/* Marks a ready listener as limited so that we only try to re-enable it when
* resources are free again. It will be queued into the specified queue.
*/
void limit_listener(struct listener *l, struct list *list)
{
if (l->state == LI_READY) {
LIST_ADDQ(list, &l->wait_queue);
EV_FD_CLR(l->fd, DIR_RD);
l->state = LI_LIMITED;
}
}
/* This function removes the specified listener's file descriptor from the
* polling lists if it is in the LI_READY or in the LI_FULL state. The listener
* enters LI_LISTEN.
*/
void disable_listener(struct listener *listener)
{
if (listener->state < LI_READY)
return;
if (listener->state == LI_READY)
EV_FD_CLR(listener->fd, DIR_RD);
if (listener->state == LI_LIMITED)
LIST_DEL(&listener->wait_queue);
listener->state = LI_LISTEN;
}
/* Marks a ready listener as full so that the session code tries to re-enable
* it upon next close() using resume_listener().
*/
void listener_full(struct listener *l)
{
if (l->state >= LI_READY) {
if (l->state == LI_LIMITED)
LIST_DEL(&l->wait_queue);
EV_FD_CLR(l->fd, DIR_RD);
l->state = LI_FULL;
}
}
/*
* This function performs a shutdown-write on a stream interface in a connected or
* init state (it does nothing for other states). It either shuts the write side
* or closes the file descriptor and marks itself as closed. The buffer flags are
* updated to reflect the new state. It does also close everything is the SI was
* marked as being in error state.
*/
void stream_sock_shutw(struct stream_interface *si)
{
si->ob->flags &= ~BF_SHUTW_NOW;
if (si->ob->flags & BF_SHUTW)
return;
si->ob->flags |= BF_SHUTW;
si->ob->wex = TICK_ETERNITY;
si->flags &= ~SI_FL_WAIT_DATA;
switch (si->state) {
case SI_ST_EST:
/* we have to shut before closing, otherwise some short messages
* may never leave the system, especially when there are remaining
* unread data in the socket input buffer, or when nolinger is set.
* However, if SI_FL_NOLINGER is explicitly set, we know there is
* no risk so we close both sides immediately.
*/
if (si->flags & SI_FL_ERR) {
/* quick close, the socket is already shut. Remove pending flags. */
si->flags &= ~SI_FL_NOLINGER;
} else if (si->flags & SI_FL_NOLINGER) {
si->flags &= ~SI_FL_NOLINGER;
setsockopt(si->fd, SOL_SOCKET, SO_LINGER,
(struct linger *) &nolinger, sizeof(struct linger));
} else {
EV_FD_CLR(si->fd, DIR_WR);
shutdown(si->fd, SHUT_WR);
if (!(si->ib->flags & (BF_SHUTR|BF_DONT_READ)))
return;
}
/* fall through */
case SI_ST_CON:
/* we may have to close a pending connection, and mark the
* response buffer as shutr
*/
fd_delete(si->fd);
/* fall through */
case SI_ST_CER:
case SI_ST_QUE:
case SI_ST_TAR:
si->state = SI_ST_DIS;
default:
si->flags &= ~SI_FL_WAIT_ROOM;
si->ib->flags |= BF_SHUTR;
si->ib->rex = TICK_ETERNITY;
si->exp = TICK_ETERNITY;
return;
}
if (si->release)
si->release(si);
}
/*
* Linux unbinds the listen socket after a SHUT_RD, and ignores SHUT_WR.
* Solaris refuses either shutdown().
* OpenBSD ignores SHUT_RD but closes upon SHUT_WR and refuses to rebind.
* So a common validation path involves SHUT_WR && listen && SHUT_RD.
* If disabling at least one listener returns an error, then the proxy
* state is set to PR_STERROR because we don't know how to resume from this.
*/
void pause_proxy(struct proxy *p)
{
struct listener *l;
for (l = p->listen; l != NULL; l = l->next) {
if (shutdown(l->fd, SHUT_WR) == 0 &&
listen(l->fd, p->backlog ? p->backlog : p->maxconn) == 0 &&
shutdown(l->fd, SHUT_RD) == 0) {
EV_FD_CLR(l->fd, DIR_RD);
if (p->state != PR_STERROR)
p->state = PR_STPAUSED;
}
else
p->state = PR_STERROR;
}
}
/* This function tries to temporarily disable a listener, depending on the OS
* capabilities. Linux unbinds the listen socket after a SHUT_RD, and ignores
* SHUT_WR. Solaris refuses either shutdown(). OpenBSD ignores SHUT_RD but
* closes upon SHUT_WR and refuses to rebind. So a common validation path
* involves SHUT_WR && listen && SHUT_RD. In case of success, the FD's polling
* is disabled. It normally returns non-zero, unless an error is reported.
