]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
db5ce514 | 2 | * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. |
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> | |
064af421 BP |
32 | #include <sys/un.h> |
33 | #include <unistd.h> | |
f89ffb0e | 34 | #include "dynamic-string.h" |
064af421 | 35 | #include "fatal-signal.h" |
f89ffb0e | 36 | #include "packets.h" |
064af421 | 37 | #include "util.h" |
064af421 | 38 | #include "vlog.h" |
f89ffb0e BP |
39 | #if AF_PACKET && __linux__ |
40 | #include <linux/if_packet.h> | |
41 | #endif | |
42 | #ifdef HAVE_NETLINK | |
43 | #include "netlink-protocol.h" | |
44 | #include "netlink-socket.h" | |
45 | #endif | |
5136ce49 | 46 | |
d98e6007 | 47 | VLOG_DEFINE_THIS_MODULE(socket_util); |
064af421 | 48 | |
e61c056c BP |
49 | /* #ifdefs make it a pain to maintain code: you have to try to build both ways. |
50 | * Thus, this file compiles all of the code regardless of the target, by | |
51 | * writing "if (LINUX)" instead of "#ifdef __linux__". */ | |
52 | #ifdef __linux__ | |
53 | #define LINUX 1 | |
54 | #else | |
55 | #define LINUX 0 | |
56 | #endif | |
57 | ||
6e170b4c BP |
58 | #ifndef O_DIRECTORY |
59 | #define O_DIRECTORY 0 | |
60 | #endif | |
61 | ||
064af421 BP |
62 | /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a |
63 | * positive errno value. */ | |
64 | int | |
65 | set_nonblocking(int fd) | |
66 | { | |
67 | int flags = fcntl(fd, F_GETFL, 0); | |
68 | if (flags != -1) { | |
69 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { | |
70 | return 0; | |
71 | } else { | |
72 | VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); | |
73 | return errno; | |
74 | } | |
75 | } else { | |
76 | VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); | |
77 | return errno; | |
78 | } | |
79 | } | |
80 | ||
0ec6cfb1 BP |
81 | static bool |
82 | rlim_is_finite(rlim_t limit) | |
83 | { | |
84 | if (limit == RLIM_INFINITY) { | |
85 | return false; | |
86 | } | |
87 | ||
88 | #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */ | |
89 | if (limit == RLIM_SAVED_CUR) { | |
90 | return false; | |
91 | } | |
92 | #endif | |
93 | ||
94 | #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */ | |
95 | if (limit == RLIM_SAVED_MAX) { | |
96 | return false; | |
97 | } | |
98 | #endif | |
99 | ||
100 | return true; | |
101 | } | |
102 | ||
064af421 BP |
103 | /* Returns the maximum valid FD value, plus 1. */ |
104 | int | |
105 | get_max_fds(void) | |
106 | { | |
107 | static int max_fds = -1; | |
108 | if (max_fds < 0) { | |
109 | struct rlimit r; | |
0ec6cfb1 | 110 | if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) { |
064af421 BP |
111 | max_fds = r.rlim_cur; |
112 | } else { | |
113 | VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); | |
114 | max_fds = 1024; | |
115 | } | |
116 | } | |
117 | return max_fds; | |
118 | } | |
119 | ||
2b35e147 BP |
120 | /* Translates 'host_name', which must be a string representation of an IP |
121 | * address, into a numeric IP address in '*addr'. Returns 0 if successful, | |
122 | * otherwise a positive errno value. */ | |
064af421 | 123 | int |
d295e8e9 | 124 | lookup_ip(const char *host_name, struct in_addr *addr) |
064af421 BP |
125 | { |
126 | if (!inet_aton(host_name, addr)) { | |
db5ce514 | 127 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2b35e147 BP |
128 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name); |
129 | return ENOENT; | |
064af421 BP |
130 | } |
131 | return 0; | |
132 | } | |
133 | ||
d31f1109 JP |
134 | /* Translates 'host_name', which must be a string representation of an IPv6 |
135 | * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, | |
136 | * otherwise a positive errno value. */ | |
137 | int | |
138 | lookup_ipv6(const char *host_name, struct in6_addr *addr) | |
139 | { | |
140 | if (inet_pton(AF_INET6, host_name, addr) != 1) { | |
141 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
142 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name); | |
143 | return ENOENT; | |
144 | } | |
145 | return 0; | |
146 | } | |
147 | ||
13f2ef97 BP |
148 | /* Translates 'host_name', which must be a host name or a string representation |
