blob: 406fef35d1054ae5675e99d0a2083d1756b126e8 [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 Tarreau96199b12012-08-24 00:46:52 +020045#if defined(CONFIG_HAP_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 Tarreaub6daedd2013-01-07 16:57:09 +010057/* Versions of splice between 2.6.25 and 2.6.27.12 were bogus and would return EAGAIN
58 * on incoming shutdowns. On these versions, we have to call recv() after such a return
59 * in order to find whether splice is OK or not. Since 2.6.27.13 we don't need to do
60 * this anymore, and we can avoid this logic by defining ASSUME_SPLICE_WORKS.
61 */
62
Willy Tarreau5bd8c372009-01-19 00:32:22 +010063/* Returns :
Willy Tarreau96199b12012-08-24 00:46:52 +020064 * -1 if splice() is not supported
65 * >= 0 to report the amount of spliced bytes.
66 * connection flags are updated (error, read0, wait_room, wait_data).
67 * The caller must have previously allocated the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +010068 */
Willy Tarreau96199b12012-08-24 00:46:52 +020069int raw_sock_to_pipe(struct connection *conn, struct pipe *pipe, unsigned int count)
Willy Tarreau5bd8c372009-01-19 00:32:22 +010070{
Willy Tarreaub6daedd2013-01-07 16:57:09 +010071#ifndef ASSUME_SPLICE_WORKS
Willy Tarreau82a04562011-12-11 22:37:06 +010072 static int splice_detects_close;
Willy Tarreaub6daedd2013-01-07 16:57:09 +010073#endif
Willy Tarreau31971e52009-09-20 12:07:52 +020074 int ret;
Willy Tarreauafad0e02012-08-09 14:45:22 +020075 int retval = 0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010076
Willy Tarreauf79c8172013-10-21 16:30:56 +020077
Willy Tarreaufd803bb2014-01-20 15:13:07 +010078 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +020079 return 0;
80
Willy Tarreau585744b2017-08-24 14:31:19 +020081 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +010082 return 0;
83
Willy Tarreauce3eda72013-12-05 00:49:40 +010084 errno = 0;
85
Willy Tarreau96199b12012-08-24 00:46:52 +020086 /* Under Linux, if FD_POLL_HUP is set, we have reached the end.
87 * Since older splice() implementations were buggy and returned
88 * EAGAIN on end of read, let's bypass the call to splice() now.
89 */
Willy Tarreau585744b2017-08-24 14:31:19 +020090 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +020091 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +020092 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +020093 goto out_read0;
94
95 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +020096 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +010097 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreauce3eda72013-12-05 00:49:40 +010098 errno = 0; /* let the caller do a getsockopt() if it wants it */
Willy Tarreau6f5d1412012-10-04 20:38:49 +020099 return retval;
100 }
101 }
Willy Tarreaua9de3332009-11-28 07:47:10 +0100102
Willy Tarreau96199b12012-08-24 00:46:52 +0200103 while (count) {
104 if (count > MAX_SPLICE_AT_ONCE)
105 count = MAX_SPLICE_AT_ONCE;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100106
Willy Tarreau585744b2017-08-24 14:31:19 +0200107 ret = splice(conn->handle.fd, NULL, pipe->prod, NULL, count,
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100108 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
109
110 if (ret <= 0) {
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100111 if (ret == 0) {
Willy Tarreau98b306b2009-01-25 11:11:32 +0100112 /* connection closed. This is only detected by
Willy Tarreau82a04562011-12-11 22:37:06 +0100113 * recent kernels (>= 2.6.27.13). If we notice
114 * it works, we store the info for later use.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100115 */
Willy Tarreaub6daedd2013-01-07 16:57:09 +0100116#ifndef ASSUME_SPLICE_WORKS
Willy Tarreau82a04562011-12-11 22:37:06 +0100117 splice_detects_close = 1;
Willy Tarreaub6daedd2013-01-07 16:57:09 +0100118#endif
Willy Tarreau96199b12012-08-24 00:46:52 +0200119 goto out_read0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100120 }
121
122 if (errno == EAGAIN) {
123 /* there are two reasons for EAGAIN :
124 * - nothing in the socket buffer (standard)
125 * - pipe is full
Willy Tarreau98b306b2009-01-25 11:11:32 +0100126 * - the connection is closed (kernel < 2.6.27.13)
Willy Tarreau96199b12012-08-24 00:46:52 +0200127 * The last case is annoying but know if we can detect it
128 * and if we can't then we rely on the call to recv() to
129 * get a valid verdict. The difference between the first
130 * two situations is problematic. Since we don't know if
131 * the pipe is full, we'll stop if the pipe is not empty.
