]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
e0edde6f | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "socket-util.h" | |
19 | #include <arpa/inet.h> | |
cea15768 | 20 | #include <assert.h> |
064af421 BP |
21 | #include <errno.h> |
22 | #include <fcntl.h> | |
f89ffb0e | 23 | #include <net/if.h> |
064af421 BP |
24 | #include <netdb.h> |
25 | #include <poll.h> | |
26 | #include <stddef.h> | |
27 | #include <stdio.h> | |
78ff0270 | 28 | #include <stdlib.h> |
064af421 BP |
29 | #include <string.h> |
30 | #include <sys/resource.h> | |
3762274e BP |
31 | #include <sys/socket.h> |
32 | #include <sys/stat.h> | |
a7a27d07 | 33 | #include <sys/uio.h> |
064af421 BP |
34 | #include <sys/un.h> |
35 | #include <unistd.h> | |
f89ffb0e | 36 | #include "dynamic-string.h" |
064af421 | 37 | #include "fatal-signal.h" |
f89ffb0e | 38 | #include "packets.h" |
fd94a42c | 39 | #include "poll-loop.h" |
064af421 | 40 | #include "util.h" |
064af421 | 41 | #include "vlog.h" |
f89ffb0e BP |
42 | #if AF_PACKET && __linux__ |
43 | #include <linux/if_packet.h> | |
44 | #endif | |
45 | #ifdef HAVE_NETLINK | |
46 | #include "netlink-protocol.h" | |
47 | #include "netlink-socket.h" | |
48 | #endif | |
5136ce49 | 49 | |
d98e6007 | 50 | VLOG_DEFINE_THIS_MODULE(socket_util); |
064af421 | 51 | |
e61c056c BP |
52 | /* #ifdefs make it a pain to maintain code: you have to try to build both ways. |
53 | * Thus, this file compiles all of the code regardless of the target, by | |
54 | * writing "if (LINUX)" instead of "#ifdef __linux__". */ | |
55 | #ifdef __linux__ | |
56 | #define LINUX 1 | |
57 | #else | |
58 | #define LINUX 0 | |
59 | #endif | |
60 | ||
6e170b4c BP |
61 | #ifndef O_DIRECTORY |
62 | #define O_DIRECTORY 0 | |
63 | #endif | |
64 | ||
c1c19657 BP |
65 | static int getsockopt_int(int fd, int level, int option, const char *optname, |
66 | int *valuep); | |
67 | ||
064af421 BP |
68 | /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a |
69 | * positive errno value. */ | |
70 | int | |
71 | set_nonblocking(int fd) | |
72 | { | |
73 | int flags = fcntl(fd, F_GETFL, 0); | |
74 | if (flags != -1) { | |
75 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { | |
76 | return 0; | |
77 | } else { | |
78 | VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); | |
79 | return errno; | |
80 | } | |
81 | } else { | |
82 | VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); | |
83 | return errno; | |
84 | } | |
85 | } | |
86 | ||
a0505c49 BP |
87 | void |
88 | xset_nonblocking(int fd) | |
89 | { | |
90 | if (set_nonblocking(fd)) { | |
91 | exit(EXIT_FAILURE); | |
92 | } | |
93 | } | |
94 | ||
6b9c1eab | 95 | int |
cea15768 EJ |
96 | set_dscp(int fd, uint8_t dscp) |
97 | { | |
eb0cb316 EM |
98 | int val; |
99 | ||
cea15768 EJ |
100 | if (dscp > 63) { |
101 | return EINVAL; | |
102 | } | |
103 | ||
eb0cb316 EM |
104 | val = dscp << 2; |
105 | if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) { | |
cea15768 EJ |
106 | return errno; |
107 | } | |
108 | ||
109 | return 0; | |
110 | } | |
111 | ||
0ec6cfb1 BP |
112 | static bool |
113 | rlim_is_finite(rlim_t limit) | |
114 | { | |
115 | if (limit == RLIM_INFINITY) { | |
116 | return false; | |
117 | } | |
118 | ||
119 | #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */ | |
120 | if (limit == RLIM_SAVED_CUR) { | |
121 | return false; | |
122 | } | |
123 | #endif | |
124 | ||
125 | #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */ | |
126 | if (limit == RLIM_SAVED_MAX) { | |
127 | return false; | |
128 | } | |
129 | #endif | |
130 | ||
131 | return true; | |
132 | } | |
133 | ||
064af421 BP |
134 | /* Returns the maximum valid FD value, plus 1. */ |
135 | int | |
136 | get_max_fds(void) | |
137 | { | |
138 | static int max_fds = -1; | |
139 | if (max_fds < 0) { | |
140 | struct rlimit r; | |
0ec6cfb1 | 141 | if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) { |
064af421 BP |
142 | max_fds = r.rlim_cur; |
143 | } else { | |
144 | VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); | |
145 | max_fds = 1024; | |
146 | } | |
147 | } | |
148 | return max_fds; | |
149 | } | |
150 | ||
2b35e147 BP |
151 | /* Translates 'host_name', which must be a string representation of an IP |
152 | * address, into a numeric IP address in '*addr'. Returns 0 if successful, | |
153 | * otherwise a positive errno value. */ | |
064af421 | 154 | int |
d295e8e9 | 155 | lookup_ip(const char *host_name, struct in_addr *addr) |
064af421 BP |
156 | { |
157 | if (!inet_aton(host_name, addr)) { | |
db5ce514 | 158 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2b35e147 BP |
159 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name); |
160 | return ENOENT; | |
064af421 BP |
161 | } |
162 | return 0; | |
163 | } | |
164 | ||
d31f1109 JP |
165 | /* Translates 'host_name', which must be a string representation of an IPv6 |
166 | * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, | |
167 | * otherwise a positive errno value. */ | |
168 | int | |
169 | lookup_ipv6(const char *host_name, struct in6_addr *addr) | |
170 | { | |
171 | if (inet_pton(AF_INET6, host_name, addr) != 1) { | |
172 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
173 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name); | |
174 | return ENOENT; | |
175 | } | |
176 | return 0; | |
177 | } | |
178 | ||
13f2ef97 BP |
179 | /* Translates 'host_name', which must be a host name or a string representation |
180 | * of an IP address, into a numeric IP address in '*addr'. Returns 0 if | |
181 | * successful, otherwise a positive errno value. | |
182 | * | |
183 | * Most Open vSwitch code should not use this because it causes deadlocks: | |