149 | * of an IP address, into a numeric IP address in '*addr'. Returns 0 if | |
150 | * successful, otherwise a positive errno value. | |
151 | * | |
152 | * Most Open vSwitch code should not use this because it causes deadlocks: | |
153 | * gethostbyname() sends out a DNS request but that starts a new flow for which | |
154 | * OVS must set up a flow, but it can't because it's waiting for a DNS reply. | |
155 | * The synchronous lookup also delays other activty. (Of course we can solve | |
156 | * this but it doesn't seem worthwhile quite yet.) */ | |
157 | int | |
158 | lookup_hostname(const char *host_name, struct in_addr *addr) | |
159 | { | |
160 | struct hostent *h; | |
161 | ||
162 | if (inet_aton(host_name, addr)) { | |
163 | return 0; | |
164 | } | |
165 | ||
166 | h = gethostbyname(host_name); | |
167 | if (h) { | |
168 | *addr = *(struct in_addr *) h->h_addr; | |
169 | return 0; | |
170 | } | |
171 | ||
172 | return (h_errno == HOST_NOT_FOUND ? ENOENT | |
173 | : h_errno == TRY_AGAIN ? EAGAIN | |
174 | : h_errno == NO_RECOVERY ? EIO | |
175 | : h_errno == NO_ADDRESS ? ENXIO | |
176 | : EINVAL); | |
177 | } | |
178 | ||
064af421 BP |
179 | /* Returns the error condition associated with socket 'fd' and resets the |
180 | * socket's error status. */ | |
181 | int | |
d295e8e9 | 182 | get_socket_error(int fd) |
064af421 BP |
183 | { |
184 | int error; | |
185 | socklen_t len = sizeof(error); | |
186 | if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) { | |
db5ce514 | 187 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
064af421 BP |
188 | error = errno; |
189 | VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error)); | |
190 | } | |
191 | return error; | |
192 | } | |
193 | ||
194 | int | |
d295e8e9 | 195 | check_connection_completion(int fd) |
064af421 BP |
196 | { |
197 | struct pollfd pfd; | |
198 | int retval; | |
199 | ||
200 | pfd.fd = fd; | |
201 | pfd.events = POLLOUT; | |
202 | do { | |
203 | retval = poll(&pfd, 1, 0); | |
204 | } while (retval < 0 && errno == EINTR); | |
205 | if (retval == 1) { | |
206 | return get_socket_error(fd); | |
207 | } else if (retval < 0) { | |
208 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); | |
209 | VLOG_ERR_RL(&rl, "poll: %s", strerror(errno)); | |
210 | return errno; | |
211 | } else { | |
212 | return EAGAIN; | |
213 | } | |
214 | } | |
215 | ||
216 | /* Drain all the data currently in the receive queue of a datagram socket (and | |
217 | * possibly additional data). There is no way to know how many packets are in | |
218 | * the receive queue, but we do know that the total number of bytes queued does | |
219 | * not exceed the receive buffer size, so we pull packets until none are left | |
220 | * or we've read that many bytes. */ | |
221 | int | |
222 | drain_rcvbuf(int fd) | |
223 | { | |
224 | socklen_t rcvbuf_len; | |
6d8ffe18 | 225 | size_t rcvbuf = 0; |
064af421 BP |
226 | |
227 | rcvbuf_len = sizeof rcvbuf; | |
228 | if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) { | |
229 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); | |
230 | VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno)); | |
231 | return errno; | |
232 | } | |
233 | while (rcvbuf > 0) { | |
234 | /* In Linux, specifying MSG_TRUNC in the flags argument causes the | |
235 | * datagram length to be returned, even if that is longer than the | |
236 | * buffer provided. Thus, we can use a 1-byte buffer to discard the | |
237 | * incoming datagram and still be able to account how many bytes were | |
238 | * removed from the receive buffer. | |
239 | * | |
240 | * On other Unix-like OSes, MSG_TRUNC has no effect in the flags | |
241 | * argument. */ | |
e61c056c | 242 | char buffer[LINUX ? 1 : 2048]; |
064af421 BP |
243 | ssize_t n_bytes = recv(fd, buffer, sizeof buffer, |
244 | MSG_TRUNC | MSG_DONTWAIT); | |
245 | if (n_bytes <= 0 || n_bytes >= rcvbuf) { | |
246 | break; | |
247 | } | |
248 | rcvbuf -= n_bytes; | |
249 | } | |
250 | return 0; | |
251 | } | |
252 | ||
253 | /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no | |
254 | * more data can be immediately read. ('fd' should therefore be in | |
255 | * non-blocking mode.)*/ | |
256 | void | |