132 * Anyway, we will almost always fill/empty the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100133 */
Willy Tarreau96199b12012-08-24 00:46:52 +0200134 if (pipe->data) {
135 /* alway stop reading until the pipe is flushed */
136 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100137 break;
138 }
139
Willy Tarreau82a04562011-12-11 22:37:06 +0100140 /* We don't know if the connection was closed,
141 * but if we know splice detects close, then we
142 * know it for sure.
Willy Tarreau98b306b2009-01-25 11:11:32 +0100143 * But if we're called upon POLLIN with an empty
Willy Tarreau82a04562011-12-11 22:37:06 +0100144 * pipe and get EAGAIN, it is suspect enough to
Willy Tarreau98b306b2009-01-25 11:11:32 +0100145 * try to fall back to the normal recv scheme
146 * which will be able to deal with the situation.
147 */
Willy Tarreaub6daedd2013-01-07 16:57:09 +0100148#ifndef ASSUME_SPLICE_WORKS
Willy Tarreau82a04562011-12-11 22:37:06 +0100149 if (splice_detects_close)
Willy Tarreaub6daedd2013-01-07 16:57:09 +0100150#endif
Willy Tarreau585744b2017-08-24 14:31:19 +0200151 fd_cant_recv(conn->handle.fd); /* we know for sure that it's EAGAIN */
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100152 break;
153 }
Willy Tarreau45b88932012-11-12 12:00:09 +0100154 else if (errno == ENOSYS || errno == EINVAL || errno == EBADF) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200155 /* splice not supported on this end, disable it.
156 * We can safely return -1 since there is no
157 * chance that any data has been piped yet.
158 */
Willy Tarreaudc340a92009-06-28 23:10:19 +0200159 return -1;
160 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200161 else if (errno == EINTR) {
162 /* try again */
163 continue;
164 }
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100165 /* here we have another error */
Willy Tarreau96199b12012-08-24 00:46:52 +0200166 conn->flags |= CO_FL_ERROR;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100167 break;
168 } /* ret <= 0 */
169
Willy Tarreau96199b12012-08-24 00:46:52 +0200170 retval += ret;
171 pipe->data += ret;
Willy Tarreau4fc90ef2013-04-06 11:29:39 +0200172 count -= ret;
Willy Tarreaubaf2a502013-01-07 16:38:26 +0100173
Willy Tarreau96199b12012-08-24 00:46:52 +0200174 if (pipe->data >= SPLICE_FULL_HINT || ret >= global.tune.recv_enough) {
175 /* We've read enough of it for this time, let's stop before
176 * being asked to poll.
177 */
Willy Tarreau61d39a02013-07-18 21:49:32 +0200178 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau585744b2017-08-24 14:31:19 +0200179 fd_done_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100180 break;
181 }
182 } /* while */
183
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200184 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && retval)
185 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau96199b12012-08-24 00:46:52 +0200186 return retval;
Willy Tarreau3eba98a2009-01-25 13:56:13 +0100187
Willy Tarreau96199b12012-08-24 00:46:52 +0200188 out_read0:
189 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200190 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100191 return retval;
192}
193
Willy Tarreau96199b12012-08-24 00:46:52 +0200194/* Send as many bytes as possible from the pipe to the connection's socket.
