blob: 3612442c16623dc4b2f15a4617346da7a428fa45 [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 Tarreau4c7e4b72020-05-27 12:58:42 +020025#include <haproxy/api.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020026#include <common/buffer.h>
Willy Tarreau83749182007-04-15 20:56:27 +020027#include <common/standard.h>
Willy Tarreau0c303ee2008-07-07 00:09:58 +020028#include <common/ticks.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020029#include <common/time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020030
Willy Tarreau8b117082012-08-06 15:06:49 +020031#include <proto/connection.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020032#include <proto/fd.h>
Willy Tarreaueb472682010-05-28 18:46:57 +020033#include <proto/freq_ctr.h>
34#include <proto/log.h>
Willy Tarreau3eba98a2009-01-25 13:56:13 +010035#include <proto/pipe.h>
Willy Tarreau75bf2c92012-08-20 17:01:35 +020036#include <proto/raw_sock.h>
Willy Tarreau73b013b2012-05-21 16:31:45 +020037#include <proto/stream_interface.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020038#include <proto/task.h>
39
Willy Tarreau5bd8c372009-01-19 00:32:22 +010040#include <types/global.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020041
Willy Tarreaub277d6e2012-05-11 16:59:14 +020042
Willy Tarreaue5733232019-05-22 19:24:06 +020043#if defined(USE_LINUX_SPLICE)
Willy Tarreau5bd8c372009-01-19 00:32:22 +010044
45/* A pipe contains 16 segments max, and it's common to see segments of 1448 bytes
46 * because of timestamps. Use this as a hint for not looping on splice().
47 */
48#define SPLICE_FULL_HINT 16*1448
49
Willy Tarreaua9de3332009-11-28 07:47:10 +010050/* how many data we attempt to splice at once when the buffer is configured for
51 * infinite forwarding */
52#define MAX_SPLICE_AT_ONCE (1<<30)
53
Willy Tarreau5bd8c372009-01-19 00:32:22 +010054/* Returns :
Willy Tarreau96199b12012-08-24 00:46:52 +020055 * -1 if splice() is not supported
56 * >= 0 to report the amount of spliced bytes.
57 * connection flags are updated (error, read0, wait_room, wait_data).
58 * The caller must have previously allocated the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +010059 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +010060int raw_sock_to_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe, unsigned int count)
Willy Tarreau5bd8c372009-01-19 00:32:22 +010061{
Willy Tarreau31971e52009-09-20 12:07:52 +020062 int ret;
Willy Tarreauafad0e02012-08-09 14:45:22 +020063 int retval = 0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010064
Willy Tarreauf79c8172013-10-21 16:30:56 +020065
Willy Tarreaufd803bb2014-01-20 15:13:07 +010066 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +020067 return 0;
68
Willy Tarreau585744b2017-08-24 14:31:19 +020069 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +010070 return 0;
71
Willy Tarreaue2a0eec2020-01-17 09:59:40 +010072 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +010073 errno = 0;
74
Willy Tarreau96199b12012-08-24 00:46:52 +020075 /* Under Linux, if FD_POLL_HUP is set, we have reached the end.
76 * Since older splice() implementations were buggy and returned
77 * EAGAIN on end of read, let's bypass the call to splice() now.
