blob: ab788bde720fa1f1a252b787ce8387f808d2efbe [file] [log] [blame]
Willy Tarreau92fb9832007-10-16 17:34:28 +02001/*
2 * UNIX SOCK_STREAM protocol layer (uxst)
3 *
Willy Tarreaueb472682010-05-28 18:46:57 +02004 * Copyright 2000-2010 Willy Tarreau <w@1wt.eu>
Willy Tarreau92fb9832007-10-16 17:34:28 +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
13#include <ctype.h>
14#include <errno.h>
15#include <fcntl.h>
Willy Tarreaud0a895d2012-09-18 17:40:35 +020016#include <pwd.h>
17#include <grp.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020018#include <stdio.h>
19#include <stdlib.h>
20#include <string.h>
21#include <syslog.h>
22#include <time.h>
23
Willy Tarreau92fb9832007-10-16 17:34:28 +020024#include <sys/socket.h>
25#include <sys/stat.h>
26#include <sys/types.h>
27#include <sys/un.h>
28
29#include <common/compat.h>
30#include <common/config.h>
31#include <common/debug.h>
Willy Tarreaud740bab2007-10-28 11:14:07 +010032#include <common/errors.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020033#include <common/mini-clist.h>
34#include <common/standard.h>
35#include <common/time.h>
36#include <common/version.h>
37
Willy Tarreau92fb9832007-10-16 17:34:28 +020038#include <types/global.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020039
Willy Tarreau47f48c42014-05-09 22:57:47 +020040#include <proto/connection.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020041#include <proto/fd.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020042#include <proto/listener.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020043#include <proto/log.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020044#include <proto/protocol.h>
Willy Tarreau92fb9832007-10-16 17:34:28 +020045#include <proto/task.h>
46
Emeric Bruncf20bf12010-10-22 16:06:11 +020047static int uxst_bind_listener(struct listener *listener, char *errmsg, int errlen);
48static int uxst_bind_listeners(struct protocol *proto, char *errmsg, int errlen);
Willy Tarreaudabf2e22007-10-28 21:59:24 +010049static int uxst_unbind_listeners(struct protocol *proto);
Willy Tarreau47f48c42014-05-09 22:57:47 +020050static int uxst_connect_server(struct connection *conn, int data, int delack);
Willy Tarreau9d5be5c2017-09-15 07:55:51 +020051static void uxst_add_listener(struct listener *listener, int port);
Willy Tarreau31794892017-09-15 07:59:31 +020052static int uxst_pause_listener(struct listener *l);
53static int uxst_get_src(int fd, struct sockaddr *sa, socklen_t salen, int dir);
54static int uxst_get_dst(int fd, struct sockaddr *sa, socklen_t salen, int dir);
Willy Tarreaudabf2e22007-10-28 21:59:24 +010055
56/* Note: must not be declared <const> as its list will be overwritten */
57static struct protocol proto_unix = {
58 .name = "unix_stream",
59 .sock_domain = PF_UNIX,
60 .sock_type = SOCK_STREAM,
61 .sock_prot = 0,
62 .sock_family = AF_UNIX,
63 .sock_addrlen = sizeof(struct sockaddr_un),
64 .l3_addrlen = sizeof(((struct sockaddr_un*)0)->sun_path),/* path len */
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020065 .accept = &listener_accept,
Willy Tarreau47f48c42014-05-09 22:57:47 +020066 .connect = &uxst_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +020067 .bind = uxst_bind_listener,
Willy Tarreaudabf2e22007-10-28 21:59:24 +010068 .bind_all = uxst_bind_listeners,
69 .unbind_all = uxst_unbind_listeners,
70 .enable_all = enable_all_listeners,
71 .disable_all = disable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +020072 .get_src = uxst_get_src,
73 .get_dst = uxst_get_dst,
Willy Tarreaufd0e0082014-07-07 21:07:51 +020074 .pause = uxst_pause_listener,
Willy Tarreau9d5be5c2017-09-15 07:55:51 +020075 .add = uxst_add_listener,
Willy Tarreaudabf2e22007-10-28 21:59:24 +010076 .listeners = LIST_HEAD_INIT(proto_unix.listeners),
77 .nb_listeners = 0,
78};
79
Willy Tarreaudabf2e22007-10-28 21:59:24 +010080/********************************
81 * 1) low-level socket functions
82 ********************************/
83
Willy Tarreau59b94792012-05-11 16:16:40 +020084/*
85 * Retrieves the source address for the socket <fd>, with <dir> indicating
86 * if we're a listener (=0) or an initiator (!=0). It returns 0 in case of
87 * success, -1 in case of error. The socket's source address is stored in
88 * <sa> for <salen> bytes.