*/
int pause_listener(struct listener *l)
{
if (l->state <= LI_PAUSED)
return 1;
if (shutdown(l->fd, SHUT_WR) != 0)
return 0; /* Solaris dies here */
if (listen(l->fd, l->backlog ? l->backlog : l->maxconn) != 0)
return 0; /* OpenBSD dies here */
if (shutdown(l->fd, SHUT_RD) != 0)
return 0; /* should always be OK */
if (l->state == LI_LIMITED)
LIST_DEL(&l->wait_queue);
EV_FD_CLR(l->fd, DIR_RD);
l->state = LI_PAUSED;
return 1;
}
/*
* This function performs a shutdown-read on a stream interface in a connected or
* init state (it does nothing for other states). It either shuts the read side
* or closes the file descriptor and marks itself as closed. The buffer flags are
* updated to reflect the new state.
*/
void stream_sock_shutr(struct stream_interface *si)
{
si->ib->flags &= ~BF_SHUTR_NOW;
if (si->ib->flags & BF_SHUTR)
return;
si->ib->flags |= BF_SHUTR;
si->ib->rex = TICK_ETERNITY;
si->flags &= ~SI_FL_WAIT_ROOM;
if (si->state != SI_ST_EST && si->state != SI_ST_CON)
return;
if (si->ob->flags & BF_SHUTW) {
fd_delete(si->fd);
si->state = SI_ST_DIS;
si->exp = TICK_ETERNITY;
return;
}
EV_FD_CLR(si->fd, DIR_RD);
return;
}
/* Returns :
* -1 if splice is not possible or not possible anymore and we must switch to
* user-land copy (eg: to_forward reached)
* 0 when we know that polling is required to get more data (EAGAIN)
* 1 for all other cases (we can safely try again, or if an activity has been
* detected (DATA/NULL/ERR))
* Sets :
* BF_READ_NULL
* BF_READ_PARTIAL
* BF_WRITE_PARTIAL (during copy)
* BF_OUT_EMPTY (during copy)
* SI_FL_ERR
* SI_FL_WAIT_ROOM
* (SI_FL_WAIT_RECV)
*
* This function automatically allocates a pipe from the pipe pool. It also
* carefully ensures to clear b->pipe whenever it leaves the pipe empty.
*/
static int stream_sock_splice_in(struct buffer *b, struct stream_interface *si)
{
static int splice_detects_close;
int fd = si->fd;
int ret;
unsigned long max;
int retval = 1;
if (!b->to_forward)
return -1;
if (!(b->flags & BF_KERN_SPLICING))
return -1;
if (b->l) {
/* We're embarrassed, there are already data pending in
* the buffer and we don't want to have them at two
* locations at a time. Let's indicate we need some
* place and ask the consumer to hurry.
*/
si->flags |= SI_FL_WAIT_ROOM;
EV_FD_CLR(fd, DIR_RD);
b->rex = TICK_ETERNITY;
b->cons->chk_snd(b->cons);
return 1;
}
if (unlikely(b->pipe == NULL)) {
if (pipes_used >= global.maxpipes || !(b->pipe = get_pipe())) {
b->flags &= ~BF_KERN_SPLICING;
return -1;
}
}
/* At this point, b->pipe is valid */
while (1) {
if (b->to_forward == BUF_INFINITE_FORWARD)
max = MAX_SPLICE_AT_ONCE;
else
max = b->to_forward;
if (!max) {
/* It looks like the buffer + the pipe already contain
* the maximum amount of data to be transferred. Try to
* send those data immediately on the other side if it
* is currently waiting.
*/
retval = -1; /* end of forwarding */
break;
}
ret = splice(fd, NULL, b->pipe->prod, NULL, max,
SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
if (ret <= 0) {
if (ret == 0) {
/* connection closed. This is only detected by
* recent kernels (>= 2.6.27.13). If we notice
* it works, we store the info for later use.
