]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
df451457 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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> | |
20 | #include <errno.h> | |
21 | #include <fcntl.h> | |
f89ffb0e | 22 | #include <net/if.h> |
064af421 BP |
23 | #include <netdb.h> |
24 | #include <poll.h> | |
25 | #include <stddef.h> | |
26 | #include <stdio.h> | |
78ff0270 | 27 | #include <stdlib.h> |
064af421 BP |
28 | #include <string.h> |
29 | #include <sys/resource.h> | |
3762274e BP |
30 | #include <sys/socket.h> |
31 | #include <sys/stat.h> | |
a7a27d07 | 32 | #include <sys/uio.h> |
064af421 BP |
33 | #include <sys/un.h> |
34 | #include <unistd.h> | |
f89ffb0e | 35 | #include "dynamic-string.h" |
064af421 | 36 | #include "fatal-signal.h" |
f89ffb0e | 37 | #include "packets.h" |
fd94a42c | 38 | #include "poll-loop.h" |
064af421 | 39 | #include "util.h" |
064af421 | 40 | #include "vlog.h" |
361906b1 | 41 | #if AF_PACKET && LINUX_DATAPATH |
f89ffb0e BP |
42 | #include <linux/if_packet.h> |
43 | #endif | |
44 | #ifdef HAVE_NETLINK | |
45 | #include "netlink-protocol.h" | |
46 | #include "netlink-socket.h" | |
47 | #endif | |
5136ce49 | 48 | |
d98e6007 | 49 | VLOG_DEFINE_THIS_MODULE(socket_util); |
064af421 | 50 | |
e61c056c BP |
51 | /* #ifdefs make it a pain to maintain code: you have to try to build both ways. |
52 | * Thus, this file compiles all of the code regardless of the target, by | |
361906b1 EJ |
53 | * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */ |
54 | #ifndef LINUX_DATAPATH | |
55 | #define LINUX_DATAPATH 0 | |
e61c056c BP |
56 | #endif |
57 | ||
6e170b4c BP |
58 | #ifndef O_DIRECTORY |
59 | #define O_DIRECTORY 0 | |
60 | #endif | |
61 | ||
c1c19657 BP |
62 | static int getsockopt_int(int fd, int level, int option, const char *optname, |
63 | int *valuep); | |
64 | ||
064af421 BP |
65 | /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a |
66 | * positive errno value. */ | |
67 | int | |
68 | set_nonblocking(int fd) | |
69 | { | |
70 | int flags = fcntl(fd, F_GETFL, 0); | |
71 | if (flags != -1) { | |
72 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { | |
73 | return 0; | |
74 | } else { | |
75 | VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); | |
76 | return errno; | |
77 | } | |
78 | } else { | |
79 | VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); | |
80 | return errno; | |
81 | } | |
82 | } | |
83 | ||
a0505c49 BP |
84 | void |
85 | xset_nonblocking(int fd) | |
86 | { | |
87 | if (set_nonblocking(fd)) { | |
88 | exit(EXIT_FAILURE); | |
89 | } | |
90 | } | |
91 | ||
6b9c1eab | 92 | int |
cea15768 EJ |
93 | set_dscp(int fd, uint8_t dscp) |
94 | { | |
eb0cb316 EM |
95 | int val; |
96 | ||
cea15768 EJ |
97 | if (dscp > 63) { |
98 | return EINVAL; | |
99 | } | |
100 | ||
eb0cb316 EM |
101 | val = dscp << 2; |
102 | if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) { | |
cea15768 EJ |
103 | return errno; |
104 | } | |
105 | ||
106 | return 0; | |
107 | } | |
108 | ||
0ec6cfb1 BP |
109 | static bool |
110 | rlim_is_finite(rlim_t limit) | |
111 | { | |
112 | if (limit == RLIM_INFINITY) { | |
113 | return false; | |
114 | } | |
115 | ||
116 | #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */ | |
117 | if (limit == RLIM_SAVED_CUR) { | |
118 | return false; | |
119 | } | |
120 | #endif | |
121 | ||
122 | #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */ | |
123 | if (limit == RLIM_SAVED_MAX) { | |
124 | return false; | |
125 | } | |
126 | #endif | |
127 | ||
128 | return true; | |
129 | } | |
130 | ||
064af421 BP |
131 | /* Returns the maximum valid FD value, plus 1. */ |
132 | int | |
133 | get_max_fds(void) | |
134 | { | |
135 | static int max_fds = -1; | |
136 | if (max_fds < 0) { | |
137 | struct rlimit r; | |
0ec6cfb1 | 138 | if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) { |
064af421 BP |
139 | max_fds = r.rlim_cur; |
140 | } else { | |
141 | VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); | |
142 | max_fds = 1024; | |
143 | } | |
144 | } | |
145 | return max_fds; | |
146 | } | |
147 | ||
2b35e147 BP |
148 | /* Translates 'host_name', which must be a string representation of an IP |
149 | * address, into a numeric IP address in '*addr'. Returns 0 if successful, | |
150 | * otherwise a positive errno value. */ | |
064af421 | 151 | int |
d295e8e9 | 152 | lookup_ip(const char *host_name, struct in_addr *addr) |
064af421 BP |
153 | { |
154 | if (!inet_aton(host_name, addr)) { | |
db5ce514 | 155 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2b35e147 BP |
156 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name); |
157 | return ENOENT; | |
064af421 BP |
158 | } |
159 | return 0; | |
160 | } | |
161 | ||
d31f1109 JP |
162 | /* Translates 'host_name', which must be a string representation of an IPv6 |
163 | * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, | |
164 | * otherwise a positive errno value. */ | |
165 | int | |
166 | lookup_ipv6(const char *host_name, struct in6_addr *addr) | |
167 | { | |
168 | if (inet_pton(AF_INET6, host_name, addr) != 1) { | |
169 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
170 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name); | |
171 | return ENOENT; | |
172 | } | |
173 | return 0; | |
174 | } | |
175 | ||
13f2ef97 BP |
176 | /* Translates 'host_name', which must be a host name or a string representation |
177 | * of an IP address, into a numeric IP address in '*addr'. Returns 0 if | |
178 | * successful, otherwise a positive errno value. | |
179 | * | |
180 | * Most Open vSwitch code should not use this because it causes deadlocks: | |
3cbb5dc7 | 181 | * getaddrinfo() sends out a DNS request but that starts a new flow for which |
13f2ef97 | 182 | * OVS must set up a flow, but it can't because it's waiting for a DNS reply. |