89 */
Willy Tarreau31794892017-09-15 07:59:31 +020090static int uxst_get_src(int fd, struct sockaddr *sa, socklen_t salen, int dir)
Willy Tarreau59b94792012-05-11 16:16:40 +020091{
92 if (dir)
93 return getsockname(fd, sa, &salen);
94 else
95 return getpeername(fd, sa, &salen);
96}
97
98
99/*
100 * Retrieves the original destination address for the socket <fd>, with <dir>
101 * indicating if we're a listener (=0) or an initiator (!=0). It returns 0 in
102 * case of success, -1 in case of error. The socket's source address is stored
103 * in <sa> for <salen> bytes.
104 */
Willy Tarreau31794892017-09-15 07:59:31 +0200105static int uxst_get_dst(int fd, struct sockaddr *sa, socklen_t salen, int dir)
Willy Tarreau59b94792012-05-11 16:16:40 +0200106{
107 if (dir)
108 return getpeername(fd, sa, &salen);
109 else
110 return getsockname(fd, sa, &salen);
111}
112
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100113
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100114/********************************
115 * 2) listener-oriented functions
116 ********************************/
117
118
Olivier Houchardf886e342017-04-05 22:24:59 +0200119static int uxst_find_compatible_fd(struct listener *l)
120{
121 struct xfer_sock_list *xfer_sock = xfer_sock_list;
122 int ret = -1;
123
124 while (xfer_sock) {
125 struct sockaddr_un *un1 = (void *)&l->addr;
126 struct sockaddr_un *un2 = (void *)&xfer_sock->addr;
127
128 /*
129 * The bound socket's path as returned by getsockaddr
130 * will be the temporary name <sockname>.XXXXX.tmp,
131 * so we can't just compare the two names
132 */
133 if (xfer_sock->addr.ss_family == AF_UNIX &&
134 strncmp(un1->sun_path, un2->sun_path,
135 strlen(un1->sun_path)) == 0) {
136 char *after_sockname = un2->sun_path +
137 strlen(un1->sun_path);
138 /* Make a reasonnable effort to check that
139 * it is indeed a haproxy-generated temporary
140 * name, it's not perfect, but probably good enough.
141 */
142 if (after_sockname[0] == '.') {
143 after_sockname++;
144 while (after_sockname[0] >= '0' &&
145 after_sockname[0] <= '9')
146 after_sockname++;
147 if (!strcmp(after_sockname, ".tmp"))
148 break;
Olivier Houchardb4dd15b2018-06-06 18:34:34 +0200149 /* abns sockets sun_path starts with a \0 */
150 } else if (un1->sun_path[0] == 0
151 && un2->sun_path[0] == 0
152 && !memcmp(&un1->sun_path[1], &un2->sun_path[1],
153 sizeof(un1->sun_path) - 1))
154 break;
Olivier Houchardf886e342017-04-05 22:24:59 +0200155 }
156 xfer_sock = xfer_sock->next;
157 }
158 if (xfer_sock != NULL) {
159 ret = xfer_sock->fd;
160 if (xfer_sock == xfer_sock_list)
161 xfer_sock_list = xfer_sock->next;
162 if (xfer_sock->prev)
163 xfer_sock->prev->next = xfer_sock->next;
164 if (xfer_sock->next)
Olivier Houchardec9516a2018-03-08 18:25:49 +0100165 xfer_sock->next->prev = xfer_sock->prev;
Olivier Houchardf886e342017-04-05 22:24:59 +0200166 free(xfer_sock);
167 }
168 return ret;
169
170}
171
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100172/* This function creates a UNIX socket associated to the listener. It changes
173 * the state from ASSIGNED to LISTEN. The socket is NOT enabled for polling.
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100174 * The return value is composed from ERR_NONE, ERR_RETRYABLE and ERR_FATAL. It
175 * may return a warning or an error message in <errmsg> if the message is at
176 * most <errlen> bytes long (including '\0'). Note that <errmsg> may be NULL if
177 * <errlen> is also zero.