*/
splice_detects_close = 1;
b->flags |= BF_READ_NULL;
retval = 1; /* no need for further polling */
break;
}
if (errno == EAGAIN) {
/* there are two reasons for EAGAIN :
* - nothing in the socket buffer (standard)
* - pipe is full
* - the connection is closed (kernel < 2.6.27.13)
* Since we don't know if pipe is full, we'll
* stop if the pipe is not empty. Anyway, we
* will almost always fill/empty the pipe.
*/
if (b->pipe->data) {
si->flags |= SI_FL_WAIT_ROOM;
retval = 1;
break;
}
/* We don't know if the connection was closed,
* but if we know splice detects close, then we
* know it for sure.
* But if we're called upon POLLIN with an empty
* pipe and get EAGAIN, it is suspect enough to
* try to fall back to the normal recv scheme
* which will be able to deal with the situation.
*/
if (splice_detects_close)
retval = 0; /* we know for sure that it's EAGAIN */
else
retval = -1;
break;
}
if (errno == ENOSYS || errno == EINVAL) {
/* splice not supported on this end, disable it */
b->flags &= ~BF_KERN_SPLICING;
si->flags &= ~SI_FL_CAP_SPLICE;
put_pipe(b->pipe);
b->pipe = NULL;
return -1;
}
/* here we have another error */
si->flags |= SI_FL_ERR;
retval = 1;
break;
} /* ret <= 0 */
if (b->to_forward != BUF_INFINITE_FORWARD)
b->to_forward -= ret;
b->total += ret;
b->pipe->data += ret;
b->flags |= BF_READ_PARTIAL;
b->flags &= ~BF_OUT_EMPTY;
if (b->pipe->data >= SPLICE_FULL_HINT ||
ret >= global.tune.recv_enough) {
/* We've read enough of it for this time. */
retval = 1;
break;
}
} /* while */
if (unlikely(!b->pipe->data)) {
put_pipe(b->pipe);
b->pipe = NULL;
}
return retval;
}
/*
* This function is called on a write event from a stream socket.
* It returns 0 if the caller needs to poll before calling it again, otherwise
* non-zero.
*/
int stream_sock_write(int fd)
{
struct stream_interface *si = fdtab[fd].owner;
struct buffer *b = si->ob;
int retval = 1;
#ifdef DEBUG_FULL
fprintf(stderr,"stream_sock_write : fd=%d, owner=%p\n", fd, fdtab[fd].owner);
#endif
retval = 1;
if (fdtab[fd].state == FD_STERROR)
goto out_error;
/* we might have been called just after an asynchronous shutw */
if (b->flags & BF_SHUTW)
goto out_wakeup;
if (likely(!(b->flags & BF_OUT_EMPTY) || si->send_proxy_ofs)) {
/* OK there are data waiting to be sent */
retval = stream_sock_write_loop(si, b);
if (retval < 0)
goto out_error;
else if (retval == 0 && si->send_proxy_ofs)
goto out_may_wakeup; /* we failed to send the PROXY string */
}
else {
/* may be we have received a connection acknowledgement in TCP mode without data */
if (likely(fdtab[fd].state == FD_STCONN)) {
/* We have no data to send to check the connection, and
* getsockopt() will not inform us whether the connection
* is still pending. So we'll reuse connect() to check the
* state of the socket. This has the advantage of givig us
* the following info :
* - error
* - connecting (EALREADY, EINPROGRESS)
* - connected (EISCONN, 0)
*/
if ((connect(fd, fdinfo[fd].peeraddr, fdinfo[fd].peerlen) == 0))
errno = 0;
if (errno == EALREADY || errno == EINPROGRESS) {
retval = 0;
goto out_may_wakeup;
}
if (errno && errno != EISCONN)
goto out_error;
/* OK we just need to indicate that we got a connection
* and that we wrote nothing.
*/
b->flags |= BF_WRITE_NULL;
fdtab[fd].state = FD_STREADY;
}
/* Funny, we were called to write something but there wasn't
* anything. We can get there, for example if we were woken up
* on a write event to finish the splice, but the send_max is 0
* so we cannot write anything from the buffer. Let's disable
* the write event and pretend we never came there.
*/
}
if (b->flags & BF_OUT_EMPTY) {
/* the connection is established but we can't write. Either the
* buffer is empty, or we just refrain from sending because the
* send_max limit was reached. Maybe we just wrote the last
* chunk and need to close.