293d49bd | 183 | * The synchronous lookup also delays other activity. (Of course we can solve |
13f2ef97 BP |
184 | * this but it doesn't seem worthwhile quite yet.) */ |
185 | int | |
186 | lookup_hostname(const char *host_name, struct in_addr *addr) | |
187 | { | |
3cbb5dc7 BP |
188 | struct addrinfo *result; |
189 | struct addrinfo hints; | |
13f2ef97 BP |
190 | |
191 | if (inet_aton(host_name, addr)) { | |
192 | return 0; | |
193 | } | |
194 | ||
3cbb5dc7 BP |
195 | memset(&hints, 0, sizeof hints); |
196 | hints.ai_family = AF_INET; | |
197 | ||
198 | switch (getaddrinfo(host_name, NULL, &hints, &result)) { | |
199 | case 0: | |
200 | *addr = ((struct sockaddr_in *) result->ai_addr)->sin_addr; | |
201 | freeaddrinfo(result); | |
13f2ef97 | 202 | return 0; |
13f2ef97 | 203 | |
44f645a5 | 204 | #ifdef EAI_ADDRFAMILY |
3cbb5dc7 | 205 | case EAI_ADDRFAMILY: |
44f645a5 | 206 | #endif |
3cbb5dc7 BP |
207 | case EAI_NONAME: |
208 | case EAI_SERVICE: | |
209 | return ENOENT; | |
210 | ||
211 | case EAI_AGAIN: | |
212 | return EAGAIN; | |
213 | ||
214 | case EAI_BADFLAGS: | |
215 | case EAI_FAMILY: | |
216 | case EAI_SOCKTYPE: | |
217 | return EINVAL; | |
218 | ||
219 | case EAI_FAIL: | |
220 | return EIO; | |
221 | ||
222 | case EAI_MEMORY: | |
223 | return ENOMEM; | |
224 | ||
44f645a5 | 225 | #ifdef EAI_NODATA |
3cbb5dc7 BP |
226 | case EAI_NODATA: |
227 | return ENXIO; | |
44f645a5 | 228 | #endif |
3cbb5dc7 BP |
229 | |
230 | case EAI_SYSTEM: | |
231 | return errno; | |
232 | ||
233 | default: | |
234 | return EPROTO; | |
235 | } | |
13f2ef97 BP |
236 | } |
237 | ||
064af421 | 238 | int |
d295e8e9 | 239 | check_connection_completion(int fd) |
064af421 | 240 | { |
d6cedfd9 | 241 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
064af421 BP |
242 | struct pollfd pfd; |
243 | int retval; | |
244 | ||
245 | pfd.fd = fd; | |
246 | pfd.events = POLLOUT; | |
247 | do { | |
248 | retval = poll(&pfd, 1, 0); | |
249 | } while (retval < 0 && errno == EINTR); | |
250 | if (retval == 1) { | |
d6cedfd9 BP |
251 | if (pfd.revents & POLLERR) { |
252 | ssize_t n = send(fd, "", 1, MSG_DONTWAIT); | |
253 | if (n < 0) { | |
254 | return errno; | |
255 | } else { | |
256 | VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded"); | |
257 | return EPROTO; | |
258 | } | |
259 | } | |
260 | return 0; | |
064af421 | 261 | } else if (retval < 0) { |
064af421 BP |
262 | VLOG_ERR_RL(&rl, "poll: %s", strerror(errno)); |
263 | return errno; | |
264 | } else { | |
265 | return EAGAIN; | |
266 | } | |
267 | } | |
268 | ||
269 | /* Drain all the data currently in the receive queue of a datagram socket (and | |
270 | * possibly additional data). There is no way to know how many packets are in | |
271 | * the receive queue, but we do know that the total number of bytes queued does | |
272 | * not exceed the receive buffer size, so we pull packets until none are left | |
273 | * or we've read that many bytes. */ | |
274 | int | |
275 | drain_rcvbuf(int fd) | |
276 | { | |
c1c19657 | 277 | int rcvbuf; |
064af421 | 278 | |
c1c19657 BP |
279 | rcvbuf = get_socket_rcvbuf(fd); |
280 | if (rcvbuf < 0) { | |
281 | return -rcvbuf; | |
064af421 | 282 | } |
c1c19657 | 283 | |
064af421 BP |
284 | while (rcvbuf > 0) { |
285 | /* In Linux, specifying MSG_TRUNC in the flags argument causes the | |
286 | * datagram length to be returned, even if that is longer than the | |
287 | * buffer provided. Thus, we can use a 1-byte buffer to discard the | |
288 | * incoming datagram and still be able to account how many bytes were | |
289 | * removed from the receive buffer. | |
290 | * | |
291 | * On other Unix-like OSes, MSG_TRUNC has no effect in the flags | |
292 | * argument. */ | |
361906b1 | 293 | char buffer[LINUX_DATAPATH ? 1 : 2048]; |
064af421 BP |
294 | ssize_t n_bytes = recv(fd, buffer, sizeof buffer, |
295 | MSG_TRUNC | MSG_DONTWAIT); | |
296 | if (n_bytes <= 0 || n_bytes >= rcvbuf) { | |
297 | break; | |
298 | } | |
299 | rcvbuf -= n_bytes; | |
300 | } | |
301 | return 0; | |
302 | } | |
303 | ||
c1c19657 BP |
304 | /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a |
305 | * negative errno value if an error occurs. */ | |
306 | int | |
307 | get_socket_rcvbuf(int sock) | |
308 | { | |
309 | int rcvbuf; | |
310 | int error; | |
311 | ||
312 | error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf); | |
313 | return error ? -error : rcvbuf; | |
314 | } | |
315 | ||
064af421 BP |
316 | /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no |
317 | * more data can be immediately read. ('fd' should therefore be in | |
318 | * non-blocking mode.)*/ | |
319 | void | |
320 | drain_fd(int fd, size_t n_packets) | |
321 | { | |
322 | for (; n_packets > 0; n_packets--) { | |
323 | /* 'buffer' only needs to be 1 byte long in most circumstances. This | |
324 | * size is defensive against the possibility that we someday want to | |
325 | * use a Linux tap device without TUN_NO_PI, in which case a buffer | |
326 | * smaller than sizeof(struct tun_pi) will give EINVAL on read. */ | |
327 | char buffer[128]; | |
328 | if (read(fd, buffer, sizeof buffer) <= 0) { | |
329 | break; | |
330 | } | |
331 | } | |
332 | } | |
333 | ||
334 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in | |
335 | * '*un_len' the size of the sockaddr_un. */ | |
336 | static void | |
6e170b4c | 337 | make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len) |
064af421 BP |
338 | { |
339 | un->sun_family = AF_UNIX; | |
71d7c22f | 340 | ovs_strzcpy(un->sun_path, name, sizeof un->sun_path); |
064af421 BP |
341 | *un_len = (offsetof(struct sockaddr_un, sun_path) |
342 | + strlen (un->sun_path) + 1); | |
343 | } | |
344 | ||
6e170b4c BP |
345 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in |
346 | * '*un_len' the size of the sockaddr_un. | |