Willy Tarreau92fb9832007-10-16 17:34:28 +0200178 */
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100179static int uxst_bind_listener(struct listener *listener, char *errmsg, int errlen)
Willy Tarreau92fb9832007-10-16 17:34:28 +0200180{
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100181 int fd;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200182 char tempname[MAXPATHLEN];
183 char backname[MAXPATHLEN];
184 struct sockaddr_un addr;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100185 const char *msg = NULL;
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100186 const char *path;
Willy Tarreau40aa0702013-03-10 23:51:38 +0100187 int ext, ready;
188 socklen_t ready_len;
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200189 int err;
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100190 int ret;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200191
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200192 err = ERR_NONE;
193
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100194 /* ensure we never return garbage */
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100195 if (errlen)
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100196 *errmsg = 0;
197
198 if (listener->state != LI_ASSIGNED)
199 return ERR_NONE; /* already bound */
200
Olivier Houchardf886e342017-04-05 22:24:59 +0200201 if (listener->fd == -1)
202 listener->fd = uxst_find_compatible_fd(listener);
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100203 path = ((struct sockaddr_un *)&listener->addr)->sun_path;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200204
Willy Tarreau40aa0702013-03-10 23:51:38 +0100205 /* if the listener already has an fd assigned, then we were offered the
206 * fd by an external process (most likely the parent), and we don't want
207 * to create a new socket. However we still want to set a few flags on
208 * the socket.
209 */
210 fd = listener->fd;
211 ext = (fd >= 0);
212 if (ext)
213 goto fd_ready;
214
Willy Tarreauccfccef2014-05-10 01:49:15 +0200215 if (path[0]) {
216 ret = snprintf(tempname, MAXPATHLEN, "%s.%d.tmp", path, pid);
217 if (ret < 0 || ret >= MAXPATHLEN) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200218 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200219 msg = "name too long for UNIX socket";
220 goto err_return;
221 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200222
Willy Tarreauccfccef2014-05-10 01:49:15 +0200223 ret = snprintf(backname, MAXPATHLEN, "%s.%d.bak", path, pid);
224 if (ret < 0 || ret >= MAXPATHLEN) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200225 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200226 msg = "name too long for UNIX socket";
227 goto err_return;
228 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200229
Willy Tarreauccfccef2014-05-10 01:49:15 +0200230 /* 2. clean existing orphaned entries */
231 if (unlink(tempname) < 0 && errno != ENOENT) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200232 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200233 msg = "error when trying to unlink previous UNIX socket";
234 goto err_return;
235 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200236
Willy Tarreauccfccef2014-05-10 01:49:15 +0200237 if (unlink(backname) < 0 && errno != ENOENT) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200238 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200239 msg = "error when trying to unlink previous UNIX socket";
240 goto err_return;
241 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200242
Willy Tarreauccfccef2014-05-10 01:49:15 +0200243 /* 3. backup existing socket */
244 if (link(path, backname) < 0 && errno != ENOENT) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200245 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200246 msg = "error when trying to preserve previous UNIX socket";
247 goto err_return;
248 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200249
Willy Tarreauccfccef2014-05-10 01:49:15 +0200250 strncpy(addr.sun_path, tempname, sizeof(addr.sun_path));
251 addr.sun_path[sizeof(addr.sun_path) - 1] = 0;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200252 }
Willy Tarreauccfccef2014-05-10 01:49:15 +0200253 else {
254 /* first char is zero, it's an abstract socket whose address
255 * is defined by all the bytes past this zero.
256 */
257 memcpy(addr.sun_path, path, sizeof(addr.sun_path));
258 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200259 addr.sun_family = AF_UNIX;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200260
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100261 fd = socket(PF_UNIX, SOCK_STREAM, 0);
262 if (fd < 0) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200263 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100264 msg = "cannot create UNIX socket";
Willy Tarreau92fb9832007-10-16 17:34:28 +0200265 goto err_unlink_back;
266 }
267
Willy Tarreau40aa0702013-03-10 23:51:38 +0100268 fd_ready:
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100269 if (fd >= global.maxsock) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200270 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100271 msg = "socket(): not enough free sockets, raise -n argument";
Willy Tarreau92fb9832007-10-16 17:34:28 +0200272 goto err_unlink_temp;
273 }
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100274
275 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200276 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100277 msg = "cannot make UNIX socket non-blocking";
Willy Tarreau92fb9832007-10-16 17:34:28 +0200278 goto err_unlink_temp;
279 }
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100280
Willy Tarreau40aa0702013-03-10 23:51:38 +0100281 if (!ext && bind(fd, (struct sockaddr *)&addr, sizeof(addr)) < 0) {
Willy Tarreau92fb9832007-10-16 17:34:28 +0200282 /* note that bind() creates the socket <tempname> on the file system */
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200283 if (errno == EADDRINUSE) {
284 /* the old process might still own it, let's retry */
285 err |= ERR_RETRYABLE | ERR_ALERT;
286 msg = "cannot listen to socket";
287 }
288 else {
289 err |= ERR_FATAL | ERR_ALERT;
290 msg = "cannot bind UNIX socket";
291 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200292 goto err_unlink_temp;
293 }
294
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100295 /* <uid> and <gid> different of -1 will be used to change the socket owner.
