/* Schedule up to <bytes> more bytes to be forwarded via the channel without
* notifying the owner task. Any data pending in the buffer are scheduled to be
* sent as well, in the limit of the number of bytes to forward. This must be
* the only method to use to schedule bytes to be forwarded. If the requested
* number is too large, it is automatically adjusted. The number of bytes taken
* into account is returned. Directly touching ->to_forward will cause lockups
* when buf->o goes down to zero if nobody is ready to push the remaining data.
*/
unsigned long long __channel_forward(struct channel *chn, unsigned long long bytes)
{
unsigned int new_forward;
unsigned int forwarded;
forwarded = chn->buf->i;
b_adv(chn->buf, chn->buf->i);
/* Note: the case below is the only case where we may return
* a byte count that does not fit into a 32-bit number.
*/
if (likely(chn->to_forward == CHN_INFINITE_FORWARD))
return bytes;
if (likely(bytes == CHN_INFINITE_FORWARD)) {
chn->to_forward = bytes;
return bytes;
}
new_forward = chn->to_forward + bytes - forwarded;
bytes = forwarded; /* at least those bytes were scheduled */
if (new_forward <= chn->to_forward) {
/* integer overflow detected, let's assume no more than 2G at once */
new_forward = MID_RANGE(new_forward);
}
if (new_forward > chn->to_forward) {
bytes += new_forward - chn->to_forward;
chn->to_forward = new_forward;
}
return bytes;
}
/*
* Add data to compress
*/
int http_compression_buffer_add_data(struct session *s, struct buffer *in, struct buffer *out)
{
struct http_msg *msg = &s->txn.rsp;
int consumed_data = 0;
int data_process_len;
int left;
int ret;
/*
* Skip data, we don't need them in the new buffer. They are results
* of CHUNK_CRLF and CHUNK_SIZE parsing.
*/
b_adv(in, msg->next);
msg->next = 0;
msg->sov = 0;
msg->sol = 0;
/*
* select the smallest size between the announced chunk size, the input
* data, and the available output buffer size
*/
data_process_len = MIN(in->i, msg->chunk_len);
data_process_len = MIN(out->size - buffer_len(out), data_process_len);
left = data_process_len - bi_contig_data(in);
if (left <= 0) {
consumed_data += ret = s->comp_algo->add_data(s->comp_ctx, bi_ptr(in), data_process_len, out);
if (ret < 0)
return -1;
} else {
consumed_data += ret = s->comp_algo->add_data(s->comp_ctx, bi_ptr(in), bi_contig_data(in), out);
if (ret < 0)
return -1;
consumed_data += ret = s->comp_algo->add_data(s->comp_ctx, in->data, left, out);
if (ret < 0)
return -1;
}
b_adv(in, data_process_len);
msg->chunk_len -= data_process_len;
return consumed_data;
}
/* If an explicit source binding is specified on the server and/or backend, and
* this source makes use of the transparent proxy, then it is extracted now and
* assigned to the session's pending connection. This function assumes that an
* outgoing connection has already been assigned to s->si[1].end.
