blob: d8a27cd617f0120f59dcd374c371f8cd40124b74 [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 Tarreau847271d2020-01-17 09:59:40 +010072 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreau256b9c52017-10-25 09:30:13 +020073 conn_refresh_polling_flags(conn);
Willy Tarreauce3eda72013-12-05 00:49:40 +010074 errno = 0;
75
Willy Tarreau96199b12012-08-24 00:46:52 +020076 /* Under Linux, if FD_POLL_HUP is set, we have reached the end.
77 * Since older splice() implementations were buggy and returned
78 * EAGAIN on end of read, let's bypass the call to splice() now.
79 */
Willy Tarreau585744b2017-08-24 14:31:19 +020080 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +020081 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +020082 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +020083 goto out_read0;
84
85 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +020086 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +010087 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreauce3eda72013-12-05 00:49:40 +010088 errno = 0; /* let the caller do a getsockopt() if it wants it */
Willy Tarreau256b9c52017-10-25 09:30:13 +020089 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +020090 }
91 }
Willy Tarreaua9de3332009-11-28 07:47:10 +010092
Willy Tarreau96199b12012-08-24 00:46:52 +020093 while (count) {
94 if (count > MAX_SPLICE_AT_ONCE)
95 count = MAX_SPLICE_AT_ONCE;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010096
Willy Tarreau585744b2017-08-24 14:31:19 +020097 ret = splice(conn->handle.fd, NULL, pipe->prod, NULL, count,
Willy Tarreau5bd8c372009-01-19 00:32:22 +010098 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
99
100 if (ret <= 0) {
Willy Tarreau38447472019-05-22 19:55:24 +0200101 if (ret == 0)
Willy Tarreau96199b12012-08-24 00:46:52 +0200102 goto out_read0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100103
104 if (errno == EAGAIN) {
105 /* there are two reasons for EAGAIN :
106 * - nothing in the socket buffer (standard)
107 * - pipe is full
Willy Tarreau38447472019-05-22 19:55:24 +0200108 * The difference between these two situations
109 * is problematic. Since we don't know if the
110 * pipe is full, we'll stop if the pipe is not
111 * empty. Anyway, we will almost always fill or
112 * empty the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100113 */
Willy Tarreau96199b12012-08-24 00:46:52 +0200114 if (pipe->data) {
115 /* alway stop reading until the pipe is flushed */
116 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100117 break;
118 }
119
Willy Tarreau38447472019-05-22 19:55:24 +0200120 fd_cant_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100121 break;
122 }
Willy Tarreau45b88932012-11-12 12:00:09 +0100123 else if (errno == ENOSYS || errno == EINVAL || errno == EBADF) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200124 /* splice not supported on this end, disable it.
125 * We can safely return -1 since there is no
126 * chance that any data has been piped yet.
127 */
Willy Tarreau256b9c52017-10-25 09:30:13 +0200128 retval = -1;
129 goto leave;
Willy Tarreaudc340a92009-06-28 23:10:19 +0200130 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200131 else if (errno == EINTR) {
132 /* try again */
133 continue;
134 }
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100135 /* here we have another error */
Willy Tarreau96199b12012-08-24 00:46:52 +0200136 conn->flags |= CO_FL_ERROR;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100137 break;
138 } /* ret <= 0 */
139
Willy Tarreau96199b12012-08-24 00:46:52 +0200140 retval += ret;
141 pipe->data += ret;
Willy Tarreau4fc90ef2013-04-06 11:29:39 +0200142 count -= ret;
Willy Tarreaubaf2a502013-01-07 16:38:26 +0100143
Willy Tarreau96199b12012-08-24 00:46:52 +0200144 if (pipe->data >= SPLICE_FULL_HINT || ret >= global.tune.recv_enough) {
145 /* We've read enough of it for this time, let's stop before
146 * being asked to poll.