296 * If <mode> is not 0, it will be used to restrict access to the socket.
297 * While it is known not to be portable on every OS, it's still useful
Willy Tarreauccfccef2014-05-10 01:49:15 +0200298 * where it works. We also don't change permissions on abstract sockets.
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100299 */
Willy Tarreauccfccef2014-05-10 01:49:15 +0200300 if (!ext && path[0] &&
Willy Tarreau40aa0702013-03-10 23:51:38 +0100301 (((listener->bind_conf->ux.uid != -1 || listener->bind_conf->ux.gid != -1) &&
302 (chown(tempname, listener->bind_conf->ux.uid, listener->bind_conf->ux.gid) == -1)) ||
303 (listener->bind_conf->ux.mode != 0 && chmod(tempname, listener->bind_conf->ux.mode) == -1))) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200304 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100305 msg = "cannot change UNIX socket ownership";
Willy Tarreaue6ad2b12007-10-18 12:45:54 +0200306 goto err_unlink_temp;
307 }
308
Willy Tarreau40aa0702013-03-10 23:51:38 +0100309 ready = 0;
310 ready_len = sizeof(ready);
311 if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &ready, &ready_len) == -1)
312 ready = 0;
313
314 if (!(ext && ready) && /* only listen if not already done by external process */
315 listen(fd, listener->backlog ? listener->backlog : listener->maxconn) < 0) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200316 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100317 msg = "cannot listen to UNIX socket";
Willy Tarreau92fb9832007-10-16 17:34:28 +0200318 goto err_unlink_temp;
319 }
320
Willy Tarreauccfccef2014-05-10 01:49:15 +0200321 /* Point of no return: we are ready, we'll switch the sockets. We don't
Willy Tarreau92fb9832007-10-16 17:34:28 +0200322 * fear loosing the socket <path> because we have a copy of it in
Willy Tarreauccfccef2014-05-10 01:49:15 +0200323 * backname. Abstract sockets are not renamed.
Willy Tarreau92fb9832007-10-16 17:34:28 +0200324 */
Willy Tarreauccfccef2014-05-10 01:49:15 +0200325 if (!ext && path[0] && rename(tempname, path) < 0) {
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200326 err |= ERR_FATAL | ERR_ALERT;
Willy Tarreaub40dc942010-11-07 12:10:51 +0100327 msg = "cannot switch final and temporary UNIX sockets";
Willy Tarreau92fb9832007-10-16 17:34:28 +0200328 goto err_rename;
329 }
330
Willy Tarreau68986ab2017-06-16 10:34:20 +0200331 /* Cleanup: only unlink if we didn't inherit the fd from the parent */
Willy Tarreauccfccef2014-05-10 01:49:15 +0200332 if (!ext && path[0])
Willy Tarreau40aa0702013-03-10 23:51:38 +0100333 unlink(backname);
Willy Tarreau92fb9832007-10-16 17:34:28 +0200334
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100335 /* the socket is now listening */
336 listener->fd = fd;
337 listener->state = LI_LISTEN;
338
Willy Tarreaua9786b62018-01-25 07:22:13 +0100339 fd_insert(fd, listener, listener->proto->accept,
340 listener->bind_conf->bind_thread[relative_pid-1] ?
341 listener->bind_conf->bind_thread[relative_pid-1] : MAX_THREADS_MASK);
342
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200343 return err;
344
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100345 err_rename:
346 ret = rename(backname, path);
347 if (ret < 0 && errno == ENOENT)
348 unlink(path);
349 err_unlink_temp:
Jan Seda7319b642014-06-26 20:44:05 +0200350 if (!ext && path[0])
Willy Tarreau40aa0702013-03-10 23:51:38 +0100351 unlink(tempname);
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100352 close(fd);
353 err_unlink_back:
Jan Seda7319b642014-06-26 20:44:05 +0200354 if (!ext && path[0])
Willy Tarreau40aa0702013-03-10 23:51:38 +0100355 unlink(backname);
Cyril Bonté1f5848a2010-11-14 17:03:19 +0100356 err_return:
Willy Tarreau40aa0702013-03-10 23:51:38 +0100357 if (msg && errlen) {
358 if (!ext)
359 snprintf(errmsg, errlen, "%s [%s]", msg, path);
360 else
361 snprintf(errmsg, errlen, "%s [fd %d]", msg, fd);
362 }
Willy Tarreau3c5efa22014-07-07 18:36:45 +0200363 return err;
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100364}
365
366/* This function closes the UNIX sockets for the specified listener.