78 */
Willy Tarreau585744b2017-08-24 14:31:19 +020079 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +020080 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +020081 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +020082 goto out_read0;
83
84 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +020085 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +010086 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreauce3eda72013-12-05 00:49:40 +010087 errno = 0; /* let the caller do a getsockopt() if it wants it */
Willy Tarreau256b9c52017-10-25 09:30:13 +020088 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +020089 }
90 }
Willy Tarreaua9de3332009-11-28 07:47:10 +010091
Willy Tarreau96199b12012-08-24 00:46:52 +020092 while (count) {
93 if (count > MAX_SPLICE_AT_ONCE)
94 count = MAX_SPLICE_AT_ONCE;
Willy Tarreau5bd8c372009-01-19 00:32:22 +010095
Willy Tarreau585744b2017-08-24 14:31:19 +020096 ret = splice(conn->handle.fd, NULL, pipe->prod, NULL, count,
Willy Tarreau5bd8c372009-01-19 00:32:22 +010097 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
98
99 if (ret <= 0) {
Willy Tarreau38447472019-05-22 19:55:24 +0200100 if (ret == 0)
Willy Tarreau96199b12012-08-24 00:46:52 +0200101 goto out_read0;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100102
103 if (errno == EAGAIN) {
104 /* there are two reasons for EAGAIN :
105 * - nothing in the socket buffer (standard)
106 * - pipe is full
Willy Tarreau38447472019-05-22 19:55:24 +0200107 * The difference between these two situations
108 * is problematic. Since we don't know if the
109 * pipe is full, we'll stop if the pipe is not
110 * empty. Anyway, we will almost always fill or
111 * empty the pipe.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100112 */
Willy Tarreau96199b12012-08-24 00:46:52 +0200113 if (pipe->data) {
Ilya Shipitsince7b00f2020-03-23 22:28:40 +0500114 /* always stop reading until the pipe is flushed */
Willy Tarreau96199b12012-08-24 00:46:52 +0200115 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100116 break;
117 }
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100118 /* socket buffer exhausted */
119 fd_cant_recv(conn->handle.fd);
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100120 break;
121 }
Willy Tarreau45b88932012-11-12 12:00:09 +0100122 else if (errno == ENOSYS || errno == EINVAL || errno == EBADF) {
Willy Tarreau96199b12012-08-24 00:46:52 +0200123 /* splice not supported on this end, disable it.
124 * We can safely return -1 since there is no
125 * chance that any data has been piped yet.
126 */
Willy Tarreau256b9c52017-10-25 09:30:13 +0200127 retval = -1;
128 goto leave;
Willy Tarreaudc340a92009-06-28 23:10:19 +0200129 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200130 else if (errno == EINTR) {
131 /* try again */
132 continue;
133 }
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100134 /* here we have another error */
Willy Tarreau96199b12012-08-24 00:46:52 +0200135 conn->flags |= CO_FL_ERROR;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100136 break;
137 } /* ret <= 0 */
138
Willy Tarreau96199b12012-08-24 00:46:52 +0200139 retval += ret;
140 pipe->data += ret;
Willy Tarreau4fc90ef2013-04-06 11:29:39 +0200141 count -= ret;
Willy Tarreaubaf2a502013-01-07 16:38:26 +0100142
Willy Tarreau96199b12012-08-24 00:46:52 +0200143 if (pipe->data >= SPLICE_FULL_HINT || ret >= global.tune.recv_enough) {
144 /* We've read enough of it for this time, let's stop before
145 * being asked to poll.
146 */
Willy Tarreau61d39a02013-07-18 21:49:32 +0200147 conn->flags |= CO_FL_WAIT_ROOM;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100148 break;
149 }
150 } /* while */
151
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200152 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && retval)
153 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200154
155 leave:
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200156 if (retval > 0) {
157 /* we count the total bytes sent, and the send rate for 32-byte
158 * blocks. The reason for the latter is that freq_ctr are
159 * limited to 4GB and that it's not enough per second.
160 */
161 _HA_ATOMIC_ADD(&global.out_bytes, retval);
162 update_freq_ctr(&global.out_32bps, (retval + 16) / 32);
163 }
Willy Tarreau96199b12012-08-24 00:46:52 +0200164 return retval;
Willy Tarreau3eba98a2009-01-25 13:56:13 +0100165
Willy Tarreau96199b12012-08-24 00:46:52 +0200166 out_read0:
167 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200168 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200169 goto leave;
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100170}
171
Willy Tarreau96199b12012-08-24 00:46:52 +0200172/* Send as many bytes as possible from the pipe to the connection's socket.
