blob: af9a91182939300621bb529eb7688fd70d05e8e5 [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
Willy Tarreauf7bc57c2012-10-03 00:19:48 +02002 * RAW transport layer over SOCK_STREAM sockets.
Willy Tarreaubaaee002006-06-26 02:48:02 +02003 *
Willy Tarreaub277d6e2012-05-11 16:59:14 +02004 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
Willy Tarreaubaaee002006-06-26 02:48:02 +02005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
Willy Tarreau6b4aad42009-01-18 21:59:13 +010013#define _GNU_SOURCE
Willy Tarreaubaaee002006-06-26 02:48:02 +020014#include <errno.h>
15#include <fcntl.h>
16#include <stdio.h>
17#include <stdlib.h>
18
19#include <sys/socket.h>
20#include <sys/stat.h>
21#include <sys/types.h>
22
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040023#include <netinet/tcp.h>
24
Willy Tarreauc7e42382012-08-24 19:22:53 +020025#include <common/buffer.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020026#include <common/compat.h>
Willy Tarreaue3ba5f02006-06-29 18:54:54 +020027#include <common/config.h>
Willy Tarreaud6f087e2008-01-18 17:20:13 +010028#include <common/debug.h>
Willy Tarreau83749182007-04-15 20:56:27 +020029#include <common/standard.h>
Willy Tarreau0c303ee2008-07-07 00:09:58 +020030#include <common/ticks.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020031#include <common/time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020032
Willy Tarreau8b117082012-08-06 15:06:49 +020033#include <proto/connection.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020034#include <proto/fd.h>
Willy Tarreaueb472682010-05-28 18:46:57 +020035#include <proto/freq_ctr.h>
36#include <proto/log.h>
Willy Tarreau3eba98a2009-01-25 13:56:13 +010037#include <proto/pipe.h>
Willy Tarreau75bf2c92012-08-20 17:01:35 +020038#include <proto/raw_sock.h>
Willy Tarreau73b013b2012-05-21 16:31:45 +020039#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040#include <proto/task.h>
41
Willy Tarreau5bd8c372009-01-19 00:32:22 +010042#include <types/global.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043
Willy Tarreaub277d6e2012-05-11 16:59:14 +020044
Willy Tarreaue5733232019-05-22 19:24:06 +020045#if defined(USE_LINUX_SPLICE)
Willy Tarreau43d8fb22011-08-22 17:12:02 +020046#include <common/splice.h>
Willy Tarreau5bd8c372009-01-19 00:32:22 +010047
48/* A pipe contains 16 segments max, and it's common to see segments of 1448 bytes
49 * because of timestamps. Use this as a hint for not looping on splice().
50 */
51#define SPLICE_FULL_HINT 16*1448
52
Willy Tarreaua9de3332009-11-28 07:47:10 +010053/* how many data we attempt to splice at once when the buffer is configured for
54 * infinite forwarding */
55#define MAX_SPLICE_AT_ONCE (1<<30)
56
Willy Tarreau5bd8c372009-01-19 00:32:22 +010057/* Returns :
Willy Tarreau96199b12012-08-24 00:46:52 +020058 * -1 if splice() is not supported
59 * >= 0 to report the amount of spliced bytes.
60 * connection flags are updated (error, read0, wait_room, wait_data).
61 * The caller must have previously allocated the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +010062 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +010063int raw_sock_to_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe, unsigned int count)
Willy Tarreau5bd8c372009-01-19 00:32:22 +010064{
Willy Tarreau31971e52009-09-20 12:07:52 +020065 int ret;
Willy Tarreauafad0e02012-08-09 14:45:22 +020066 int retval = 0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010067
Willy Tarreauf79c8172013-10-21 16:30:56 +020068
Willy Tarreaufd803bb2014-01-20 15:13:07 +010069 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +020070 return 0;
71
Willy Tarreau585744b2017-08-24 14:31:19 +020072 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +010073 return 0;
74
Willy Tarreaue2a0eec2020-01-17 09:59:40 +010075 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +010076 errno = 0;
77
Willy Tarreau96199b12012-08-24 00:46:52 +020078 /* Under Linux, if FD_POLL_HUP is set, we have reached the end.
79 * Since older splice() implementations were buggy and returned
80 * EAGAIN on end of read, let's bypass the call to splice() now.