257 | drain_fd(int fd, size_t n_packets) | |
258 | { | |
259 | for (; n_packets > 0; n_packets--) { | |
260 | /* 'buffer' only needs to be 1 byte long in most circumstances. This | |
261 | * size is defensive against the possibility that we someday want to | |
262 | * use a Linux tap device without TUN_NO_PI, in which case a buffer | |
263 | * smaller than sizeof(struct tun_pi) will give EINVAL on read. */ | |
264 | char buffer[128]; | |
265 | if (read(fd, buffer, sizeof buffer) <= 0) { | |
266 | break; | |
267 | } | |
268 | } | |
269 | } | |
270 | ||
271 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in | |
272 | * '*un_len' the size of the sockaddr_un. */ | |
273 | static void | |
6e170b4c | 274 | make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len) |
064af421 BP |
275 | { |
276 | un->sun_family = AF_UNIX; | |
71d7c22f | 277 | ovs_strzcpy(un->sun_path, name, sizeof un->sun_path); |
064af421 BP |
278 | *un_len = (offsetof(struct sockaddr_un, sun_path) |
279 | + strlen (un->sun_path) + 1); | |
280 | } | |
281 | ||
6e170b4c BP |
282 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in |
283 | * '*un_len' the size of the sockaddr_un. | |
284 | * | |
285 | * Returns 0 on success, otherwise a positive errno value. On success, | |
286 | * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller | |
287 | * should close after using '*un' to bind or connect. On failure, '*dirfdp' is | |
288 | * -1. */ | |
289 | static int | |
290 | make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, | |
291 | int *dirfdp) | |
292 | { | |
293 | enum { MAX_UN_LEN = sizeof un->sun_path - 1 }; | |
294 | ||
295 | *dirfdp = -1; | |
296 | if (strlen(name) > MAX_UN_LEN) { | |
297 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
298 | ||
299 | if (LINUX) { | |
300 | /* 'name' is too long to fit in a sockaddr_un, but we have a | |
301 | * workaround for that on Linux: shorten it by opening a file | |
302 | * descriptor for the directory part of the name and indirecting | |
303 | * through /proc/self/fd/<dirfd>/<basename>. */ | |
304 | char *dir, *base; | |
305 | char *short_name; | |
306 | int dirfd; | |
307 | ||
308 | dir = dir_name(name); | |
309 | base = base_name(name); | |
310 | ||
311 | dirfd = open(dir, O_DIRECTORY | O_RDONLY); | |
312 | if (dirfd < 0) { | |
08871bf6 JP |
313 | free(base); |
314 | free(dir); | |
6e170b4c BP |
315 | return errno; |
316 | } | |
317 | ||
318 | short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base); | |
319 | free(dir); | |
320 | free(base); | |
321 | ||
322 | if (strlen(short_name) <= MAX_UN_LEN) { | |
323 | make_sockaddr_un__(short_name, un, un_len); | |
324 | free(short_name); | |
325 | *dirfdp = dirfd; | |
326 | return 0; | |
327 | } | |
328 | free(short_name); | |
329 | close(dirfd); | |
330 | ||
331 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
332 | "%d bytes (even shortened)", name, MAX_UN_LEN); | |
333 | } else { | |
334 | /* 'name' is too long and we have no workaround. */ | |
335 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
336 | "%d bytes", name, MAX_UN_LEN); | |
337 | } | |
338 | ||
339 | return ENAMETOOLONG; | |
340 | } else { | |
341 | make_sockaddr_un__(name, un, un_len); | |
342 | return 0; | |
343 | } | |
344 | } | |
345 | ||
e2bd0593 BP |
346 | /* Binds Unix domain socket 'fd' to a file with permissions 0700. */ |
347 | static int | |
348 | bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len) | |
349 | { | |
e2bd0593 BP |
350 | /* According to _Unix Network Programming_, umask should affect bind(). */ |
351 | mode_t old_umask = umask(0077); | |
352 | int error = bind(fd, sun, sun_len) ? errno : 0; | |
353 | umask(old_umask); | |
354 | return error; | |
e2bd0593 BP |
355 | } |
356 | ||
064af421 BP |
357 | /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or |
358 | * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and | |
359 | * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock' | |
360 | * is true, the socket is made non-blocking. If 'passcred' is true, the socket | |
361 | * is configured to receive SCM_CREDENTIALS control messages. | |
362 | * | |
363 | * Returns the socket's fd if successful, otherwise a negative errno value. */ | |