173 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100174int raw_sock_from_pipe(struct connection *conn, void *xprt_ctx, struct pipe *pipe)
Willy Tarreau96199b12012-08-24 00:46:52 +0200175{
176 int ret, done;
177
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100178 if (!conn_ctrl_ready(conn))
179 return 0;
180
Willy Tarreau585744b2017-08-24 14:31:19 +0200181 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200182 return 0;
183
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100184 if (conn->flags & CO_FL_SOCK_WR_SH) {
185 /* it's already closed */
186 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
187 errno = EPIPE;
188 return 0;
189 }
190
Willy Tarreau96199b12012-08-24 00:46:52 +0200191 done = 0;
192 while (pipe->data) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200193 ret = splice(pipe->cons, NULL, conn->handle.fd, NULL, pipe->data,
Willy Tarreau96199b12012-08-24 00:46:52 +0200194 SPLICE_F_MOVE|SPLICE_F_NONBLOCK);
195
196 if (ret <= 0) {
197 if (ret == 0 || errno == EAGAIN) {
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100198 fd_cant_send(conn->handle.fd);
Willy Tarreau96199b12012-08-24 00:46:52 +0200199 break;
200 }
201 else if (errno == EINTR)
202 continue;
203
204 /* here we have another error */
205 conn->flags |= CO_FL_ERROR;
206 break;
207 }
208
209 done += ret;
210 pipe->data -= ret;
211 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200212 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200213 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200214 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200215
Willy Tarreau96199b12012-08-24 00:46:52 +0200216 return done;
217}
218
Willy Tarreaue5733232019-05-22 19:24:06 +0200219#endif /* USE_LINUX_SPLICE */
Willy Tarreau6b4aad42009-01-18 21:59:13 +0100220
221
Willy Tarreau2ba44652012-08-20 17:30:32 +0200222/* Receive up to <count> bytes from connection <conn>'s socket and store them
Willy Tarreauabf08d92014-01-14 11:31:27 +0100223 * into buffer <buf>. Only one call to recv() is performed, unless the
Willy Tarreau2ba44652012-08-20 17:30:32 +0200224 * buffer wraps, in which case a second call may be performed. The connection's
225 * flags are updated with whatever special event is detected (error, read0,
226 * empty). The caller is responsible for taking care of those events and
227 * avoiding the call if inappropriate. The function does not call the
228 * connection's polling update function, so the caller is responsible for this.
Willy Tarreauce3eda72013-12-05 00:49:40 +0100229 * errno is cleared before starting so that the caller knows that if it spots an
230 * error without errno, it's pending and can be retrieved via getsockopt(SO_ERROR).
Willy Tarreau2ba44652012-08-20 17:30:32 +0200231 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100232static 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 +0200233{
Willy Tarreaubfc4d772018-07-18 11:22:03 +0200234 ssize_t ret;
235 size_t try, done = 0;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200236
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100237 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200238 return 0;
239
Willy Tarreau585744b2017-08-24 14:31:19 +0200240 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100241 return 0;
242
Willy Tarreaue2a0eec2020-01-17 09:59:40 +0100243 conn->flags &= ~CO_FL_WAIT_ROOM;
Willy Tarreauce3eda72013-12-05 00:49:40 +0100244 errno = 0;
245
Willy Tarreau585744b2017-08-24 14:31:19 +0200246 if (unlikely(!(fdtab[conn->handle.fd].ev & FD_POLL_IN))) {
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200247 /* stop here if we reached the end of data */
Willy Tarreau585744b2017-08-24 14:31:19 +0200248 if ((fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) == FD_POLL_HUP)
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200249 goto read0;
250
251 /* report error on POLL_ERR before connection establishment */
Willy Tarreau585744b2017-08-24 14:31:19 +0200252 if ((fdtab[conn->handle.fd].ev & FD_POLL_ERR) && (conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100253 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200254 goto leave;
Willy Tarreau6f5d1412012-10-04 20:38:49 +0200255 }
256 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200257
Willy Tarreau2ba44652012-08-20 17:30:32 +0200258 /* read the largest possible block. For this, we perform only one call
259 * to recv() unless the buffer wraps and we exactly fill the first hunk,
260 * in which case we accept to do it once again. A new attempt is made on
261 * EINTR too.
