_FIELD + skip_fields);
if (!*b) {
truncated_line(linenum, line);
continue;
}
e = field_stop(b + 1);
/* we have field TIME_FIELD in [b]..[e-1] */
p = b;
err = 0;
for (f = 0; f < 4 && *p; f++) {
tps = str2ic(p);
if (tps < 0) {
tps = -1;
err = 1;
}
SKIP_CHAR(p, '/');
}
if (f < 4) {
parse_err++;
continue;
}
test &= (tps >= filter_time_resp) ^ !!(filter & FILT_INVERT_TIME_RESP);
}
if (filter & FILT_ERRORS_ONLY) {
/* only report erroneous status codes */
b = field_start(line, STATUS_FIELD + skip_fields);
if (!*b) {
truncated_line(linenum, line);
continue;
}
if (*b == '-') {
test &= !!(filter & FILT_INVERT_ERRORS);
} else {
val = strl2ui(b, 3);
test &= (val >= 500 && val <= 599) ^ !!(filter & FILT_INVERT_ERRORS);
}
}
if (filter & (FILT_ACC_COUNT|FILT_ACC_DELAY)) {
b = field_start(line, ACCEPT_FIELD + skip_fields);
if (!*b) {
truncated_line(linenum, line);
continue;
}
tot++;
val = convert_date(b);
//printf("date=%s => %d\n", b, val);
if (val < 0) {
parse_err++;
continue;
}
t2 = insert_value(&timers[0], &t, val);
t2->count++;
continue;
}
if (filter & (FILT_GRAPH_TIMERS|FILT_PERCENTILE)) {
int f;
b = field_start(line, TIME_FIELD + skip_fields);
if (!*b) {
truncated_line(linenum, line);
continue;
}
e = field_stop(b + 1);
/* we have field TIME_FIELD in [b]..[e-1] */
p = b;
err = 0;
for (f = 0; f < 5 && *p; f++) {
array[f] = str2ic(p);
if (array[f] < 0) {
array[f] = -1;
err = 1;
}
SKIP_CHAR(p, '/');
}
if (f < 5) {
parse_err++;
continue;
}
/* if we find at least one negative time, we count one error
* with a time equal to the total session time. This will
* emphasize quantum timing effects associated to known
* timeouts. Note that on some buggy machines, it is possible
* that the total time is negative, hence the reason to reset
* it.
*/
if (filter & FILT_GRAPH_TIMERS) {
if (err) {
if (array[4] < 0)
array[4] = -1;
t2 = insert_timer(&timers[0], &t, array[4]); // total time
t2->count++;
} else {
int v;
t2 = insert_timer(&timers[1], &t, array[0]); t2->count++; // req
t2 = insert_timer(&timers[2], &t, array[2]); t2->count++; // conn
t2 = insert_timer(&timers[3], &t, array[3]); t2->count++; // resp
v = array[4] - array[0] - array[1] - array[2] - array[3]; // data time
if (v < 0 && !(filter & FILT_QUIET))
fprintf(stderr, "ERR: %s (%d %d %d %d %d => %d)\n",
line, array[0], array[1], array[2], array[3], array[4], v);
t2 = insert_timer(&timers[4], &t, v); t2->count++;
tot++;
}
} else { /* percentile */
if (err) {
if (array[4] < 0)
array[4] = -1;
t2 = insert_value(&timers[0], &t, array[4]); // total time
t2->count++;
} else {
int v;
t2 = insert_value(&timers[1], &t, array[0]); t2->count++; // req
t2 = insert_value(&timers[2], &t, array[2]); t2->count++; // conn
t2 = insert_value(&timers[3], &t, array[3]); t2->count++; // resp
v = array[4] - array[0] - array[1] - array[2] - array[3]; // data time
if (v < 0 && !(filter & FILT_QUIET))
fprintf(stderr, "ERR: %s (%d %d %d %d %d => %d)\n",
line, array[0], array[1], array[2], array[3], array[4], v);
t2 = insert_value(&timers[4], &t, v); t2->count++;
tot++;
}
}
continue;
}
test ^= filter_invert;
if (!test)
continue;
/* all other cases mean we just want to count lines */
tot++;
if (!(filter & FILT_COUNT_ONLY))
puts(line);
}
if (t)
free(t);
if (filter & FILT_COUNT_ONLY) {
printf("%d\n", tot);
exit(0);
}
if (filter & FILT_ERRORS_ONLY)
exit(0);
if (filter & (FILT_ACC_COUNT|FILT_ACC_DELAY)) {
/* sort and count all timers. Output will look like this :
* <accept_date> <delta_ms from previous one> <nb entries>
*/
n = eb32_first(&timers[0]);
if (n)
last = n->key;
while (n) {
unsigned int d, h, m, s, ms;
t = container_of(n, struct timer, node);
h = n->key;
d = h - last;
last = h;
if (d >= filter_acc_delay && t->count >= filter_acc_count) {
ms = h % 1000; h = h / 1000;
s = h % 60; h = h / 60;
m = h % 60; h = h / 60;
tot++;
printf("%02d:%02d:%02d.%03d %d %d %d\n", h, m, s, ms, last, d, t->count);
}
n = eb32_next(n);
}
}
/*
* 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;
}