]>
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> | |
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" |
064af421 | 38 | #include "util.h" |
064af421 | 39 | #include "vlog.h" |
f89ffb0e BP |
40 | #if AF_PACKET && __linux__ |
41 | #include <linux/if_packet.h> | |
42 | #endif | |
43 | #ifdef HAVE_NETLINK | |
44 | #include "netlink-protocol.h" | |
45 | #include "netlink-socket.h" | |
46 | #endif | |
5136ce49 | 47 | |
d98e6007 | 48 | VLOG_DEFINE_THIS_MODULE(socket_util); |
064af421 | 49 | |
e61c056c BP |
50 | /* #ifdefs make it a pain to maintain code: you have to try to build both ways. |
51 | * Thus, this file compiles all of the code regardless of the target, by | |
52 | * writing "if (LINUX)" instead of "#ifdef __linux__". */ | |
53 | #ifdef __linux__ | |
54 | #define LINUX 1 | |
55 | #else | |
56 | #define LINUX 0 | |
57 | #endif | |
58 | ||
6e170b4c BP |
59 | #ifndef O_DIRECTORY |
60 | #define O_DIRECTORY 0 | |
61 | #endif | |
62 | ||
c1c19657 BP |
63 | static int getsockopt_int(int fd, int level, int option, const char *optname, |
64 | int *valuep); | |
65 | ||
064af421 BP |
66 | /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a |
67 | * positive errno value. */ | |
68 | int | |
69 | set_nonblocking(int fd) | |
70 | { | |
71 | int flags = fcntl(fd, F_GETFL, 0); | |
72 | if (flags != -1) { | |
73 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { | |
74 | return 0; | |
75 | } else { | |
76 | VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); | |
77 | return errno; | |
78 | } | |
79 | } else { | |
80 | VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); | |
81 | return errno; | |
82 | } | |
83 | } | |
84 | ||
cea15768 EJ |
85 | static int |
86 | set_dscp(int fd, uint8_t dscp) | |
87 | { | |
88 | if (dscp > 63) { | |
89 | return EINVAL; | |
90 | } | |
91 | ||
92 | dscp = dscp << 2; | |
93 | if (setsockopt(fd, IPPROTO_IP, IP_TOS, &dscp, sizeof dscp)) { | |
94 | return errno; | |
95 | } | |
96 | ||
97 | return 0; | |
98 | } | |
99 | ||
0ec6cfb1 BP |
100 | static bool |
101 | rlim_is_finite(rlim_t limit) | |
102 | { | |
103 | if (limit == RLIM_INFINITY) { | |
104 | return false; | |
105 | } | |
106 | ||
107 | #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */ | |
108 | if (limit == RLIM_SAVED_CUR) { | |
109 | return false; | |
110 | } | |
111 | #endif | |
112 | ||
113 | #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */ | |
114 | if (limit == RLIM_SAVED_MAX) { | |
115 | return false; | |
116 | } | |
117 | #endif | |
118 | ||
119 | return true; | |
120 | } | |
121 | ||
064af421 BP |
122 | /* Returns the maximum valid FD value, plus 1. */ |
123 | int | |
124 | get_max_fds(void) | |
125 | { | |
126 | static int max_fds = -1; | |
127 | if (max_fds < 0) { | |
128 | struct rlimit r; | |
0ec6cfb1 | 129 | if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) { |
064af421 BP |
130 | max_fds = r.rlim_cur; |
131 | } else { | |
132 | VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); | |
133 | max_fds = 1024; | |
134 | } | |
135 | } | |
136 | return max_fds; | |
137 | } | |
138 | ||
2b35e147 BP |
139 | /* Translates 'host_name', which must be a string representation of an IP |
140 | * address, into a numeric IP address in '*addr'. Returns 0 if successful, | |
141 | * otherwise a positive errno value. */ | |
064af421 | 142 | int |
d295e8e9 | 143 | lookup_ip(const char *host_name, struct in_addr *addr) |
064af421 BP |
144 | { |
145 | if (!inet_aton(host_name, addr)) { | |
db5ce514 | 146 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2b35e147 BP |
147 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name); |
148 | return ENOENT; | |
064af421 BP |