262 */
Willy Tarreauabf08d92014-01-14 11:31:27 +0100263 while (count > 0) {
Willy Tarreau591d4452018-06-15 17:21:00 +0200264 try = b_contig_space(buf);
265 if (!try)
266 break;
267
Willy Tarreauabf08d92014-01-14 11:31:27 +0100268 if (try > count)
269 try = count;
270
Willy Tarreau8f9c72d2018-06-07 18:46:28 +0200271 ret = recv(conn->handle.fd, b_tail(buf), try, 0);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200272
273 if (ret > 0) {
Olivier Houchardacd14032018-06-28 18:17:23 +0200274 b_add(buf, ret);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200275 done += ret;
276 if (ret < try) {
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100277 /* socket buffer exhausted */
278 fd_cant_recv(conn->handle.fd);
279
Willy Tarreau2ba44652012-08-20 17:30:32 +0200280 /* unfortunately, on level-triggered events, POLL_HUP
281 * is generally delivered AFTER the system buffer is
Willy Tarreau68128712017-03-13 12:04:34 +0100282 * empty, unless the poller supports POLL_RDHUP. If
283 * we know this is the case, we don't try to read more
284 * as we know there's no more available. Similarly, if
285 * there's no problem with lingering we don't even try
286 * to read an unlikely close from the client since we'll
287 * close first anyway.
Willy Tarreau2ba44652012-08-20 17:30:32 +0200288 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200289 if (fdtab[conn->handle.fd].ev & FD_POLL_HUP)
Willy Tarreau2ba44652012-08-20 17:30:32 +0200290 goto read0;
Willy Tarreau6c11bd22014-01-24 00:54:27 +0100291
Willy Tarreau585744b2017-08-24 14:31:19 +0200292 if ((!fdtab[conn->handle.fd].linger_risk) ||
Willy Tarreau68128712017-03-13 12:04:34 +0100293 (cur_poller.flags & HAP_POLL_F_RDHUP)) {
Willy Tarreau68128712017-03-13 12:04:34 +0100294 break;
295 }
Willy Tarreau2ba44652012-08-20 17:30:32 +0200296 }
297 count -= ret;
Willy Tarreau716bec22020-02-20 11:04:40 +0100298
299 if (flags & CO_RFL_READ_ONCE)
300 break;
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 Tarreau8dd348c2020-02-28 14:09:12 +0100306 /* socket buffer exhausted */
307 fd_cant_recv(conn->handle.fd);
Willy Tarreau2ba44652012-08-20 17:30:32 +0200308 break;
309 }
310 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100311 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200312 break;
313 }
314 }
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200315
316 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done)
317 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200318
319 leave:
Willy Tarreau2ba44652012-08-20 17:30:32 +0200320 return done;
321
322 read0:
323 conn_sock_read0(conn);
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200324 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreaudebdc4b2012-12-07 00:01:33 +0100325
326 /* Now a final check for a possible asynchronous low-level error
327 * report. This can happen when a connection receives a reset
328 * after a shutdown, both POLL_HUP and POLL_ERR are queued, and
329 * we might have come from there by just checking POLL_HUP instead
330 * of recv()'s return value 0, so we have no way to tell there was
331 * an error without checking.
332 */
Willy Tarreau585744b2017-08-24 14:31:19 +0200333 if (unlikely(fdtab[conn->handle.fd].ev & FD_POLL_ERR))
Willy Tarreau26f4a042013-12-04 23:44:10 +0100334 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau256b9c52017-10-25 09:30:13 +0200335 goto leave;
Willy Tarreau2ba44652012-08-20 17:30:32 +0200336}
337
338
Willy Tarreau787db9a2018-06-14 18:31:46 +0200339/* Send up to <count> pending bytes from buffer <buf> to connection <conn>'s
340 * socket. <flags> may contain some CO_SFL_* flags to hint the system about
341 * other pending data for example, but this flag is ignored at the moment.