*/
if (((b->flags & (BF_SHUTW|BF_HIJACK|BF_SHUTW_NOW)) == BF_SHUTW_NOW) &&
(si->state == SI_ST_EST)) {
stream_sock_shutw(si);
goto out_wakeup;
}
if ((b->flags & (BF_SHUTW|BF_SHUTW_NOW|BF_FULL|BF_HIJACK)) == 0)
si->flags |= SI_FL_WAIT_DATA;
EV_FD_CLR(fd, DIR_WR);
b->wex = TICK_ETERNITY;
}
out_may_wakeup:
if (b->flags & BF_WRITE_ACTIVITY) {
/* update timeout if we have written something */
if ((b->flags & (BF_OUT_EMPTY|BF_SHUTW|BF_WRITE_PARTIAL)) == BF_WRITE_PARTIAL)
b->wex = tick_add_ifset(now_ms, b->wto);
out_wakeup:
if (tick_isset(si->ib->rex) && !(si->flags & SI_FL_INDEP_STR)) {
/* Note: to prevent the client from expiring read timeouts
* during writes, we refresh it. We only do this if the
* interface is not configured for "independant streams",
* because for some applications it's better not to do this,
* for instance when continuously exchanging small amounts
* of data which can full the socket buffers long before a
* write timeout is detected.
*/
si->ib->rex = tick_add_ifset(now_ms, si->ib->rto);
}
/* the producer might be waiting for more room to store data */
if (likely((b->flags & (BF_SHUTW|BF_WRITE_PARTIAL|BF_FULL|BF_DONT_READ)) == BF_WRITE_PARTIAL &&
(b->prod->flags & SI_FL_WAIT_ROOM)))
b->prod->chk_rcv(b->prod);
/* we have to wake up if there is a special event or if we don't have
* any more data to forward and it's not planned to send any more.
*/
if (likely((b->flags & (BF_WRITE_NULL|BF_WRITE_ERROR|BF_SHUTW)) ||
((b->flags & BF_OUT_EMPTY) && !b->to_forward) ||
si->state != SI_ST_EST ||
b->prod->state != SI_ST_EST))
task_wakeup(si->owner, TASK_WOKEN_IO);
}
fdtab[fd].ev &= ~FD_POLL_OUT;
return retval;
out_error:
/* Write error on the file descriptor. We mark the FD as STERROR so
* that we don't use it anymore. The error is reported to the stream
* interface which will take proper action. We must not perturbate the
* buffer because the stream interface wants to ensure transparent
* connection retries.
*/
fdtab[fd].state = FD_STERROR;
fdtab[fd].ev &= ~FD_POLL_STICKY;
EV_FD_REM(fd);
si->flags |= SI_FL_ERR;
task_wakeup(si->owner, TASK_WOKEN_IO);
return 1;
}
/*
* this function is called on a read event from a stream socket.
* It returns 0 if we have a high confidence that we will not be
* able to read more data without polling first. Returns non-zero
* otherwise.
*/
int stream_sock_read(int fd) {
struct stream_interface *si = fdtab[fd].owner;
struct buffer *b = si->ib;
int ret, max, retval, cur_read;
int read_poll = MAX_READ_POLL_LOOPS;
#ifdef DEBUG_FULL
fprintf(stderr,"stream_sock_read : fd=%d, ev=0x%02x, owner=%p\n", fd, fdtab[fd].ev, fdtab[fd].owner);
#endif
retval = 1;
/* stop immediately on errors. Note that we DON'T want to stop on
* POLL_ERR, as the poller might report a write error while there
* are still data available in the recv buffer. This typically
* happens when we send too large a request to a backend server
* which rejects it before reading it all.
*/
if (fdtab[fd].state == FD_STERROR)
goto out_error;
/* stop here if we reached the end of data */
if ((fdtab[fd].ev & (FD_POLL_IN|FD_POLL_HUP)) == FD_POLL_HUP)
goto out_shutdown_r;
/* maybe we were called immediately after an asynchronous shutr */
if (b->flags & BF_SHUTR)
goto out_wakeup;
#if defined(CONFIG_HAP_LINUX_SPLICE)
if (b->to_forward >= MIN_SPLICE_FORWARD && b->flags & BF_KERN_SPLICING) {
/* Under Linux, if FD_POLL_HUP is set, we have reached the end.
