blob: f17d45d78399f516b802b3cdc240f383e7de22af [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 Tarreau5bd8c372009-01-19 00:32:22 +010046
47/* A pipe contains 16 segments max, and it's common to see segments of 1448 bytes
48 * because of timestamps. Use this as a hint for not looping on splice().
49 */
50#define SPLICE_FULL_HINT 16*1448
51
Willy Tarreaua9de3332009-11-28 07:47:10 +010052/* how many data we attempt to splice at once when the buffer is configured for
53 * infinite forwarding */
54#define MAX_SPLICE_AT_ONCE (1<<30)
55
Willy Tarreau5bd8c372009-01-19 00:32:22 +010056/* Returns :
Willy Tarreau96199b12012-08-24 00:46:52 +020057 * -1 if splice() is not supported
58 * >= 0 to report the amount of spliced bytes.
59 * connection flags are updated (error, read0, wait_room, wait_data).
60 * The caller must have previously allocated the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +010061 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +010062int raw_sock_to_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe, unsigned int count)
Willy Tarreau5bd8c372009-01-19 00:32:22 +010063{
Willy Tarreau31971e52009-09-20 12:07:52 +020064 int ret;
Willy Tarreauafad0e02012-08-09 14:45:22 +020065 int retval = 0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010066
Willy Tarreauf79c8172013-10-21 16:30:56 +020067
Willy Tarreaufd803bb2014-01-20 15:13:07 +010068 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +020069 return 0;
70
Willy Tarreau585744b2017-08-24 14:31:19 +020071 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +010072 return 0;
73
Willy Tarreaue2a0eec2020-01-17 09:59:40 +010074 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +010075 errno = 0;
76
Willy Tarreau96199b12012-08-24 00:46:52 +020077 /* Under Linux, if FD_POLL_HUP is set, we have reached the end.
78 * Since older splice() implementations were buggy and returned
79 * EAGAIN on end of read, let's bypass the call to splice() now.
80 */
Willy Tarreau585744b2017-08-24 14:31:19 +020081 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +020082 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +020083 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +020084 goto out_read0;
85
86 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +020087 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +010088 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreauce3eda72013-12-05 00:49:40 +010089 errno = 0; /* let the caller do a getsockopt() if it wants it */
Willy Tarreau256b9c52017-10-25 09:30:13 +020090 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +020091 }
92 }
Willy Tarreaua9de3332009-11-28 07:47:10 +010093
Willy Tarreau96199b12012-08-24 00:46:52 +020094 while (count) {
95 if (count > MAX_SPLICE_AT_ONCE)
96 count = MAX_SPLICE_AT_ONCE;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010097
Willy Tarreau585744b2017-08-24 14:31:19 +020098 ret = splice(conn->handle.fd, NULL, pipe->prod, NULL, count,
Willy Tarreau5bd8c372009-01-19 00:32:22 +010099 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
100
101 if (ret <= 0) {
Willy Tarreau38447472019-05-22 19:55:24 +0200102 if (ret == 0)
Willy Tarreau96199b12012-08-24 00:46:52 +0200103 goto out_read0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100104
105 if (errno == EAGAIN) {
106 /* there are two reasons for EAGAIN :
107 * - nothing in the socket buffer (standard)
108 * - pipe is full
Willy Tarreau38447472019-05-22 19:55:24 +0200109 * The difference between these two situations
110 * is problematic. Since we don't know if the
111 * pipe is full, we'll stop if the pipe is not
112 * empty. Anyway, we will almost always fill or
113 * empty the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100114 */
Willy Tarreau96199b12012-08-24 00:46:52 +0200115 if (pipe->data) {
116 /* alway stop reading until the pipe is flushed */
117 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100118 break;
119 }
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100120 /* socket buffer exhausted */
121 fd_cant_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100122 break;
123 }
Willy Tarreau45b88932012-11-12 12:00:09 +0100124 else if (errno == ENOSYS || errno == EINVAL || errno == EBADF) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200125 /* splice not supported on this end, disable it.
126 * We can safely return -1 since there is no
127 * chance that any data has been piped yet.
128 */
Willy Tarreau256b9c52017-10-25 09:30:13 +0200129 retval = -1;
130 goto leave;
Willy Tarreaudc340a92009-06-28 23:10:19 +0200131 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200132 else if (errno == EINTR) {
133 /* try again */
134 continue;
135 }
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100136 /* here we have another error */
Willy Tarreau96199b12012-08-24 00:46:52 +0200137 conn->flags |= CO_FL_ERROR;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100138 break;
139 } /* ret <= 0 */
140
Willy Tarreau96199b12012-08-24 00:46:52 +0200141 retval += ret;
142 pipe->data += ret;
Willy Tarreau4fc90ef2013-04-06 11:29:39 +0200143 count -= ret;
Willy Tarreaubaf2a502013-01-07 16:38:26 +0100144
Willy Tarreau96199b12012-08-24 00:46:52 +0200145 if (pipe->data >= SPLICE_FULL_HINT || ret >= global.tune.recv_enough) {
146 /* We've read enough of it for this time, let's stop before
147 * being asked to poll.