367 * The listener enters the LI_ASSIGNED state. It always returns ERR_NONE.
368 */
369static int uxst_unbind_listener(struct listener *listener)
370{
Willy Tarreaube58c382011-07-24 18:28:10 +0200371 if (listener->state > LI_ASSIGNED) {
372 unbind_listener(listener);
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100373 }
374 return ERR_NONE;
375}
376
Willy Tarreau32282382017-09-15 07:44:44 +0200377/* Add <listener> to the list of unix stream listeners (port is ignored). The
378 * listener's state is automatically updated from LI_INIT to LI_ASSIGNED.
379 * The number of listeners for the protocol is updated.
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100380 */
Willy Tarreau9d5be5c2017-09-15 07:55:51 +0200381static void uxst_add_listener(struct listener *listener, int port)
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100382{
383 if (listener->state != LI_INIT)
384 return;
385 listener->state = LI_ASSIGNED;
386 listener->proto = &proto_unix;
387 LIST_ADDQ(&proto_unix.listeners, &listener->proto_list);
388 proto_unix.nb_listeners++;
389}
390
Willy Tarreaufd0e0082014-07-07 21:07:51 +0200391/* Pause a listener. Returns < 0 in case of failure, 0 if the listener
392 * was totally stopped, or > 0 if correctly paused. Nothing is done for
393 * plain unix sockets since currently it's the new process which handles
394 * the renaming. Abstract sockets are completely unbound.
395 */
Willy Tarreau31794892017-09-15 07:59:31 +0200396static int uxst_pause_listener(struct listener *l)
Willy Tarreaufd0e0082014-07-07 21:07:51 +0200397{
398 if (((struct sockaddr_un *)&l->addr)->sun_path[0])
399 return 1;
400
Christopher Faulet510c0d62018-03-16 10:04:47 +0100401 /* Listener's lock already held. Call lockless version of
402 * unbind_listener. */
403 do_unbind_listener(l, 1);
Willy Tarreaufd0e0082014-07-07 21:07:51 +0200404 return 0;
405}
406
Willy Tarreau47f48c42014-05-09 22:57:47 +0200407
408/*
409 * This function initiates a UNIX connection establishment to the target assigned
410 * to connection <conn> using (si->{target,addr.to}). The source address is ignored
411 * and will be selected by the system. conn->target may point either to a valid
412 * server or to a backend, depending on conn->target. Only OBJ_TYPE_PROXY and
413 * OBJ_TYPE_SERVER are supported. The <data> parameter is a boolean indicating
414 * whether there are data waiting for being sent or not, in order to adjust data
415 * write polling and on some platforms. The <delack> argument is ignored.
416 *
417 * Note that a pending send_proxy message accounts for data.
418 *
419 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +0200420 * - SF_ERR_NONE if everything's OK
421 * - SF_ERR_SRVTO if there are no more servers
422 * - SF_ERR_SRVCL if the connection was refused by the server
423 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
424 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
425 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +0100426 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau47f48c42014-05-09 22:57:47 +0200427 *
Willy Tarreaue7dff022015-04-03 01:14:29 +0200428 * The connection's fd is inserted only when SF_ERR_NONE is returned, otherwise
Willy Tarreau47f48c42014-05-09 22:57:47 +0200429 * it's invalid and the caller has nothing to do.