* Since older splice() implementations were buggy and returned
* EAGAIN on end of read, let's bypass the call to splice() now.
*/
if (fdtab[fd].ev & FD_POLL_HUP)
goto out_shutdown_r;
retval = stream_sock_splice_in(b, si);
if (retval >= 0) {
if (si->flags & SI_FL_ERR)
goto out_error;
if (b->flags & BF_READ_NULL)
goto out_shutdown_r;
goto out_wakeup;
}
/* splice not possible (anymore), let's go on on standard copy */
}
#endif
cur_read = 0;
while (1) {
max = buffer_max_len(b) - b->l;
if (max <= 0) {
b->flags |= BF_FULL;
si->flags |= SI_FL_WAIT_ROOM;
break;
}
/*
* 1. compute the maximum block size we can read at once.
*/
if (b->l == 0) {
/* let's realign the buffer to optimize I/O */
b->r = b->w = b->lr = b->data;
}
else if (b->r > b->w) {
/* remaining space wraps at the end, with a moving limit */
if (max > b->data + b->size - b->r)
max = b->data + b->size - b->r;
}
/* else max is already OK */
/*
* 2. read the largest possible block
*/
ret = recv(fd, b->r, max, 0);
if (ret > 0) {
b->r += ret;
b->l += ret;
cur_read += ret;
/* if we're allowed to directly forward data, we must update send_max */
if (b->to_forward && !(b->flags & (BF_SHUTW|BF_SHUTW_NOW))) {
unsigned long fwd = ret;
if (b->to_forward != BUF_INFINITE_FORWARD) {
if (fwd > b->to_forward)
fwd = b->to_forward;
b->to_forward -= fwd;
}
b->send_max += fwd;
b->flags &= ~BF_OUT_EMPTY;
}
if (fdtab[fd].state == FD_STCONN)
fdtab[fd].state = FD_STREADY;
b->flags |= BF_READ_PARTIAL;
if (b->r == b->data + b->size) {
b->r = b->data; /* wrap around the buffer */
}
b->total += ret;
if (b->l >= buffer_max_len(b)) {
/* The buffer is now full, there's no point in going through
* the loop again.
*/
if (!(b->flags & BF_STREAMER_FAST) && (cur_read == b->l)) {
b->xfer_small = 0;
b->xfer_large++;
if (b->xfer_large >= 3) {
/* we call this buffer a fast streamer if it manages
* to be filled in one call 3 consecutive times.
*/
b->flags |= (BF_STREAMER | BF_STREAMER_FAST);
//fputc('+', stderr);
}
}
else if ((b->flags & (BF_STREAMER | BF_STREAMER_FAST)) &&
(cur_read <= b->size / 2)) {
b->xfer_large = 0;
b->xfer_small++;
if (b->xfer_small >= 2) {
/* if the buffer has been at least half full twice,
* we receive faster than we send, so at least it
* is not a "fast streamer".
*/
b->flags &= ~BF_STREAMER_FAST;
//fputc('-', stderr);
}
}
else {
b->xfer_small = 0;
b->xfer_large = 0;
}
b->flags |= BF_FULL;
si->flags |= SI_FL_WAIT_ROOM;
break;
}
/* if too many bytes were missing from last read, it means that
* it's pointless trying to read again because the system does
* not have them in buffers. BTW, if FD_POLL_HUP was present,
* it means that we have reached the end and that the connection
* is closed.
*/
if (ret < max) {
if ((b->flags & (BF_STREAMER | BF_STREAMER_FAST)) &&
(cur_read <= b->size / 2)) {
b->xfer_large = 0;
b->xfer_small++;
if (b->xfer_small >= 3) {
/* we have read less than half of the buffer in
* one pass, and this happened at least 3 times.
* This is definitely not a streamer.
*/
b->flags &= ~(BF_STREAMER | BF_STREAMER_FAST);
//fputc('!', stderr);
}
}
/* unfortunately, on level-triggered events, POLL_HUP
* is generally delivered AFTER the system buffer is
* empty, so this one might never match.
*/
if (fdtab[fd].ev & FD_POLL_HUP)
goto out_shutdown_r;
/* if a streamer has read few data, it may be because we
* have exhausted system buffers. It's not worth trying
* again.