364 | int | |
67a4917b | 365 | make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, |
064af421 BP |
366 | const char *bind_path, const char *connect_path) |
367 | { | |
368 | int error; | |
369 | int fd; | |
370 | ||
371 | fd = socket(PF_UNIX, style, 0); | |
372 | if (fd < 0) { | |
373 | return -errno; | |
374 | } | |
375 | ||
376 | /* Set nonblocking mode right away, if we want it. This prevents blocking | |
377 | * in connect(), if connect_path != NULL. (In turn, that's a corner case: | |
378 | * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only | |
379 | * if a backlog of un-accepted connections has built up in the kernel.) */ | |
380 | if (nonblock) { | |
381 | int flags = fcntl(fd, F_GETFL, 0); | |
382 | if (flags == -1) { | |
6e170b4c | 383 | error = errno; |
064af421 BP |
384 | goto error; |
385 | } | |
386 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { | |
6e170b4c | 387 | error = errno; |
064af421 BP |
388 | goto error; |
389 | } | |
390 | } | |
391 | ||
392 | if (bind_path) { | |
393 | struct sockaddr_un un; | |
394 | socklen_t un_len; | |
6e170b4c BP |
395 | int dirfd; |
396 | ||
397 | if (unlink(bind_path) && errno != ENOENT) { | |
398 | VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); | |
064af421 BP |
399 | } |
400 | fatal_signal_add_file_to_unlink(bind_path); | |
6e170b4c BP |
401 | |
402 | error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd); | |
e2bd0593 BP |
403 | if (!error) { |
404 | error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len); | |
6e170b4c BP |
405 | } |
406 | if (dirfd >= 0) { | |
407 | close(dirfd); | |
408 | } | |
409 | if (error) { | |
064af421 BP |
410 | goto error; |
411 | } | |
412 | } | |
413 | ||
414 | if (connect_path) { | |
415 | struct sockaddr_un un; | |
416 | socklen_t un_len; | |
6e170b4c BP |
417 | int dirfd; |
418 | ||
419 | error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd); | |
420 | if (!error | |
421 | && connect(fd, (struct sockaddr*) &un, un_len) | |
064af421 | 422 | && errno != EINPROGRESS) { |
6e170b4c BP |
423 | error = errno; |
424 | } | |
425 | if (dirfd >= 0) { | |
426 | close(dirfd); | |
427 | } | |
428 | if (error) { | |
064af421 BP |
429 | goto error; |
430 | } | |
431 | } | |
432 | ||
433 | #ifdef SCM_CREDENTIALS | |
434 | if (passcred) { | |
435 | int enable = 1; | |
436 | if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) { | |
6e170b4c | 437 | error = errno; |
064af421 BP |
438 | goto error; |
439 | } | |
440 | } | |
441 | #endif | |
442 | ||
443 | return fd; | |
444 | ||
445 | error: | |
6e170b4c BP |
446 | if (error == EAGAIN) { |
447 | error = EPROTO; | |
448 | } | |
064af421 BP |
449 | if (bind_path) { |
450 | fatal_signal_remove_file_to_unlink(bind_path); | |
451 | } | |
064af421 BP |
452 | close(fd); |
453 | return -error; | |
454 | } | |
455 | ||
456 | int | |
457 | get_unix_name_len(socklen_t sun_len) | |
458 | { | |
459 | return (sun_len >= offsetof(struct sockaddr_un, sun_path) | |
460 | ? sun_len - offsetof(struct sockaddr_un, sun_path) | |
461 | : 0); | |
462 | } | |
463 | ||
dbba996b BP |
464 | ovs_be32 |
465 | guess_netmask(ovs_be32 ip_) | |
064af421 | 466 | { |
dbba996b | 467 | uint32_t ip = ntohl(ip_); |
064af421 BP |
468 | return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ |
469 | : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ | |
470 | : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ | |
471 | : htonl(0)); /* ??? */ | |
472 | } | |
473 | ||
52f8a75e BP |
474 | /* Parses 'target', which should be a string in the format "<host>[:<port>]". |
475 | * <host> is required. If 'default_port' is nonzero then <port> is optional | |
476 | * and defaults to 'default_port'. | |
4f2eb9a7 | 477 | * |
52f8a75e BP |
478 | * On success, returns true and stores the parsed remote address into '*sinp'. |
479 | * On failure, logs an error, stores zeros into '*sinp', and returns false. */ | |
480 | bool | |
481 | inet_parse_active(const char *target_, uint16_t default_port, | |
482 | struct sockaddr_in *sinp) | |
78ff0270 BP |
483 | { |
484 | char *target = xstrdup(target_); | |
485 | char *save_ptr = NULL; | |
486 | const char *host_name; | |