195 */
196int raw_sock_from_pipe(struct connection *conn, struct pipe *pipe)
197{
198 int ret, done;
199
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100200 if (!conn_ctrl_ready(conn))
201 return 0;
202
Willy Tarreau585744b2017-08-24 14:31:19 +0200203 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200204 return 0;
205
Willy Tarreau96199b12012-08-24 00:46:52 +0200206 done = 0;
207 while (pipe->data) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200208 ret = splice(pipe->cons, NULL, conn->handle.fd, NULL, pipe->data,
Willy Tarreau96199b12012-08-24 00:46:52 +0200209 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
210
211 if (ret <= 0) {
212 if (ret == 0 || errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200213 fd_cant_send(conn->handle.fd);
Willy Tarreau96199b12012-08-24 00:46:52 +0200214 break;
215 }
216 else if (errno == EINTR)
217 continue;
218
219 /* here we have another error */
220 conn->flags |= CO_FL_ERROR;
221 break;
222 }
223
224 done += ret;
225 pipe->data -= ret;
226 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200227 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
228 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau96199b12012-08-24 00:46:52 +0200229 return done;
230}
231
Willy Tarreau6b4aad42009-01-18 21:59:13 +0100232#endif /* CONFIG_HAP_LINUX_SPLICE */
233
234
Willy Tarreau2ba44652012-08-20 17:30:32 +0200235/* Receive up to <count> bytes from connection <conn>'s socket and store them
Willy Tarreauabf08d92014-01-14 11:31:27 +0100236 * into buffer <buf>. Only one call to recv() is performed, unless the
Willy Tarreau2ba44652012-08-20 17:30:32 +0200237 * buffer wraps, in which case a second call may be performed. The connection's
238 * flags are updated with whatever special event is detected (error, read0,
239 * empty). The caller is responsible for taking care of those events and
240 * avoiding the call if inappropriate. The function does not call the
241 * connection's polling update function, so the caller is responsible for this.
Willy Tarreauce3eda72013-12-05 00:49:40 +0100242 * errno is cleared before starting so that the caller knows that if it spots an
243 * error without errno, it's pending and can be retrieved via getsockopt(SO_ERROR).
Willy Tarreau2ba44652012-08-20 17:30:32 +0200244 */
245static int raw_sock_to_buf(struct connection *conn, struct buffer *buf, int count)
246{
247 int ret, done = 0;
Willy Tarreauabf08d92014-01-14 11:31:27 +0100248 int try;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200249
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100250 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200251 return 0;
252
Willy Tarreau585744b2017-08-24 14:31:19 +0200253 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100254 return 0;
255
Willy Tarreauce3eda72013-12-05 00:49:40 +0100256 errno = 0;
257
Willy Tarreau585744b2017-08-24 14:31:19 +0200258 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200259 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +0200260 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200261 goto read0;
262
263 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +0200264 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100265 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200266 return done;
267 }
268 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200269
Willy Tarreauabf08d92014-01-14 11:31:27 +0100270 /* let's realign the buffer to optimize I/O */
271 if (buffer_empty(buf))
Willy Tarreau2ba44652012-08-20 17:30:32 +0200272 buf->p = buf->data;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200273
274 /* read the largest possible block. For this, we perform only one call
275 * to recv() unless the buffer wraps and we exactly fill the first hunk,
276 * in which case we accept to do it once again. A new attempt is made on
277 * EINTR too.