81 */
Willy Tarreau585744b2017-08-24 14:31:19 +020082 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +020083 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +020084 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +020085 goto out_read0;
86
87 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +020088 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +010089 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreauce3eda72013-12-05 00:49:40 +010090 errno = 0; /* let the caller do a getsockopt() if it wants it */
Willy Tarreau256b9c52017-10-25 09:30:13 +020091 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +020092 }
93 }
Willy Tarreaua9de3332009-11-28 07:47:10 +010094
Willy Tarreau96199b12012-08-24 00:46:52 +020095 while (count) {
96 if (count > MAX_SPLICE_AT_ONCE)
97 count = MAX_SPLICE_AT_ONCE;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010098
Willy Tarreau585744b2017-08-24 14:31:19 +020099 ret = splice(conn->handle.fd, NULL, pipe->prod, NULL, count,
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100100 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
101
102 if (ret <= 0) {
Willy Tarreau38447472019-05-22 19:55:24 +0200103 if (ret == 0)
Willy Tarreau96199b12012-08-24 00:46:52 +0200104 goto out_read0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100105
106 if (errno == EAGAIN) {
107 /* there are two reasons for EAGAIN :
108 * - nothing in the socket buffer (standard)
109 * - pipe is full
Willy Tarreau38447472019-05-22 19:55:24 +0200110 * The difference between these two situations
111 * is problematic. Since we don't know if the
112 * pipe is full, we'll stop if the pipe is not
113 * empty. Anyway, we will almost always fill or
114 * empty the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100115 */
Willy Tarreau96199b12012-08-24 00:46:52 +0200116 if (pipe->data) {
117 /* alway stop reading until the pipe is flushed */
118 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100119 break;
120 }
121
Willy Tarreau38447472019-05-22 19:55:24 +0200122 fd_cant_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100123 break;
124 }
Willy Tarreau45b88932012-11-12 12:00:09 +0100125 else if (errno == ENOSYS || errno == EINVAL || errno == EBADF) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200126 /* splice not supported on this end, disable it.
127 * We can safely return -1 since there is no
128 * chance that any data has been piped yet.
129 */
Willy Tarreau256b9c52017-10-25 09:30:13 +0200130 retval = -1;
131 goto leave;
Willy Tarreaudc340a92009-06-28 23:10:19 +0200132 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200133 else if (errno == EINTR) {
134 /* try again */
135 continue;
136 }
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100137 /* here we have another error */
Willy Tarreau96199b12012-08-24 00:46:52 +0200138 conn->flags |= CO_FL_ERROR;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100139 break;
140 } /* ret <= 0 */
141
Willy Tarreau96199b12012-08-24 00:46:52 +0200142 retval += ret;
143 pipe->data += ret;
Willy Tarreau4fc90ef2013-04-06 11:29:39 +0200144 count -= ret;
Willy Tarreaubaf2a502013-01-07 16:38:26 +0100145
Willy Tarreau96199b12012-08-24 00:46:52 +0200146 if (pipe->data >= SPLICE_FULL_HINT || ret >= global.tune.recv_enough) {
147 /* We've read enough of it for this time, let's stop before
148 * being asked to poll.
149 */
Willy Tarreau61d39a02013-07-18 21:49:32 +0200150 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau585744b2017-08-24 14:31:19 +0200151 fd_done_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100152 break;
153 }
154 } /* while */
155
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200156 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && retval)
157 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200158
159 leave:
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200160 if (retval > 0) {
161 /* we count the total bytes sent, and the send rate for 32-byte
162 * blocks. The reason for the latter is that freq_ctr are
163 * limited to 4GB and that it's not enough per second.
164 */
165 _HA_ATOMIC_ADD(&global.out_bytes, retval);
166 update_freq_ctr(&global.out_32bps, (retval + 16) / 32);
167 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200168 return retval;
Willy Tarreau3eba98a2009-01-25 13:56:13 +0100169
Willy Tarreau96199b12012-08-24 00:46:52 +0200170 out_read0:
171 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200172 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200173 goto leave;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100174}
175
Willy Tarreau96199b12012-08-24 00:46:52 +0200176/* Send as many bytes as possible from the pipe to the connection's socket.