*/
if (b->flags & BF_STREAMER)
break;
/* generally if we read something smaller than 1 or 2 MSS,
* it means that either we have exhausted the system's
* buffers (streamer or question-response protocol) or
* that the connection will be closed. Streamers are
* easily detected so we return early. For other cases,
* it's still better to perform a last read to be sure,
* because it may save one complete poll/read/wakeup cycle
* in case of shutdown.
*/
if (ret < MIN_RET_FOR_READ_LOOP && b->flags & BF_STREAMER)
break;
/* if we read a large block smaller than what we requested,
* it's almost certain we'll never get anything more.
*/
if (ret >= global.tune.recv_enough)
break;
}
if ((b->flags & BF_READ_DONTWAIT) || --read_poll <= 0)
break;
}
else if (ret == 0) {
/* connection closed */
goto out_shutdown_r;
}
else if (errno == EAGAIN) {
/* Ignore EAGAIN but inform the poller that there is
* nothing to read left if we did not read much, ie
* less than what we were still expecting to read.
* But we may have done some work justifying to notify
* the task.
*/
if (cur_read < MIN_RET_FOR_READ_LOOP)
retval = 0;
break;
}
else {
goto out_error;
}
} /* while (1) */
out_wakeup:
/* We might have some data the consumer is waiting for.
* We can do fast-forwarding, but we avoid doing this for partial
* buffers, because it is very likely that it will be done again
* immediately afterwards once the following data is parsed (eg:
* HTTP chunking).
*/
if (b->pipe || /* always try to send spliced data */
(b->send_max == b->l && (b->cons->flags & SI_FL_WAIT_DATA))) {
int last_len = b->pipe ? b->pipe->data : 0;
b->cons->chk_snd(b->cons);
/* check if the consumer has freed some space */
if (!(b->flags & BF_FULL) &&
(!last_len || !b->pipe || b->pipe->data < last_len))
si->flags &= ~SI_FL_WAIT_ROOM;
}
if (si->flags & SI_FL_WAIT_ROOM) {
EV_FD_CLR(fd, DIR_RD);
b->rex = TICK_ETERNITY;
}
else if ((b->flags & (BF_SHUTR|BF_READ_PARTIAL|BF_FULL|BF_DONT_READ|BF_READ_NOEXP)) == BF_READ_PARTIAL)
b->rex = tick_add_ifset(now_ms, b->rto);
/* we have to wake up if there is a special event or if we don't have
* any more data to forward.
*/
if ((b->flags & (BF_READ_NULL|BF_READ_ERROR)) ||
si->state != SI_ST_EST ||
(si->flags & SI_FL_ERR) ||
((b->flags & BF_READ_PARTIAL) && (!b->to_forward || b->cons->state != SI_ST_EST)))
task_wakeup(si->owner, TASK_WOKEN_IO);
if (b->flags & BF_READ_ACTIVITY)
b->flags &= ~BF_READ_DONTWAIT;
fdtab[fd].ev &= ~FD_POLL_IN;
return retval;
out_shutdown_r:
/* we received a shutdown */
fdtab[fd].ev &= ~FD_POLL_HUP;
b->flags |= BF_READ_NULL;
if (b->flags & BF_AUTO_CLOSE)
buffer_shutw_now(b);
stream_sock_shutr(si);
goto out_wakeup;
out_error:
/* Read error on the file descriptor. We mark the FD as STERROR so
* that we don't use it anymore. The error is reported to the stream
* interface which will take proper action. We must not perturbate the
* buffer because the stream interface wants to ensure transparent
* connection retries.
*/
fdtab[fd].state = FD_STERROR;
fdtab[fd].ev &= ~FD_POLL_STICKY;
EV_FD_REM(fd);
si->flags |= SI_FL_ERR;
retval = 1;
goto out_wakeup;
}
/* Finish a session accept() for a proxy (TCP or HTTP). It returns a negative
* value in case of a critical failure which must cause the listener to be
* disabled, a positive value in case of success, or zero if it is a success
* but the session must be closed ASAP (eg: monitoring).
*/
int frontend_accept(struct session *s)
{
int cfd = s->si[0].fd;
tv_zero(&s->logs.tv_request);
s->logs.t_queue = -1;
s->logs.t_connect = -1;
s->logs.t_data = -1;
s->logs.t_close = 0;
s->logs.bytes_in = s->logs.bytes_out = 0;
s->logs.prx_queue_size = 0; /* we get the number of pending conns before us */
s->logs.srv_queue_size = 0; /* we will get this number soon */
/* FIXME: the logs are horribly complicated now, because they are
* defined in <p>, <p>, and later <be> and <be>.