149 | } |
150 | return 0; | |
151 | } | |
152 | ||
d31f1109 JP |
153 | /* Translates 'host_name', which must be a string representation of an IPv6 |
154 | * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, | |
155 | * otherwise a positive errno value. */ | |
156 | int | |
157 | lookup_ipv6(const char *host_name, struct in6_addr *addr) | |
158 | { | |
159 | if (inet_pton(AF_INET6, host_name, addr) != 1) { | |
160 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
161 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name); | |
162 | return ENOENT; | |
163 | } | |
164 | return 0; | |
165 | } | |
166 | ||
13f2ef97 BP |
167 | /* Translates 'host_name', which must be a host name or a string representation |
168 | * of an IP address, into a numeric IP address in '*addr'. Returns 0 if | |
169 | * successful, otherwise a positive errno value. | |
170 | * | |
171 | * Most Open vSwitch code should not use this because it causes deadlocks: | |
172 | * gethostbyname() sends out a DNS request but that starts a new flow for which | |
173 | * OVS must set up a flow, but it can't because it's waiting for a DNS reply. | |
174 | * The synchronous lookup also delays other activty. (Of course we can solve | |
175 | * this but it doesn't seem worthwhile quite yet.) */ | |
176 | int | |
177 | lookup_hostname(const char *host_name, struct in_addr *addr) | |
178 | { | |
179 | struct hostent *h; | |
180 | ||
181 | if (inet_aton(host_name, addr)) { | |
182 | return 0; | |
183 | } | |
184 | ||
185 | h = gethostbyname(host_name); | |
186 | if (h) { | |
187 | *addr = *(struct in_addr *) h->h_addr; | |
188 | return 0; | |
189 | } | |
190 | ||
191 | return (h_errno == HOST_NOT_FOUND ? ENOENT | |
192 | : h_errno == TRY_AGAIN ? EAGAIN | |
193 | : h_errno == NO_RECOVERY ? EIO | |
194 | : h_errno == NO_ADDRESS ? ENXIO | |
195 | : EINVAL); | |
196 | } | |
197 | ||
064af421 BP |
198 | /* Returns the error condition associated with socket 'fd' and resets the |
199 | * socket's error status. */ | |
200 | int | |
d295e8e9 | 201 | get_socket_error(int fd) |
064af421 BP |
202 | { |
203 | int error; | |
c1c19657 BP |
204 | |
205 | if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) { | |
064af421 | 206 | error = errno; |
064af421 BP |
207 | } |
208 | return error; | |
209 | } | |
210 | ||
211 | int | |
d295e8e9 | 212 | check_connection_completion(int fd) |
064af421 BP |
213 | { |
214 | struct pollfd pfd; | |
215 | int retval; | |
216 | ||
217 | pfd.fd = fd; | |
218 | pfd.events = POLLOUT; | |
219 | do { | |
220 | retval = poll(&pfd, 1, 0); | |
221 | } while (retval < 0 && errno == EINTR); | |
222 | if (retval == 1) { | |
223 | return get_socket_error(fd); | |
224 | } else if (retval < 0) { | |
225 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); | |
226 | VLOG_ERR_RL(&rl, "poll: %s", strerror(errno)); | |
227 | return errno; | |
228 | } else { | |
229 | return EAGAIN; | |
230 | } | |
231 | } | |
232 | ||
233 | /* Drain all the data currently in the receive queue of a datagram socket (and | |
234 | * possibly additional data). There is no way to know how many packets are in | |
235 | * the receive queue, but we do know that the total number of bytes queued does | |
236 | * not exceed the receive buffer size, so we pull packets until none are left | |
237 | * or we've read that many bytes. */ | |
238 | int | |
239 | drain_rcvbuf(int fd) | |
240 | { | |
c1c19657 | 241 | int rcvbuf; |
064af421 | 242 | |
c1c19657 BP |
243 | rcvbuf = get_socket_rcvbuf(fd); |
244 | if (rcvbuf < 0) { | |
245 | return -rcvbuf; | |
064af421 | 246 | } |
c1c19657 | 247 | |
064af421 BP |
248 | while (rcvbuf > 0) { |