184 | * gethostbyname() sends out a DNS request but that starts a new flow for which | |
185 | * OVS must set up a flow, but it can't because it's waiting for a DNS reply. | |
293d49bd | 186 | * The synchronous lookup also delays other activity. (Of course we can solve |
13f2ef97 BP |
187 | * this but it doesn't seem worthwhile quite yet.) */ |
188 | int | |
189 | lookup_hostname(const char *host_name, struct in_addr *addr) | |
190 | { | |
191 | struct hostent *h; | |
192 | ||
193 | if (inet_aton(host_name, addr)) { | |
194 | return 0; | |
195 | } | |
196 | ||
197 | h = gethostbyname(host_name); | |
198 | if (h) { | |
199 | *addr = *(struct in_addr *) h->h_addr; | |
200 | return 0; | |
201 | } | |
202 | ||
203 | return (h_errno == HOST_NOT_FOUND ? ENOENT | |
204 | : h_errno == TRY_AGAIN ? EAGAIN | |
205 | : h_errno == NO_RECOVERY ? EIO | |
206 | : h_errno == NO_ADDRESS ? ENXIO | |
207 | : EINVAL); | |
208 | } | |
209 | ||
064af421 BP |
210 | /* Returns the error condition associated with socket 'fd' and resets the |
211 | * socket's error status. */ | |
212 | int | |
d295e8e9 | 213 | get_socket_error(int fd) |
064af421 BP |
214 | { |
215 | int error; | |
c1c19657 BP |
216 | |
217 | if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) { | |
064af421 | 218 | error = errno; |
064af421 BP |
219 | } |
220 | return error; | |
221 | } | |
222 | ||
223 | int | |
d295e8e9 | 224 | check_connection_completion(int fd) |
064af421 BP |
225 | { |
226 | struct pollfd pfd; | |
227 | int retval; | |
228 | ||
229 | pfd.fd = fd; | |
230 | pfd.events = POLLOUT; | |
231 | do { | |
232 | retval = poll(&pfd, 1, 0); | |
233 | } while (retval < 0 && errno == EINTR); | |
234 | if (retval == 1) { | |
235 | return get_socket_error(fd); | |
236 | } else if (retval < 0) { | |
237 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); | |
238 | VLOG_ERR_RL(&rl, "poll: %s", strerror(errno)); | |
239 | return errno; | |
240 | } else { | |
241 | return EAGAIN; | |
242 | } | |
243 | } | |
244 | ||
245 | /* Drain all the data currently in the receive queue of a datagram socket (and | |
246 | * possibly additional data). There is no way to know how many packets are in | |
247 | * the receive queue, but we do know that the total number of bytes queued does | |
248 | * not exceed the receive buffer size, so we pull packets until none are left | |
249 | * or we've read that many bytes. */ | |
250 | int | |
251 | drain_rcvbuf(int fd) | |
252 | { | |
c1c19657 | 253 | int rcvbuf; |
064af421 | 254 | |
c1c19657 BP |
255 | rcvbuf = get_socket_rcvbuf(fd); |
256 | if (rcvbuf < 0) { | |
257 | return -rcvbuf; | |
064af421 | 258 | } |
c1c19657 | 259 | |
064af421 BP |
260 | while (rcvbuf > 0) { |
261 | /* In Linux, specifying MSG_TRUNC in the flags argument causes the | |
262 | * datagram length to be returned, even if that is longer than the | |
263 | * buffer provided. Thus, we can use a 1-byte buffer to discard the | |
264 | * incoming datagram and still be able to account how many bytes were | |
265 | * removed from the receive buffer. | |
266 | * | |
267 | * On other Unix-like OSes, MSG_TRUNC has no effect in the flags | |
268 | * argument. */ | |
e61c056c | 269 | char buffer[LINUX ? 1 : 2048]; |
064af421 BP |
270 | ssize_t n_bytes = recv(fd, buffer, sizeof buffer, |
271 | MSG_TRUNC | MSG_DONTWAIT); | |
272 | if (n_bytes <= 0 || n_bytes >= rcvbuf) { | |
273 | break; | |
274 | } | |
275 | rcvbuf -= n_bytes; | |
276 | } | |
277 | return 0; | |
278 | } | |
279 | ||
c1c19657 BP |
280 | /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a |
281 | * negative errno value if an error occurs. */ | |
282 | int | |
283 | get_socket_rcvbuf(int sock) | |
284 | { | |
285 | int rcvbuf; | |
286 | int error; | |
287 | ||
288 | error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf); | |
289 | return error ? -error : rcvbuf; | |
290 | } | |
291 | ||
064af421 BP |
292 | /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no |
293 | * more data can be immediately read. ('fd' should therefore be in | |
294 | * non-blocking mode.)*/ | |
295 | void | |
296 | drain_fd(int fd, size_t n_packets) | |
297 | { | |
298 | for (; n_packets > 0; n_packets--) { | |
299 | /* 'buffer' only needs to be 1 byte long in most circumstances. This | |
300 | * size is defensive against the possibility that we someday want to | |
301 | * use a Linux tap device without TUN_NO_PI, in which case a buffer | |
302 | * smaller than sizeof(struct tun_pi) will give EINVAL on read. */ | |
303 | char buffer[128]; | |
304 | if (read(fd, buffer, sizeof buffer) <= 0) { | |
305 | break; | |
306 | } | |
307 | } | |
308 | } | |
309 | ||
310 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in | |
311 | * '*un_len' the size of the sockaddr_un. */ | |
312 | static void | |
6e170b4c | 313 | make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len) |
064af421 BP |
314 | { |
315 | un->sun_family = AF_UNIX; | |
71d7c22f | 316 | ovs_strzcpy(un->sun_path, name, sizeof un->sun_path); |
064af421 BP |
317 | *un_len = (offsetof(struct sockaddr_un, sun_path) |
318 | + strlen (un->sun_path) + 1); | |
319 | } | |
320 | ||
6e170b4c BP |
321 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in |
322 | * '*un_len' the size of the sockaddr_un. | |
323 | * | |
324 | * Returns 0 on success, otherwise a positive errno value. On success, | |
325 | * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller | |
326 | * should close after using '*un' to bind or connect. On failure, '*dirfdp' is | |
327 | * -1. */ | |
328 | static int | |
329 | make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, | |