347 | * | |
348 | * Returns 0 on success, otherwise a positive errno value. On success, | |
349 | * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller | |
350 | * should close after using '*un' to bind or connect. On failure, '*dirfdp' is | |
351 | * -1. */ | |
352 | static int | |
353 | make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, | |
354 | int *dirfdp) | |
355 | { | |
356 | enum { MAX_UN_LEN = sizeof un->sun_path - 1 }; | |
357 | ||
358 | *dirfdp = -1; | |
359 | if (strlen(name) > MAX_UN_LEN) { | |
360 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
361 | ||
361906b1 | 362 | if (LINUX_DATAPATH) { |
6e170b4c BP |
363 | /* 'name' is too long to fit in a sockaddr_un, but we have a |
364 | * workaround for that on Linux: shorten it by opening a file | |
365 | * descriptor for the directory part of the name and indirecting | |
366 | * through /proc/self/fd/<dirfd>/<basename>. */ | |
367 | char *dir, *base; | |
368 | char *short_name; | |
369 | int dirfd; | |
370 | ||
371 | dir = dir_name(name); | |
372 | base = base_name(name); | |
373 | ||
374 | dirfd = open(dir, O_DIRECTORY | O_RDONLY); | |
375 | if (dirfd < 0) { | |
08871bf6 JP |
376 | free(base); |
377 | free(dir); | |
6e170b4c BP |
378 | return errno; |
379 | } | |
380 | ||
381 | short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base); | |
382 | free(dir); | |
383 | free(base); | |
384 | ||
385 | if (strlen(short_name) <= MAX_UN_LEN) { | |
386 | make_sockaddr_un__(short_name, un, un_len); | |
387 | free(short_name); | |
388 | *dirfdp = dirfd; | |
389 | return 0; | |
390 | } | |
391 | free(short_name); | |
392 | close(dirfd); | |
393 | ||
394 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
395 | "%d bytes (even shortened)", name, MAX_UN_LEN); | |
396 | } else { | |
397 | /* 'name' is too long and we have no workaround. */ | |
398 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
399 | "%d bytes", name, MAX_UN_LEN); | |
400 | } | |
401 | ||
402 | return ENAMETOOLONG; | |
403 | } else { | |
404 | make_sockaddr_un__(name, un, un_len); | |
405 | return 0; | |
406 | } | |
407 | } | |
408 | ||
e2bd0593 BP |
409 | /* Binds Unix domain socket 'fd' to a file with permissions 0700. */ |
410 | static int | |
411 | bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len) | |
412 | { | |
e2bd0593 BP |
413 | /* According to _Unix Network Programming_, umask should affect bind(). */ |
414 | mode_t old_umask = umask(0077); | |
415 | int error = bind(fd, sun, sun_len) ? errno : 0; | |
416 | umask(old_umask); | |
417 | return error; | |
e2bd0593 BP |
418 | } |
419 | ||
064af421 BP |
420 | /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or |
421 | * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and | |
422 | * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock' | |
5ca92d1d | 423 | * is true, the socket is made non-blocking. |
064af421 BP |
424 | * |
425 | * Returns the socket's fd if successful, otherwise a negative errno value. */ | |
426 | int | |
5ca92d1d | 427 | make_unix_socket(int style, bool nonblock, |
064af421 BP |
428 | const char *bind_path, const char *connect_path) |
429 | { | |
430 | int error; | |
431 | int fd; | |
432 | ||
433 | fd = socket(PF_UNIX, style, 0); | |
434 | if (fd < 0) { | |
435 | return -errno; | |
436 | } | |
437 | ||
438 | /* Set nonblocking mode right away, if we want it. This prevents blocking | |
439 | * in connect(), if connect_path != NULL. (In turn, that's a corner case: | |
440 | * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only | |
441 | * if a backlog of un-accepted connections has built up in the kernel.) */ | |
442 | if (nonblock) { | |
b36db114 BP |
443 | error = set_nonblocking(fd); |
444 | if (error) { | |
064af421 BP |
445 | goto error; |
446 | } | |
447 | } | |
448 | ||
449 | if (bind_path) { | |
450 | struct sockaddr_un un; | |
451 | socklen_t un_len; | |
6e170b4c BP |
452 | int dirfd; |
453 | ||
454 | if (unlink(bind_path) && errno != ENOENT) { | |
455 | VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); | |
064af421 BP |
456 | } |
457 | fatal_signal_add_file_to_unlink(bind_path); | |
6e170b4c BP |
458 | |
459 | error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd); | |
e2bd0593 BP |
460 | if (!error) { |
461 | error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len); | |
6e170b4c BP |
462 | } |
463 | if (dirfd >= 0) { | |
464 | close(dirfd); | |
465 | } | |
466 | if (error) { | |
064af421 BP |
467 | goto error; |
468 | } | |
469 | } | |
470 | ||
471 | if (connect_path) { | |
472 | struct sockaddr_un un; | |
473 | socklen_t un_len; | |
6e170b4c BP |
474 | int dirfd; |
475 | ||
476 | error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd); | |
477 | if (!error | |
478 | && connect(fd, (struct sockaddr*) &un, un_len) | |
064af421 | 479 | && errno != EINPROGRESS) { |
6e170b4c BP |
480 | error = errno; |
481 | } | |
482 | if (dirfd >= 0) { | |
483 | close(dirfd); | |
484 | } | |
485 | if (error) { | |
064af421 BP |
486 | goto error; |
487 | } | |
488 | } | |
489 | ||
064af421 BP |
490 | return fd; |
491 | ||
492 | error: | |
6e170b4c BP |
493 | if (error == EAGAIN) { |
494 | error = EPROTO; | |
495 | } | |
064af421 | 496 | if (bind_path) { |
e36d0b53 | 497 | fatal_signal_unlink_file_now(bind_path); |
064af421 | 498 | } |
064af421 BP |
499 | close(fd); |
500 | return -error; | |
501 | } | |
502 | ||
503 | int | |
504 | get_unix_name_len(socklen_t sun_len) | |
505 | { | |
506 | return (sun_len >= offsetof(struct sockaddr_un, sun_path) | |
507 | ? sun_len - offsetof(struct sockaddr_un, sun_path) | |
508 | : 0); | |
509 | } | |
510 | ||
dbba996b BP |
511 | ovs_be32 |
512 | guess_netmask(ovs_be32 ip_) | |
064af421 | 513 | { |
dbba996b | 514 | uint32_t ip = ntohl(ip_); |
064af421 BP |