147 */
Willy Tarreau61d39a02013-07-18 21:49:32 +0200148 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau585744b2017-08-24 14:31:19 +0200149 fd_done_recv(conn->handle.fd);
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 Tarreau256b9c52017-10-25 09:30:13 +0200186 conn_refresh_polling_flags(conn);
Willy Tarreau96199b12012-08-24 00:46:52 +0200187 done = 0;
188 while (pipe->data) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200189 ret = splice(pipe->cons, NULL, conn->handle.fd, NULL, pipe->data,
Willy Tarreau96199b12012-08-24 00:46:52 +0200190 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
191
192 if (ret <= 0) {
193 if (ret == 0 || errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200194 fd_cant_send(conn->handle.fd);
Willy Tarreau96199b12012-08-24 00:46:52 +0200195 break;
196 }
197 else if (errno == EINTR)
198 continue;
199
200 /* here we have another error */
201 conn->flags |= CO_FL_ERROR;
202 break;
203 }
204
205 done += ret;
206 pipe->data -= ret;
207 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200208 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
209 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200210
Willy Tarreau96199b12012-08-24 00:46:52 +0200211 return done;
212}
213
Willy Tarreaue5733232019-05-22 19:24:06 +0200214#endif /* USE_LINUX_SPLICE */
Willy Tarreau6b4aad42009-01-18 21:59:13 +0100215
216
Willy Tarreau2ba44652012-08-20 17:30:32 +0200217/* Receive up to <count> bytes from connection <conn>'s socket and store them
Willy Tarreauabf08d92014-01-14 11:31:27 +0100218 * into buffer <buf>. Only one call to recv() is performed, unless the
Willy Tarreau2ba44652012-08-20 17:30:32 +0200219 * buffer wraps, in which case a second call may be performed. The connection's
220 * flags are updated with whatever special event is detected (error, read0,
221 * empty). The caller is responsible for taking care of those events and
222 * avoiding the call if inappropriate. The function does not call the
223 * connection's polling update function, so the caller is responsible for this.
Willy Tarreauce3eda72013-12-05 00:49:40 +0100224 * errno is cleared before starting so that the caller knows that if it spots an
225 * error without errno, it's pending and can be retrieved via getsockopt(SO_ERROR).
Willy Tarreau2ba44652012-08-20 17:30:32 +0200226 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100227static 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 +0200228{
Willy Tarreaubfc4d772018-07-18 11:22:03 +0200229 ssize_t ret;
230 size_t try, done = 0;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200231
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100232 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200233 return 0;
234
Willy Tarreau585744b2017-08-24 14:31:19 +0200235 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100236 return 0;
237
Willy Tarreau847271d2020-01-17 09:59:40 +0100238 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200239 conn_refresh_polling_flags(conn);
Willy Tarreauce3eda72013-12-05 00:49:40 +0100240 errno = 0;
241
Willy Tarreau585744b2017-08-24 14:31:19 +0200242 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200243 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +0200244 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200245 goto read0;
246
247 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +0200248 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100249 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200250 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200251 }
252 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200253
Willy Tarreau2ba44652012-08-20 17:30:32 +0200254 /* read the largest possible block. For this, we perform only one call
255 * to recv() unless the buffer wraps and we exactly fill the first hunk,
256 * in which case we accept to do it once again. A new attempt is made on
257 * EINTR too.
258 */
Willy Tarreauabf08d92014-01-14 11:31:27 +0100259 while (count > 0) {
Willy Tarreau591d4452018-06-15 17:21:00 +0200260 try = b_contig_space(buf);
261 if (!try)
262 break;
263
Willy Tarreauabf08d92014-01-14 11:31:27 +0100264 if (try > count)
265 try = count;
266
Willy Tarreau8f9c72d2018-06-07 18:46:28 +0200267 ret = recv(conn->handle.fd, b_tail(buf), try, 0);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200268
269 if (ret > 0) {
Olivier Houchardacd14032018-06-28 18:17:23 +0200270 b_add(buf, ret);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200271 done += ret;
272 if (ret < try) {
273 /* unfortunately, on level-triggered events, POLL_HUP
274 * is generally delivered AFTER the system buffer is
Willy Tarreau68128712017-03-13 12:04:34 +0100275 * empty, unless the poller supports POLL_RDHUP. If
276 * we know this is the case, we don't try to read more
277 * as we know there's no more available. Similarly, if
278 * there's no problem with lingering we don't even try
279 * to read an unlikely close from the client since we'll
280 * close first anyway.