Willy Tarreau5368d802012-08-21 18:22:06 +0200342 * Only one call to send() is performed, unless the buffer wraps, in which case
343 * a second call may be performed. The connection's flags are updated with
344 * whatever special event is detected (error, empty). The caller is responsible
345 * for taking care of those events and avoiding the call if inappropriate. The
346 * function does not call the connection's polling update function, so the caller
Willy Tarreau787db9a2018-06-14 18:31:46 +0200347 * is responsible for this. It's up to the caller to update the buffer's contents
348 * based on the return value.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200349 */
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100350static 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 +0100351{
Willy Tarreau787db9a2018-06-14 18:31:46 +0200352 ssize_t ret;
353 size_t try, done;
354 int send_flag;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200355
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100356 if (!conn_ctrl_ready(conn))
357 return 0;
358
Willy Tarreau585744b2017-08-24 14:31:19 +0200359 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200360 return 0;
361
Willy Tarreaua8c7e8e2020-01-23 18:17:55 +0100362 if (conn->flags & CO_FL_SOCK_WR_SH) {
363 /* it's already closed */
364 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH;
365 errno = EPIPE;
366 return 0;
367 }
368
Willy Tarreau5368d802012-08-21 18:22:06 +0200369 done = 0;
370 /* send the largest possible block. For this we perform only one call
371 * to send() unless the buffer wraps and we exactly fill the first hunk,
372 * in which case we accept to do it once again.
Willy Tarreau5bd8c372009-01-19 00:32:22 +0100373 */
Willy Tarreau787db9a2018-06-14 18:31:46 +0200374 while (count) {
375 try = b_contig_data(buf, done);
376 if (try > count)
377 try = count;
Willy Tarreauf890dc92008-12-13 21:12:26 +0100378
Willy Tarreau5368d802012-08-21 18:22:06 +0200379 send_flag = MSG_DONTWAIT | MSG_NOSIGNAL;
Willy Tarreau787db9a2018-06-14 18:31:46 +0200380 if (try < count || flags & CO_SFL_MSG_MORE)
Willy Tarreau7e4086d2014-02-02 01:44:13 +0100381 send_flag |= MSG_MORE;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200382
Willy Tarreau787db9a2018-06-14 18:31:46 +0200383 ret = send(conn->handle.fd, b_peek(buf, done), try, send_flag);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200384
385 if (ret > 0) {
Willy Tarreau787db9a2018-06-14 18:31:46 +0200386 count -= ret;
Willy Tarreau5368d802012-08-21 18:22:06 +0200387 done += ret;
Willy Tarreaub38903c2008-11-23 21:33:29 +0100388
Willy Tarreauab3e1d32007-06-03 14:10:36 +0200389 /* if the system buffer is full, don't insist */
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100390 if (ret < try) {
391 fd_cant_send(conn->handle.fd);
Willy Tarreau6996e152007-04-30 14:37:43 +0200392 break;
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100393 }
Willy Tarreau3381bf82020-01-17 17:39:35 +0100394 if (!count)
Willy Tarreau3110eb72020-02-21 10:21:46 +0100395 fd_stop_send(conn->handle.fd);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200396 }
Willy Tarreau034c88c2017-01-23 23:36:45 +0100397 else if (ret == 0 || errno == EAGAIN || errno == ENOTCONN || errno == EINPROGRESS) {
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100398 /* nothing written, we need to poll for write first */
Willy Tarreau8dd348c2020-02-28 14:09:12 +0100399 fd_cant_send(conn->handle.fd);
Willy Tarreau5368d802012-08-21 18:22:06 +0200400 break;
Willy Tarreau83749182007-04-15 20:56:27 +0200401 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200402 else if (errno != EINTR) {
Willy Tarreau26f4a042013-12-04 23:44:10 +0100403 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau5368d802012-08-21 18:22:06 +0200404 break;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200405 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200406 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200407 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && done) {
Willy Tarreau665e6ee2012-10-04 20:20:46 +0200408 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200409 }
Willy Tarreau256b9c52017-10-25 09:30:13 +0200410
Willy Tarreau7cf0e452019-05-23 11:39:14 +0200411 if (done > 0) {
412 /* we count the total bytes sent, and the send rate for 32-byte
413 * blocks. The reason for the latter is that freq_ctr are
414 * limited to 4GB and that it's not enough per second.