515 | return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ |
516 | : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ | |
517 | : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ | |
518 | : htonl(0)); /* ??? */ | |
519 | } | |
520 | ||
52f8a75e BP |
521 | /* Parses 'target', which should be a string in the format "<host>[:<port>]". |
522 | * <host> is required. If 'default_port' is nonzero then <port> is optional | |
523 | * and defaults to 'default_port'. | |
4f2eb9a7 | 524 | * |
52f8a75e BP |
525 | * On success, returns true and stores the parsed remote address into '*sinp'. |
526 | * On failure, logs an error, stores zeros into '*sinp', and returns false. */ | |
527 | bool | |
528 | inet_parse_active(const char *target_, uint16_t default_port, | |
529 | struct sockaddr_in *sinp) | |
78ff0270 BP |
530 | { |
531 | char *target = xstrdup(target_); | |
532 | char *save_ptr = NULL; | |
533 | const char *host_name; | |
534 | const char *port_string; | |
52f8a75e | 535 | bool ok = false; |
78ff0270 BP |
536 | |
537 | /* Defaults. */ | |
52f8a75e BP |
538 | sinp->sin_family = AF_INET; |
539 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
540 | |
541 | /* Tokenize. */ | |
542 | host_name = strtok_r(target, ":", &save_ptr); | |
543 | port_string = strtok_r(NULL, ":", &save_ptr); | |
544 | if (!host_name) { | |
52f8a75e | 545 | VLOG_ERR("%s: bad peer name format", target_); |
78ff0270 BP |
546 | goto exit; |
547 | } | |
548 | ||
549 | /* Look up IP, port. */ | |
52f8a75e | 550 | if (lookup_ip(host_name, &sinp->sin_addr)) { |
78ff0270 BP |
551 | goto exit; |
552 | } | |
553 | if (port_string && atoi(port_string)) { | |
52f8a75e | 554 | sinp->sin_port = htons(atoi(port_string)); |
1901968e BP |
555 | } else if (!default_port) { |
556 | VLOG_ERR("%s: port number must be specified", target_); | |
52f8a75e BP |
557 | goto exit; |
558 | } | |
559 | ||
560 | ok = true; | |
561 | ||
562 | exit: | |
563 | if (!ok) { | |
564 | memset(sinp, 0, sizeof *sinp); | |
565 | } | |
566 | free(target); | |
567 | return ok; | |
568 | } | |
569 | ||
570 | /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to | |
571 | * 'target', which should be a string in the format "<host>[:<port>]". <host> | |
572 | * is required. If 'default_port' is nonzero then <port> is optional and | |
573 | * defaults to 'default_port'. | |
574 | * | |
575 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
576 | * | |
577 | * On success, returns 0 (indicating connection complete) or EAGAIN (indicating | |
578 | * connection in progress), in which case the new file descriptor is stored | |
579 | * into '*fdp'. On failure, returns a positive errno value other than EAGAIN | |
580 | * and stores -1 into '*fdp'. | |
581 | * | |
582 | * If 'sinp' is non-null, then on success the target address is stored into | |
f125905c MM |
583 | * '*sinp'. |
584 | * | |
cea15768 EJ |
585 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
586 | * should be in the range [0, 63] and will automatically be shifted to the | |
587 | * appropriately place in the IP tos field. */ | |
52f8a75e BP |
588 | int |
589 | inet_open_active(int style, const char *target, uint16_t default_port, | |
f125905c | 590 | struct sockaddr_in *sinp, int *fdp, uint8_t dscp) |
52f8a75e BP |
591 | { |
592 | struct sockaddr_in sin; | |
593 | int fd = -1; | |
594 | int error; | |
595 | ||
596 | /* Parse. */ | |
597 | if (!inet_parse_active(target, default_port, &sin)) { | |
1901968e BP |
598 | error = EAFNOSUPPORT; |
599 | goto exit; | |
78ff0270 BP |
600 | } |
601 | ||
602 | /* Create non-blocking socket. */ | |
4f2eb9a7 | 603 | fd = socket(AF_INET, style, 0); |
78ff0270 | 604 | if (fd < 0) { |
52f8a75e | 605 | VLOG_ERR("%s: socket: %s", target, strerror(errno)); |
78ff0270 BP |
606 | error = errno; |
607 | goto exit; | |
608 | } | |
609 | error = set_nonblocking(fd); | |
610 | if (error) { | |
a4efa3fc | 611 | goto exit; |
78ff0270 BP |
612 | } |
613 | ||
cea15768 EJ |
614 | /* The dscp bits must be configured before connect() to ensure that the TOS |
615 | * field is set during the connection establishment. If set after | |
616 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
617 | error = set_dscp(fd, dscp); | |
618 | if (error) { | |
619 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 620 | goto exit; |
f125905c MM |
621 | } |
622 | ||
78ff0270 BP |
623 | /* Connect. */ |
624 | error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno; | |
625 | if (error == EINPROGRESS) { | |
626 | error = EAGAIN; | |
78ff0270 BP |
627 | } |
628 | ||
78ff0270 BP |
629 | exit: |
630 | if (!error || error == EAGAIN) { | |
631 | if (sinp) { | |
632 | *sinp = sin; | |
633 | } | |
a4efa3fc EJ |
634 | } else if (fd >= 0) { |
635 | close(fd); | |
ce78d135 | 636 | fd = -1; |
78ff0270 | 637 | } |
a4efa3fc | 638 | *fdp = fd; |
78ff0270 BP |
639 | return error; |
640 | } | |
641 | ||
d98fa503 | 642 | /* Parses 'target', which should be a string in the format "[<port>][:<ip>]": |
e1bd3bee BP |
643 | * |
644 | * - If 'default_port' is -1, then <port> is required. Otherwise, if | |
645 | * <port> is omitted, then 'default_port' is used instead. | |
646 | * | |
647 | * - If <port> (or 'default_port', if used) is 0, then no port is bound | |
648 | * and the TCP/IP stack will select a port. | |
649 | * | |
650 | * - If <ip> is omitted then the IP address is wildcarded. | |
4f2eb9a7 | 651 | * |
d98fa503 BP |
652 | * If successful, stores the address into '*sinp' and returns true; otherwise |
653 | * zeros '*sinp' and returns false. */ | |
654 | bool | |
995337c4 | 655 | inet_parse_passive(const char *target_, int default_port, |
d98fa503 | 656 | struct sockaddr_in *sinp) |
78ff0270 BP |
657 | { |
658 | char *target = xstrdup(target_); | |
659 | char *string_ptr = target; | |
78ff0270 BP |