330 | int *dirfdp) | |
331 | { | |
332 | enum { MAX_UN_LEN = sizeof un->sun_path - 1 }; | |
333 | ||
334 | *dirfdp = -1; | |
335 | if (strlen(name) > MAX_UN_LEN) { | |
336 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
337 | ||
338 | if (LINUX) { | |
339 | /* 'name' is too long to fit in a sockaddr_un, but we have a | |
340 | * workaround for that on Linux: shorten it by opening a file | |
341 | * descriptor for the directory part of the name and indirecting | |
342 | * through /proc/self/fd/<dirfd>/<basename>. */ | |
343 | char *dir, *base; | |
344 | char *short_name; | |
345 | int dirfd; | |
346 | ||
347 | dir = dir_name(name); | |
348 | base = base_name(name); | |
349 | ||
350 | dirfd = open(dir, O_DIRECTORY | O_RDONLY); | |
351 | if (dirfd < 0) { | |
08871bf6 JP |
352 | free(base); |
353 | free(dir); | |
6e170b4c BP |
354 | return errno; |
355 | } | |
356 | ||
357 | short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base); | |
358 | free(dir); | |
359 | free(base); | |
360 | ||
361 | if (strlen(short_name) <= MAX_UN_LEN) { | |
362 | make_sockaddr_un__(short_name, un, un_len); | |
363 | free(short_name); | |
364 | *dirfdp = dirfd; | |
365 | return 0; | |
366 | } | |
367 | free(short_name); | |
368 | close(dirfd); | |
369 | ||
370 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
371 | "%d bytes (even shortened)", name, MAX_UN_LEN); | |
372 | } else { | |
373 | /* 'name' is too long and we have no workaround. */ | |
374 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
375 | "%d bytes", name, MAX_UN_LEN); | |
376 | } | |
377 | ||
378 | return ENAMETOOLONG; | |
379 | } else { | |
380 | make_sockaddr_un__(name, un, un_len); | |
381 | return 0; | |
382 | } | |
383 | } | |
384 | ||
e2bd0593 BP |
385 | /* Binds Unix domain socket 'fd' to a file with permissions 0700. */ |
386 | static int | |
387 | bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len) | |
388 | { | |
e2bd0593 BP |
389 | /* According to _Unix Network Programming_, umask should affect bind(). */ |
390 | mode_t old_umask = umask(0077); | |
391 | int error = bind(fd, sun, sun_len) ? errno : 0; | |
392 | umask(old_umask); | |
393 | return error; | |
e2bd0593 BP |
394 | } |
395 | ||
064af421 BP |
396 | /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or |
397 | * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and | |
398 | * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock' | |
5ca92d1d | 399 | * is true, the socket is made non-blocking. |
064af421 BP |
400 | * |
401 | * Returns the socket's fd if successful, otherwise a negative errno value. */ | |
402 | int | |
5ca92d1d | 403 | make_unix_socket(int style, bool nonblock, |
064af421 BP |
404 | const char *bind_path, const char *connect_path) |
405 | { | |
406 | int error; | |
407 | int fd; | |
408 | ||
409 | fd = socket(PF_UNIX, style, 0); | |
410 | if (fd < 0) { | |
411 | return -errno; | |
412 | } | |
413 | ||
414 | /* Set nonblocking mode right away, if we want it. This prevents blocking | |
415 | * in connect(), if connect_path != NULL. (In turn, that's a corner case: | |
416 | * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only | |
417 | * if a backlog of un-accepted connections has built up in the kernel.) */ | |
418 | if (nonblock) { | |
419 | int flags = fcntl(fd, F_GETFL, 0); | |
420 | if (flags == -1) { | |
6e170b4c | 421 | error = errno; |
064af421 BP |
422 | goto error; |
423 | } | |
424 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { | |
6e170b4c | 425 | error = errno; |
064af421 BP |
426 | goto error; |
427 | } | |
428 | } | |
429 | ||
430 | if (bind_path) { | |
431 | struct sockaddr_un un; | |
432 | socklen_t un_len; | |
6e170b4c BP |
433 | int dirfd; |
434 | ||
435 | if (unlink(bind_path) && errno != ENOENT) { | |
436 | VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); | |
064af421 BP |
437 | } |
438 | fatal_signal_add_file_to_unlink(bind_path); | |
6e170b4c BP |
439 | |
440 | error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd); | |
e2bd0593 BP |
441 | if (!error) { |
442 | error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len); | |
6e170b4c BP |
443 | } |
444 | if (dirfd >= 0) { | |
445 | close(dirfd); | |
446 | } | |
447 | if (error) { | |
064af421 BP |
448 | goto error; |
449 | } | |
450 | } | |
451 | ||
452 | if (connect_path) { | |
453 | struct sockaddr_un un; | |
454 | socklen_t un_len; | |
6e170b4c BP |
455 | int dirfd; |
456 | ||
457 | error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd); | |
458 | if (!error | |
459 | && connect(fd, (struct sockaddr*) &un, un_len) | |
064af421 | 460 | && errno != EINPROGRESS) { |
6e170b4c BP |
461 | error = errno; |
462 | } | |
463 | if (dirfd >= 0) { | |
464 | close(dirfd); | |
465 | } | |
466 | if (error) { | |
064af421 BP |
467 | goto error; |
468 | } | |
469 | } | |
470 | ||
064af421 BP |
471 | return fd; |
472 | ||
473 | error: | |
6e170b4c BP |
474 | if (error == EAGAIN) { |
475 | error = EPROTO; | |
476 | } | |
064af421 | 477 | if (bind_path) { |
e36d0b53 | 478 | fatal_signal_unlink_file_now(bind_path); |
064af421 | 479 | } |
064af421 BP |
480 | close(fd); |
481 | return -error; | |
482 | } | |
483 | ||
484 | int | |
485 | get_unix_name_len(socklen_t sun_len) | |
486 | { | |
487 | return (sun_len >= offsetof(struct sockaddr_un, sun_path) | |
488 | ? sun_len - offsetof(struct sockaddr_un, sun_path) | |
489 | : 0); | |
490 | } | |
491 | ||
dbba996b BP |
492 | ovs_be32 |
493 | guess_netmask(ovs_be32 ip_) | |
064af421 | 494 | { |
dbba996b | 495 | uint32_t ip = ntohl(ip_); |
064af421 BP |
496 | return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ |
497 | : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ | |
498 | : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ | |
499 | : htonl(0)); /* ??? */ | |
500 | } | |
501 | ||
52f8a75e BP |
502 | /* Parses 'target', which should be a string in the format "<host>[:<port>]". |
503 | * <host> is required. If 'default_port' is nonzero then <port> is optional | |
504 | * and defaults to 'default_port'. | |
4f2eb9a7 | 505 | * |
52f8a75e BP |
506 | * On success, returns true and stores the parsed remote address into '*sinp'. |
507 | * On failure, logs an error, stores zeros into '*sinp', and returns false. */ | |
508 | bool | |
509 | inet_parse_active(const char *target_, uint16_t default_port, | |
510 | struct sockaddr_in *sinp) | |
78ff0270 BP |
511 | { |
512 | char *target = xstrdup(target_); | |
513 | char *save_ptr = NULL; | |
514 | const char *host_name; | |
515 | const char *port_string; | |
52f8a75e | 516 | bool ok = false; |
78ff0270 BP |
517 | |
518 | /* Defaults. */ | |
52f8a75e BP |
519 | sinp->sin_family = AF_INET; |
520 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
521 | |
522 | /* Tokenize. */ | |
523 | host_name = strtok_r(target, ":", &save_ptr); | |
524 | port_string = strtok_r(NULL, ":", &save_ptr); | |
525 | if (!host_name) { | |
52f8a75e | 526 | VLOG_ERR("%s: bad peer name format", target_); |
78ff0270 BP |
527 | goto exit; |
528 | } | |
529 | ||
530 | /* Look up IP, port. */ | |
52f8a75e | 531 | if (lookup_ip(host_name, &sinp->sin_addr)) { |
78ff0270 BP |
532 | goto exit; |
533 | } | |
534 | if (port_string && atoi(port_string)) { | |
52f8a75e | 535 | sinp->sin_port = htons(atoi(port_string)); |
1901968e BP |
536 | } else if (!default_port) { |
537 | VLOG_ERR("%s: port number must be specified", target_); | |
52f8a75e BP |
538 | goto exit; |
539 | } | |
540 | ||
541 | ok = true; | |
542 | ||
543 | exit: | |
544 | if (!ok) { | |
545 | memset(sinp, 0, sizeof *sinp); | |
546 | } | |
547 | free(target); | |
548 | return ok; | |
549 | } | |
550 | ||
551 | /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to | |
552 | * 'target', which should be a string in the format "<host>[:<port>]". <host> | |
553 | * is required. If 'default_port' is nonzero then <port> is optional and | |
554 | * defaults to 'default_port'. | |
555 | * | |
556 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
557 | * | |
558 | * On success, returns 0 (indicating connection complete) or EAGAIN (indicating | |
559 | * connection in progress), in which case the new file descriptor is stored | |
560 | * into '*fdp'. On failure, returns a positive errno value other than EAGAIN | |
561 | * and stores -1 into '*fdp'. | |
562 | * | |
563 | * If 'sinp' is non-null, then on success the target address is stored into | |
f125905c MM |
564 | * '*sinp'. |
565 | * | |
cea15768 EJ |
566 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
567 | * should be in the range [0, 63] and will automatically be shifted to the | |
568 | * appropriately place in the IP tos field. */ | |
52f8a75e BP |
569 | int |
570 | inet_open_active(int style, const char *target, uint16_t default_port, | |
f125905c | 571 | struct sockaddr_in *sinp, int *fdp, uint8_t dscp) |
52f8a75e BP |
572 | { |
573 | struct sockaddr_in sin; | |
574 | int fd = -1; | |
575 | int error; | |
576 | ||
577 | /* Parse. */ | |
578 | if (!inet_parse_active(target, default_port, &sin)) { | |
1901968e BP |
579 | error = EAFNOSUPPORT; |
580 | goto exit; | |
78ff0270 BP |
581 | } |
582 | ||
583 | /* Create non-blocking socket. */ | |
4f2eb9a7 | 584 | fd = socket(AF_INET, style, 0); |
78ff0270 | 585 | if (fd < 0) { |
52f8a75e | 586 | VLOG_ERR("%s: socket: %s", target, strerror(errno)); |
78ff0270 BP |
587 | error = errno; |
588 | goto exit; | |
589 | } | |
590 | error = set_nonblocking(fd); | |
591 | if (error) { | |
a4efa3fc | 592 | goto exit; |
78ff0270 BP |
593 | } |
594 | ||
cea15768 EJ |
595 | /* The dscp bits must be configured before connect() to ensure that the TOS |
596 | * field is set during the connection establishment. If set after | |
597 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
598 | error = set_dscp(fd, dscp); | |
599 | if (error) { | |
600 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 601 | goto exit; |
f125905c MM |
602 | } |
603 | ||
78ff0270 BP |
604 | /* Connect. */ |
605 | error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno; | |
606 | if (error == EINPROGRESS) { | |
607 | error = EAGAIN; | |
78ff0270 BP |
608 | } |
609 | ||
78ff0270 BP |
610 | exit: |
611 | if (!error || error == EAGAIN) { | |
612 | if (sinp) { | |
613 | *sinp = sin; | |
614 | } | |
a4efa3fc EJ |
615 | } else if (fd >= 0) { |
616 | close(fd); | |
78ff0270 | 617 | } |
a4efa3fc | 618 | *fdp = fd; |
78ff0270 BP |
619 | return error; |
620 | } | |
621 | ||
d98fa503 | 622 | /* Parses 'target', which should be a string in the format "[<port>][:<ip>]": |
e1bd3bee BP |
623 | * |
624 | * - If 'default_port' is -1, then <port> is required. Otherwise, if | |
625 | * <port> is omitted, then 'default_port' is used instead. | |
626 | * | |
627 | * - If <port> (or 'default_port', if used) is 0, then no port is bound | |
628 | * and the TCP/IP stack will select a port. | |
629 | * | |
630 | * - If <ip> is omitted then the IP address is wildcarded. | |
4f2eb9a7 | 631 | * |
d98fa503 BP |
632 | * If successful, stores the address into '*sinp' and returns true; otherwise |
633 | * zeros '*sinp' and returns false. */ | |
634 | bool | |
995337c4 | 635 | inet_parse_passive(const char *target_, int default_port, |
d98fa503 | 636 | struct sockaddr_in *sinp) |
78ff0270 BP |