Willy Tarreau2ba44652012-08-20 17:30:32 +0200281 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200282 if (fdtab[conn->handle.fd].ev & FD_POLL_HUP)
Willy Tarreau2ba44652012-08-20 17:30:32 +0200283 goto read0;
Willy Tarreau6c11bd22014-01-24 00:54:27 +0100284
Willy Tarreau585744b2017-08-24 14:31:19 +0200285 if ((!fdtab[conn->handle.fd].linger_risk) ||
Willy Tarreau68128712017-03-13 12:04:34 +0100286 (cur_poller.flags & HAP_POLL_F_RDHUP)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200287 fd_done_recv(conn->handle.fd);
Willy Tarreau68128712017-03-13 12:04:34 +0100288 break;
289 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200290 }
291 count -= ret;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200292 }
293 else if (ret == 0) {
294 goto read0;
295 }
Joshua M. Clulow07249032014-03-03 13:48:42 -0800296 else if (errno == EAGAIN || errno == ENOTCONN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200297 fd_cant_recv(conn->handle.fd);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200298 break;
299 }
300 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100301 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200302 break;
303 }
304 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200305
306 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
307 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200308
309 leave:
Willy Tarreau2ba44652012-08-20 17:30:32 +0200310 return done;
311
312 read0:
313 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200314 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreaudebdc4b2012-12-07 00:01:33 +0100315
316 /* Now a final check for a possible asynchronous low-level error
317 * report. This can happen when a connection receives a reset
318 * after a shutdown, both POLL_HUP and POLL_ERR are queued, and
319 * we might have come from there by just checking POLL_HUP instead
320 * of recv()'s return value 0, so we have no way to tell there was
321 * an error without checking.
322 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200323 if (unlikely(fdtab[conn->handle.fd].ev & FD_POLL_ERR))
Willy Tarreau26f4a042013-12-04 23:44:10 +0100324 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200325 goto leave;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200326}
327
328
Willy Tarreau787db9a2018-06-14 18:31:46 +0200329/* Send up to <count> pending bytes from buffer <buf> to connection <conn>'s
330 * socket. <flags> may contain some CO_SFL_* flags to hint the system about
331 * other pending data for example, but this flag is ignored at the moment.
Willy Tarreau5368d802012-08-21 18:22:06 +0200332 * Only one call to send() is performed, unless the buffer wraps, in which case
333 * a second call may be performed. The connection's flags are updated with
334 * whatever special event is detected (error, empty). The caller is responsible
335 * for taking care of those events and avoiding the call if inappropriate. The
336 * function does not call the connection's polling update function, so the caller
Willy Tarreau787db9a2018-06-14 18:31:46 +0200337 * is responsible for this. It's up to the caller to update the buffer's contents
338 * based on the return value.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200339 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100340static 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 +0100341{
Willy Tarreau787db9a2018-06-14 18:31:46 +0200342 ssize_t ret;
343 size_t try, done;
344 int send_flag;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200345
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100346 if (!conn_ctrl_ready(conn))
347 return 0;
348
Willy Tarreau585744b2017-08-24 14:31:19 +0200349 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200350 return 0;
351
Willy Tarreau256b9c52017-10-25 09:30:13 +0200352 conn_refresh_polling_flags(conn);
Willy Tarreau5368d802012-08-21 18:22:06 +0200353 done = 0;
354 /* send the largest possible block. For this we perform only one call
355 * to send() unless the buffer wraps and we exactly fill the first hunk,
356 * in which case we accept to do it once again.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100357 */
Willy Tarreau787db9a2018-06-14 18:31:46 +0200358 while (count) {
359 try = b_contig_data(buf, done);
360 if (try > count)
361 try = count;
Willy Tarreauf890dc92008-12-13 21:12:26 +0100362
Willy Tarreau5368d802012-08-21 18:22:06 +0200363 send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
Willy Tarreau787db9a2018-06-14 18:31:46 +0200364 if (try < count || flags & CO_SFL_MSG_MORE)
Willy Tarreau7e4086d2014-02-02 01:44:13 +0100365 send_flag |= MSG_MORE;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200366