430 */
Willy Tarreau31794892017-09-15 07:59:31 +0200431static int uxst_connect_server(struct connection *conn, int data, int delack)
Willy Tarreau47f48c42014-05-09 22:57:47 +0200432{
433 int fd;
434 struct server *srv;
435 struct proxy *be;
436
Willy Tarreau7bb21532014-05-10 09:48:28 +0200437 conn->flags = 0;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200438
439 switch (obj_type(conn->target)) {
440 case OBJ_TYPE_PROXY:
441 be = objt_proxy(conn->target);
442 srv = NULL;
443 break;
444 case OBJ_TYPE_SERVER:
445 srv = objt_server(conn->target);
446 be = srv->proxy;
447 break;
448 default:
449 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200450 return SF_ERR_INTERNAL;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200451 }
452
Willy Tarreau585744b2017-08-24 14:31:19 +0200453 if ((fd = conn->handle.fd = socket(PF_UNIX, SOCK_STREAM, 0)) == -1) {
Willy Tarreau47f48c42014-05-09 22:57:47 +0200454 qfprintf(stderr, "Cannot get a server socket.\n");
455
456 if (errno == ENFILE) {
457 conn->err_code = CO_ER_SYS_FDLIM;
458 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100459 "Proxy %s reached system FD limit (maxsock=%d). Please check system tunables.\n",
460 be->id, global.maxsock);
Willy Tarreau47f48c42014-05-09 22:57:47 +0200461 }
462 else if (errno == EMFILE) {
463 conn->err_code = CO_ER_PROC_FDLIM;
464 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100465 "Proxy %s reached process FD limit (maxsock=%d). Please check 'ulimit-n' and restart.\n",
466 be->id, global.maxsock);
Willy Tarreau47f48c42014-05-09 22:57:47 +0200467 }
468 else if (errno == ENOBUFS || errno == ENOMEM) {
469 conn->err_code = CO_ER_SYS_MEMLIM;
470 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100471 "Proxy %s reached system memory limit (maxsock=%d). Please check system tunables.\n",
472 be->id, global.maxsock);
Willy Tarreau47f48c42014-05-09 22:57:47 +0200473 }
474 else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
475 conn->err_code = CO_ER_NOPROTO;
476 }
477 else
478 conn->err_code = CO_ER_SOCK_ERR;
479
480 /* this is a resource error */
481 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200482 return SF_ERR_RESOURCE;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200483 }
484
485 if (fd >= global.maxsock) {
486 /* do not log anything there, it's a normal condition when this option
487 * is used to serialize connections to a server !
488 */
Christopher Faulet767a84b2017-11-24 16:50:31 +0100489 ha_alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
Willy Tarreau47f48c42014-05-09 22:57:47 +0200490 close(fd);
491 conn->err_code = CO_ER_CONF_FDLIM;
492 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200493 return SF_ERR_PRXCOND; /* it is a configuration limit */
Willy Tarreau47f48c42014-05-09 22:57:47 +0200494 }
495
496 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
497 qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
498 close(fd);
499 conn->err_code = CO_ER_SOCK_ERR;
500 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200501 return SF_ERR_INTERNAL;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200502 }
503
504 /* if a send_proxy is there, there are data */
505 data |= conn->send_proxy_ofs;
506
507 if (global.tune.server_sndbuf)
508 setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));
509
510 if (global.tune.server_rcvbuf)
511 setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));
512
Willy Tarreau7bb21532014-05-10 09:48:28 +0200513 if (connect(fd, (struct sockaddr *)&conn->addr.to, get_addr_len(&conn->addr.to)) == -1) {
Willy Tarreau94841792017-01-25 14:27:38 +0100514 if (errno == EINPROGRESS || errno == EALREADY) {
Willy Tarreau7bb21532014-05-10 09:48:28 +0200515 conn->flags |= CO_FL_WAIT_L4_CONN;
516 }
Willy Tarreau94841792017-01-25 14:27:38 +0100517 else if (errno == EISCONN) {
518 conn->flags &= ~CO_FL_WAIT_L4_CONN;
519 }
Willy Tarreau7bb21532014-05-10 09:48:28 +0200520 else if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
Willy Tarreau47f48c42014-05-09 22:57:47 +0200521 char *msg;
522 if (errno == EAGAIN || errno == EADDRNOTAVAIL) {
Lukas Tribus9f256d42016-01-26 20:33:14 +0100523 msg = "can't connect to destination unix socket, check backlog size on the server";
Willy Tarreau47f48c42014-05-09 22:57:47 +0200524 conn->err_code = CO_ER_FREE_PORTS;
525 }
526 else {
527 msg = "local address already in use";
528 conn->err_code = CO_ER_ADDR_INUSE;
529 }
530
531 qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
532 close(fd);
533 send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
534 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200535 return SF_ERR_RESOURCE;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200536 }
537 else if (errno == ETIMEDOUT) {
538 close(fd);
539 conn->err_code = CO_ER_SOCK_ERR;
540 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200541 return SF_ERR_SRVTO;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200542 }
543 else { // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
544 close(fd);
545 conn->err_code = CO_ER_SOCK_ERR;
546 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200547 return SF_ERR_SRVCL;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200548 }
549 }
Willy Tarreau7bb21532014-05-10 09:48:28 +0200550 else {
551 /* connect() already succeeded, which is quite usual for unix
Willy Tarreau94841792017-01-25 14:27:38 +0100552 * sockets. Let's avoid a second connect() probe to complete it.