415 */
416 _HA_ATOMIC_ADD(&global.out_bytes, done);
417 update_freq_ctr(&global.out_32bps, (done + 16) / 32);
418 }
Willy Tarreau5368d802012-08-21 18:22:06 +0200419 return done;
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100420}
Willy Tarreau6996e152007-04-30 14:37:43 +0200421
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100422/* Called from the upper layer, to subscribe <es> to events <event_type>. The
423 * event subscriber <es> is not allowed to change from a previous call as long
424 * as at least one event is still subscribed. The <event_type> must only be a
425 * combination of SUB_RETRY_RECV and SUB_RETRY_SEND. It always returns 0.
426 */
427static int raw_sock_subscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100428{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100429 return conn_subscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100430}
431
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100432/* Called from the upper layer, to unsubscribe <es> from events <event_type>.
433 * The <es> pointer is not allowed to differ from the one passed to the
434 * subscribe() call. It always returns zero.
435 */
436static int raw_sock_unsubscribe(struct connection *conn, void *xprt_ctx, int event_type, struct wait_event *es)
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100437{
Willy Tarreauee1a6fc2020-01-17 07:52:13 +0100438 return conn_unsubscribe(conn, xprt_ctx, event_type, es);
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100439}
Willy Tarreau0c2fc1f2009-01-18 15:30:37 +0100440
Olivier Houchard5149b592019-05-23 17:47:36 +0200441/* We can't have an underlying XPRT, so just return -1 to signify failure */
442static int raw_sock_remove_xprt(struct connection *conn, void *xprt_ctx, void *toremove_ctx, const struct xprt_ops *newops, void *newctx)
443{
444 /* This is the lowest xprt we can have, so if we get there we didn't
445 * find the xprt we wanted to remove, that's a bug
446 */
447 BUG_ON(1);
448 return -1;
449}
450
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200451/* transport-layer operations for RAW sockets */
Willy Tarreaud9f5cca2016-12-22 21:08:52 +0100452static struct xprt_ops raw_sock = {
Willy Tarreauc5788912012-08-24 18:12:41 +0200453 .snd_buf = raw_sock_from_buf,
454 .rcv_buf = raw_sock_to_buf,
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100455 .subscribe = raw_sock_subscribe,
456 .unsubscribe = raw_sock_unsubscribe,
Olivier Houchard5149b592019-05-23 17:47:36 +0200457 .remove_xprt = raw_sock_remove_xprt,
Willy Tarreaue5733232019-05-22 19:24:06 +0200458#if defined(USE_LINUX_SPLICE)
Willy Tarreau96199b12012-08-24 00:46:52 +0200459 .rcv_pipe = raw_sock_to_pipe,
460 .snd_pipe = raw_sock_from_pipe,
461#endif
Willy Tarreauc5788912012-08-24 18:12:41 +0200462 .shutr = NULL,
463 .shutw = NULL,
464 .close = NULL,
Willy Tarreau8e0bb0a2016-11-24 16:58:12 +0100465 .name = "RAW",
Willy Tarreau5c979a92012-05-07 17:15:39 +0200466};
Willy Tarreaubaaee002006-06-26 02:48:02 +0200467
Willy Tarreau13e14102016-12-22 20:25:26 +0100468
469__attribute__((constructor))
Olivier Houchard0d005932017-08-14 15:59:44 +0200470static void __raw_sock_init(void)
Willy Tarreau13e14102016-12-22 20:25:26 +0100471{
472 xprt_register(XPRT_RAW, &raw_sock);
473}
474
Willy Tarreaubaaee002006-06-26 02:48:02 +0200475/*
476 * Local variables:
477 * c-indent-level: 8
478 * c-basic-offset: 8
479 * End:
480 */