249 | /* In Linux, specifying MSG_TRUNC in the flags argument causes the | |
250 | * datagram length to be returned, even if that is longer than the | |
251 | * buffer provided. Thus, we can use a 1-byte buffer to discard the | |
252 | * incoming datagram and still be able to account how many bytes were | |
253 | * removed from the receive buffer. | |
254 | * | |
255 | * On other Unix-like OSes, MSG_TRUNC has no effect in the flags | |
256 | * argument. */ | |
e61c056c | 257 | char buffer[LINUX ? 1 : 2048]; |
064af421 BP |
258 | ssize_t n_bytes = recv(fd, buffer, sizeof buffer, |
259 | MSG_TRUNC | MSG_DONTWAIT); | |
260 | if (n_bytes <= 0 || n_bytes >= rcvbuf) { | |
261 | break; | |
262 | } | |
263 | rcvbuf -= n_bytes; | |
264 | } | |
265 | return 0; | |
266 | } | |
267 | ||
c1c19657 BP |
268 | /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a |
269 | * negative errno value if an error occurs. */ | |
270 | int | |
271 | get_socket_rcvbuf(int sock) | |
272 | { | |
273 | int rcvbuf; | |
274 | int error; | |
275 | ||
276 | error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf); | |
277 | return error ? -error : rcvbuf; | |
278 | } | |
279 | ||
064af421 BP |
280 | /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no |
281 | * more data can be immediately read. ('fd' should therefore be in | |
282 | * non-blocking mode.)*/ | |
283 | void | |
284 | drain_fd(int fd, size_t n_packets) | |
285 | { | |
286 | for (; n_packets > 0; n_packets--) { | |
287 | /* 'buffer' only needs to be 1 byte long in most circumstances. This | |
288 | * size is defensive against the possibility that we someday want to | |
289 | * use a Linux tap device without TUN_NO_PI, in which case a buffer | |
290 | * smaller than sizeof(struct tun_pi) will give EINVAL on read. */ | |
291 | char buffer[128]; | |
292 | if (read(fd, buffer, sizeof buffer) <= 0) { | |
293 | break; | |
294 | } | |
295 | } | |
296 | } | |
297 | ||
298 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in | |
299 | * '*un_len' the size of the sockaddr_un. */ | |
300 | static void | |
6e170b4c | 301 | make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len) |
064af421 BP |
302 | { |
303 | un->sun_family = AF_UNIX; | |
71d7c22f | 304 | ovs_strzcpy(un->sun_path, name, sizeof un->sun_path); |
064af421 BP |
305 | *un_len = (offsetof(struct sockaddr_un, sun_path) |
306 | + strlen (un->sun_path) + 1); | |
307 | } | |
308 | ||
6e170b4c BP |
309 | /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in |
310 | * '*un_len' the size of the sockaddr_un. | |
311 | * | |
312 | * Returns 0 on success, otherwise a positive errno value. On success, | |
313 | * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller | |
314 | * should close after using '*un' to bind or connect. On failure, '*dirfdp' is | |
315 | * -1. */ | |
316 | static int | |
317 | make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, | |
318 | int *dirfdp) | |
319 | { | |
320 | enum { MAX_UN_LEN = sizeof un->sun_path - 1 }; | |
321 | ||
322 | *dirfdp = -1; | |
323 | if (strlen(name) > MAX_UN_LEN) { | |
324 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
325 | ||
326 | if (LINUX) { | |
327 | /* 'name' is too long to fit in a sockaddr_un, but we have a | |
328 | * workaround for that on Linux: shorten it by opening a file | |
329 | * descriptor for the directory part of the name and indirecting | |
330 | * through /proc/self/fd/<dirfd>/<basename>. */ | |
331 | char *dir, *base; | |
332 | char *short_name; | |
333 | int dirfd; | |
334 | ||
335 | dir = dir_name(name); | |
336 | base = base_name(name); | |
337 | ||
338 | dirfd = open(dir, O_DIRECTORY | O_RDONLY); | |
339 | if (dirfd < 0) { | |
08871bf6 JP |