637 | { |
638 | char *target = xstrdup(target_); | |
639 | char *string_ptr = target; | |
78ff0270 BP |
640 | const char *host_name; |
641 | const char *port_string; | |
d98fa503 BP |
642 | bool ok = false; |
643 | int port; | |
78ff0270 BP |
644 | |
645 | /* Address defaults. */ | |
d98fa503 BP |
646 | memset(sinp, 0, sizeof *sinp); |
647 | sinp->sin_family = AF_INET; | |
648 | sinp->sin_addr.s_addr = htonl(INADDR_ANY); | |
649 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
650 | |
651 | /* Parse optional port number. */ | |
652 | port_string = strsep(&string_ptr, ":"); | |
e1bd3bee | 653 | if (port_string && str_to_int(port_string, 10, &port)) { |
d98fa503 | 654 | sinp->sin_port = htons(port); |
e1bd3bee | 655 | } else if (default_port < 0) { |
1901968e | 656 | VLOG_ERR("%s: port number must be specified", target_); |
1901968e | 657 | goto exit; |
78ff0270 BP |
658 | } |
659 | ||
660 | /* Parse optional bind IP. */ | |
661 | host_name = strsep(&string_ptr, ":"); | |
d98fa503 BP |
662 | if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { |
663 | goto exit; | |
664 | } | |
665 | ||
666 | ok = true; | |
667 | ||
668 | exit: | |
669 | if (!ok) { | |
670 | memset(sinp, 0, sizeof *sinp); | |
671 | } | |
672 | free(target); | |
673 | return ok; | |
674 | } | |
675 | ||
676 | ||
677 | /* Opens a non-blocking IPv4 socket of the specified 'style', binds to | |
678 | * 'target', and listens for incoming connections. Parses 'target' in the same | |
679 | * way was inet_parse_passive(). | |
680 | * | |
681 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
682 | * | |
683 | * For TCP, the socket will have SO_REUSEADDR turned on. | |
684 | * | |
685 | * On success, returns a non-negative file descriptor. On failure, returns a | |
686 | * negative errno value. | |
687 | * | |
688 | * If 'sinp' is non-null, then on success the bound address is stored into | |
f125905c MM |
689 | * '*sinp'. |
690 | * | |
cea15768 EJ |
691 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
692 | * should be in the range [0, 63] and will automatically be shifted to the | |
693 | * appropriately place in the IP tos field. */ | |
d98fa503 BP |
694 | int |
695 | inet_open_passive(int style, const char *target, int default_port, | |
f125905c | 696 | struct sockaddr_in *sinp, uint8_t dscp) |
d98fa503 BP |
697 | { |
698 | struct sockaddr_in sin; | |
699 | int fd = 0, error; | |
700 | unsigned int yes = 1; | |
701 | ||
702 | if (!inet_parse_passive(target, default_port, &sin)) { | |
ca286ba9 | 703 | return -EAFNOSUPPORT; |
78ff0270 BP |
704 | } |
705 | ||
706 | /* Create non-blocking socket, set SO_REUSEADDR. */ | |
4f2eb9a7 | 707 | fd = socket(AF_INET, style, 0); |
78ff0270 BP |
708 | if (fd < 0) { |
709 | error = errno; | |
d98fa503 | 710 | VLOG_ERR("%s: socket: %s", target, strerror(error)); |
ca286ba9 | 711 | return -error; |
78ff0270 BP |
712 | } |
713 | error = set_nonblocking(fd); | |
714 | if (error) { | |
d98fa503 | 715 | goto error; |
78ff0270 | 716 | } |
4f2eb9a7 BP |
717 | if (style == SOCK_STREAM |
718 | && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { | |
78ff0270 | 719 | error = errno; |
d98fa503 BP |
720 | VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error)); |
721 | goto error; | |
78ff0270 BP |
722 | } |
723 | ||
724 | /* Bind. */ | |
725 | if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) { | |
726 | error = errno; | |
d98fa503 BP |
727 | VLOG_ERR("%s: bind: %s", target, strerror(error)); |
728 | goto error; | |
78ff0270 BP |
729 | } |
730 | ||
cea15768 EJ |
731 | /* The dscp bits must be configured before connect() to ensure that the TOS |
732 | * field is set during the connection establishment. If set after | |
733 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
734 | error = set_dscp(fd, dscp); | |
735 | if (error) { | |
736 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 737 | goto error; |
f125905c MM |
738 | } |
739 | ||
78ff0270 | 740 | /* Listen. */ |
041dc07f | 741 | if (style == SOCK_STREAM && listen(fd, 10) < 0) { |
78ff0270 | 742 | error = errno; |
d98fa503 BP |
743 | VLOG_ERR("%s: listen: %s", target, strerror(error)); |
744 | goto error; | |
78ff0270 | 745 | } |
36775dad BP |
746 | |
747 | if (sinp) { | |
748 | socklen_t sin_len = sizeof sin; | |
749 | if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){ | |
750 | error = errno; | |
d98fa503 BP |
751 | VLOG_ERR("%s: getsockname: %s", target, strerror(error)); |
752 | goto error; | |
36775dad BP |
753 | } |
754 | if (sin.sin_family != AF_INET || sin_len != sizeof sin) { | |
ca286ba9 | 755 | error = EAFNOSUPPORT; |
d98fa503 BP |
756 | VLOG_ERR("%s: getsockname: invalid socket name", target); |
757 | goto error; | |
36775dad BP |
758 | } |
759 | *sinp = sin; | |
760 | } | |
761 | ||
d98fa503 | 762 | return fd; |
78ff0270 | 763 | |
d98fa503 | 764 | error: |
78ff0270 | 765 | close(fd); |
ca286ba9 | 766 | return -error; |
78ff0270 BP |
767 | } |
768 | ||
a8b5f8b4 BP |
769 | /* Returns a readable and writable fd for /dev/null, if successful, otherwise |
770 | * a negative errno value. The caller must not close the returned fd (because | |
771 | * the same fd will be handed out to subsequent callers). */ | |
772 | int | |
773 | get_null_fd(void) | |
774 | { | |
775 | static int null_fd = -1; | |
776 | if (null_fd < 0) { | |
777 | null_fd = open("/dev/null", O_RDWR); | |
778 | if (null_fd < 0) { | |
779 | int error = errno; | |
780 | VLOG_ERR("could not open /dev/null: %s", strerror(error)); | |
781 | return -error; | |
782 | } | |
783 | } | |
784 | return null_fd; | |
785 | } | |
786 | ||
064af421 BP |
787 | int |