177 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100178int raw_sock_from_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe)
Willy Tarreau96199b12012-08-24 00:46:52 +0200179{
180 int ret, done;
181
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100182 if (!conn_ctrl_ready(conn))
183 return 0;
184
Willy Tarreau585744b2017-08-24 14:31:19 +0200185 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200186 return 0;
187
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100188 if (conn->flags & CO_FL_SOCK_WR_SH) {
189 /* it's already closed */
190 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
191 errno = EPIPE;
192 return 0;
193 }
194
Willy Tarreau96199b12012-08-24 00:46:52 +0200195 done = 0;
196 while (pipe->data) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200197 ret = splice(pipe->cons, NULL, conn->handle.fd, NULL, pipe->data,
Willy Tarreau96199b12012-08-24 00:46:52 +0200198 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
199
200 if (ret <= 0) {
201 if (ret == 0 || errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200202 fd_cant_send(conn->handle.fd);
Willy Tarreau96199b12012-08-24 00:46:52 +0200203 break;
204 }
205 else if (errno == EINTR)
206 continue;
207
208 /* here we have another error */
209 conn->flags |= CO_FL_ERROR;
210 break;
211 }
212
213 done += ret;
214 pipe->data -= ret;
215 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200216 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200217 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200218 fd_cond_recv(conn->handle.fd);
219 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200220
Willy Tarreau96199b12012-08-24 00:46:52 +0200221 return done;
222}
223
Willy Tarreaue5733232019-05-22 19:24:06 +0200224#endif /* USE_LINUX_SPLICE */
Willy Tarreau6b4aad42009-01-18 21:59:13 +0100225
226
Willy Tarreau2ba44652012-08-20 17:30:32 +0200227/* Receive up to <count> bytes from connection <conn>'s socket and store them
Willy Tarreauabf08d92014-01-14 11:31:27 +0100228 * into buffer <buf>. Only one call to recv() is performed, unless the
Willy Tarreau2ba44652012-08-20 17:30:32 +0200229 * buffer wraps, in which case a second call may be performed. The connection's
230 * flags are updated with whatever special event is detected (error, read0,
231 * empty). The caller is responsible for taking care of those events and
232 * avoiding the call if inappropriate. The function does not call the
233 * connection's polling update function, so the caller is responsible for this.
Willy Tarreauce3eda72013-12-05 00:49:40 +0100234 * errno is cleared before starting so that the caller knows that if it spots an
235 * error without errno, it's pending and can be retrieved via getsockopt(SO_ERROR).
Willy Tarreau2ba44652012-08-20 17:30:32 +0200236 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100237static size_t raw_sock_to_buf(struct connection *conn, void *xprt_ctx, struct buffer *buf, size_t count, int flags)
Willy Tarreau2ba44652012-08-20 17:30:32 +0200238{
Willy Tarreaubfc4d772018-07-18 11:22:03 +0200239 ssize_t ret;
240 size_t try, done = 0;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200241
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100242 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200243 return 0;
244
Willy Tarreau585744b2017-08-24 14:31:19 +0200245 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100246 return 0;
247
Willy Tarreaue2a0eec2020-01-17 09:59:40 +0100248 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +0100249 errno = 0;
250
Willy Tarreau585744b2017-08-24 14:31:19 +0200251 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200252 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +0200253 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200254 goto read0;
255
256 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +0200257 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100258 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200259 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200260 }
261 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200262
Willy Tarreau2ba44652012-08-20 17:30:32 +0200263 /* read the largest possible block. For this, we perform only one call
264 * to recv() unless the buffer wraps and we exactly fill the first hunk,
265 * in which case we accept to do it once again. A new attempt is made on
266 * EINTR too.
267 */
Willy Tarreauabf08d92014-01-14 11:31:27 +0100268 while (count > 0) {
Willy Tarreau591d4452018-06-15 17:21:00 +0200269 try = b_contig_space(buf);
270 if (!try)
271 break;
272
Willy Tarreauabf08d92014-01-14 11:31:27 +0100273 if (try > count)
274 try = count;
275
Willy Tarreau8f9c72d2018-06-07 18:46:28 +0200276 ret = recv(conn->handle.fd, b_tail(buf), try, 0);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200277
278 if (ret > 0) {
Olivier Houchardacd14032018-06-28 18:17:23 +0200279 b_add(buf, ret);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200280 done += ret;
281 if (ret < try) {
282 /* unfortunately, on level-triggered events, POLL_HUP
283 * is generally delivered AFTER the system buffer is
Willy Tarreau68128712017-03-13 12:04:34 +0100284 * empty, unless the poller supports POLL_RDHUP. If
285 * we know this is the case, we don't try to read more
286 * as we know there's no more available. Similarly, if
287 * there's no problem with lingering we don't even try
288 * to read an unlikely close from the client since we'll
289 * close first anyway.