340 | free(base); |
341 | free(dir); | |
6e170b4c BP |
342 | return errno; |
343 | } | |
344 | ||
345 | short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base); | |
346 | free(dir); | |
347 | free(base); | |
348 | ||
349 | if (strlen(short_name) <= MAX_UN_LEN) { | |
350 | make_sockaddr_un__(short_name, un, un_len); | |
351 | free(short_name); | |
352 | *dirfdp = dirfd; | |
353 | return 0; | |
354 | } | |
355 | free(short_name); | |
356 | close(dirfd); | |
357 | ||
358 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
359 | "%d bytes (even shortened)", name, MAX_UN_LEN); | |
360 | } else { | |
361 | /* 'name' is too long and we have no workaround. */ | |
362 | VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " | |
363 | "%d bytes", name, MAX_UN_LEN); | |
364 | } | |
365 | ||
366 | return ENAMETOOLONG; | |
367 | } else { | |
368 | make_sockaddr_un__(name, un, un_len); | |
369 | return 0; | |
370 | } | |
371 | } | |
372 | ||
e2bd0593 BP |
373 | /* Binds Unix domain socket 'fd' to a file with permissions 0700. */ |
374 | static int | |
375 | bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len) | |
376 | { | |
e2bd0593 BP |
377 | /* According to _Unix Network Programming_, umask should affect bind(). */ |
378 | mode_t old_umask = umask(0077); | |
379 | int error = bind(fd, sun, sun_len) ? errno : 0; | |
380 | umask(old_umask); | |
381 | return error; | |
e2bd0593 BP |
382 | } |
383 | ||
064af421 BP |
384 | /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or |
385 | * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and | |
386 | * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock' | |
387 | * is true, the socket is made non-blocking. If 'passcred' is true, the socket | |
388 | * is configured to receive SCM_CREDENTIALS control messages. | |
389 | * | |
390 | * Returns the socket's fd if successful, otherwise a negative errno value. */ | |
391 | int | |
67a4917b | 392 | make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, |
064af421 BP |
393 | const char *bind_path, const char *connect_path) |
394 | { | |
395 | int error; | |
396 | int fd; | |
397 | ||
398 | fd = socket(PF_UNIX, style, 0); | |
399 | if (fd < 0) { | |
400 | return -errno; | |
401 | } | |
402 | ||
403 | /* Set nonblocking mode right away, if we want it. This prevents blocking | |
404 | * in connect(), if connect_path != NULL. (In turn, that's a corner case: | |
405 | * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only | |
406 | * if a backlog of un-accepted connections has built up in the kernel.) */ | |
407 | if (nonblock) { | |
408 | int flags = fcntl(fd, F_GETFL, 0); | |
409 | if (flags == -1) { | |
6e170b4c | 410 | error = errno; |
064af421 BP |
411 | goto error; |
412 | } | |
413 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { | |
6e170b4c | 414 | error = errno; |
064af421 BP |
415 | goto error; |
416 | } | |
417 | } | |
418 | ||
419 | if (bind_path) { | |
420 | struct sockaddr_un un; | |
421 | socklen_t un_len; | |
6e170b4c BP |
422 | int dirfd; |
423 | ||
424 | if (unlink(bind_path) && errno != ENOENT) { | |
425 | VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); | |
064af421 BP |
426 | } |
427 | fatal_signal_add_file_to_unlink(bind_path); | |
6e170b4c BP |
428 | |
429 | error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd); | |
e2bd0593 BP |
430 | if (!error) { |
431 | error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len); | |
6e170b4c BP |
432 | } |
433 | if (dirfd >= 0) { | |
434 | close(dirfd); | |
435 | } | |
436 | if (error) { | |
064af421 BP |
437 | goto error; |
438 | } | |
439 | } | |
440 | ||
441 | if (connect_path) { | |
442 | struct sockaddr_un un; | |