660 | const char *host_name; |
661 | const char *port_string; | |
d98fa503 BP |
662 | bool ok = false; |
663 | int port; | |
78ff0270 BP |
664 | |
665 | /* Address defaults. */ | |
d98fa503 BP |
666 | memset(sinp, 0, sizeof *sinp); |
667 | sinp->sin_family = AF_INET; | |
668 | sinp->sin_addr.s_addr = htonl(INADDR_ANY); | |
669 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
670 | |
671 | /* Parse optional port number. */ | |
672 | port_string = strsep(&string_ptr, ":"); | |
e1bd3bee | 673 | if (port_string && str_to_int(port_string, 10, &port)) { |
d98fa503 | 674 | sinp->sin_port = htons(port); |
e1bd3bee | 675 | } else if (default_port < 0) { |
1901968e | 676 | VLOG_ERR("%s: port number must be specified", target_); |
1901968e | 677 | goto exit; |
78ff0270 BP |
678 | } |
679 | ||
680 | /* Parse optional bind IP. */ | |
681 | host_name = strsep(&string_ptr, ":"); | |
d98fa503 BP |
682 | if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { |
683 | goto exit; | |
684 | } | |
685 | ||
686 | ok = true; | |
687 | ||
688 | exit: | |
689 | if (!ok) { | |
690 | memset(sinp, 0, sizeof *sinp); | |
691 | } | |
692 | free(target); | |
693 | return ok; | |
694 | } | |
695 | ||
696 | ||
697 | /* Opens a non-blocking IPv4 socket of the specified 'style', binds to | |
698 | * 'target', and listens for incoming connections. Parses 'target' in the same | |
699 | * way was inet_parse_passive(). | |
700 | * | |
701 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
702 | * | |
703 | * For TCP, the socket will have SO_REUSEADDR turned on. | |
704 | * | |
705 | * On success, returns a non-negative file descriptor. On failure, returns a | |
706 | * negative errno value. | |
707 | * | |
708 | * If 'sinp' is non-null, then on success the bound address is stored into | |
f125905c MM |
709 | * '*sinp'. |
710 | * | |
cea15768 EJ |
711 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
712 | * should be in the range [0, 63] and will automatically be shifted to the | |
713 | * appropriately place in the IP tos field. */ | |
d98fa503 BP |
714 | int |
715 | inet_open_passive(int style, const char *target, int default_port, | |
f125905c | 716 | struct sockaddr_in *sinp, uint8_t dscp) |
d98fa503 | 717 | { |
df451457 | 718 | bool kernel_chooses_port; |
d98fa503 BP |
719 | struct sockaddr_in sin; |
720 | int fd = 0, error; | |
721 | unsigned int yes = 1; | |
722 | ||
723 | if (!inet_parse_passive(target, default_port, &sin)) { | |
ca286ba9 | 724 | return -EAFNOSUPPORT; |
78ff0270 BP |
725 | } |
726 | ||
727 | /* Create non-blocking socket, set SO_REUSEADDR. */ | |
4f2eb9a7 | 728 | fd = socket(AF_INET, style, 0); |
78ff0270 BP |
729 | if (fd < 0) { |
730 | error = errno; | |
d98fa503 | 731 | VLOG_ERR("%s: socket: %s", target, strerror(error)); |
ca286ba9 | 732 | return -error; |
78ff0270 BP |
733 | } |
734 | error = set_nonblocking(fd); | |
735 | if (error) { | |
d98fa503 | 736 | goto error; |
78ff0270 | 737 | } |
4f2eb9a7 BP |
738 | if (style == SOCK_STREAM |
739 | && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { | |
78ff0270 | 740 | error = errno; |
d98fa503 BP |
741 | VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error)); |
742 | goto error; | |
78ff0270 BP |
743 | } |
744 | ||
745 | /* Bind. */ | |
746 | if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) { | |
747 | error = errno; | |
d98fa503 BP |
748 | VLOG_ERR("%s: bind: %s", target, strerror(error)); |
749 | goto error; | |
78ff0270 BP |
750 | } |
751 | ||
cea15768 EJ |
752 | /* The dscp bits must be configured before connect() to ensure that the TOS |
753 | * field is set during the connection establishment. If set after | |
754 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
755 | error = set_dscp(fd, dscp); | |
756 | if (error) { | |
757 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 758 | goto error; |
f125905c MM |
759 | } |
760 | ||
78ff0270 | 761 | /* Listen. */ |
041dc07f | 762 | if (style == SOCK_STREAM && listen(fd, 10) < 0) { |
78ff0270 | 763 | error = errno; |
d98fa503 BP |
764 | VLOG_ERR("%s: listen: %s", target, strerror(error)); |
765 | goto error; | |
78ff0270 | 766 | } |
36775dad | 767 | |
df451457 BP |
768 | kernel_chooses_port = sin.sin_port == htons(0); |
769 | if (sinp || kernel_chooses_port) { | |
36775dad | 770 | socklen_t sin_len = sizeof sin; |
df451457 | 771 | if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0) { |
36775dad | 772 | error = errno; |
d98fa503 BP |
773 | VLOG_ERR("%s: getsockname: %s", target, strerror(error)); |
774 | goto error; | |
36775dad BP |
775 | } |
776 | if (sin.sin_family != AF_INET || sin_len != sizeof sin) { | |
ca286ba9 | 777 | error = EAFNOSUPPORT; |
d98fa503 BP |
778 | VLOG_ERR("%s: getsockname: invalid socket name", target); |
779 | goto error; | |
36775dad | 780 | } |
df451457 BP |
781 | if (sinp) { |
782 | *sinp = sin; | |
783 | } | |
784 | if (kernel_chooses_port) { | |
785 | VLOG_INFO("%s: listening on port %"PRIu16, | |
786 | target, ntohs(sin.sin_port)); | |
787 | } | |
36775dad BP |
788 | } |
789 | ||
d98fa503 | 790 | return fd; |
78ff0270 | 791 | |
d98fa503 | 792 | error: |
78ff0270 | 793 | close(fd); |
ca286ba9 | 794 | return -error; |
78ff0270 BP |
795 | } |
796 | ||
a8b5f8b4 BP |
797 | /* Returns a readable and writable fd for /dev/null, if successful, otherwise |
798 | * a negative errno value. The caller must not close the returned fd (because | |
799 | * the same fd will be handed out to subsequent callers). */ | |
800 | int | |
801 | get_null_fd(void) | |
802 | { | |
803 | static int null_fd = -1; | |
804 | if (null_fd < 0) { | |
805 | null_fd = open("/dev/null", O_RDWR); | |
806 | if (null_fd < 0) { | |
807 | int error = errno; | |
808 | VLOG_ERR("could not open /dev/null: %s", strerror(error)); | |
809 | return -error; | |
810 | } | |