Willy Tarreau2ba44652012-08-20 17:30:32 +0200290 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200291 if (fdtab[conn->handle.fd].ev & FD_POLL_HUP)
Willy Tarreau2ba44652012-08-20 17:30:32 +0200292 goto read0;
Willy Tarreau6c11bd22014-01-24 00:54:27 +0100293
Willy Tarreau585744b2017-08-24 14:31:19 +0200294 if ((!fdtab[conn->handle.fd].linger_risk) ||
Willy Tarreau68128712017-03-13 12:04:34 +0100295 (cur_poller.flags & HAP_POLL_F_RDHUP)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200296 fd_done_recv(conn->handle.fd);
Willy Tarreau68128712017-03-13 12:04:34 +0100297 break;
298 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200299 }
300 count -= ret;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200301 }
302 else if (ret == 0) {
303 goto read0;
304 }
Joshua M. Clulow07249032014-03-03 13:48:42 -0800305 else if (errno == EAGAIN || errno == ENOTCONN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200306 fd_cant_recv(conn->handle.fd);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200307 break;
308 }
309 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100310 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200311 break;
312 }
313 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200314
315 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
316 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200317
318 leave:
Willy Tarreau2ba44652012-08-20 17:30:32 +0200319 return done;
320
321 read0:
322 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200323 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreaudebdc4b2012-12-07 00:01:33 +0100324
325 /* Now a final check for a possible asynchronous low-level error
326 * report. This can happen when a connection receives a reset
327 * after a shutdown, both POLL_HUP and POLL_ERR are queued, and
328 * we might have come from there by just checking POLL_HUP instead
329 * of recv()'s return value 0, so we have no way to tell there was
330 * an error without checking.
331 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200332 if (unlikely(fdtab[conn->handle.fd].ev & FD_POLL_ERR))
Willy Tarreau26f4a042013-12-04 23:44:10 +0100333 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200334 goto leave;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200335}
336
337
Willy Tarreau787db9a2018-06-14 18:31:46 +0200338/* Send up to <count> pending bytes from buffer <buf> to connection <conn>'s
339 * socket. <flags> may contain some CO_SFL_* flags to hint the system about
340 * other pending data for example, but this flag is ignored at the moment.
Willy Tarreau5368d802012-08-21 18:22:06 +0200341 * Only one call to send() is performed, unless the buffer wraps, in which case
342 * a second call may be performed. The connection's flags are updated with
343 * whatever special event is detected (error, empty). The caller is responsible
344 * for taking care of those events and avoiding the call if inappropriate. The
345 * function does not call the connection's polling update function, so the caller
Willy Tarreau787db9a2018-06-14 18:31:46 +0200346 * is responsible for this. It's up to the caller to update the buffer's contents
347 * based on the return value.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200348 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100349static size_t raw_sock_from_buf(struct connection *conn, void *xprt_ctx, const struct buffer *buf, size_t count, int flags)
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100350{
Willy Tarreau787db9a2018-06-14 18:31:46 +0200351 ssize_t ret;
352 size_t try, done;
353 int send_flag;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200354
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100355 if (!conn_ctrl_ready(conn))
356 return 0;
357
Willy Tarreau585744b2017-08-24 14:31:19 +0200358 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200359 return 0;
360
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100361 if (conn->flags & CO_FL_SOCK_WR_SH) {
362 /* it's already closed */
363 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
364 errno = EPIPE;
365 return 0;
366 }
367
Willy Tarreau5368d802012-08-21 18:22:06 +0200368 done = 0;
369 /* send the largest possible block. For this we perform only one call
370 * to send() unless the buffer wraps and we exactly fill the first hunk,
371 * in which case we accept to do it once again.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100372 */
Willy Tarreau787db9a2018-06-14 18:31:46 +0200373 while (count) {
374 try = b_contig_data(buf, done);
375 if (try > count)
376 try = count;
Willy Tarreauf890dc92008-12-13 21:12:26 +0100377
Willy Tarreau5368d802012-08-21 18:22:06 +0200378 send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
Willy Tarreau787db9a2018-06-14 18:31:46 +0200379 if (try < count || flags & CO_SFL_MSG_MORE)
Willy Tarreau7e4086d2014-02-02 01:44:13 +0100380 send_flag |= MSG_MORE;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200381
Willy Tarreau787db9a2018-06-14 18:31:46 +0200382 ret = send(conn->handle.fd, b_peek(buf, done), try, send_flag);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200383
384 if (ret > 0) {
Willy Tarreau787db9a2018-06-14 18:31:46 +0200385 count -= ret;
Willy Tarreau5368d802012-08-21 18:22:06 +0200386 done += ret;
Willy Tarreaub38903c2008-11-23 21:33:29 +0100387
Willy Tarreauab3e1d32007-06-03 14:10:36 +0200388 /* if the system buffer is full, don't insist */
Willy Tarreau5368d802012-08-21 18:22:06 +0200389 if (ret < try)
Willy Tarreau6996e152007-04-30 14:37:43 +0200390 break;
Willy Tarreau3381bf82020-01-17 17:39:35 +0100391 if (!count)
Olivier Houchardac814742020-01-15 19:16:23 +0100392 conn_xprt_stop_send(conn);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200393 }
Willy Tarreau034c88c2017-01-23 23:36:45 +0100394 else if (ret == 0 || errno == EAGAIN || errno == ENOTCONN || errno == EINPROGRESS) {
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100395 /* nothing written, we need to poll for write first */
Willy Tarreau585744b2017-08-24 14:31:19 +0200396 fd_cant_send(conn->handle.fd);
Willy Tarreau5368d802012-08-21 18:22:06 +0200397 break;
Willy Tarreau83749182007-04-15 20:56:27 +0200398 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200399 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100400 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau5368d802012-08-21 18:22:06 +0200401 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200402 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200403 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200404 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200405 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200406 fd_cond_recv(conn->handle.fd);
407 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200408
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200409 if (done > 0) {
410 /* we count the total bytes sent, and the send rate for 32-byte
411 * blocks. The reason for the latter is that freq_ctr are
412 * limited to 4GB and that it's not enough per second.