443 | socklen_t un_len; | |
6e170b4c BP |
444 | int dirfd; |
445 | ||
446 | error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd); | |
447 | if (!error | |
448 | && connect(fd, (struct sockaddr*) &un, un_len) | |
064af421 | 449 | && errno != EINPROGRESS) { |
6e170b4c BP |
450 | error = errno; |
451 | } | |
452 | if (dirfd >= 0) { | |
453 | close(dirfd); | |
454 | } | |
455 | if (error) { | |
064af421 BP |
456 | goto error; |
457 | } | |
458 | } | |
459 | ||
460 | #ifdef SCM_CREDENTIALS | |
461 | if (passcred) { | |
462 | int enable = 1; | |
463 | if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) { | |
6e170b4c | 464 | error = errno; |
064af421 BP |
465 | goto error; |
466 | } | |
467 | } | |
468 | #endif | |
469 | ||
470 | return fd; | |
471 | ||
472 | error: | |
6e170b4c BP |
473 | if (error == EAGAIN) { |
474 | error = EPROTO; | |
475 | } | |
064af421 | 476 | if (bind_path) { |
e36d0b53 | 477 | fatal_signal_unlink_file_now(bind_path); |
064af421 | 478 | } |
064af421 BP |
479 | close(fd); |
480 | return -error; | |
481 | } | |
482 | ||
483 | int | |
484 | get_unix_name_len(socklen_t sun_len) | |
485 | { | |
486 | return (sun_len >= offsetof(struct sockaddr_un, sun_path) | |
487 | ? sun_len - offsetof(struct sockaddr_un, sun_path) | |
488 | : 0); | |
489 | } | |
490 | ||
dbba996b BP |
491 | ovs_be32 |
492 | guess_netmask(ovs_be32 ip_) | |
064af421 | 493 | { |
dbba996b | 494 | uint32_t ip = ntohl(ip_); |
064af421 BP |
495 | return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ |
496 | : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ | |
497 | : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ | |
498 | : htonl(0)); /* ??? */ | |
499 | } | |
500 | ||
52f8a75e BP |
501 | /* Parses 'target', which should be a string in the format "<host>[:<port>]". |
502 | * <host> is required. If 'default_port' is nonzero then <port> is optional | |
503 | * and defaults to 'default_port'. | |
4f2eb9a7 | 504 | * |
52f8a75e BP |
505 | * On success, returns true and stores the parsed remote address into '*sinp'. |
506 | * On failure, logs an error, stores zeros into '*sinp', and returns false. */ | |
507 | bool | |
508 | inet_parse_active(const char *target_, uint16_t default_port, | |
509 | struct sockaddr_in *sinp) | |
78ff0270 BP |
510 | { |
511 | char *target = xstrdup(target_); | |
512 | char *save_ptr = NULL; | |
513 | const char *host_name; | |
514 | const char *port_string; | |
52f8a75e | 515 | bool ok = false; |
78ff0270 BP |
516 | |
517 | /* Defaults. */ | |
52f8a75e BP |
518 | sinp->sin_family = AF_INET; |
519 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
520 | |
521 | /* Tokenize. */ | |
522 | host_name = strtok_r(target, ":", &save_ptr); | |
523 | port_string = strtok_r(NULL, ":", &save_ptr); | |
524 | if (!host_name) { | |
52f8a75e | 525 | VLOG_ERR("%s: bad peer name format", target_); |
78ff0270 BP |
526 | goto exit; |
527 | } | |
528 | ||
529 | /* Look up IP, port. */ | |
52f8a75e | 530 | if (lookup_ip(host_name, &sinp->sin_addr)) { |
78ff0270 BP |
531 | goto exit; |
532 | } | |
533 | if (port_string && atoi(port_string)) { | |
52f8a75e | 534 | sinp->sin_port = htons(atoi(port_string)); |
1901968e BP |
535 | } else if (!default_port) { |
536 | VLOG_ERR("%s: port number must be specified", target_); | |
52f8a75e BP |
537 | goto exit; |
538 | } | |
539 | ||
540 | ok = true; | |
541 | ||
542 | exit: | |
543 | if (!ok) { | |
544 | memset(sinp, 0, sizeof *sinp); | |
545 | } | |
546 | free(target); | |
547 | return ok; | |
548 | } | |
549 | ||
550 | /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to | |
551 | * 'target', which should be a string in the format "<host>[:<port>]". <host> | |