*/
static void assign_tproxy_address(struct session *s)
{
#if defined(CONFIG_HAP_CTTPROXY) || defined(CONFIG_HAP_TRANSPARENT)
struct server *srv = objt_server(s->target);
struct conn_src *src;
struct connection *cli_conn;
struct connection *srv_conn = objt_conn(s->si[1].end);
if (srv && srv->conn_src.opts & CO_SRC_BIND)
src = &srv->conn_src;
else if (s->be->conn_src.opts & CO_SRC_BIND)
src = &s->be->conn_src;
else
return;
switch (src->opts & CO_SRC_TPROXY_MASK) {
case CO_SRC_TPROXY_ADDR:
srv_conn->addr.from = src->tproxy_addr;
break;
case CO_SRC_TPROXY_CLI:
case CO_SRC_TPROXY_CIP:
/* FIXME: what can we do if the client connects in IPv6 or unix socket ? */
cli_conn = objt_conn(s->si[0].end);
if (cli_conn)
srv_conn->addr.from = cli_conn->addr.from;
else
memset(&srv_conn->addr.from, 0, sizeof(srv_conn->addr.from));
break;
case CO_SRC_TPROXY_DYN:
if (src->bind_hdr_occ) {
char *vptr;
int vlen;
int rewind;
/* bind to the IP in a header */
((struct sockaddr_in *)&srv_conn->addr.from)->sin_family = AF_INET;
((struct sockaddr_in *)&srv_conn->addr.from)->sin_port = 0;
((struct sockaddr_in *)&srv_conn->addr.from)->sin_addr.s_addr = 0;
b_rew(s->req.buf, rewind = http_hdr_rewind(&s->txn.req));
if (http_get_hdr(&s->txn.req, src->bind_hdr_name, src->bind_hdr_len,
&s->txn.hdr_idx, src->bind_hdr_occ, NULL, &vptr, &vlen)) {
((struct sockaddr_in *)&srv_conn->addr.from)->sin_addr.s_addr =
htonl(inetaddr_host_lim(vptr, vptr + vlen));
}
b_adv(s->req.buf, rewind);
}
break;
default:
memset(&srv_conn->addr.from, 0, sizeof(srv_conn->addr.from));
}
#endif
}
/* 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;
}
/*
* Add data to compress
*/
int http_compression_buffer_add_data(struct session *s, struct buffer *in, struct buffer *out)
{
struct http_msg *msg = &s->txn.rsp;
int consumed_data = 0;
int data_process_len;
int block1, block2;
/*
* Temporarily skip already parsed data and chunks to jump to the
* actual data block. It is fixed before leaving.
*/
b_adv(in, msg->next);
/*
* select the smallest size between the announced chunk size, the input
* data, and the available output buffer size. The compressors are
* assumed to be able to process all the bytes we pass to them at once.
*/
data_process_len = MIN(in->i, msg->chunk_len);
data_process_len = MIN(out->size - buffer_len(out), data_process_len);
block1 = data_process_len;
if (block1 > bi_contig_data(in))
block1 = bi_contig_data(in);
block2 = data_process_len - block1;
/* compressors return < 0 upon error or the amount of bytes read */
consumed_data = s->comp_algo->add_data(s->comp_ctx, bi_ptr(in), block1, out);
if (consumed_data >= 0 && block2 > 0) {
consumed_data = s->comp_algo->add_data(s->comp_ctx, in->data, block2, out);
if (consumed_data >= 0)
consumed_data += block1;
}
/* restore original buffer pointer */
b_rew(in, msg->next);
if (consumed_data > 0) {
msg->next += consumed_data;
msg->chunk_len -= consumed_data;
}
return consumed_data;
}
/* RDP Cookie HASH. */
struct server *get_server_rch(struct session *s)
{
unsigned int hash = 0;
struct proxy *px = s->be;
unsigned long len;
int ret;
struct sample smp;
int rewind;
/* tot_weight appears to mean srv_count */
if (px->lbprm.tot_weight == 0)
return NULL;
memset(&smp, 0, sizeof(smp));
b_rew(s->req.buf, rewind = s->req.buf->o);
ret = fetch_rdp_cookie_name(s, &smp, px->hh_name, px->hh_len);
len = smp.data.str.len;
b_adv(s->req.buf, rewind);
if (ret == 0 || (smp.flags & SMP_F_MAY_CHANGE) || len == 0)
return NULL;
/* note: we won't hash if there's only one server left */
if (px->lbprm.tot_used == 1)
goto hash_done;
/* Found a the hh_name in the headers.
* we will compute the hash based on this value ctx.val.
*/
hash = gen_hash(px, smp.data.str.str, len);
if ((px->lbprm.algo & BE_LB_HASH_MOD) == BE_LB_HMOD_AVAL)
hash = full_hash(hash);
hash_done:
if (px->lbprm.algo & BE_LB_LKUP_CHTREE)
return chash_get_server_hash(px, hash);
else
return map_get_server_hash(px, hash);
}
/*
* Init HTTP compression
*/
int http_compression_buffer_init(struct session *s, struct buffer *in, struct buffer *out)
{
struct http_msg *msg = &s->txn.rsp;
int left;
/* not enough space */
if (in->size - buffer_len(in) < 40)
return -1;
/*
* Skip data, we don't need them in the new buffer. They are results
* of CHUNK_CRLF and CHUNK_SIZE parsing.