811 | } | |
812 | return null_fd; | |
813 | } | |
814 | ||
064af421 BP |
815 | int |
816 | read_fully(int fd, void *p_, size_t size, size_t *bytes_read) | |
817 | { | |
818 | uint8_t *p = p_; | |
819 | ||
820 | *bytes_read = 0; | |
821 | while (size > 0) { | |
822 | ssize_t retval = read(fd, p, size); | |
823 | if (retval > 0) { | |
824 | *bytes_read += retval; | |
825 | size -= retval; | |
826 | p += retval; | |
827 | } else if (retval == 0) { | |
828 | return EOF; | |
829 | } else if (errno != EINTR) { | |
830 | return errno; | |
831 | } | |
832 | } | |
833 | return 0; | |
834 | } | |
835 | ||
836 | int | |
837 | write_fully(int fd, const void *p_, size_t size, size_t *bytes_written) | |
838 | { | |
839 | const uint8_t *p = p_; | |
840 | ||
841 | *bytes_written = 0; | |
842 | while (size > 0) { | |
843 | ssize_t retval = write(fd, p, size); | |
844 | if (retval > 0) { | |
845 | *bytes_written += retval; | |
846 | size -= retval; | |
847 | p += retval; | |
848 | } else if (retval == 0) { | |
849 | VLOG_WARN("write returned 0"); | |
850 | return EPROTO; | |
851 | } else if (errno != EINTR) { | |
852 | return errno; | |
853 | } | |
854 | } | |
855 | return 0; | |
856 | } | |
8e71cf88 BP |
857 | |
858 | /* Given file name 'file_name', fsyncs the directory in which it is contained. | |
859 | * Returns 0 if successful, otherwise a positive errno value. */ | |
860 | int | |
861 | fsync_parent_dir(const char *file_name) | |
862 | { | |
863 | int error = 0; | |
864 | char *dir; | |
865 | int fd; | |
866 | ||
867 | dir = dir_name(file_name); | |
868 | fd = open(dir, O_RDONLY); | |
869 | if (fd >= 0) { | |
870 | if (fsync(fd)) { | |
871 | if (errno == EINVAL || errno == EROFS) { | |
872 | /* This directory does not support synchronization. Not | |
873 | * really an error. */ | |
874 | } else { | |
875 | error = errno; | |
876 | VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); | |
877 | } | |
878 | } | |
879 | close(fd); | |
880 | } else { | |
881 | error = errno; | |
882 | VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); | |
883 | } | |
884 | free(dir); | |
885 | ||
886 | return error; | |
887 | } | |
26efd256 BP |
888 | |
889 | /* Obtains the modification time of the file named 'file_name' to the greatest | |
890 | * supported precision. If successful, stores the mtime in '*mtime' and | |
891 | * returns 0. On error, returns a positive errno value and stores zeros in | |
892 | * '*mtime'. */ | |
893 | int | |
894 | get_mtime(const char *file_name, struct timespec *mtime) | |
895 | { | |
896 | struct stat s; | |
897 | ||
898 | if (!stat(file_name, &s)) { | |
899 | mtime->tv_sec = s.st_mtime; | |
900 | ||
901 | #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC | |
902 | mtime->tv_nsec = s.st_mtim.tv_nsec; | |
903 | #elif HAVE_STRUCT_STAT_ST_MTIMENSEC | |
904 | mtime->tv_nsec = s.st_mtimensec; | |
905 | #else | |
906 | mtime->tv_nsec = 0; | |
907 | #endif | |
908 | ||
909 | return 0; | |
910 | } else { | |
911 | mtime->tv_sec = mtime->tv_nsec = 0; | |
912 | return errno; | |
913 | } | |
914 | } | |
915 | ||
279c9e03 BP |
916 | void |
917 | xpipe(int fds[2]) | |
918 | { | |
919 | if (pipe(fds)) { | |
920 | VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); | |
921 | } | |
922 | } | |
f89ffb0e | 923 | |
c0d95206 EM |
924 | void |
925 | xpipe_nonblocking(int fds[2]) | |
926 | { | |
927 | xpipe(fds); | |
928 | xset_nonblocking(fds[0]); | |
929 | xset_nonblocking(fds[1]); | |
930 | } | |
931 | ||
fd94a42c BP |
932 | void |
933 | xsocketpair(int domain, int type, int protocol, int fds[2]) | |
934 | { | |
935 | if (socketpair(domain, type, protocol, fds)) { | |
936 | VLOG_FATAL("failed to create socketpair (%s)", strerror(errno)); | |
937 | } | |
938 | } | |
939 | ||
f89ffb0e | 940 | static int |
c1c19657 | 941 | getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) |
f89ffb0e | 942 | { |
c1c19657 BP |
943 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
944 | socklen_t len; | |
945 | int value; | |
946 | int error; | |
f89ffb0e | 947 | |
c1c19657 BP |
948 | len = sizeof value; |
949 | if (getsockopt(fd, level, option, &value, &len)) { | |
950 | error = errno; | |
951 | VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error)); | |
952 | } else if (len != sizeof value) { | |
953 | error = EINVAL; | |
954 | VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)", | |
955 | optname, (unsigned int) len, sizeof value); | |
956 | } else { | |
957 | error = 0; | |
958 | } | |
959 | ||
960 | *valuep = error ? 0 : value; | |
961 | return error; | |
f89ffb0e BP |
962 | } |
963 | ||
964 | static void | |
965 | describe_sockaddr(struct ds *string, int fd, | |
966 | int (*getaddr)(int, struct sockaddr *, socklen_t *)) | |
967 | { | |
968 | struct sockaddr_storage ss; | |
969 | socklen_t len = sizeof ss; | |
970 | ||
971 | if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { | |
972 | if (ss.ss_family == AF_INET) { | |
973 | struct sockaddr_in sin; | |
974 | ||
975 | memcpy(&sin, &ss, sizeof sin); | |
976 | ds_put_format(string, IP_FMT":%"PRIu16, | |
ed36537e | 977 | IP_ARGS(sin.sin_addr.s_addr), ntohs(sin.sin_port)); |
f89ffb0e BP |
978 | } else if (ss.ss_family == AF_UNIX) { |
979 | struct sockaddr_un sun; | |
980 | const char *null; | |
981 | size_t maxlen; | |
982 | ||
983 | memcpy(&sun, &ss, sizeof sun); | |
984 | maxlen = len - offsetof(struct sockaddr_un, sun_path); | |
985 | null = memchr(sun.sun_path, '\0', maxlen); | |
986 | ds_put_buffer(string, sun.sun_path, | |
987 | null ? null - sun.sun_path : maxlen); | |
988 | } | |
989 | #ifdef HAVE_NETLINK | |
990 | else if (ss.ss_family == AF_NETLINK) { | |
991 | int protocol; | |
992 | ||
993 | /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version | |
994 | * of the Linux kernel headers in use at build time. */ | |