148 */
Willy Tarreau61d39a02013-07-18 21:49:32 +0200149 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100150 break;
151 }
152 } /* while */
153
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200154 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && retval)
155 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200156
157 leave:
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200158 if (retval > 0) {
159 /* we count the total bytes sent, and the send rate for 32-byte
160 * blocks. The reason for the latter is that freq_ctr are
161 * limited to 4GB and that it's not enough per second.
162 */
163 _HA_ATOMIC_ADD(&global.out_bytes, retval);
164 update_freq_ctr(&global.out_32bps, (retval + 16) / 32);
165 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200166 return retval;
Willy Tarreau3eba98a2009-01-25 13:56:13 +0100167
Willy Tarreau96199b12012-08-24 00:46:52 +0200168 out_read0:
169 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200170 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200171 goto leave;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100172}
173
Willy Tarreau96199b12012-08-24 00:46:52 +0200174/* Send as many bytes as possible from the pipe to the connection's socket.
175 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100176int raw_sock_from_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe)
Willy Tarreau96199b12012-08-24 00:46:52 +0200177{
178 int ret, done;
179
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100180 if (!conn_ctrl_ready(conn))
181 return 0;
182
Willy Tarreau585744b2017-08-24 14:31:19 +0200183 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200184 return 0;
185
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100186 if (conn->flags & CO_FL_SOCK_WR_SH) {
187 /* it's already closed */
188 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
189 errno = EPIPE;
190 return 0;
191 }
192
Willy Tarreau96199b12012-08-24 00:46:52 +0200193 done = 0;
194 while (pipe->data) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200195 ret = splice(pipe->cons, NULL, conn->handle.fd, NULL, pipe->data,
Willy Tarreau96199b12012-08-24 00:46:52 +0200196 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
197
198 if (ret <= 0) {
199 if (ret == 0 || errno == EAGAIN) {
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100200 fd_cant_send(conn->handle.fd);
Willy Tarreau96199b12012-08-24 00:46:52 +0200201 break;
202 }
203 else if (errno == EINTR)
204 continue;
205
206 /* here we have another error */
207 conn->flags |= CO_FL_ERROR;
208 break;
209 }
210
211 done += ret;
212 pipe->data -= ret;
213 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200214 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200215 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200216 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200217
Willy Tarreau96199b12012-08-24 00:46:52 +0200218 return done;
219}
220
Willy Tarreaue5733232019-05-22 19:24:06 +0200221#endif /* USE_LINUX_SPLICE */
Willy Tarreau6b4aad42009-01-18 21:59:13 +0100222
223
Willy Tarreau2ba44652012-08-20 17:30:32 +0200224/* Receive up to <count> bytes from connection <conn>'s socket and store them
Willy Tarreauabf08d92014-01-14 11:31:27 +0100225 * into buffer <buf>. Only one call to recv() is performed, unless the
Willy Tarreau2ba44652012-08-20 17:30:32 +0200226 * buffer wraps, in which case a second call may be performed. The connection's
227 * flags are updated with whatever special event is detected (error, read0,
228 * empty). The caller is responsible for taking care of those events and
229 * avoiding the call if inappropriate. The function does not call the
230 * connection's polling update function, so the caller is responsible for this.
Willy Tarreauce3eda72013-12-05 00:49:40 +0100231 * errno is cleared before starting so that the caller knows that if it spots an
232 * error without errno, it's pending and can be retrieved via getsockopt(SO_ERROR).
Willy Tarreau2ba44652012-08-20 17:30:32 +0200233 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100234static 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 +0200235{
Willy Tarreaubfc4d772018-07-18 11:22:03 +0200236 ssize_t ret;
237 size_t try, done = 0;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200238
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100239 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200240 return 0;
241
Willy Tarreau585744b2017-08-24 14:31:19 +0200242 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100243 return 0;
244
Willy Tarreaue2a0eec2020-01-17 09:59:40 +0100245 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +0100246 errno = 0;
247
Willy Tarreau585744b2017-08-24 14:31:19 +0200248 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200249 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +0200250 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200251 goto read0;
252
253 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +0200254 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100255 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200256 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200257 }
258 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200259
Willy Tarreau2ba44652012-08-20 17:30:32 +0200260 /* read the largest possible block. For this, we perform only one call
261 * to recv() unless the buffer wraps and we exactly fill the first hunk,
262 * in which case we accept to do it once again. A new attempt is made on
263 * EINTR too.