*/
b_adv(in, msg->next);
msg->next = 0;
msg->sov = 0;
msg->sol = 0;
out->size = global.tune.bufsize;
out->i = 0;
out->o = 0;
out->p = out->data;
/* copy output data */
if (in->o > 0) {
left = in->o - bo_contig_data(in);
memcpy(out->data, bo_ptr(in), bo_contig_data(in));
out->p += bo_contig_data(in);
if (left > 0) { /* second part of the buffer */
memcpy(out->p, in->data, left);
out->p += left;
}
out->o = in->o;
}
out->i += http_emit_chunk_size(out->p, 0, 0);
return 0;
}
/* Tries to copy character <c> into the channel's buffer after some length
* controls. The chn->o and to_forward pointers are updated. If the channel
* input is closed, -2 is returned. If there is not enough room left in the
* buffer, -1 is returned. Otherwise the number of bytes copied is returned
* (1). 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_putchr(struct channel *chn, char c)
{
if (unlikely(channel_input_closed(chn)))
return -2;
if (channel_full(chn)) {
chn->flags |= CF_WAKE_WRITE;
return -1;
}
*bi_end(chn->buf) = c;
chn->buf->i++;
chn->flags |= CF_READ_PARTIAL;
if (chn->to_forward >= 1) {
if (chn->to_forward != CHN_INFINITE_FORWARD)
chn->to_forward--;
b_adv(chn->buf, 1);
}
chn->total++;
return 1;
}
/*
* Flush data in process, and write the header and footer of the chunk. Upon
* success, in and out buffers are swapped to avoid a copy.
*/
int http_compression_buffer_end(struct session *s, struct buffer **in, struct buffer **out, int end)
{
int to_forward, forwarded;
int left;
struct http_msg *msg = &s->txn.rsp;
struct buffer *ib = *in, *ob = *out;
#ifdef USE_ZLIB
int ret;
/* flush data here */
if (end)
ret = s->comp_algo->flush(s->comp_ctx, ob, Z_FINISH); /* end of data */
else
ret = s->comp_algo->flush(s->comp_ctx, ob, Z_SYNC_FLUSH); /* end of buffer */
if (ret < 0)
return -1; /* flush failed */
#endif /* USE_ZLIB */
if (ob->i > 8) {
/* more than a chunk size => some data were emitted */
char *tail = ob->p + ob->i;
/* write real size at the begining of the chunk, no need of wrapping */
http_emit_chunk_size(ob->p, ob->i - 8, 0);
/* chunked encoding requires CRLF after data */
*tail++ = '\r';
*tail++ = '\n';
if (!(msg->flags & HTTP_MSGF_TE_CHNK) && msg->chunk_len == 0) {
/* End of data, 0<CRLF><CRLF> is needed but we're not
* in chunked mode on input so we must add it ourselves.
*/
memcpy(tail, "0\r\n\r\n", 5);
tail += 5;
}
ob->i = tail - ob->p;
} else {
/* no data were sent, cancel the chunk size */
ob->i = 0;
}
to_forward = ob->i;
/* update input rate */
forwarded = ib->o - ob->o;
if (s->comp_ctx && s->comp_ctx->cur_lvl > 0) {
update_freq_ctr(&global.comp_bps_in, forwarded);
s->fe->fe_counters.comp_in += forwarded;
s->be->be_counters.comp_in += forwarded;
} else {
s->fe->fe_counters.comp_byp += forwarded;
s->be->be_counters.comp_byp += forwarded;
}
/* copy the remaining data in the tmp buffer. */
if (ib->i > 0) {
left = ib->i - bi_contig_data(ib);
memcpy(bi_end(ob), bi_ptr(ib), bi_contig_data(ib));
ob->i += bi_contig_data(ib);
if (left > 0) {
memcpy(bi_end(ob), ib->data, left);
ob->i += left;
}
}
/* swap the buffers */
*in = ob;
*out = ib;
if (s->comp_ctx && s->comp_ctx->cur_lvl > 0) {
update_freq_ctr(&global.comp_bps_out, to_forward);
s->fe->fe_counters.comp_out += to_forward;
s->be->be_counters.comp_out += to_forward;
}
/* forward the new chunk without remaining data */
b_adv(ob, to_forward);
return to_forward;
}
/*
* Flush data in process, and write the header and footer of the chunk. Upon
* success, in and out buffers are swapped to avoid a copy.