Willy Tarreau787db9a2018-06-14 18:31:46 +0200367 ret = send(conn->handle.fd, b_peek(buf, done), try, send_flag);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200368
369 if (ret > 0) {
Willy Tarreau787db9a2018-06-14 18:31:46 +0200370 count -= ret;
Willy Tarreau5368d802012-08-21 18:22:06 +0200371 done += ret;
Willy Tarreaub38903c2008-11-23 21:33:29 +0100372
Olivier Houchardf24502b2019-01-17 19:09:11 +0100373 /* A send succeeded, so we can consier ourself connected */
374 conn->flags |= CO_FL_CONNECTED;
Willy Tarreauab3e1d32007-06-03 14:10:36 +0200375 /* if the system buffer is full, don't insist */
Willy Tarreau5368d802012-08-21 18:22:06 +0200376 if (ret < try)
Willy Tarreau6996e152007-04-30 14:37:43 +0200377 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200378 }
Willy Tarreau034c88c2017-01-23 23:36:45 +0100379 else if (ret == 0 || errno == EAGAIN || errno == ENOTCONN || errno == EINPROGRESS) {
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100380 /* nothing written, we need to poll for write first */
Willy Tarreau585744b2017-08-24 14:31:19 +0200381 fd_cant_send(conn->handle.fd);
Willy Tarreau5368d802012-08-21 18:22:06 +0200382 break;
Willy Tarreau83749182007-04-15 20:56:27 +0200383 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200384 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100385 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau5368d802012-08-21 18:22:06 +0200386 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200387 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200388 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200389 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
390 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200391
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200392 if (done > 0) {
393 /* we count the total bytes sent, and the send rate for 32-byte
394 * blocks. The reason for the latter is that freq_ctr are
395 * limited to 4GB and that it's not enough per second.
396 */
397 _HA_ATOMIC_ADD(&global.out_bytes, done);
398 update_freq_ctr(&global.out_32bps, (done + 16) / 32);
399 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200400 return done;
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100401}
Willy Tarreau6996e152007-04-30 14:37:43 +0200402
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100403static int raw_sock_subscribe(struct connection *conn, void *xprt_ctx, int event_type, void *param)
404{
405 return conn_subscribe(conn, xprt_ctx, event_type, param);
406}
407
408static int raw_sock_unsubscribe(struct connection *conn, void *xprt_ctx, int event_type, void *param)
409{
410 return conn_unsubscribe(conn, xprt_ctx, event_type, param);
411}
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100412
Olivier Houchard5149b592019-05-23 17:47:36 +0200413/* We can't have an underlying XPRT, so just return -1 to signify failure */
414static int raw_sock_remove_xprt(struct connection *conn, void *xprt_ctx, void *toremove_ctx, const struct xprt_ops *newops, void *newctx)
415{
416 /* This is the lowest xprt we can have, so if we get there we didn't
417 * find the xprt we wanted to remove, that's a bug
418 */
419 BUG_ON(1);
420 return -1;
421}
422
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200423/* transport-layer operations for RAW sockets */
Willy Tarreaud9f5cca2016-12-22 21:08:52 +0100424static struct xprt_ops raw_sock = {
Willy Tarreauc5788912012-08-24 18:12:41 +0200425 .snd_buf = raw_sock_from_buf,
426 .rcv_buf = raw_sock_to_buf,
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100427 .subscribe = raw_sock_subscribe,
428 .unsubscribe = raw_sock_unsubscribe,
Olivier Houchard5149b592019-05-23 17:47:36 +0200429 .remove_xprt = raw_sock_remove_xprt,
Willy Tarreaue5733232019-05-22 19:24:06 +0200430#if defined(USE_LINUX_SPLICE)
Willy Tarreau96199b12012-08-24 00:46:52 +0200431 .rcv_pipe = raw_sock_to_pipe,
432 .snd_pipe = raw_sock_from_pipe,
433#endif
Willy Tarreauc5788912012-08-24 18:12:41 +0200434 .shutr = NULL,
435 .shutw = NULL,
436 .close = NULL,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100437 .name = "RAW",
Willy Tarreau5c979a92012-05-07 17:15:39 +0200438};
Willy Tarreaubaaee002006-06-26 02:48:02 +0200439
Willy Tarreau13e14102016-12-22 20:25:26 +0100440
441__attribute__((constructor))
Olivier Houchard0d005932017-08-14 15:59:44 +0200442static void __raw_sock_init(void)
Willy Tarreau13e14102016-12-22 20:25:26 +0100443{
444 xprt_register(XPRT_RAW, &raw_sock);
445}
446
Willy Tarreaubaaee002006-06-26 02:48:02 +0200447/*
448 * Local variables:
449 * c-indent-level: 8
450 * c-basic-offset: 8
451 * End:
452 */