278 */
Willy Tarreauabf08d92014-01-14 11:31:27 +0100279 while (count > 0) {
280 /* first check if we have some room after p+i */
281 try = buf->data + buf->size - (buf->p + buf->i);
282 /* otherwise continue between data and p-o */
283 if (try <= 0) {
284 try = buf->p - (buf->data + buf->o);
285 if (try <= 0)
286 break;
287 }
288 if (try > count)
289 try = count;
290
Willy Tarreau585744b2017-08-24 14:31:19 +0200291 ret = recv(conn->handle.fd, bi_end(buf), try, 0);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200292
293 if (ret > 0) {
294 buf->i += ret;
295 done += ret;
296 if (ret < try) {
297 /* unfortunately, on level-triggered events, POLL_HUP
298 * is generally delivered AFTER the system buffer is
Willy Tarreau68128712017-03-13 12:04:34 +0100299 * empty, unless the poller supports POLL_RDHUP. If
300 * we know this is the case, we don't try to read more
301 * as we know there's no more available. Similarly, if
302 * there's no problem with lingering we don't even try
303 * to read an unlikely close from the client since we'll
304 * close first anyway.
Willy Tarreau2ba44652012-08-20 17:30:32 +0200305 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200306 if (fdtab[conn->handle.fd].ev & FD_POLL_HUP)
Willy Tarreau2ba44652012-08-20 17:30:32 +0200307 goto read0;
Willy Tarreau6c11bd22014-01-24 00:54:27 +0100308
Willy Tarreau585744b2017-08-24 14:31:19 +0200309 if ((!fdtab[conn->handle.fd].linger_risk) ||
Willy Tarreau68128712017-03-13 12:04:34 +0100310 (cur_poller.flags & HAP_POLL_F_RDHUP)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200311 fd_done_recv(conn->handle.fd);
Willy Tarreau68128712017-03-13 12:04:34 +0100312 break;
313 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200314 }
315 count -= ret;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200316 }
317 else if (ret == 0) {
318 goto read0;
319 }
Joshua M. Clulow07249032014-03-03 13:48:42 -0800320 else if (errno == EAGAIN || errno == ENOTCONN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200321 fd_cant_recv(conn->handle.fd);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200322 break;
323 }
324 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100325 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200326 break;
327 }
328 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200329
330 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
331 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200332 return done;
333
334 read0:
335 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200336 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreaudebdc4b2012-12-07 00:01:33 +0100337
338 /* Now a final check for a possible asynchronous low-level error
339 * report. This can happen when a connection receives a reset
340 * after a shutdown, both POLL_HUP and POLL_ERR are queued, and
341 * we might have come from there by just checking POLL_HUP instead
342 * of recv()'s return value 0, so we have no way to tell there was
343 * an error without checking.
344 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200345 if (unlikely(fdtab[conn->handle.fd].ev & FD_POLL_ERR))
Willy Tarreau26f4a042013-12-04 23:44:10 +0100346 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200347 return done;
348}
349
350
Willy Tarreau5368d802012-08-21 18:22:06 +0200351/* Send all pending bytes from buffer <buf> to connection <conn>'s socket.
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100352 * <flags> may contain some CO_SFL_* flags to hint the system about other
353 * pending data for example.
