/* Tries to copy block <blk> at once into the channel's buffer after length
* controls. The chn->o and to_forward pointers are updated. If the channel
* input is closed, -2 is returned. If the block is too large for this buffer,
* -3 is returned. If there is not enough room left in the buffer, -1 is
* returned. Otherwise the number of bytes copied is returned (0 being a valid
* number). Channel flag READ_PARTIAL is updated if some data can be
* transferred. Channel flag CF_WAKE_WRITE is set if the write fails because
* the buffer is full.
*/
int bi_putblk(struct channel *chn, const char *blk, int len)
{
int max;
if (unlikely(channel_input_closed(chn)))
return -2;
max = buffer_max_len(chn);
if (unlikely(len > max - buffer_len(chn->buf))) {
/* we can't write this chunk right now because the buffer is
* almost full or because the block is too large. Return the
* available space or -2 if impossible.
*/
if (len > max)
return -3;
chn->flags |= CF_WAKE_WRITE;
return -1;
}
if (unlikely(len == 0))
return 0;
/* OK so the data fits in the buffer in one or two blocks */
max = buffer_contig_space(chn->buf);
memcpy(bi_end(chn->buf), blk, MIN(len, max));
if (len > max)
memcpy(chn->buf->data, blk + max, len - max);
chn->buf->i += len;
chn->total += len;
if (chn->to_forward) {
unsigned long fwd = len;
if (chn->to_forward != CHN_INFINITE_FORWARD) {
if (fwd > chn->to_forward)
fwd = chn->to_forward;
chn->to_forward -= fwd;
}
b_adv(chn->buf, fwd);
}
/* notify that some data was read from the SI into the buffer */
chn->flags |= CF_READ_PARTIAL;
return len;
}
/*
* This function is called to send buffer data to a stream socket.
* It returns -1 in case of unrecoverable error, 0 if the caller needs to poll
* before calling it again, otherwise 1. If a pipe was associated with the
* buffer and it empties it, it releases it as well.
*/
static int stream_sock_write_loop(struct stream_interface *si, struct buffer *b)
{
int write_poll = MAX_WRITE_POLL_LOOPS;
int retval = 1;
int ret, max;
if (unlikely(si->send_proxy_ofs)) {
/* The target server expects a PROXY line to be sent first.
* If the send_proxy_ofs is negative, it corresponds to the
* offset to start sending from then end of the proxy string
* (which is recomputed every time since it's constant). If
* it is positive, it means we have to send from the start.
*/
ret = make_proxy_line(trash, sizeof(trash),
&b->prod->addr.from, &b->prod->addr.to);
if (!ret)
return -1;
if (si->send_proxy_ofs > 0)
si->send_proxy_ofs = -ret; /* first call */
/* we have to send trash from (ret+sp for -sp bytes) */
ret = send(si->fd, trash + ret + si->send_proxy_ofs, -si->send_proxy_ofs,
(b->flags & BF_OUT_EMPTY) ? 0 : MSG_MORE);
if (ret > 0) {
if (fdtab[si->fd].state == FD_STCONN)
fdtab[si->fd].state = FD_STREADY;
si->send_proxy_ofs += ret; /* becomes zero once complete */
b->flags |= BF_WRITE_NULL; /* connect() succeeded */
}
else if (ret == 0 || errno == EAGAIN) {
/* nothing written, we need to poll for write first */
return 0;
}
else {
/* bad, we got an error */
return -1;
}
}
#if defined(CONFIG_HAP_LINUX_SPLICE)
while (b->pipe) {
ret = splice(b->pipe->cons, NULL, si->fd, NULL, b->pipe->data,
SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
if (ret <= 0) {
if (ret == 0 || errno == EAGAIN) {
retval = 0;
return retval;
}
/* here we have another error */
retval = -1;
return retval;
}
b->flags |= BF_WRITE_PARTIAL;
b->pipe->data -= ret;
if (!b->pipe->data) {
put_pipe(b->pipe);
b->pipe = NULL;
break;
}
if (--write_poll <= 0)
return retval;
/* The only reason we did not empty the pipe is that the output
* buffer is full.
*/
return 0;
}
/* At this point, the pipe is empty, but we may still have data pending
* in the normal buffer.
*/
#endif
if (!b->send_max) {
b->flags |= BF_OUT_EMPTY;
return retval;
}
/* when we're in this loop, we already know that there is no spliced
* data left, and that there are sendable buffered data.
*/
while (1) {
if (b->r > b->w)
max = b->r - b->w;
else
max = b->data + b->size - b->w;
/* limit the amount of outgoing data if required */
if (max > b->send_max)
max = b->send_max;
/* check if we want to inform the kernel that we're interested in
* sending more data after this call. We want this if :
* - we're about to close after this last send and want to merge
* the ongoing FIN with the last segment.
* - we know we can't send everything at once and must get back
* here because of unaligned data
* - there is still a finite amount of data to forward
* The test is arranged so that the most common case does only 2
* tests.