487 | const char *port_string; | |
52f8a75e | 488 | bool ok = false; |
78ff0270 BP |
489 | |
490 | /* Defaults. */ | |
52f8a75e BP |
491 | sinp->sin_family = AF_INET; |
492 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
493 | |
494 | /* Tokenize. */ | |
495 | host_name = strtok_r(target, ":", &save_ptr); | |
496 | port_string = strtok_r(NULL, ":", &save_ptr); | |
497 | if (!host_name) { | |
52f8a75e | 498 | VLOG_ERR("%s: bad peer name format", target_); |
78ff0270 BP |
499 | goto exit; |
500 | } | |
501 | ||
502 | /* Look up IP, port. */ | |
52f8a75e | 503 | if (lookup_ip(host_name, &sinp->sin_addr)) { |
78ff0270 BP |
504 | goto exit; |
505 | } | |
506 | if (port_string && atoi(port_string)) { | |
52f8a75e | 507 | sinp->sin_port = htons(atoi(port_string)); |
1901968e BP |
508 | } else if (!default_port) { |
509 | VLOG_ERR("%s: port number must be specified", target_); | |
52f8a75e BP |
510 | goto exit; |
511 | } | |
512 | ||
513 | ok = true; | |
514 | ||
515 | exit: | |
516 | if (!ok) { | |
517 | memset(sinp, 0, sizeof *sinp); | |
518 | } | |
519 | free(target); | |
520 | return ok; | |
521 | } | |
522 | ||
523 | /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to | |
524 | * 'target', which should be a string in the format "<host>[:<port>]". <host> | |
525 | * is required. If 'default_port' is nonzero then <port> is optional and | |
526 | * defaults to 'default_port'. | |
527 | * | |
528 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
529 | * | |
530 | * On success, returns 0 (indicating connection complete) or EAGAIN (indicating | |
531 | * connection in progress), in which case the new file descriptor is stored | |
532 | * into '*fdp'. On failure, returns a positive errno value other than EAGAIN | |
533 | * and stores -1 into '*fdp'. | |
534 | * | |
535 | * If 'sinp' is non-null, then on success the target address is stored into | |
536 | * '*sinp'. */ | |
537 | int | |
538 | inet_open_active(int style, const char *target, uint16_t default_port, | |
539 | struct sockaddr_in *sinp, int *fdp) | |
540 | { | |
541 | struct sockaddr_in sin; | |
542 | int fd = -1; | |
543 | int error; | |
544 | ||
545 | /* Parse. */ | |
546 | if (!inet_parse_active(target, default_port, &sin)) { | |
1901968e BP |
547 | error = EAFNOSUPPORT; |
548 | goto exit; | |
78ff0270 BP |
549 | } |
550 | ||
551 | /* Create non-blocking socket. */ | |
4f2eb9a7 | 552 | fd = socket(AF_INET, style, 0); |
78ff0270 | 553 | if (fd < 0) { |
52f8a75e | 554 | VLOG_ERR("%s: socket: %s", target, strerror(errno)); |
78ff0270 BP |
555 | error = errno; |
556 | goto exit; | |
557 | } | |
558 | error = set_nonblocking(fd); | |
559 | if (error) { | |
560 | goto exit_close; | |
561 | } | |
562 | ||
563 | /* Connect. */ | |
564 | error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno; | |
565 | if (error == EINPROGRESS) { | |
566 | error = EAGAIN; | |
567 | } else if (error && error != EAGAIN) { | |
568 | goto exit_close; | |
569 | } | |
570 | ||
571 | /* Success: error is 0 or EAGAIN. */ | |
572 | goto exit; | |
573 | ||
574 | exit_close: | |
575 | close(fd); | |
576 | exit: | |
577 | if (!error || error == EAGAIN) { | |
578 | if (sinp) { | |
579 | *sinp = sin; | |
580 | } | |
581 | *fdp = fd; | |
582 | } else { | |
583 | *fdp = -1; | |
584 | } | |
78ff0270 BP |
585 | return error; |
586 | } | |
587 | ||
d98fa503 | 588 | /* Parses 'target', which should be a string in the format "[<port>][:<ip>]": |
e1bd3bee BP |
589 | * |
590 | * - If 'default_port' is -1, then <port> is required. Otherwise, if | |
591 | * <port> is omitted, then 'default_port' is used instead. | |
592 | * | |
593 | * - If <port> (or 'default_port', if used) is 0, then no port is bound | |
594 | * and the TCP/IP stack will select a port. | |
595 | * | |
596 | * - If <ip> is omitted then the IP address is wildcarded. | |
4f2eb9a7 | 597 | * |
d98fa503 BP |
598 | * If successful, stores the address into '*sinp' and returns true; otherwise |
599 | * zeros '*sinp' and returns false. */ | |
600 | bool | |
601 | inet_parse_passive(const char *target_, uint16_t default_port, | |
602 | struct sockaddr_in *sinp) | |
78ff0270 BP |
603 | { |
604 | char *target = xstrdup(target_); | |