264 */
Willy Tarreauabf08d92014-01-14 11:31:27 +0100265 while (count > 0) {
Willy Tarreau591d4452018-06-15 17:21:00 +0200266 try = b_contig_space(buf);
267 if (!try)
268 break;
269
Willy Tarreauabf08d92014-01-14 11:31:27 +0100270 if (try > count)
271 try = count;
272
Willy Tarreau8f9c72d2018-06-07 18:46:28 +0200273 ret = recv(conn->handle.fd, b_tail(buf), try, 0);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200274
275 if (ret > 0) {
Olivier Houchardacd14032018-06-28 18:17:23 +0200276 b_add(buf, ret);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200277 done += ret;
278 if (ret < try) {
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100279 /* socket buffer exhausted */
280 fd_cant_recv(conn->handle.fd);
281
Willy Tarreau2ba44652012-08-20 17:30:32 +0200282 /* 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 Tarreau68128712017-03-13 12:04:34 +0100296 break;
297 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200298 }
299 count -= ret;
Willy Tarreau716bec22020-02-20 11:04:40 +0100300
301 if (flags & CO_RFL_READ_ONCE)
302 break;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200303 }
304 else if (ret == 0) {
305 goto read0;
306 }
Joshua M. Clulow07249032014-03-03 13:48:42 -0800307 else if (errno == EAGAIN || errno == ENOTCONN) {
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100308 /* socket buffer exhausted */
309 fd_cant_recv(conn->handle.fd);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200310 break;
311 }
312 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100313 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200314 break;
315 }
316 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200317
318 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
319 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200320
321 leave:
Willy Tarreau2ba44652012-08-20 17:30:32 +0200322 return done;
323
324 read0:
325 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200326 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreaudebdc4b2012-12-07 00:01:33 +0100327
328 /* Now a final check for a possible asynchronous low-level error
329 * report. This can happen when a connection receives a reset
330 * after a shutdown, both POLL_HUP and POLL_ERR are queued, and
331 * we might have come from there by just checking POLL_HUP instead
332 * of recv()'s return value 0, so we have no way to tell there was
333 * an error without checking.
334 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200335 if (unlikely(fdtab[conn->handle.fd].ev & FD_POLL_ERR))
Willy Tarreau26f4a042013-12-04 23:44:10 +0100336 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200337 goto leave;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200338}
339
340
Willy Tarreau787db9a2018-06-14 18:31:46 +0200341/* Send up to <count> pending bytes from buffer <buf> to connection <conn>'s
342 * socket. <flags> may contain some CO_SFL_* flags to hint the system about
343 * other pending data for example, but this flag is ignored at the moment.
Willy Tarreau5368d802012-08-21 18:22:06 +0200344 * Only one call to send() is performed, unless the buffer wraps, in which case
345 * a second call may be performed. The connection's flags are updated with
346 * whatever special event is detected (error, empty). The caller is responsible
347 * for taking care of those events and avoiding the call if inappropriate. The
348 * function does not call the connection's polling update function, so the caller
Willy Tarreau787db9a2018-06-14 18:31:46 +0200349 * is responsible for this. It's up to the caller to update the buffer's contents
350 * based on the return value.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200351 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100352static 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 +0100353{
Willy Tarreau787db9a2018-06-14 18:31:46 +0200354 ssize_t ret;
355 size_t try, done;
356 int send_flag;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200357
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100358 if (!conn_ctrl_ready(conn))
359 return 0;
360
Willy Tarreau585744b2017-08-24 14:31:19 +0200361 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200362 return 0;
363
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100364 if (conn->flags & CO_FL_SOCK_WR_SH) {
365 /* it's already closed */
366 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
367 errno = EPIPE;
368 return 0;
369 }
370
Willy Tarreau5368d802012-08-21 18:22:06 +0200371 done = 0;
372 /* send the largest possible block. For this we perform only one call
373 * to send() unless the buffer wraps and we exactly fill the first hunk,
374 * in which case we accept to do it once again.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100375 */
Willy Tarreau787db9a2018-06-14 18:31:46 +0200376 while (count) {
377 try = b_contig_data(buf, done);
378 if (try > count)
379 try = count;
Willy Tarreauf890dc92008-12-13 21:12:26 +0100380
Willy Tarreau5368d802012-08-21 18:22:06 +0200381 send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
Willy Tarreau787db9a2018-06-14 18:31:46 +0200382 if (try < count || flags & CO_SFL_MSG_MORE)
Willy Tarreau7e4086d2014-02-02 01:44:13 +0100383 send_flag |= MSG_MORE;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200384
Willy Tarreau787db9a2018-06-14 18:31:46 +0200385 ret = send(conn->handle.fd, b_peek(buf, done), try, send_flag);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200386