995 | #ifndef SO_PROTOCOL | |
996 | #define SO_PROTOCOL 38 | |
997 | #endif | |
998 | ||
c1c19657 BP |
999 | if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", |
1000 | &protocol)) { | |
f89ffb0e BP |
1001 | switch (protocol) { |
1002 | case NETLINK_ROUTE: | |
1003 | ds_put_cstr(string, "NETLINK_ROUTE"); | |
1004 | break; | |
1005 | ||
1006 | case NETLINK_GENERIC: | |
1007 | ds_put_cstr(string, "NETLINK_GENERIC"); | |
1008 | break; | |
1009 | ||
1010 | default: | |
1011 | ds_put_format(string, "AF_NETLINK family %d", protocol); | |
1012 | break; | |
1013 | } | |
1014 | } else { | |
1015 | ds_put_cstr(string, "AF_NETLINK"); | |
1016 | } | |
1017 | } | |
1018 | #endif | |
361906b1 | 1019 | #if AF_PACKET && LINUX_DATAPATH |
f89ffb0e BP |
1020 | else if (ss.ss_family == AF_PACKET) { |
1021 | struct sockaddr_ll sll; | |
1022 | ||
1023 | memcpy(&sll, &ss, sizeof sll); | |
1024 | ds_put_cstr(string, "AF_PACKET"); | |
1025 | if (sll.sll_ifindex) { | |
1026 | char name[IFNAMSIZ]; | |
1027 | ||
1028 | if (if_indextoname(sll.sll_ifindex, name)) { | |
1029 | ds_put_format(string, "(%s)", name); | |
1030 | } else { | |
1031 | ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); | |
1032 | } | |
1033 | } | |
1034 | if (sll.sll_protocol) { | |
1035 | ds_put_format(string, "(protocol=0x%"PRIu16")", | |
1036 | ntohs(sll.sll_protocol)); | |
1037 | } | |
1038 | } | |
1039 | #endif | |
1040 | else if (ss.ss_family == AF_UNSPEC) { | |
1041 | ds_put_cstr(string, "AF_UNSPEC"); | |
1042 | } else { | |
1043 | ds_put_format(string, "AF_%d", (int) ss.ss_family); | |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | ||
361906b1 | 1049 | #ifdef LINUX_DATAPATH |
f89ffb0e BP |
1050 | static void |
1051 | put_fd_filename(struct ds *string, int fd) | |
1052 | { | |
1053 | char buf[1024]; | |
1054 | char *linkname; | |
1055 | int n; | |
1056 | ||
1057 | linkname = xasprintf("/proc/self/fd/%d", fd); | |
1058 | n = readlink(linkname, buf, sizeof buf); | |
1059 | if (n > 0) { | |
1060 | ds_put_char(string, ' '); | |
1061 | ds_put_buffer(string, buf, n); | |
1062 | if (n > sizeof buf) { | |
1063 | ds_put_cstr(string, "..."); | |
1064 | } | |
1065 | } | |
1066 | free(linkname); | |
1067 | } | |
1068 | #endif | |
1069 | ||
1070 | /* Returns a malloc()'d string describing 'fd', for use in logging. */ | |
1071 | char * | |
1072 | describe_fd(int fd) | |
1073 | { | |
1074 | struct ds string; | |
1075 | struct stat s; | |
1076 | ||
1077 | ds_init(&string); | |
1078 | if (fstat(fd, &s)) { | |
1079 | ds_put_format(&string, "fstat failed (%s)", strerror(errno)); | |
1080 | } else if (S_ISSOCK(s.st_mode)) { | |
1081 | describe_sockaddr(&string, fd, getsockname); | |
1082 | ds_put_cstr(&string, "<->"); | |
1083 | describe_sockaddr(&string, fd, getpeername); | |
1084 | } else { | |
1085 | ds_put_cstr(&string, (isatty(fd) ? "tty" | |
1086 | : S_ISDIR(s.st_mode) ? "directory" | |
1087 | : S_ISCHR(s.st_mode) ? "character device" | |
1088 | : S_ISBLK(s.st_mode) ? "block device" | |
1089 | : S_ISREG(s.st_mode) ? "file" | |
1090 | : S_ISFIFO(s.st_mode) ? "FIFO" | |
1091 | : S_ISLNK(s.st_mode) ? "symbolic link" | |
1092 | : "unknown")); | |
361906b1 | 1093 | #ifdef LINUX_DATAPATH |
f89ffb0e BP |
1094 | put_fd_filename(&string, fd); |
1095 | #endif | |
1096 | } | |
1097 | return ds_steal_cstr(&string); | |
1098 | } | |
fd94a42c BP |
1099 | |
1100 | /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'. | |
1101 | * The caller must ensure that the total does not exceed SIZE_MAX. */ | |
1102 | size_t | |
1103 | iovec_len(const struct iovec iovs[], size_t n_iovs) | |
1104 | { | |
1105 | size_t len = 0; | |
1106 | size_t i; | |
1107 | ||
1108 | for (i = 0; i < n_iovs; i++) { | |
1109 | len += iovs[i].iov_len; | |
1110 | } | |
1111 | return len; | |
1112 | } | |
1113 | ||
1114 | /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */ | |
1115 | bool | |
1116 | iovec_is_empty(const struct iovec iovs[], size_t n_iovs) | |
1117 | { | |
1118 | size_t i; | |
1119 | ||
1120 | for (i = 0; i < n_iovs; i++) { | |
1121 | if (iovs[i].iov_len) { | |
1122 | return false; | |
1123 | } | |
1124 | } | |
1125 | return true; | |
1126 | } | |
1127 | ||
1128 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1129 | * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes | |
1130 | * successfully sent or -1 if an error occurred. On error, sets errno | |
1131 | * appropriately. */ | |
1132 | int | |
1133 | send_iovec_and_fds(int sock, | |
1134 | const struct iovec *iovs, size_t n_iovs, | |
1135 | const int fds[], size_t n_fds) | |
1136 | { | |
cb22974d | 1137 | ovs_assert(sock >= 0); |
fd94a42c BP |
1138 | if (n_fds > 0) { |
1139 | union { | |
1140 | struct cmsghdr cm; | |
1141 | char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; | |
1142 | } cmsg; | |
1143 | struct msghdr msg; | |
1144 | ||
cb22974d BP |
1145 | ovs_assert(!iovec_is_empty(iovs, n_iovs)); |
1146 | ovs_assert(n_fds <= SOUTIL_MAX_FDS); | |
fd94a42c BP |
1147 | |
1148 | memset(&cmsg, 0, sizeof cmsg); | |
1149 | cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds); | |
1150 | cmsg.cm.cmsg_level = SOL_SOCKET; | |
1151 | cmsg.cm.cmsg_type = SCM_RIGHTS; | |
1152 | memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds); | |
1153 | ||
1154 | msg.msg_name = NULL; | |
1155 | msg.msg_namelen = 0; | |
f3b1ab6f | 1156 | msg.msg_iov = CONST_CAST(struct iovec *, iovs); |
fd94a42c BP |
1157 | msg.msg_iovlen = n_iovs; |
1158 | msg.msg_control = &cmsg.cm; | |
1159 | msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds); | |
1160 | msg.msg_flags = 0; | |
1161 | ||
1162 | return sendmsg(sock, &msg, 0); | |
1163 | } else { | |
1164 | return writev(sock, iovs, n_iovs); | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1169 | * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the | |