*/
int http_compression_buffer_end(struct session *s, struct buffer **in, struct buffer **out, int end)
{
int to_forward;
int left;
struct http_msg *msg = &s->txn.rsp;
struct buffer *ib = *in, *ob = *out;
#ifdef USE_ZLIB
int ret;
/* flush data here */
if (end)
ret = s->comp_algo->flush(s->comp_ctx, ob, Z_FINISH); /* end of data */
else
ret = s->comp_algo->flush(s->comp_ctx, ob, Z_SYNC_FLUSH); /* end of buffer */
if (ret < 0)
return -1; /* flush failed */
#endif /* USE_ZLIB */
if (ob->i > 8) {
/* more than a chunk size => some data were emitted */
char *tail = ob->p + ob->i;
/* write real size at the begining of the chunk, no need of wrapping */
http_emit_chunk_size(ob->p, ob->i - 8, 0);
/* chunked encoding requires CRLF after data */
*tail++ = '\r';
*tail++ = '\n';
/* At the end of data, we must write the empty chunk 0<CRLF>,
* and terminate the trailers section with a last <CRLF>. If
* we're forwarding a chunked-encoded response, we'll have a
* trailers section after the empty chunk which needs to be
* forwarded and which will provide the last CRLF. Otherwise
* we write it ourselves.
*/
if (msg->msg_state >= HTTP_MSG_TRAILERS) {
memcpy(tail, "0\r\n", 3);
tail += 3;
if (msg->msg_state >= HTTP_MSG_DONE) {
memcpy(tail, "\r\n", 2);
tail += 2;
}
}
ob->i = tail - ob->p;
} else {
/* no data were sent, cancel the chunk size */
ob->i = 0;
}
to_forward = ob->i;
/* update input rate */
if (s->comp_ctx && s->comp_ctx->cur_lvl > 0) {
update_freq_ctr(&global.comp_bps_in, msg->next);
s->fe->fe_counters.comp_in += msg->next;
s->be->be_counters.comp_in += msg->next;
} else {
s->fe->fe_counters.comp_byp += msg->next;
s->be->be_counters.comp_byp += msg->next;
}
/* copy the remaining data in the tmp buffer. */
b_adv(ib, msg->next);
msg->next = 0;
if (ib->i > 0) {
left = ib->i - bi_contig_data(ib);
memcpy(bi_end(ob), bi_ptr(ib), bi_contig_data(ib));
ob->i += bi_contig_data(ib);
if (left > 0) {
memcpy(bi_end(ob), ib->data, left);
ob->i += left;
}
}
/* swap the buffers */
*in = ob;
*out = ib;
if (s->comp_ctx && s->comp_ctx->cur_lvl > 0) {
update_freq_ctr(&global.comp_bps_out, to_forward);
s->fe->fe_counters.comp_out += to_forward;
s->be->be_counters.comp_out += to_forward;
}
/* forward the new chunk without remaining data */
b_adv(ob, to_forward);
return to_forward;
}
/*
* This is the callback which is called by the connection layer to receive data
* into the buffer from the connection. It iterates over the transport layer's
* rcv_buf function.
*/
static void si_conn_recv_cb(struct connection *conn)
{
struct stream_interface *si = conn->owner;
struct channel *chn = si->ib;
int ret, max, cur_read;
int read_poll = MAX_READ_POLL_LOOPS;
/* 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 (conn->flags & CO_FL_ERROR)
return;
/* stop here if we reached the end of data */
if (conn_data_read0_pending(conn))
goto out_shutdown_r;
/* maybe we were called immediately after an asynchronous shutr */
if (chn->flags & CF_SHUTR)
return;
cur_read = 0;
if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) && !chn->buf->o &&
global.tune.idle_timer &&
(unsigned short)(now_ms - chn->last_read) >= global.tune.idle_timer) {
/* The buffer was empty and nothing was transferred for more
* than one second. This was caused by a pause and not by
* congestion. Reset any streaming mode to reduce latency.
*/
chn->xfer_small = 0;
chn->xfer_large = 0;
chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
}
/* First, let's see if we may splice data across the channel without
* using a buffer.
*/
if (conn->xprt->rcv_pipe &&
(chn->pipe || chn->to_forward >= MIN_SPLICE_FORWARD) &&
chn->flags & CF_KERN_SPLICING) {
if (buffer_not_empty(chn->buf)) {
/* 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.