552 | * is required. If 'default_port' is nonzero then <port> is optional and | |
553 | * defaults to 'default_port'. | |
554 | * | |
555 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
556 | * | |
557 | * On success, returns 0 (indicating connection complete) or EAGAIN (indicating | |
558 | * connection in progress), in which case the new file descriptor is stored | |
559 | * into '*fdp'. On failure, returns a positive errno value other than EAGAIN | |
560 | * and stores -1 into '*fdp'. | |
561 | * | |
562 | * If 'sinp' is non-null, then on success the target address is stored into | |
f125905c MM |
563 | * '*sinp'. |
564 | * | |
cea15768 EJ |
565 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
566 | * should be in the range [0, 63] and will automatically be shifted to the | |
567 | * appropriately place in the IP tos field. */ | |
52f8a75e BP |
568 | int |
569 | inet_open_active(int style, const char *target, uint16_t default_port, | |
f125905c | 570 | struct sockaddr_in *sinp, int *fdp, uint8_t dscp) |
52f8a75e BP |
571 | { |
572 | struct sockaddr_in sin; | |
573 | int fd = -1; | |
574 | int error; | |
575 | ||
576 | /* Parse. */ | |
577 | if (!inet_parse_active(target, default_port, &sin)) { | |
1901968e BP |
578 | error = EAFNOSUPPORT; |
579 | goto exit; | |
78ff0270 BP |
580 | } |
581 | ||
582 | /* Create non-blocking socket. */ | |
4f2eb9a7 | 583 | fd = socket(AF_INET, style, 0); |
78ff0270 | 584 | if (fd < 0) { |
52f8a75e | 585 | VLOG_ERR("%s: socket: %s", target, strerror(errno)); |
78ff0270 BP |
586 | error = errno; |
587 | goto exit; | |
588 | } | |
589 | error = set_nonblocking(fd); | |
590 | if (error) { | |
a4efa3fc | 591 | goto exit; |
78ff0270 BP |
592 | } |
593 | ||
cea15768 EJ |
594 | /* The dscp bits must be configured before connect() to ensure that the TOS |
595 | * field is set during the connection establishment. If set after | |
596 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
597 | error = set_dscp(fd, dscp); | |
598 | if (error) { | |
599 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 600 | goto exit; |
f125905c MM |
601 | } |
602 | ||
78ff0270 BP |
603 | /* Connect. */ |
604 | error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno; | |
605 | if (error == EINPROGRESS) { | |
606 | error = EAGAIN; | |
78ff0270 BP |
607 | } |
608 | ||
78ff0270 BP |
609 | exit: |
610 | if (!error || error == EAGAIN) { | |
611 | if (sinp) { | |
612 | *sinp = sin; | |
613 | } | |
a4efa3fc EJ |
614 | } else if (fd >= 0) { |
615 | close(fd); | |
78ff0270 | 616 | } |
a4efa3fc | 617 | *fdp = fd; |
78ff0270 BP |
618 | return error; |
619 | } | |
620 | ||
d98fa503 | 621 | /* Parses 'target', which should be a string in the format "[<port>][:<ip>]": |
e1bd3bee BP |
622 | * |
623 | * - If 'default_port' is -1, then <port> is required. Otherwise, if | |
624 | * <port> is omitted, then 'default_port' is used instead. | |
625 | * | |
626 | * - If <port> (or 'default_port', if used) is 0, then no port is bound | |
627 | * and the TCP/IP stack will select a port. | |
628 | * | |
629 | * - If <ip> is omitted then the IP address is wildcarded. | |
4f2eb9a7 | 630 | * |
d98fa503 BP |
631 | * If successful, stores the address into '*sinp' and returns true; otherwise |
632 | * zeros '*sinp' and returns false. */ | |
633 | bool | |
995337c4 | 634 | inet_parse_passive(const char *target_, int default_port, |
d98fa503 | 635 | struct sockaddr_in *sinp) |
78ff0270 BP |
636 | { |
637 | char *target = xstrdup(target_); | |
638 | char *string_ptr = target; | |
78ff0270 BP |
639 | const char *host_name; |
640 | const char *port_string; | |
d98fa503 BP |