Willy Tarreau7bb21532014-05-10 09:48:28 +0200553 */
554 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau7bb21532014-05-10 09:48:28 +0200555 }
Willy Tarreau47f48c42014-05-09 22:57:47 +0200556
557 conn->flags |= CO_FL_ADDR_TO_SET;
558
559 /* Prepare to send a few handshakes related to the on-wire protocol. */
560 if (conn->send_proxy_ofs)
561 conn->flags |= CO_FL_SEND_PROXY;
562
563 conn_ctrl_init(conn); /* registers the FD */
Willy Tarreau7bb21532014-05-10 09:48:28 +0200564 fdtab[fd].linger_risk = 0; /* no need to disable lingering */
Willy Tarreau47f48c42014-05-09 22:57:47 +0200565
566 if (conn_xprt_init(conn) < 0) {
Willy Tarreau8c829012017-10-05 18:02:11 +0200567 conn_full_close(conn);
Willy Tarreau47f48c42014-05-09 22:57:47 +0200568 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200569 return SF_ERR_RESOURCE;
Willy Tarreau47f48c42014-05-09 22:57:47 +0200570 }
571
Willy Tarreau94841792017-01-25 14:27:38 +0100572 if (conn->flags & (CO_FL_HANDSHAKE | CO_FL_WAIT_L4_CONN)) {
573 conn_sock_want_send(conn); /* for connect status, proxy protocol or SSL */
574 }
575 else {
576 /* If there's no more handshake, we need to notify the data
577 * layer when the connection is already OK otherwise we'll have
578 * no other opportunity to do it later (eg: health checks).
579 */
580 data = 1;
581 }
582
Willy Tarreau47f48c42014-05-09 22:57:47 +0200583 if (data)
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200584 conn_xprt_want_send(conn); /* prepare to send data if any */
Willy Tarreau47f48c42014-05-09 22:57:47 +0200585
Willy Tarreaue7dff022015-04-03 01:14:29 +0200586 return SF_ERR_NONE; /* connection is OK */
Willy Tarreau47f48c42014-05-09 22:57:47 +0200587}
588
589
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100590/********************************
591 * 3) protocol-oriented functions
592 ********************************/
593
594
Willy Tarreau92fb9832007-10-16 17:34:28 +0200595/* This function creates all UNIX sockets bound to the protocol entry <proto>.
596 * It is intended to be used as the protocol's bind_all() function.
597 * The sockets will be registered but not added to any fd_set, in order not to
598 * loose them across the fork(). A call to uxst_enable_listeners() is needed
599 * to complete initialization.
600 *
601 * The return value is composed from ERR_NONE, ERR_RETRYABLE and ERR_FATAL.
602 */
Emeric Bruncf20bf12010-10-22 16:06:11 +0200603static int uxst_bind_listeners(struct protocol *proto, char *errmsg, int errlen)
Willy Tarreau92fb9832007-10-16 17:34:28 +0200604{
605 struct listener *listener;
606 int err = ERR_NONE;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200607
608 list_for_each_entry(listener, &proto->listeners, proto_list) {
Emeric Bruncf20bf12010-10-22 16:06:11 +0200609 err |= uxst_bind_listener(listener, errmsg, errlen);
610 if (err & ERR_ABORT)
611 break;
Willy Tarreau92fb9832007-10-16 17:34:28 +0200612 }
Willy Tarreau92fb9832007-10-16 17:34:28 +0200613 return err;
614}
615
Willy Tarreau92fb9832007-10-16 17:34:28 +0200616
617/* This function stops all listening UNIX sockets bound to the protocol
618 * <proto>. It does not detaches them from the protocol.
619 * It always returns ERR_NONE.