Willy Tarreau5368d802012-08-21 18:22:06 +0200354 * Only one call to send() is performed, unless the buffer wraps, in which case
355 * a second call may be performed. The connection's flags are updated with
356 * whatever special event is detected (error, empty). The caller is responsible
357 * for taking care of those events and avoiding the call if inappropriate. The
358 * function does not call the connection's polling update function, so the caller
359 * is responsible for this.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200360 */
Willy Tarreau5368d802012-08-21 18:22:06 +0200361static int raw_sock_from_buf(struct connection *conn, struct buffer *buf, int flags)
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100362{
Willy Tarreau5368d802012-08-21 18:22:06 +0200363 int ret, try, done, send_flag;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200364
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100365 if (!conn_ctrl_ready(conn))
366 return 0;
367
Willy Tarreau585744b2017-08-24 14:31:19 +0200368 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200369 return 0;
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 Tarreau5368d802012-08-21 18:22:06 +0200376 while (buf->o) {
377 try = buf->o;
Willy Tarreau89fa7062012-03-02 16:13:16 +0100378 /* outgoing data may wrap at the end */
Willy Tarreau5368d802012-08-21 18:22:06 +0200379 if (buf->data + try > buf->p)
380 try = buf->data + try - buf->p;
Willy Tarreauf890dc92008-12-13 21:12:26 +0100381
Willy Tarreau5368d802012-08-21 18:22:06 +0200382 send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
Willy Tarreau1049b1f2014-02-02 01:51:17 +0100383 if (try < buf->o || flags & CO_SFL_MSG_MORE)
Willy Tarreau7e4086d2014-02-02 01:44:13 +0100384 send_flag |= MSG_MORE;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200385
Willy Tarreau585744b2017-08-24 14:31:19 +0200386 ret = send(conn->handle.fd, bo_ptr(buf), try, send_flag);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200387
388 if (ret > 0) {
Willy Tarreau5368d802012-08-21 18:22:06 +0200389 buf->o -= ret;
390 done += ret;
Willy Tarreaub38903c2008-11-23 21:33:29 +0100391
Willy Tarreau5fb38032012-12-16 19:39:09 +0100392 if (likely(buffer_empty(buf)))
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100393 /* optimize data alignment in the buffer */
Willy Tarreau5368d802012-08-21 18:22:06 +0200394 buf->p = buf->data;
Willy Tarreau83749182007-04-15 20:56:27 +0200395
Willy Tarreauab3e1d32007-06-03 14:10:36 +0200396 /* if the system buffer is full, don't insist */
Willy Tarreau5368d802012-08-21 18:22:06 +0200397 if (ret < try)
Willy Tarreau6996e152007-04-30 14:37:43 +0200398 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200399 }
Willy Tarreau0ea0cf62012-11-11 20:38:30 +0100400 else if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100401 /* nothing written, we need to poll for write first */
Willy Tarreau585744b2017-08-24 14:31:19 +0200402 fd_cant_send(conn->handle.fd);
Willy Tarreau5368d802012-08-21 18:22:06 +0200403 break;
Willy Tarreau83749182007-04-15 20:56:27 +0200404 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200405 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100406 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau5368d802012-08-21 18:22:06 +0200407 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200408 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200409 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200410 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
411 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau5368d802012-08-21 18:22:06 +0200412 return done;
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100413}
Willy Tarreau6996e152007-04-30 14:37:43 +0200414
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100415
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200416/* transport-layer operations for RAW sockets */
Willy Tarreaud9f5cca2016-12-22 21:08:52 +0100417static struct xprt_ops raw_sock = {
Willy Tarreauc5788912012-08-24 18:12:41 +0200418 .snd_buf = raw_sock_from_buf,
419 .rcv_buf = raw_sock_to_buf,
Willy Tarreau96199b12012-08-24 00:46:52 +0200420#if defined(CONFIG_HAP_LINUX_SPLICE)
421 .rcv_pipe = raw_sock_to_pipe,
422 .snd_pipe = raw_sock_from_pipe,
423#endif
Willy Tarreauc5788912012-08-24 18:12:41 +0200424 .shutr = NULL,
425 .shutw = NULL,
426 .close = NULL,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100427 .name = "RAW",
Willy Tarreau5c979a92012-05-07 17:15:39 +0200428};
Willy Tarreaubaaee002006-06-26 02:48:02 +0200429
Willy Tarreau13e14102016-12-22 20:25:26 +0100430
431__attribute__((constructor))
Olivier Houchard0d005932017-08-14 15:59:44 +0200432static void __raw_sock_init(void)
Willy Tarreau13e14102016-12-22 20:25:26 +0100433{
434 xprt_register(XPRT_RAW, &raw_sock);
435}
436
Willy Tarreaubaaee002006-06-26 02:48:02 +0200437/*
438 * Local variables:
439 * c-indent-level: 8
440 * c-basic-offset: 8
441 * End:
442 */