413 */
414 _HA_ATOMIC_ADD(&global.out_bytes, done);
415 update_freq_ctr(&global.out_32bps, (done + 16) / 32);
416 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200417 return done;
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100418}
Willy Tarreau6996e152007-04-30 14:37:43 +0200419
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100420/* Called from the upper layer, to subscribe <es> to events <event_type>. The
421 * event subscriber <es> is not allowed to change from a previous call as long
422 * as at least one event is still subscribed. The <event_type> must only be a
423 * combination of SUB_RETRY_RECV and SUB_RETRY_SEND. It always returns 0.
424 */
425static int raw_sock_subscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100426{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100427 return conn_subscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100428}
429
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100430/* Called from the upper layer, to unsubscribe <es> from events <event_type>.
431 * The <es> pointer is not allowed to differ from the one passed to the
432 * subscribe() call. It always returns zero.
433 */
434static int raw_sock_unsubscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100435{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100436 return conn_unsubscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100437}
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100438
Olivier Houchard5149b592019-05-23 17:47:36 +0200439/* We can't have an underlying XPRT, so just return -1 to signify failure */
440static int raw_sock_remove_xprt(struct connection *conn, void *xprt_ctx, void *toremove_ctx, const struct xprt_ops *newops, void *newctx)
441{
442 /* This is the lowest xprt we can have, so if we get there we didn't
443 * find the xprt we wanted to remove, that's a bug
444 */
445 BUG_ON(1);
446 return -1;
447}
448
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200449/* transport-layer operations for RAW sockets */
Willy Tarreaud9f5cca2016-12-22 21:08:52 +0100450static struct xprt_ops raw_sock = {
Willy Tarreauc5788912012-08-24 18:12:41 +0200451 .snd_buf = raw_sock_from_buf,
452 .rcv_buf = raw_sock_to_buf,
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100453 .subscribe = raw_sock_subscribe,
454 .unsubscribe = raw_sock_unsubscribe,
Olivier Houchard5149b592019-05-23 17:47:36 +0200455 .remove_xprt = raw_sock_remove_xprt,
Willy Tarreaue5733232019-05-22 19:24:06 +0200456#if defined(USE_LINUX_SPLICE)
Willy Tarreau96199b12012-08-24 00:46:52 +0200457 .rcv_pipe = raw_sock_to_pipe,
458 .snd_pipe = raw_sock_from_pipe,
459#endif
Willy Tarreauc5788912012-08-24 18:12:41 +0200460 .shutr = NULL,
461 .shutw = NULL,
462 .close = NULL,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100463 .name = "RAW",
Willy Tarreau5c979a92012-05-07 17:15:39 +0200464};
Willy Tarreaubaaee002006-06-26 02:48:02 +0200465
Willy Tarreau13e14102016-12-22 20:25:26 +0100466
467__attribute__((constructor))
Olivier Houchard0d005932017-08-14 15:59:44 +0200468static void __raw_sock_init(void)
Willy Tarreau13e14102016-12-22 20:25:26 +0100469{
470 xprt_register(XPRT_RAW, &raw_sock);
471}
472
Willy Tarreaubaaee002006-06-26 02:48:02 +0200473/*
474 * Local variables:
475 * c-indent-level: 8
476 * c-basic-offset: 8
477 * End:
478 */