620 */
621static int uxst_unbind_listeners(struct protocol *proto)
622{
623 struct listener *listener;
624
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100625 list_for_each_entry(listener, &proto->listeners, proto_list)
626 uxst_unbind_listener(listener);
Willy Tarreau92fb9832007-10-16 17:34:28 +0200627 return ERR_NONE;
628}
629
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200630/* parse the "mode" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +0200631static int bind_parse_mode(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200632{
Willy Tarreaua1a247b2017-10-04 14:43:44 +0200633 char *endptr;
634
635 conf->ux.mode = strtol(args[cur_arg + 1], &endptr, 8);
636
637 if (!*args[cur_arg + 1] || *endptr) {
638 memprintf(err, "'%s' : missing or invalid mode '%s' (octal integer expected)", args[cur_arg], args[cur_arg + 1]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200639 return ERR_ALERT | ERR_FATAL;
640 }
641
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200642 return 0;
643}
644
645/* parse the "gid" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +0200646static int bind_parse_gid(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200647{
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200648 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200649 memprintf(err, "'%s' : missing value", args[cur_arg]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200650 return ERR_ALERT | ERR_FATAL;
651 }
652
Willy Tarreau290e63a2012-09-20 18:07:14 +0200653 conf->ux.gid = atol(args[cur_arg + 1]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200654 return 0;
655}
656
657/* parse the "group" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +0200658static int bind_parse_group(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200659{
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200660 struct group *group;
661
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200662 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200663 memprintf(err, "'%s' : missing group name", args[cur_arg]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200664 return ERR_ALERT | ERR_FATAL;
665 }
666
667 group = getgrnam(args[cur_arg + 1]);
668 if (!group) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200669 memprintf(err, "'%s' : unknown group name '%s'", args[cur_arg], args[cur_arg + 1]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200670 return ERR_ALERT | ERR_FATAL;
671 }
672
Willy Tarreau290e63a2012-09-20 18:07:14 +0200673 conf->ux.gid = group->gr_gid;
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200674 return 0;
675}
676
677/* parse the "uid" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +0200678static int bind_parse_uid(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200679{
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200680 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200681 memprintf(err, "'%s' : missing value", args[cur_arg]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200682 return ERR_ALERT | ERR_FATAL;
683 }
684
Willy Tarreau290e63a2012-09-20 18:07:14 +0200685 conf->ux.uid = atol(args[cur_arg + 1]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200686 return 0;
687}
688
689/* parse the "user" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +0200690static int bind_parse_user(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200691{
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200692 struct passwd *user;
693
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200694 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200695 memprintf(err, "'%s' : missing user name", args[cur_arg]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200696 return ERR_ALERT | ERR_FATAL;
697 }
698
699 user = getpwnam(args[cur_arg + 1]);
700 if (!user) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +0200701 memprintf(err, "'%s' : unknown user name '%s'", args[cur_arg], args[cur_arg + 1]);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200702 return ERR_ALERT | ERR_FATAL;
703 }
704
Willy Tarreau290e63a2012-09-20 18:07:14 +0200705 conf->ux.uid = user->pw_uid;
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200706 return 0;
707}
708
709/* Note: must not be declared <const> as its list will be overwritten.
710 * Please take care of keeping this list alphabetically sorted, doing so helps
711 * all code contributors.
712 * Optional keywords are also declared with a NULL ->parse() function so that
713 * the config parser can report an appropriate error when a known keyword was
714 * not enabled.
715 */
Willy Tarreau51fb7652012-09-18 18:24:39 +0200716static struct bind_kw_list bind_kws = { "UNIX", { }, {
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200717 { "gid", bind_parse_gid, 1 }, /* set the socket's gid */
718 { "group", bind_parse_group, 1 }, /* set the socket's gid from the group name */
719 { "mode", bind_parse_mode, 1 }, /* set the socket's mode (eg: 0644)*/
720 { "uid", bind_parse_uid, 1 }, /* set the socket's uid */
721 { "user", bind_parse_user, 1 }, /* set the socket's uid from the user name */
722 { NULL, NULL, 0 },
723}};
Willy Tarreaudabf2e22007-10-28 21:59:24 +0100724
725/********************************
726 * 4) high-level functions
727 ********************************/
728
Willy Tarreau92fb9832007-10-16 17:34:28 +0200729__attribute__((constructor))
730static void __uxst_protocol_init(void)
731{
732 protocol_register(&proto_unix);
Willy Tarreaud0a895d2012-09-18 17:40:35 +0200733 bind_register_keywords(&bind_kws);
Willy Tarreau92fb9832007-10-16 17:34:28 +0200734}
735
736
737/*
738 * Local variables:
739 * c-indent-level: 8
740 * c-basic-offset: 8
741 * End:
742 */