788 | read_fully(int fd, void *p_, size_t size, size_t *bytes_read) | |
789 | { | |
790 | uint8_t *p = p_; | |
791 | ||
792 | *bytes_read = 0; | |
793 | while (size > 0) { | |
794 | ssize_t retval = read(fd, p, size); | |
795 | if (retval > 0) { | |
796 | *bytes_read += retval; | |
797 | size -= retval; | |
798 | p += retval; | |
799 | } else if (retval == 0) { | |
800 | return EOF; | |
801 | } else if (errno != EINTR) { | |
802 | return errno; | |
803 | } | |
804 | } | |
805 | return 0; | |
806 | } | |
807 | ||
808 | int | |
809 | write_fully(int fd, const void *p_, size_t size, size_t *bytes_written) | |
810 | { | |
811 | const uint8_t *p = p_; | |
812 | ||
813 | *bytes_written = 0; | |
814 | while (size > 0) { | |
815 | ssize_t retval = write(fd, p, size); | |
816 | if (retval > 0) { | |
817 | *bytes_written += retval; | |
818 | size -= retval; | |
819 | p += retval; | |
820 | } else if (retval == 0) { | |
821 | VLOG_WARN("write returned 0"); | |
822 | return EPROTO; | |
823 | } else if (errno != EINTR) { | |
824 | return errno; | |
825 | } | |
826 | } | |
827 | return 0; | |
828 | } | |
8e71cf88 BP |
829 | |
830 | /* Given file name 'file_name', fsyncs the directory in which it is contained. | |
831 | * Returns 0 if successful, otherwise a positive errno value. */ | |
832 | int | |
833 | fsync_parent_dir(const char *file_name) | |
834 | { | |
835 | int error = 0; | |
836 | char *dir; | |
837 | int fd; | |
838 | ||
839 | dir = dir_name(file_name); | |
840 | fd = open(dir, O_RDONLY); | |
841 | if (fd >= 0) { | |
842 | if (fsync(fd)) { | |
843 | if (errno == EINVAL || errno == EROFS) { | |
844 | /* This directory does not support synchronization. Not | |
845 | * really an error. */ | |
846 | } else { | |
847 | error = errno; | |
848 | VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); | |
849 | } | |
850 | } | |
851 | close(fd); | |
852 | } else { | |
853 | error = errno; | |
854 | VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); | |
855 | } | |
856 | free(dir); | |
857 | ||
858 | return error; | |
859 | } | |
26efd256 BP |
860 | |
861 | /* Obtains the modification time of the file named 'file_name' to the greatest | |
862 | * supported precision. If successful, stores the mtime in '*mtime' and | |
863 | * returns 0. On error, returns a positive errno value and stores zeros in | |
864 | * '*mtime'. */ | |
865 | int | |
866 | get_mtime(const char *file_name, struct timespec *mtime) | |
867 | { | |
868 | struct stat s; | |
869 | ||
870 | if (!stat(file_name, &s)) { | |
871 | mtime->tv_sec = s.st_mtime; | |
872 | ||
873 | #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC | |
874 | mtime->tv_nsec = s.st_mtim.tv_nsec; | |
875 | #elif HAVE_STRUCT_STAT_ST_MTIMENSEC | |
876 | mtime->tv_nsec = s.st_mtimensec; | |
877 | #else | |
878 | mtime->tv_nsec = 0; | |
879 | #endif | |
880 | ||
881 | return 0; | |
882 | } else { | |
883 | mtime->tv_sec = mtime->tv_nsec = 0; | |
884 | return errno; | |
885 | } | |
886 | } | |
887 | ||
279c9e03 BP |
888 | void |
889 | xpipe(int fds[2]) | |
890 | { | |
891 | if (pipe(fds)) { | |
892 | VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); | |
893 | } | |
894 | } | |
f89ffb0e | 895 | |
fd94a42c BP |
896 | void |
897 | xsocketpair(int domain, int type, int protocol, int fds[2]) | |
898 | { | |
899 | if (socketpair(domain, type, protocol, fds)) { | |
900 | VLOG_FATAL("failed to create socketpair (%s)", strerror(errno)); | |
901 | } | |
902 | } | |
903 | ||
f89ffb0e | 904 | static int |
c1c19657 | 905 | getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) |
f89ffb0e | 906 | { |
c1c19657 BP |
907 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
908 | socklen_t len; | |
909 | int value; | |
910 | int error; | |
f89ffb0e | 911 | |
c1c19657 BP |
912 | len = sizeof value; |
913 | if (getsockopt(fd, level, option, &value, &len)) { | |
914 | error = errno; | |
915 | VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error)); | |
916 | } else if (len != sizeof value) { | |
917 | error = EINVAL; | |
918 | VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)", | |
919 | optname, (unsigned int) len, sizeof value); | |
920 | } else { | |
921 | error = 0; | |
922 | } | |
923 | ||
924 | *valuep = error ? 0 : value; | |
925 | return error; | |
f89ffb0e BP |
926 | } |
927 | ||
928 | static void | |
929 | describe_sockaddr(struct ds *string, int fd, | |
930 | int (*getaddr)(int, struct sockaddr *, socklen_t *)) | |
931 | { | |
932 | struct sockaddr_storage ss; | |
933 | socklen_t len = sizeof ss; | |
934 | ||
935 | if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { | |
936 | if (ss.ss_family == AF_INET) { | |
937 | struct sockaddr_in sin; | |
938 | ||
939 | memcpy(&sin, &ss, sizeof sin); | |
940 | ds_put_format(string, IP_FMT":%"PRIu16, | |
941 | IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
942 | } else if (ss.ss_family == AF_UNIX) { | |
943 | struct sockaddr_un sun; | |
944 | const char *null; | |
945 | size_t maxlen; | |
946 | ||
947 | memcpy(&sun, &ss, sizeof sun); | |
948 | maxlen = len - offsetof(struct sockaddr_un, sun_path); | |
949 | null = memchr(sun.sun_path, '\0', maxlen); | |
950 | ds_put_buffer(string, sun.sun_path, | |
951 | null ? null - sun.sun_path : maxlen); | |
952 | } | |
953 | #ifdef HAVE_NETLINK | |
954 | else if (ss.ss_family == AF_NETLINK) { | |
955 | int protocol; | |
956 | ||
957 | /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version | |
958 | * of the Linux kernel headers in use at build time. */ | |
959 | #ifndef SO_PROTOCOL | |
960 | #define SO_PROTOCOL 38 | |
961 | #endif | |
962 | ||
c1c19657 BP |