605 | char *string_ptr = target; | |
78ff0270 BP |
606 | const char *host_name; |
607 | const char *port_string; | |
d98fa503 BP |
608 | bool ok = false; |
609 | int port; | |
78ff0270 BP |
610 | |
611 | /* Address defaults. */ | |
d98fa503 BP |
612 | memset(sinp, 0, sizeof *sinp); |
613 | sinp->sin_family = AF_INET; | |
614 | sinp->sin_addr.s_addr = htonl(INADDR_ANY); | |
615 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
616 | |
617 | /* Parse optional port number. */ | |
618 | port_string = strsep(&string_ptr, ":"); | |
e1bd3bee | 619 | if (port_string && str_to_int(port_string, 10, &port)) { |
d98fa503 | 620 | sinp->sin_port = htons(port); |
e1bd3bee | 621 | } else if (default_port < 0) { |
1901968e | 622 | VLOG_ERR("%s: port number must be specified", target_); |
1901968e | 623 | goto exit; |
78ff0270 BP |
624 | } |
625 | ||
626 | /* Parse optional bind IP. */ | |
627 | host_name = strsep(&string_ptr, ":"); | |
d98fa503 BP |
628 | if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { |
629 | goto exit; | |
630 | } | |
631 | ||
632 | ok = true; | |
633 | ||
634 | exit: | |
635 | if (!ok) { | |
636 | memset(sinp, 0, sizeof *sinp); | |
637 | } | |
638 | free(target); | |
639 | return ok; | |
640 | } | |
641 | ||
642 | ||
643 | /* Opens a non-blocking IPv4 socket of the specified 'style', binds to | |
644 | * 'target', and listens for incoming connections. Parses 'target' in the same | |
645 | * way was inet_parse_passive(). | |
646 | * | |
647 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
648 | * | |
649 | * For TCP, the socket will have SO_REUSEADDR turned on. | |
650 | * | |
651 | * On success, returns a non-negative file descriptor. On failure, returns a | |
652 | * negative errno value. | |
653 | * | |
654 | * If 'sinp' is non-null, then on success the bound address is stored into | |
655 | * '*sinp'. */ | |
656 | int | |
657 | inet_open_passive(int style, const char *target, int default_port, | |
658 | struct sockaddr_in *sinp) | |
659 | { | |
660 | struct sockaddr_in sin; | |
661 | int fd = 0, error; | |
662 | unsigned int yes = 1; | |
663 | ||
664 | if (!inet_parse_passive(target, default_port, &sin)) { | |
665 | return EAFNOSUPPORT; | |
78ff0270 BP |
666 | } |
667 | ||
668 | /* Create non-blocking socket, set SO_REUSEADDR. */ | |
4f2eb9a7 | 669 | fd = socket(AF_INET, style, 0); |
78ff0270 BP |
670 | if (fd < 0) { |
671 | error = errno; | |
d98fa503 BP |
672 | VLOG_ERR("%s: socket: %s", target, strerror(error)); |
673 | return error; | |
78ff0270 BP |
674 | } |
675 | error = set_nonblocking(fd); | |
676 | if (error) { | |
d98fa503 | 677 | goto error; |
78ff0270 | 678 | } |
4f2eb9a7 BP |
679 | if (style == SOCK_STREAM |
680 | && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { | |
78ff0270 | 681 | error = errno; |
d98fa503 BP |
682 | VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error)); |
683 | goto error; | |
78ff0270 BP |
684 | } |
685 | ||
686 | /* Bind. */ | |
687 | if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) { | |
688 | error = errno; | |
d98fa503 BP |
689 | VLOG_ERR("%s: bind: %s", target, strerror(error)); |
690 | goto error; | |
78ff0270 BP |
691 | } |
692 | ||
693 | /* Listen. */ | |
694 | if (listen(fd, 10) < 0) { | |
695 | error = errno; | |
d98fa503 BP |
696 | VLOG_ERR("%s: listen: %s", target, strerror(error)); |
697 | goto error; | |
78ff0270 | 698 | } |
36775dad BP |
699 | |
700 | if (sinp) { | |
701 | socklen_t sin_len = sizeof sin; | |
702 | if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){ | |
703 | error = errno; | |
d98fa503 BP |
704 | VLOG_ERR("%s: getsockname: %s", target, strerror(error)); |
705 | goto error; | |
36775dad BP |
706 | } |
707 | if (sin.sin_family != AF_INET || sin_len != sizeof sin) { | |
d98fa503 BP |
708 | VLOG_ERR("%s: getsockname: invalid socket name", target); |
709 | goto error; | |
36775dad BP |
710 | } |
711 | *sinp = sin; | |
712 | } | |
713 | ||
d98fa503 | 714 | return fd; |
78ff0270 | 715 | |
d98fa503 | 716 | error: |
78ff0270 | 717 | close(fd); |
d98fa503 | 718 | return error; |
78ff0270 BP |
719 | } |
720 | ||
a8b5f8b4 BP |
721 | /* Returns a readable and writable fd for /dev/null, if successful, otherwise |