*/
goto abort_splice;
}
if (unlikely(chn->pipe == NULL)) {
if (pipes_used >= global.maxpipes || !(chn->pipe = get_pipe())) {
chn->flags &= ~CF_KERN_SPLICING;
goto abort_splice;
}
}
ret = conn->xprt->rcv_pipe(conn, chn->pipe, chn->to_forward);
if (ret < 0) {
/* splice not supported on this end, let's disable it */
chn->flags &= ~CF_KERN_SPLICING;
goto abort_splice;
}
if (ret > 0) {
if (chn->to_forward != CHN_INFINITE_FORWARD)
chn->to_forward -= ret;
chn->total += ret;
cur_read += ret;
chn->flags |= CF_READ_PARTIAL;
}
if (conn_data_read0_pending(conn))
goto out_shutdown_r;
if (conn->flags & CO_FL_ERROR)
return;
if (conn->flags & CO_FL_WAIT_ROOM) {
/* the pipe is full or we have read enough data that it
* could soon be full. Let's stop before needing to poll.
*/
si->flags |= SI_FL_WAIT_ROOM;
__conn_data_stop_recv(conn);
}
/* splice not possible (anymore), let's go on on standard copy */
}
abort_splice:
if (chn->pipe && unlikely(!chn->pipe->data)) {
put_pipe(chn->pipe);
chn->pipe = NULL;
}
/* Important note : if we're called with POLL_IN|POLL_HUP, it means the read polling
* was enabled, which implies that the recv buffer was not full. So we have a guarantee
* that if such an event is not handled above in splice, it will be handled here by
* recv().
*/
while (!(conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_DATA_RD_SH | CO_FL_WAIT_ROOM | CO_FL_HANDSHAKE))) {
max = bi_avail(chn);
if (!max) {
si->flags |= SI_FL_WAIT_ROOM;
break;
}
ret = conn->xprt->rcv_buf(conn, chn->buf, max);
if (ret <= 0)
break;
cur_read += ret;
/* if we're allowed to directly forward data, we must update ->o */
if (chn->to_forward && !(chn->flags & (CF_SHUTW|CF_SHUTW_NOW))) {
unsigned long fwd = ret;
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);
}
chn->flags |= CF_READ_PARTIAL;
chn->total += ret;
if (channel_full(chn)) {
si->flags |= SI_FL_WAIT_ROOM;
break;
}
if ((chn->flags & CF_READ_DONTWAIT) || --read_poll <= 0) {
si->flags |= SI_FL_WAIT_ROOM;
__conn_data_stop_recv(conn);
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.
*/
if (ret < max) {
/* if a streamer has read few data, it may be because we
* have exhausted system buffers. It's not worth trying
* again.
*/
if (chn->flags & CF_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;
}
} /* while !flags */
if (conn->flags & CO_FL_ERROR)
return;
if (cur_read) {
if ((chn->flags & (CF_STREAMER | CF_STREAMER_FAST)) &&
(cur_read <= chn->buf->size / 2)) {
chn->xfer_large = 0;
chn->xfer_small++;
if (chn->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.
*/
chn->flags &= ~(CF_STREAMER | CF_STREAMER_FAST);
}
else if (chn->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".
*/
chn->flags &= ~CF_STREAMER_FAST;
}
}
else if (!(chn->flags & CF_STREAMER_FAST) &&
(cur_read >= chn->buf->size - global.tune.maxrewrite)) {
/* we read a full buffer at once */
chn->xfer_small = 0;
chn->xfer_large++;
if (chn->xfer_large >= 3) {
/* we call this buffer a fast streamer if it manages
* to be filled in one call 3 consecutive times.
*/
chn->flags |= (CF_STREAMER | CF_STREAMER_FAST);
}
}
else {
chn->xfer_small = 0;
chn->xfer_large = 0;
}
chn->last_read = now_ms;
}
if (conn_data_read0_pending(conn))
/* connection closed */
goto out_shutdown_r;
return;
out_shutdown_r:
/* we received a shutdown */
chn->flags |= CF_READ_NULL;
if (chn->flags & CF_AUTO_CLOSE)
channel_shutw_now(chn);
stream_sock_read0(si);
conn_data_read0(conn);
return;
}