*/
if (MSG_NOSIGNAL && MSG_MORE) {
unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
if ((!(b->flags & BF_NEVER_WAIT) &&
((b->to_forward && b->to_forward != BUF_INFINITE_FORWARD) ||
(b->flags & BF_EXPECT_MORE))) ||
((b->flags & (BF_SHUTW|BF_SHUTW_NOW|BF_HIJACK)) == BF_SHUTW_NOW && (max == b->send_max)) ||
(max != b->l && max != b->send_max)) {
send_flag |= MSG_MORE;
}
/* this flag has precedence over the rest */
if (b->flags & BF_SEND_DONTWAIT)
send_flag &= ~MSG_MORE;
ret = send(si->fd, b->w, max, send_flag);
} else {
int skerr;
socklen_t lskerr = sizeof(skerr);
ret = getsockopt(si->fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr);
if (ret == -1 || skerr)
ret = -1;
else
ret = send(si->fd, b->w, max, MSG_DONTWAIT);
}
if (ret > 0) {
if (fdtab[si->fd].state == FD_STCONN)
fdtab[si->fd].state = FD_STREADY;
b->flags |= BF_WRITE_PARTIAL;
b->w += ret;
if (b->w == b->data + b->size)
b->w = b->data; /* wrap around the buffer */
b->l -= ret;
if (likely(b->l < buffer_max_len(b)))
b->flags &= ~BF_FULL;
if (likely(!b->l))
/* optimize data alignment in the buffer */
b->r = b->w = b->lr = b->data;
b->send_max -= ret;
if (!b->send_max) {
/* Always clear both flags once everything has been sent, they're one-shot */
b->flags &= ~(BF_EXPECT_MORE | BF_SEND_DONTWAIT);
if (likely(!b->pipe))
b->flags |= BF_OUT_EMPTY;
break;
}
/* if the system buffer is full, don't insist */
if (ret < max)
break;
if (--write_poll <= 0)
break;
}
else if (ret == 0 || errno == EAGAIN) {
/* nothing written, we need to poll for write first */
retval = 0;
break;
}
else {
/* bad, we got an error */
retval = -1;
break;
}
} /* while (1) */
return retval;
}
/*
* 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;
}
/* Return the version of the SSL protocol in the request. It supports both
* SSLv3 (TLSv1) header format for any message, and SSLv2 header format for
* the hello message. The SSLv3 format is described in RFC 2246 p49, and the
* SSLv2 format is described here, and completed p67 of RFC 2246 :
* http://wp.netscape.com/eng/security/SSL_2.html
*
* Note: this decoder only works with non-wrapping data.
*/
static int
smp_fetch_req_ssl_ver(struct proxy *px, struct session *s, void *l7, unsigned int opt,
const struct arg *args, struct sample *smp, const char *kw)
{
int version, bleft, msg_len;
const unsigned char *data;
if (!s || !s->req)
return 0;
msg_len = 0;
bleft = s->req->buf->i;
if (!bleft)
goto too_short;
data = (const unsigned char *)s->req->buf->p;
if ((*data >= 0x14 && *data <= 0x17) || (*data == 0xFF)) {
/* SSLv3 header format */
if (bleft < 5)
goto too_short;
version = (data[1] << 16) + data[2]; /* version: major, minor */
msg_len = (data[3] << 8) + data[4]; /* record length */
/* format introduced with SSLv3 */
if (version < 0x00030000)
goto not_ssl;
/* message length between 1 and 2^14 + 2048 */
if (msg_len < 1 || msg_len > ((1<<14) + 2048))
goto not_ssl;
bleft -= 5; data += 5;
} else {
/* SSLv2 header format, only supported for hello (msg type 1) */
int rlen, plen, cilen, silen, chlen;
if (*data & 0x80) {
if (bleft < 3)
goto too_short;
/* short header format : 15 bits for length */
rlen = ((data[0] & 0x7F) << 8) | data[1];
plen = 0;
bleft -= 2; data += 2;
} else {
if (bleft < 4)
goto too_short;
/* long header format : 14 bits for length + pad length */
rlen = ((data[0] & 0x3F) << 8) | data[1];
plen = data[2];
bleft -= 3; data += 2;
}
if (*data != 0x01)
goto not_ssl;
bleft--; data++;
if (bleft < 8)
goto too_short;
version = (data[0] << 16) + data[1]; /* version: major, minor */
cilen = (data[2] << 8) + data[3]; /* cipher len, multiple of 3 */
silen = (data[4] << 8) + data[5]; /* session_id_len: 0 or 16 */
chlen = (data[6] << 8) + data[7]; /* 16<=challenge length<=32 */
bleft -= 8; data += 8;
if (cilen % 3 != 0)
goto not_ssl;
if (silen && silen != 16)
goto not_ssl;
if (chlen < 16 || chlen > 32)
goto not_ssl;
if (rlen != 9 + cilen + silen + chlen)
goto not_ssl;
/* focus on the remaining data length */
msg_len = cilen + silen + chlen + plen;
}
/* We could recursively check that the buffer ends exactly on an SSL
* fragment boundary and that a possible next segment is still SSL,
* but that's a bit pointless. However, we could still check that
* all the part of the request which fits in a buffer is already
* there.