722 | * a negative errno value. The caller must not close the returned fd (because | |
723 | * the same fd will be handed out to subsequent callers). */ | |
724 | int | |
725 | get_null_fd(void) | |
726 | { | |
727 | static int null_fd = -1; | |
728 | if (null_fd < 0) { | |
729 | null_fd = open("/dev/null", O_RDWR); | |
730 | if (null_fd < 0) { | |
731 | int error = errno; | |
732 | VLOG_ERR("could not open /dev/null: %s", strerror(error)); | |
733 | return -error; | |
734 | } | |
735 | } | |
736 | return null_fd; | |
737 | } | |
738 | ||
064af421 BP |
739 | int |
740 | read_fully(int fd, void *p_, size_t size, size_t *bytes_read) | |
741 | { | |
742 | uint8_t *p = p_; | |
743 | ||
744 | *bytes_read = 0; | |
745 | while (size > 0) { | |
746 | ssize_t retval = read(fd, p, size); | |
747 | if (retval > 0) { | |
748 | *bytes_read += retval; | |
749 | size -= retval; | |
750 | p += retval; | |
751 | } else if (retval == 0) { | |
752 | return EOF; | |
753 | } else if (errno != EINTR) { | |
754 | return errno; | |
755 | } | |
756 | } | |
757 | return 0; | |
758 | } | |
759 | ||
760 | int | |
761 | write_fully(int fd, const void *p_, size_t size, size_t *bytes_written) | |
762 | { | |
763 | const uint8_t *p = p_; | |
764 | ||
765 | *bytes_written = 0; | |
766 | while (size > 0) { | |
767 | ssize_t retval = write(fd, p, size); | |
768 | if (retval > 0) { | |
769 | *bytes_written += retval; | |
770 | size -= retval; | |
771 | p += retval; | |
772 | } else if (retval == 0) { | |
773 | VLOG_WARN("write returned 0"); | |
774 | return EPROTO; | |
775 | } else if (errno != EINTR) { | |
776 | return errno; | |
777 | } | |
778 | } | |
779 | return 0; | |
780 | } | |
8e71cf88 BP |
781 | |
782 | /* Given file name 'file_name', fsyncs the directory in which it is contained. | |
783 | * Returns 0 if successful, otherwise a positive errno value. */ | |
784 | int | |
785 | fsync_parent_dir(const char *file_name) | |
786 | { | |
787 | int error = 0; | |
788 | char *dir; | |
789 | int fd; | |
790 | ||
791 | dir = dir_name(file_name); | |
792 | fd = open(dir, O_RDONLY); | |
793 | if (fd >= 0) { | |
794 | if (fsync(fd)) { | |
795 | if (errno == EINVAL || errno == EROFS) { | |
796 | /* This directory does not support synchronization. Not | |
797 | * really an error. */ | |
798 | } else { | |
799 | error = errno; | |
800 | VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); | |
801 | } | |
802 | } | |
803 | close(fd); | |
804 | } else { | |
805 | error = errno; | |
806 | VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); | |
807 | } | |
808 | free(dir); | |
809 | ||
810 | return error; | |
811 | } | |
26efd256 BP |
812 | |
813 | /* Obtains the modification time of the file named 'file_name' to the greatest | |
814 | * supported precision. If successful, stores the mtime in '*mtime' and | |
815 | * returns 0. On error, returns a positive errno value and stores zeros in | |
816 | * '*mtime'. */ | |
817 | int | |
818 | get_mtime(const char *file_name, struct timespec *mtime) | |
819 | { | |
820 | struct stat s; | |
821 | ||
822 | if (!stat(file_name, &s)) { | |
823 | mtime->tv_sec = s.st_mtime; | |
824 | ||
825 | #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC | |
826 | mtime->tv_nsec = s.st_mtim.tv_nsec; | |
827 | #elif HAVE_STRUCT_STAT_ST_MTIMENSEC | |
828 | mtime->tv_nsec = s.st_mtimensec; | |
829 | #else | |
830 | mtime->tv_nsec = 0; | |
831 | #endif | |
832 | ||
833 | return 0; | |
834 | } else { | |
835 | mtime->tv_sec = mtime->tv_nsec = 0; | |
836 | return errno; | |
837 | } | |
838 | } | |
839 | ||
279c9e03 BP |
840 | void |
841 | xpipe(int fds[2]) | |
842 | { | |
843 | if (pipe(fds)) { | |
844 | VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); | |
845 | } | |
846 | } | |
f89ffb0e BP |
847 | |
848 | static int | |
849 | getsockopt_int(int fd, int level, int optname, int *valuep) | |
850 | { | |
851 | socklen_t len = sizeof *valuep; | |
852 | ||
853 | return (getsockopt(fd, level, optname, valuep, &len) ? errno | |
854 | : len == sizeof *valuep ? 0 | |
855 | : EINVAL); | |
856 | } | |
857 | ||
858 | static void | |
859 | describe_sockaddr(struct ds *string, int fd, | |
860 | int (*getaddr)(int, struct sockaddr *, socklen_t *)) | |