963 | if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", |
964 | &protocol)) { | |
f89ffb0e BP |
965 | switch (protocol) { |
966 | case NETLINK_ROUTE: | |
967 | ds_put_cstr(string, "NETLINK_ROUTE"); | |
968 | break; | |
969 | ||
970 | case NETLINK_GENERIC: | |
971 | ds_put_cstr(string, "NETLINK_GENERIC"); | |
972 | break; | |
973 | ||
974 | default: | |
975 | ds_put_format(string, "AF_NETLINK family %d", protocol); | |
976 | break; | |
977 | } | |
978 | } else { | |
979 | ds_put_cstr(string, "AF_NETLINK"); | |
980 | } | |
981 | } | |
982 | #endif | |
983 | #if AF_PACKET && __linux__ | |
984 | else if (ss.ss_family == AF_PACKET) { | |
985 | struct sockaddr_ll sll; | |
986 | ||
987 | memcpy(&sll, &ss, sizeof sll); | |
988 | ds_put_cstr(string, "AF_PACKET"); | |
989 | if (sll.sll_ifindex) { | |
990 | char name[IFNAMSIZ]; | |
991 | ||
992 | if (if_indextoname(sll.sll_ifindex, name)) { | |
993 | ds_put_format(string, "(%s)", name); | |
994 | } else { | |
995 | ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); | |
996 | } | |
997 | } | |
998 | if (sll.sll_protocol) { | |
999 | ds_put_format(string, "(protocol=0x%"PRIu16")", | |
1000 | ntohs(sll.sll_protocol)); | |
1001 | } | |
1002 | } | |
1003 | #endif | |
1004 | else if (ss.ss_family == AF_UNSPEC) { | |
1005 | ds_put_cstr(string, "AF_UNSPEC"); | |
1006 | } else { | |
1007 | ds_put_format(string, "AF_%d", (int) ss.ss_family); | |
1008 | } | |
1009 | } | |
1010 | } | |
1011 | ||
1012 | ||
1013 | #ifdef __linux__ | |
1014 | static void | |
1015 | put_fd_filename(struct ds *string, int fd) | |
1016 | { | |
1017 | char buf[1024]; | |
1018 | char *linkname; | |
1019 | int n; | |
1020 | ||
1021 | linkname = xasprintf("/proc/self/fd/%d", fd); | |
1022 | n = readlink(linkname, buf, sizeof buf); | |
1023 | if (n > 0) { | |
1024 | ds_put_char(string, ' '); | |
1025 | ds_put_buffer(string, buf, n); | |
1026 | if (n > sizeof buf) { | |
1027 | ds_put_cstr(string, "..."); | |
1028 | } | |
1029 | } | |
1030 | free(linkname); | |
1031 | } | |
1032 | #endif | |
1033 | ||
1034 | /* Returns a malloc()'d string describing 'fd', for use in logging. */ | |
1035 | char * | |
1036 | describe_fd(int fd) | |
1037 | { | |
1038 | struct ds string; | |
1039 | struct stat s; | |
1040 | ||
1041 | ds_init(&string); | |
1042 | if (fstat(fd, &s)) { | |
1043 | ds_put_format(&string, "fstat failed (%s)", strerror(errno)); | |
1044 | } else if (S_ISSOCK(s.st_mode)) { | |
1045 | describe_sockaddr(&string, fd, getsockname); | |
1046 | ds_put_cstr(&string, "<->"); | |
1047 | describe_sockaddr(&string, fd, getpeername); | |
1048 | } else { | |
1049 | ds_put_cstr(&string, (isatty(fd) ? "tty" | |
1050 | : S_ISDIR(s.st_mode) ? "directory" | |
1051 | : S_ISCHR(s.st_mode) ? "character device" | |
1052 | : S_ISBLK(s.st_mode) ? "block device" | |
1053 | : S_ISREG(s.st_mode) ? "file" | |
1054 | : S_ISFIFO(s.st_mode) ? "FIFO" | |
1055 | : S_ISLNK(s.st_mode) ? "symbolic link" | |
1056 | : "unknown")); | |
1057 | #ifdef __linux__ | |
1058 | put_fd_filename(&string, fd); | |
1059 | #endif | |
1060 | } | |
1061 | return ds_steal_cstr(&string); | |
1062 | } | |
fd94a42c BP |
1063 | |
1064 | /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'. | |
1065 | * The caller must ensure that the total does not exceed SIZE_MAX. */ | |
1066 | size_t | |
1067 | iovec_len(const struct iovec iovs[], size_t n_iovs) | |
1068 | { | |
1069 | size_t len = 0; | |
1070 | size_t i; | |
1071 | ||
1072 | for (i = 0; i < n_iovs; i++) { | |
1073 | len += iovs[i].iov_len; | |
1074 | } | |
1075 | return len; | |
1076 | } | |
1077 | ||
1078 | /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */ | |
1079 | bool | |
1080 | iovec_is_empty(const struct iovec iovs[], size_t n_iovs) | |
1081 | { | |
1082 | size_t i; | |
1083 | ||
1084 | for (i = 0; i < n_iovs; i++) { | |
1085 | if (iovs[i].iov_len) { | |
1086 | return false; | |
1087 | } | |
1088 | } | |
1089 | return true; | |
1090 | } | |
1091 | ||
1092 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1093 | * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes | |
1094 | * successfully sent or -1 if an error occurred. On error, sets errno | |
1095 | * appropriately. */ | |
1096 | int | |
1097 | send_iovec_and_fds(int sock, | |
1098 | const struct iovec *iovs, size_t n_iovs, | |
1099 | const int fds[], size_t n_fds) | |
1100 | { | |
1101 | assert(sock >= 0); | |
1102 | if (n_fds > 0) { | |
1103 | union { | |
1104 | struct cmsghdr cm; | |
1105 | char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; | |
1106 | } cmsg; | |
1107 | struct msghdr msg; | |
1108 | ||
1109 | assert(!iovec_is_empty(iovs, n_iovs)); | |
1110 | assert(n_fds <= SOUTIL_MAX_FDS); | |
1111 | ||
1112 | memset(&cmsg, 0, sizeof cmsg); | |
1113 | cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds); | |
1114 | cmsg.cm.cmsg_level = SOL_SOCKET; | |
1115 | cmsg.cm.cmsg_type = SCM_RIGHTS; | |
1116 | memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds); | |
1117 | ||
1118 | msg.msg_name = NULL; | |
1119 | msg.msg_namelen = 0; | |
1120 | msg.msg_iov = (struct iovec *) iovs; | |
1121 | msg.msg_iovlen = n_iovs; | |
1122 | msg.msg_control = &cmsg.cm; | |
1123 | msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds); | |
1124 | msg.msg_flags = 0; | |
1125 | ||
1126 | return sendmsg(sock, &msg, 0); | |
1127 | } else { | |
1128 | return writev(sock, iovs, n_iovs); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1133 | * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the | |
1134 | * first 'skip_bytes' of data in the iovecs are not sent, and none of the file | |
1135 | * descriptors are sent. The function continues to retry sending until an | |