*/
s->do_log = sess_log;
/* default error reporting function, may be changed by analysers */
s->srv_error = default_srv_error;
/* Adjust some socket options */
if (s->listener->addr.ss_family == AF_INET || s->listener->addr.ss_family == AF_INET6) {
if (setsockopt(cfd, IPPROTO_TCP, TCP_NODELAY,
(char *) &one, sizeof(one)) == -1)
goto out_return;
if (s->fe->options & PR_O_TCP_CLI_KA)
setsockopt(cfd, SOL_SOCKET, SO_KEEPALIVE,
(char *) &one, sizeof(one));
if (s->fe->options & PR_O_TCP_NOLING)
setsockopt(cfd, SOL_SOCKET, SO_LINGER,
(struct linger *) &nolinger, sizeof(struct linger));
#if defined(TCP_MAXSEG)
if (s->listener->maxseg < 0) {
/* we just want to reduce the current MSS by that value */
int mss;
socklen_t mss_len = sizeof(mss);
if (getsockopt(cfd, IPPROTO_TCP, TCP_MAXSEG, &mss, &mss_len) == 0) {
mss += s->listener->maxseg; /* remember, it's < 0 */
setsockopt(cfd, IPPROTO_TCP, TCP_MAXSEG, &mss, sizeof(mss));
}
}
#endif
}
if (global.tune.client_sndbuf)
setsockopt(cfd, SOL_SOCKET, SO_SNDBUF, &global.tune.client_sndbuf, sizeof(global.tune.client_sndbuf));
if (global.tune.client_rcvbuf)
setsockopt(cfd, SOL_SOCKET, SO_RCVBUF, &global.tune.client_rcvbuf, sizeof(global.tune.client_rcvbuf));
if (s->fe->mode == PR_MODE_HTTP) {
/* the captures are only used in HTTP frontends */
if (unlikely(s->fe->nb_req_cap > 0 &&
(s->txn.req.cap = pool_alloc2(s->fe->req_cap_pool)) == NULL))
goto out_return; /* no memory */
if (unlikely(s->fe->nb_rsp_cap > 0 &&
(s->txn.rsp.cap = pool_alloc2(s->fe->rsp_cap_pool)) == NULL))
goto out_free_reqcap; /* no memory */
}
if (s->fe->acl_requires & ACL_USE_L7_ANY) {
/* we have to allocate header indexes only if we know
* that we may make use of them. This of course includes
* (mode == PR_MODE_HTTP).
*/
s->txn.hdr_idx.size = global.tune.max_http_hdr;
if (unlikely((s->txn.hdr_idx.v = pool_alloc2(pool2_hdr_idx)) == NULL))
goto out_free_rspcap; /* no memory */
/* and now initialize the HTTP transaction state */
http_init_txn(s);
}
if ((s->fe->mode == PR_MODE_TCP || s->fe->mode == PR_MODE_HTTP)
&& (!LIST_ISEMPTY(&s->fe->logsrvs))) {
if (likely(s->fe->to_log)) {
/* we have the client ip */
if (s->logs.logwait & LW_CLIP)
if (!(s->logs.logwait &= ~LW_CLIP))
s->do_log(s);
}
else {
char pn[INET6_ADDRSTRLEN], sn[INET6_ADDRSTRLEN];
if (!(s->flags & SN_FRT_ADDR_SET))
get_frt_addr(s);
switch (addr_to_str(&s->req->prod->addr.from, pn, sizeof(pn))) {
case AF_INET:
case AF_INET6:
addr_to_str(&s->req->prod->addr.to, sn, sizeof(sn));
send_log(s->fe, LOG_INFO, "Connect from %s:%d to %s:%d (%s/%s)\n",
pn, get_host_port(&s->req->prod->addr.from),
sn, get_host_port(&s->req->prod->addr.to),
s->fe->id, (s->fe->mode == PR_MODE_HTTP) ? "HTTP" : "TCP");
break;
case AF_UNIX:
/* UNIX socket, only the destination is known */
send_log(s->fe, LOG_INFO, "Connect to unix:%d (%s/%s)\n",
s->listener->luid,
s->fe->id, (s->fe->mode == PR_MODE_HTTP) ? "HTTP" : "TCP");
break;
}
}
}
if (unlikely((global.mode & MODE_DEBUG) && (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)))) {
char pn[INET6_ADDRSTRLEN];
int len = 0;
if (!(s->flags & SN_FRT_ADDR_SET))
get_frt_addr(s);
switch (addr_to_str(&s->req->prod->addr.from, pn, sizeof(pn))) {