387 if (ret > 0) {
Willy Tarreau787db9a2018-06-14 18:31:46 +0200388 count -= ret;
Willy Tarreau5368d802012-08-21 18:22:06 +0200389 done += ret;
Willy Tarreaub38903c2008-11-23 21:33:29 +0100390
Willy Tarreauab3e1d32007-06-03 14:10:36 +0200391 /* if the system buffer is full, don't insist */
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100392 if (ret < try) {
393 fd_cant_send(conn->handle.fd);
Willy Tarreau6996e152007-04-30 14:37:43 +0200394 break;
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100395 }
Willy Tarreau3381bf82020-01-17 17:39:35 +0100396 if (!count)
Willy Tarreau3110eb72020-02-21 10:21:46 +0100397 fd_stop_send(conn->handle.fd);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200398 }
Willy Tarreau034c88c2017-01-23 23:36:45 +0100399 else if (ret == 0 || errno == EAGAIN || errno == ENOTCONN || errno == EINPROGRESS) {
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100400 /* nothing written, we need to poll for write first */
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100401 fd_cant_send(conn->handle.fd);
Willy Tarreau5368d802012-08-21 18:22:06 +0200402 break;
Willy Tarreau83749182007-04-15 20:56:27 +0200403 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200404 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100405 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau5368d802012-08-21 18:22:06 +0200406 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200407 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200408 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200409 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200410 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200411 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200412
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200413 if (done > 0) {
414 /* we count the total bytes sent, and the send rate for 32-byte
415 * blocks. The reason for the latter is that freq_ctr are
416 * limited to 4GB and that it's not enough per second.
417 */
418 _HA_ATOMIC_ADD(&global.out_bytes, done);
419 update_freq_ctr(&global.out_32bps, (done + 16) / 32);
420 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200421 return done;
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100422}
Willy Tarreau6996e152007-04-30 14:37:43 +0200423
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100424/* Called from the upper layer, to subscribe <es> to events <event_type>. The
425 * event subscriber <es> is not allowed to change from a previous call as long
426 * as at least one event is still subscribed. The <event_type> must only be a
427 * combination of SUB_RETRY_RECV and SUB_RETRY_SEND. It always returns 0.
428 */
429static int raw_sock_subscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100430{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100431 return conn_subscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100432}
433
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100434/* Called from the upper layer, to unsubscribe <es> from events <event_type>.
435 * The <es> pointer is not allowed to differ from the one passed to the
436 * subscribe() call. It always returns zero.
437 */
438static int raw_sock_unsubscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100439{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100440 return conn_unsubscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100441}
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100442
Olivier Houchard5149b592019-05-23 17:47:36 +0200443/* We can't have an underlying XPRT, so just return -1 to signify failure */
444static int raw_sock_remove_xprt(struct connection *conn, void *xprt_ctx, void *toremove_ctx, const struct xprt_ops *newops, void *newctx)
445{
446 /* This is the lowest xprt we can have, so if we get there we didn't
447 * find the xprt we wanted to remove, that's a bug
448 */
449 BUG_ON(1);
450 return -1;
451}
452
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200453/* transport-layer operations for RAW sockets */
Willy Tarreaud9f5cca2016-12-22 21:08:52 +0100454static struct xprt_ops raw_sock = {
Willy Tarreauc5788912012-08-24 18:12:41 +0200455 .snd_buf = raw_sock_from_buf,
456 .rcv_buf = raw_sock_to_buf,
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100457 .subscribe = raw_sock_subscribe,
458 .unsubscribe = raw_sock_unsubscribe,
Olivier Houchard5149b592019-05-23 17:47:36 +0200459 .remove_xprt = raw_sock_remove_xprt,
Willy Tarreaue5733232019-05-22 19:24:06 +0200460#if defined(USE_LINUX_SPLICE)
Willy Tarreau96199b12012-08-24 00:46:52 +0200461 .rcv_pipe = raw_sock_to_pipe,
462 .snd_pipe = raw_sock_from_pipe,
463#endif
Willy Tarreauc5788912012-08-24 18:12:41 +0200464 .shutr = NULL,
465 .shutw = NULL,
466 .close = NULL,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100467 .name = "RAW",
Willy Tarreau5c979a92012-05-07 17:15:39 +0200468};
Willy Tarreaubaaee002006-06-26 02:48:02 +0200469
Willy Tarreau13e14102016-12-22 20:25:26 +0100470
471__attribute__((constructor))
Olivier Houchard0d005932017-08-14 15:59:44 +0200472static void __raw_sock_init(void)
Willy Tarreau13e14102016-12-22 20:25:26 +0100473{
474 xprt_register(XPRT_RAW, &raw_sock);
475}
476
Willy Tarreaubaaee002006-06-26 02:48:02 +0200477/*
478 * Local variables:
479 * c-indent-level: 8
480 * c-basic-offset: 8
481 * End:
482 */