641 | bool ok = false; |
642 | int port; | |
78ff0270 BP |
643 | |
644 | /* Address defaults. */ | |
d98fa503 BP |
645 | memset(sinp, 0, sizeof *sinp); |
646 | sinp->sin_family = AF_INET; | |
647 | sinp->sin_addr.s_addr = htonl(INADDR_ANY); | |
648 | sinp->sin_port = htons(default_port); | |
78ff0270 BP |
649 | |
650 | /* Parse optional port number. */ | |
651 | port_string = strsep(&string_ptr, ":"); | |
e1bd3bee | 652 | if (port_string && str_to_int(port_string, 10, &port)) { |
d98fa503 | 653 | sinp->sin_port = htons(port); |
e1bd3bee | 654 | } else if (default_port < 0) { |
1901968e | 655 | VLOG_ERR("%s: port number must be specified", target_); |
1901968e | 656 | goto exit; |
78ff0270 BP |
657 | } |
658 | ||
659 | /* Parse optional bind IP. */ | |
660 | host_name = strsep(&string_ptr, ":"); | |
d98fa503 BP |
661 | if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { |
662 | goto exit; | |
663 | } | |
664 | ||
665 | ok = true; | |
666 | ||
667 | exit: | |
668 | if (!ok) { | |
669 | memset(sinp, 0, sizeof *sinp); | |
670 | } | |
671 | free(target); | |
672 | return ok; | |
673 | } | |
674 | ||
675 | ||
676 | /* Opens a non-blocking IPv4 socket of the specified 'style', binds to | |
677 | * 'target', and listens for incoming connections. Parses 'target' in the same | |
678 | * way was inet_parse_passive(). | |
679 | * | |
680 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
681 | * | |
682 | * For TCP, the socket will have SO_REUSEADDR turned on. | |
683 | * | |
684 | * On success, returns a non-negative file descriptor. On failure, returns a | |
685 | * negative errno value. | |
686 | * | |
687 | * If 'sinp' is non-null, then on success the bound address is stored into | |
f125905c MM |
688 | * '*sinp'. |
689 | * | |
cea15768 EJ |
690 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
691 | * should be in the range [0, 63] and will automatically be shifted to the | |
692 | * appropriately place in the IP tos field. */ | |
d98fa503 BP |
693 | int |
694 | inet_open_passive(int style, const char *target, int default_port, | |
f125905c | 695 | struct sockaddr_in *sinp, uint8_t dscp) |
d98fa503 BP |
696 | { |
697 | struct sockaddr_in sin; | |
698 | int fd = 0, error; | |
699 | unsigned int yes = 1; | |
700 | ||
701 | if (!inet_parse_passive(target, default_port, &sin)) { | |
ca286ba9 | 702 | return -EAFNOSUPPORT; |
78ff0270 BP |
703 | } |
704 | ||
705 | /* Create non-blocking socket, set SO_REUSEADDR. */ | |
4f2eb9a7 | 706 | fd = socket(AF_INET, style, 0); |
78ff0270 BP |
707 | if (fd < 0) { |
708 | error = errno; | |
d98fa503 | 709 | VLOG_ERR("%s: socket: %s", target, strerror(error)); |
ca286ba9 | 710 | return -error; |
78ff0270 BP |
711 | } |
712 | error = set_nonblocking(fd); | |
713 | if (error) { | |
d98fa503 | 714 | goto error; |
78ff0270 | 715 | } |
4f2eb9a7 BP |
716 | if (style == SOCK_STREAM |
717 | && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { | |
78ff0270 | 718 | error = errno; |
d98fa503 BP |
719 | VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error)); |
720 | goto error; | |
78ff0270 BP |
721 | } |
722 | ||
723 | /* Bind. */ | |
724 | if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) { | |
725 | error = errno; | |
d98fa503 BP |
726 | VLOG_ERR("%s: bind: %s", target, strerror(error)); |
727 | goto error; | |
78ff0270 BP |
728 | } |
729 | ||
cea15768 EJ |
730 | /* The dscp bits must be configured before connect() to ensure that the TOS |
731 | * field is set during the connection establishment. If set after | |
732 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
733 | error = set_dscp(fd, dscp); | |
734 | if (error) { | |