case AF_INET:
case AF_INET6:
len = sprintf(trash, "%08x:%s.accept(%04x)=%04x from [%s:%d]\n",
s->uniq_id, s->fe->id, (unsigned short)s->listener->fd, (unsigned short)cfd,
pn, get_host_port(&s->req->prod->addr.from));
break;
case AF_UNIX:
/* UNIX socket, only the destination is known */
len = sprintf(trash, "%08x:%s.accept(%04x)=%04x from [unix:%d]\n",
s->uniq_id, s->fe->id, (unsigned short)s->listener->fd, (unsigned short)cfd,
s->listener->luid);
break;
}
write(1, trash, len);
}
if (s->fe->mode == PR_MODE_HTTP)
s->req->flags |= BF_READ_DONTWAIT; /* one read is usually enough */
/* note: this should not happen anymore since there's always at least the switching rules */
if (!s->req->analysers) {
buffer_auto_connect(s->req); /* don't wait to establish connection */
buffer_auto_close(s->req); /* let the producer forward close requests */
}
s->req->rto = s->fe->timeout.client;
s->rep->wto = s->fe->timeout.client;
fdtab[cfd].flags = FD_FL_TCP | FD_FL_TCP_NODELAY;
if (s->fe->options & PR_O_TCP_NOLING)
fdtab[cfd].flags |= FD_FL_TCP_NOLING;
if (unlikely((s->fe->mode == PR_MODE_HTTP && (s->flags & SN_MONITOR)) ||
(s->fe->mode == PR_MODE_HEALTH && ((s->fe->options2 & PR_O2_CHK_ANY) == PR_O2_HTTP_CHK)))) {
/* Either we got a request from a monitoring system on an HTTP instance,
* or we're in health check mode with the 'httpchk' option enabled. In
* both cases, we return a fake "HTTP/1.0 200 OK" response and we exit.
*/
struct chunk msg;
chunk_initstr(&msg, "HTTP/1.0 200 OK\r\n\r\n");
stream_int_retnclose(&s->si[0], &msg); /* forge a 200 response */
s->req->analysers = 0;
s->task->expire = s->rep->wex;
EV_FD_CLR(cfd, DIR_RD);
}
else if (unlikely(s->fe->mode == PR_MODE_HEALTH)) { /* health check mode, no client reading */
struct chunk msg;
chunk_initstr(&msg, "OK\n");
stream_int_retnclose(&s->si[0], &msg); /* forge an "OK" response */
s->req->analysers = 0;
s->task->expire = s->rep->wex;
EV_FD_CLR(cfd, DIR_RD);
}
/* everything's OK, let's go on */
return 1;
/* Error unrolling */
out_free_rspcap:
pool_free2(s->fe->rsp_cap_pool, s->txn.rsp.cap);
out_free_reqcap:
pool_free2(s->fe->req_cap_pool, s->txn.req.cap);
out_return:
return -1;
}
/* 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)) {
EV_FD_CLR(l->fd, DIR_RD);
l->state = LI_FULL;
return 0;
}
if (p && p->fe_sps_lim) {
int max = freq_ctr_remain(&p->fe_sess_per_sec, p->fe_sps_lim, 0);
if (max_accept > max)
max_accept = max;
}
while ((!p || p->feconn < p->maxconn) && actconn < global.maxconn && max_accept--) {
struct sockaddr_storage addr;
socklen_t laddr = sizeof(addr);
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);
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);
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);
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);
return 0;
}
jobs++;
actconn++;
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).
*/
jobs--;
actconn--;
l->nbconn--;
if (ret == 0) /* successful termination */
continue;
if (p) {
disable_listener(l);
p->state = PR_STIDLE;
}
return 0;
}
if (l->nbconn >= l->maxconn) {
EV_FD_CLR(l->fd, DIR_RD);
l->state = LI_FULL;
}
} /* end of while (p->feconn < p->maxconn) */
return 0;
}