1170 | * first 'skip_bytes' of data in the iovecs are not sent, and none of the file | |
1171 | * descriptors are sent. The function continues to retry sending until an | |
1172 | * error (other than EINTR) occurs or all the data and fds are sent. | |
1173 | * | |
1174 | * Returns 0 if all the data and fds were successfully sent, otherwise a | |
1175 | * positive errno value. Regardless of success, stores the number of bytes | |
1176 | * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte | |
1177 | * is sent, then all the fds have been sent.) | |
1178 | * | |
1179 | * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */ | |
1180 | int | |
1181 | send_iovec_and_fds_fully(int sock, | |
1182 | const struct iovec iovs[], size_t n_iovs, | |
1183 | const int fds[], size_t n_fds, | |
1184 | size_t skip_bytes, size_t *bytes_sent) | |
1185 | { | |
1186 | *bytes_sent = 0; | |
1187 | while (n_iovs > 0) { | |
1188 | int retval; | |
1189 | ||
1190 | if (skip_bytes) { | |
1191 | retval = skip_bytes; | |
1192 | skip_bytes = 0; | |
1193 | } else if (!*bytes_sent) { | |
1194 | retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds); | |
1195 | } else { | |
1196 | retval = writev(sock, iovs, n_iovs); | |
1197 | } | |
1198 | ||
1199 | if (retval > 0) { | |
1200 | *bytes_sent += retval; | |
1201 | while (retval > 0) { | |
1202 | const uint8_t *base = iovs->iov_base; | |
1203 | size_t len = iovs->iov_len; | |
1204 | ||
1205 | if (retval < len) { | |
1206 | size_t sent; | |
1207 | int error; | |
1208 | ||
1209 | error = write_fully(sock, base + retval, len - retval, | |
1210 | &sent); | |
1211 | *bytes_sent += sent; | |
1212 | retval += sent; | |
1213 | if (error) { | |
1214 | return error; | |
1215 | } | |
1216 | } | |
1217 | retval -= len; | |
1218 | iovs++; | |
1219 | n_iovs--; | |
1220 | } | |
1221 | } else if (retval == 0) { | |
1222 | if (iovec_is_empty(iovs, n_iovs)) { | |
1223 | break; | |
1224 | } | |
1225 | VLOG_WARN("send returned 0"); | |
1226 | return EPROTO; | |
1227 | } else if (errno != EINTR) { | |
1228 | return errno; | |
1229 | } | |
1230 | } | |
1231 | ||
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors | |
1236 | * in 'fds' on Unix domain socket 'sock'. The function continues to retry | |
1237 | * sending until an error (other than EAGAIN or EINTR) occurs or all the data | |
1238 | * and fds are sent. Upon EAGAIN, the function blocks until the socket is | |
1239 | * ready for more data. | |
1240 | * | |
1241 | * Returns 0 if all the data and fds were successfully sent, otherwise a | |
1242 | * positive errno value. */ | |
1243 | int | |
1244 | send_iovec_and_fds_fully_block(int sock, | |
1245 | const struct iovec iovs[], size_t n_iovs, | |
1246 | const int fds[], size_t n_fds) | |
1247 | { | |
1248 | size_t sent = 0; | |
1249 | ||
1250 | for (;;) { | |
1251 | int error; | |
1252 | ||
1253 | error = send_iovec_and_fds_fully(sock, iovs, n_iovs, | |
1254 | fds, n_fds, sent, &sent); | |
1255 | if (error != EAGAIN) { | |
1256 | return error; | |
1257 | } | |
1258 | poll_fd_wait(sock, POLLOUT); | |
1259 | poll_block(); | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of | |
1264 | * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'. | |
1265 | * | |
1266 | * - Upon success, returns the number of bytes of data copied into 'data' | |
1267 | * and stores the number of received file descriptors into '*n_fdsp'. | |
1268 | * | |
1269 | * - On failure, returns a negative errno value and stores 0 in | |
1270 | * '*n_fdsp'. | |
1271 | * | |
1272 | * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */ | |
1273 | int | |
1274 | recv_data_and_fds(int sock, | |
1275 | void *data, size_t size, | |
1276 | int fds[SOUTIL_MAX_FDS], size_t *n_fdsp) | |
1277 | { | |
1278 | union { | |
1279 | struct cmsghdr cm; | |
1280 | char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; | |
1281 | } cmsg; | |
1282 | struct msghdr msg; | |
1283 | int retval; | |
1284 | struct cmsghdr *p; | |
1285 | size_t i; | |
1286 | ||
1287 | *n_fdsp = 0; | |
1288 | ||
1289 | do { | |
1290 | struct iovec iov; | |
1291 | ||
1292 | iov.iov_base = data; | |
1293 | iov.iov_len = size; | |
1294 | ||
1295 | msg.msg_name = NULL; | |
1296 | msg.msg_namelen = 0; | |
1297 | msg.msg_iov = &iov; | |
1298 | msg.msg_iovlen = 1; | |
1299 | msg.msg_control = &cmsg.cm; | |
1300 | msg.msg_controllen = sizeof cmsg.control; | |
1301 | msg.msg_flags = 0; | |
1302 | ||
1303 | retval = recvmsg(sock, &msg, 0); | |
1304 | } while (retval < 0 && errno == EINTR); | |
1305 | if (retval <= 0) { | |
1306 | return retval < 0 ? -errno : 0; | |
1307 | } | |
1308 | ||
1309 | for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) { | |
1310 | if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) { | |
1311 | VLOG_ERR("unexpected control message %d:%d", | |
1312 | p->cmsg_level, p->cmsg_type); | |
1313 | goto error; | |
1314 | } else if (*n_fdsp) { | |
1315 | VLOG_ERR("multiple SCM_RIGHTS received"); | |
1316 | goto error; | |
1317 | } else { | |
1318 | size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds; | |
1319 | const int *fds_data = (const int *) CMSG_DATA(p); | |
1320 | ||
cb22974d | 1321 | ovs_assert(n_fds > 0); |
fd94a42c BP |
1322 | if (n_fds > SOUTIL_MAX_FDS) { |
1323 | VLOG_ERR("%zu fds received but only %d supported", | |
1324 | n_fds, SOUTIL_MAX_FDS); | |
1325 | for (i = 0; i < n_fds; i++) { | |
1326 | close(fds_data[i]); | |
1327 | } | |
1328 | goto error; | |
1329 | } | |
1330 | ||
1331 | *n_fdsp = n_fds; | |
1332 | memcpy(fds, fds_data, n_fds * sizeof *fds); | |
1333 | } | |
1334 | } | |
1335 | ||
1336 | return retval; | |
1337 | ||
1338 | error: | |
1339 | for (i = 0; i < *n_fdsp; i++) { | |
1340 | close(fds[i]); | |
1341 | } | |
1342 | *n_fdsp = 0; | |
1343 | return EPROTO; | |
1344 | } |