*/
if (msg_len > buffer_max_len(s->req) + s->req->buf->data - s->req->buf->p)
msg_len = buffer_max_len(s->req) + s->req->buf->data - s->req->buf->p;
if (bleft < msg_len)
goto too_short;
/* OK that's enough. We have at least the whole message, and we have
* the protocol version.
*/
smp->type = SMP_T_UINT;
smp->data.uint = version;
smp->flags = SMP_F_VOLATILE;
return 1;
too_short:
smp->flags = SMP_F_MAY_CHANGE;
not_ssl:
return 0;
}
/*
* This function is called to send buffer data to a stream socket.
* It returns -1 in case of unrecoverable error, 0 if the caller needs to poll
* before calling it again, otherwise 1. If a pipe was associated with the
* buffer and it empties it, it releases it as well.
*/
static int stream_sock_write_loop(struct stream_interface *si, struct buffer *b)
{
int write_poll = MAX_WRITE_POLL_LOOPS;
int retval = 1;
int ret, max;
#if defined(CONFIG_HAP_LINUX_SPLICE)
while (b->pipe) {
ret = splice(b->pipe->cons, NULL, si->fd, NULL, b->pipe->data,
SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
if (ret <= 0) {
if (ret == 0 || errno == EAGAIN) {
retval = 0;
return retval;
}
/* here we have another error */
retval = -1;
return retval;
}
b->flags |= BF_WRITE_PARTIAL;
b->pipe->data -= ret;
if (!b->pipe->data) {
put_pipe(b->pipe);
b->pipe = NULL;
break;
}
if (--write_poll <= 0)
return retval;
}
/* At this point, the pipe is empty, but we may still have data pending
* in the normal buffer.
*/
#endif
if (!b->send_max) {
b->flags |= BF_OUT_EMPTY;
return retval;
}
/* when we're in this loop, we already know that there is no spliced
* data left, and that there are sendable buffered data.
*/
while (1) {
if (b->r > b->w)
max = b->r - b->w;
else
max = b->data + b->size - b->w;
/* limit the amount of outgoing data if required */
if (max > b->send_max)
max = b->send_max;
/* check if we want to inform the kernel that we're interested in
* sending more data after this call. We want this if :
* - we're about to close after this last send and want to merge
* the ongoing FIN with the last segment.
* - we know we can't send everything at once and must get back
* here because of unaligned data
* - there is still a finite amount of data to forward
* The test is arranged so that the most common case does only 2
* tests.
*/
if (MSG_NOSIGNAL && MSG_MORE) {
unsigned int send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
if (((b->to_forward && b->to_forward != BUF_INFINITE_FORWARD) ||
((b->flags & (BF_SHUTW|BF_SHUTW_NOW|BF_HIJACK)) == BF_SHUTW_NOW && (max == b->send_max)) ||
(max != b->l && max != b->send_max))
&& (fdtab[si->fd].flags & FD_FL_TCP)) {
send_flag |= MSG_MORE;
}
else if (b->flags & BF_EXPECT_MORE) {
/* it was forced on the buffer, this flag is one-shoot */
b->flags &= ~BF_EXPECT_MORE;
send_flag |= MSG_MORE;
}
/* this flag has precedence over the rest */
if (b->flags & BF_SEND_DONTWAIT)
send_flag &= ~MSG_MORE;
ret = send(si->fd, b->w, max, send_flag);
/* disable it only once everything has been sent */
if (ret == max && (b->flags & BF_SEND_DONTWAIT))
b->flags &= ~BF_SEND_DONTWAIT;
} else {
int skerr;
socklen_t lskerr = sizeof(skerr);
ret = getsockopt(si->fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr);
if (ret == -1 || skerr)
ret = -1;
else
ret = send(si->fd, b->w, max, MSG_DONTWAIT);
}
if (ret > 0) {
if (fdtab[si->fd].state == FD_STCONN)
fdtab[si->fd].state = FD_STREADY;
b->flags |= BF_WRITE_PARTIAL;
b->w += ret;
if (b->w == b->data + b->size)
b->w = b->data; /* wrap around the buffer */
b->l -= ret;
if (likely(b->l < buffer_max_len(b)))
b->flags &= ~BF_FULL;
if (likely(!b->l))
/* optimize data alignment in the buffer */
b->r = b->w = b->lr = b->data;
b->send_max -= ret;
if (!b->send_max) {
if (likely(!b->pipe))
b->flags |= BF_OUT_EMPTY;
break;
}
/* if the system buffer is full, don't insist */
if (ret < max)
break;
if (--write_poll <= 0)
break;
}
else if (ret == 0 || errno == EAGAIN) {
/* nothing written, we need to poll for write first */
retval = 0;
break;
}
else {
/* bad, we got an error */
retval = -1;
break;
}
} /* while (1) */
return retval;
}