861 | { | |
862 | struct sockaddr_storage ss; | |
863 | socklen_t len = sizeof ss; | |
864 | ||
865 | if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { | |
866 | if (ss.ss_family == AF_INET) { | |
867 | struct sockaddr_in sin; | |
868 | ||
869 | memcpy(&sin, &ss, sizeof sin); | |
870 | ds_put_format(string, IP_FMT":%"PRIu16, | |
871 | IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
872 | } else if (ss.ss_family == AF_UNIX) { | |
873 | struct sockaddr_un sun; | |
874 | const char *null; | |
875 | size_t maxlen; | |
876 | ||
877 | memcpy(&sun, &ss, sizeof sun); | |
878 | maxlen = len - offsetof(struct sockaddr_un, sun_path); | |
879 | null = memchr(sun.sun_path, '\0', maxlen); | |
880 | ds_put_buffer(string, sun.sun_path, | |
881 | null ? null - sun.sun_path : maxlen); | |
882 | } | |
883 | #ifdef HAVE_NETLINK | |
884 | else if (ss.ss_family == AF_NETLINK) { | |
885 | int protocol; | |
886 | ||
887 | /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version | |
888 | * of the Linux kernel headers in use at build time. */ | |
889 | #ifndef SO_PROTOCOL | |
890 | #define SO_PROTOCOL 38 | |
891 | #endif | |
892 | ||
893 | if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, &protocol)) { | |
894 | switch (protocol) { | |
895 | case NETLINK_ROUTE: | |
896 | ds_put_cstr(string, "NETLINK_ROUTE"); | |
897 | break; | |
898 | ||
899 | case NETLINK_GENERIC: | |
900 | ds_put_cstr(string, "NETLINK_GENERIC"); | |
901 | break; | |
902 | ||
903 | default: | |
904 | ds_put_format(string, "AF_NETLINK family %d", protocol); | |
905 | break; | |
906 | } | |
907 | } else { | |
908 | ds_put_cstr(string, "AF_NETLINK"); | |
909 | } | |
910 | } | |
911 | #endif | |
912 | #if AF_PACKET && __linux__ | |
913 | else if (ss.ss_family == AF_PACKET) { | |
914 | struct sockaddr_ll sll; | |
915 | ||
916 | memcpy(&sll, &ss, sizeof sll); | |
917 | ds_put_cstr(string, "AF_PACKET"); | |
918 | if (sll.sll_ifindex) { | |
919 | char name[IFNAMSIZ]; | |
920 | ||
921 | if (if_indextoname(sll.sll_ifindex, name)) { | |
922 | ds_put_format(string, "(%s)", name); | |
923 | } else { | |
924 | ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); | |
925 | } | |
926 | } | |
927 | if (sll.sll_protocol) { | |
928 | ds_put_format(string, "(protocol=0x%"PRIu16")", | |
929 | ntohs(sll.sll_protocol)); | |
930 | } | |
931 | } | |
932 | #endif | |
933 | else if (ss.ss_family == AF_UNSPEC) { | |
934 | ds_put_cstr(string, "AF_UNSPEC"); | |
935 | } else { | |
936 | ds_put_format(string, "AF_%d", (int) ss.ss_family); | |
937 | } | |
938 | } | |
939 | } | |
940 | ||
941 | ||
942 | #ifdef __linux__ | |
943 | static void | |
944 | put_fd_filename(struct ds *string, int fd) | |
945 | { | |
946 | char buf[1024]; | |
947 | char *linkname; | |
948 | int n; | |
949 | ||
950 | linkname = xasprintf("/proc/self/fd/%d", fd); | |
951 | n = readlink(linkname, buf, sizeof buf); | |
952 | if (n > 0) { | |
953 | ds_put_char(string, ' '); | |
954 | ds_put_buffer(string, buf, n); | |
955 | if (n > sizeof buf) { | |
956 | ds_put_cstr(string, "..."); | |
957 | } | |
958 | } | |
959 | free(linkname); | |
960 | } | |
961 | #endif | |
962 | ||
963 | /* Returns a malloc()'d string describing 'fd', for use in logging. */ | |
964 | char * | |
965 | describe_fd(int fd) | |
966 | { | |
967 | struct ds string; | |
968 | struct stat s; | |
969 | ||
970 | ds_init(&string); | |
971 | if (fstat(fd, &s)) { | |
972 | ds_put_format(&string, "fstat failed (%s)", strerror(errno)); | |
973 | } else if (S_ISSOCK(s.st_mode)) { | |
974 | describe_sockaddr(&string, fd, getsockname); | |
975 | ds_put_cstr(&string, "<->"); | |
976 | describe_sockaddr(&string, fd, getpeername); | |
977 | } else { | |
978 | ds_put_cstr(&string, (isatty(fd) ? "tty" | |
979 | : S_ISDIR(s.st_mode) ? "directory" | |
980 | : S_ISCHR(s.st_mode) ? "character device" | |
981 | : S_ISBLK(s.st_mode) ? "block device" | |
982 | : S_ISREG(s.st_mode) ? "file" | |
983 | : S_ISFIFO(s.st_mode) ? "FIFO" | |
984 | : S_ISLNK(s.st_mode) ? "symbolic link" | |
985 | : "unknown")); | |
986 | #ifdef __linux__ | |
987 | put_fd_filename(&string, fd); | |
988 | #endif | |
989 | } | |
990 | return ds_steal_cstr(&string); | |
991 | } |