1136 | * error (other than EINTR) occurs or all the data and fds are sent. | |
1137 | * | |
1138 | * Returns 0 if all the data and fds were successfully sent, otherwise a | |
1139 | * positive errno value. Regardless of success, stores the number of bytes | |
1140 | * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte | |
1141 | * is sent, then all the fds have been sent.) | |
1142 | * | |
1143 | * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */ | |
1144 | int | |
1145 | send_iovec_and_fds_fully(int sock, | |
1146 | const struct iovec iovs[], size_t n_iovs, | |
1147 | const int fds[], size_t n_fds, | |
1148 | size_t skip_bytes, size_t *bytes_sent) | |
1149 | { | |
1150 | *bytes_sent = 0; | |
1151 | while (n_iovs > 0) { | |
1152 | int retval; | |
1153 | ||
1154 | if (skip_bytes) { | |
1155 | retval = skip_bytes; | |
1156 | skip_bytes = 0; | |
1157 | } else if (!*bytes_sent) { | |
1158 | retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds); | |
1159 | } else { | |
1160 | retval = writev(sock, iovs, n_iovs); | |
1161 | } | |
1162 | ||
1163 | if (retval > 0) { | |
1164 | *bytes_sent += retval; | |
1165 | while (retval > 0) { | |
1166 | const uint8_t *base = iovs->iov_base; | |
1167 | size_t len = iovs->iov_len; | |
1168 | ||
1169 | if (retval < len) { | |
1170 | size_t sent; | |
1171 | int error; | |
1172 | ||
1173 | error = write_fully(sock, base + retval, len - retval, | |
1174 | &sent); | |
1175 | *bytes_sent += sent; | |
1176 | retval += sent; | |
1177 | if (error) { | |
1178 | return error; | |
1179 | } | |
1180 | } | |
1181 | retval -= len; | |
1182 | iovs++; | |
1183 | n_iovs--; | |
1184 | } | |
1185 | } else if (retval == 0) { | |
1186 | if (iovec_is_empty(iovs, n_iovs)) { | |
1187 | break; | |
1188 | } | |
1189 | VLOG_WARN("send returned 0"); | |
1190 | return EPROTO; | |
1191 | } else if (errno != EINTR) { | |
1192 | return errno; | |
1193 | } | |
1194 | } | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1200 | * in 'fds' on Unix domain socket 'sock'. The function continues to retry | |
1201 | * sending until an error (other than EAGAIN or EINTR) occurs or all the data | |
1202 | * and fds are sent. Upon EAGAIN, the function blocks until the socket is | |
1203 | * ready for more data. | |
1204 | * | |
1205 | * Returns 0 if all the data and fds were successfully sent, otherwise a | |
1206 | * positive errno value. */ | |
1207 | int | |
1208 | send_iovec_and_fds_fully_block(int sock, | |
1209 | const struct iovec iovs[], size_t n_iovs, | |
1210 | const int fds[], size_t n_fds) | |
1211 | { | |
1212 | size_t sent = 0; | |
1213 | ||
1214 | for (;;) { | |
1215 | int error; | |
1216 | ||
1217 | error = send_iovec_and_fds_fully(sock, iovs, n_iovs, | |
1218 | fds, n_fds, sent, &sent); | |
1219 | if (error != EAGAIN) { | |
1220 | return error; | |
1221 | } | |
1222 | poll_fd_wait(sock, POLLOUT); | |
1223 | poll_block(); | |
1224 | } | |
1225 | } | |
1226 | ||
1227 | /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of | |
1228 | * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'. | |
1229 | * | |
1230 | * - Upon success, returns the number of bytes of data copied into 'data' | |
1231 | * and stores the number of received file descriptors into '*n_fdsp'. | |
1232 | * | |
1233 | * - On failure, returns a negative errno value and stores 0 in | |
1234 | * '*n_fdsp'. | |
1235 | * | |
1236 | * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */ | |
1237 | int | |
1238 | recv_data_and_fds(int sock, | |
1239 | void *data, size_t size, | |
1240 | int fds[SOUTIL_MAX_FDS], size_t *n_fdsp) | |
1241 | { | |
1242 | union { | |
1243 | struct cmsghdr cm; | |
1244 | char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; | |
1245 | } cmsg; | |
1246 | struct msghdr msg; | |
1247 | int retval; | |
1248 | struct cmsghdr *p; | |
1249 | size_t i; | |
1250 | ||
1251 | *n_fdsp = 0; | |
1252 | ||
1253 | do { | |
1254 | struct iovec iov; | |
1255 | ||
1256 | iov.iov_base = data; | |
1257 | iov.iov_len = size; | |
1258 | ||
1259 | msg.msg_name = NULL; | |
1260 | msg.msg_namelen = 0; | |
1261 | msg.msg_iov = &iov; | |
1262 | msg.msg_iovlen = 1; | |
1263 | msg.msg_control = &cmsg.cm; | |
1264 | msg.msg_controllen = sizeof cmsg.control; | |
1265 | msg.msg_flags = 0; | |
1266 | ||
1267 | retval = recvmsg(sock, &msg, 0); | |
1268 | } while (retval < 0 && errno == EINTR); | |
1269 | if (retval <= 0) { | |
1270 | return retval < 0 ? -errno : 0; | |
1271 | } | |
1272 | ||
1273 | for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) { | |
1274 | if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) { | |
1275 | VLOG_ERR("unexpected control message %d:%d", | |
1276 | p->cmsg_level, p->cmsg_type); | |
1277 | goto error; | |
1278 | } else if (*n_fdsp) { | |
1279 | VLOG_ERR("multiple SCM_RIGHTS received"); | |
1280 | goto error; | |
1281 | } else { | |
1282 | size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds; | |
1283 | const int *fds_data = (const int *) CMSG_DATA(p); | |
1284 | ||
1285 | assert(n_fds > 0); | |
1286 | if (n_fds > SOUTIL_MAX_FDS) { | |
1287 | VLOG_ERR("%zu fds received but only %d supported", | |
1288 | n_fds, SOUTIL_MAX_FDS); | |
1289 | for (i = 0; i < n_fds; i++) { | |
1290 | close(fds_data[i]); | |
1291 | } | |
1292 | goto error; | |
1293 | } | |
1294 | ||
1295 | *n_fdsp = n_fds; | |
1296 | memcpy(fds, fds_data, n_fds * sizeof *fds); | |
1297 | } | |
1298 | } | |
1299 | ||
1300 | return retval; | |
1301 | ||
1302 | error: | |
1303 | for (i = 0; i < *n_fdsp; i++) { | |
1304 | close(fds[i]); | |
1305 | } | |
1306 | *n_fdsp = 0; | |
1307 | return EPROTO; | |
1308 | } |