735 | VLOG_ERR("%s: socket: %s", target, strerror(error)); | |
ef8a3d14 | 736 | goto error; |
f125905c MM |
737 | } |
738 | ||
78ff0270 | 739 | /* Listen. */ |
041dc07f | 740 | if (style == SOCK_STREAM && listen(fd, 10) < 0) { |
78ff0270 | 741 | error = errno; |
d98fa503 BP |
742 | VLOG_ERR("%s: listen: %s", target, strerror(error)); |
743 | goto error; | |
78ff0270 | 744 | } |
36775dad BP |
745 | |
746 | if (sinp) { | |
747 | socklen_t sin_len = sizeof sin; | |
748 | if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){ | |
749 | error = errno; | |
d98fa503 BP |
750 | VLOG_ERR("%s: getsockname: %s", target, strerror(error)); |
751 | goto error; | |
36775dad BP |
752 | } |
753 | if (sin.sin_family != AF_INET || sin_len != sizeof sin) { | |
ca286ba9 | 754 | error = EAFNOSUPPORT; |
d98fa503 BP |
755 | VLOG_ERR("%s: getsockname: invalid socket name", target); |
756 | goto error; | |
36775dad BP |
757 | } |
758 | *sinp = sin; | |
759 | } | |
760 | ||
d98fa503 | 761 | return fd; |
78ff0270 | 762 | |
d98fa503 | 763 | error: |
78ff0270 | 764 | close(fd); |
ca286ba9 | 765 | return -error; |
78ff0270 BP |
766 | } |
767 | ||
a8b5f8b4 BP |
768 | /* Returns a readable and writable fd for /dev/null, if successful, otherwise |
769 | * a negative errno value. The caller must not close the returned fd (because | |
770 | * the same fd will be handed out to subsequent callers). */ | |
771 | int | |
772 | get_null_fd(void) | |
773 | { | |
774 | static int null_fd = -1; | |
775 | if (null_fd < 0) { | |
776 | null_fd = open("/dev/null", O_RDWR); | |
777 | if (null_fd < 0) { | |
778 | int error = errno; | |
779 | VLOG_ERR("could not open /dev/null: %s", strerror(error)); | |
780 | return -error; | |
781 | } | |
782 | } | |
783 | return null_fd; | |
784 | } | |
785 | ||
064af421 BP |
786 | int |
787 | read_fully(int fd, void *p_, size_t size, size_t *bytes_read) | |
788 | { | |
789 | uint8_t *p = p_; | |
790 | ||
791 | *bytes_read = 0; | |
792 | while (size > 0) { | |
793 | ssize_t retval = read(fd, p, size); | |
794 | if (retval > 0) { | |
795 | *bytes_read += retval; | |
796 | size -= retval; | |
797 | p += retval; | |
798 | } else if (retval == 0) { | |
799 | return EOF; | |
800 | } else if (errno != EINTR) { | |
801 | return errno; | |
802 | } | |
803 | } | |
804 | return 0; | |
805 | } | |
806 | ||
807 | int | |
808 | write_fully(int fd, const void *p_, size_t size, size_t *bytes_written) | |
809 | { | |
810 | const uint8_t *p = p_; | |
811 | ||
812 | *bytes_written = 0; | |
813 | while (size > 0) { | |
814 | ssize_t retval = write(fd, p, size); | |
815 | if (retval > 0) { | |
816 | *bytes_written += retval; | |
817 | size -= retval; | |
818 | p += retval; | |
819 | } else if (retval == 0) { | |
820 | VLOG_WARN("write returned 0"); | |
821 | return EPROTO; | |
822 | } else if (errno != EINTR) { | |
823 | return errno; | |
824 | } | |
825 | } | |
826 | return 0; | |
827 | } | |
8e71cf88 BP |
828 | |
829 | /* Given file name 'file_name', fsyncs the directory in which it is contained. | |
830 | * Returns 0 if successful, otherwise a positive errno value. */ | |
831 | int | |
832 | fsync_parent_dir(const char *file_name) | |
833 | { | |
834 | int error = 0; | |
835 | char *dir; | |
836 | int fd; | |
837 | ||
838 | dir = dir_name(file_name); | |
839 | fd = open(dir, O_RDONLY); | |
840 | if (fd >= 0) { | |
841 | if (fsync(fd)) { | |
842 | if (errno == EINVAL || errno == EROFS) { | |
843 | /* This directory does not support synchronization. Not | |
844 | * really an error. */ | |
845 | } else { | |
846 | error = errno; | |
847 | VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); | |
848 | } | |
849 | } | |
850 | close(fd); | |
851 | } else { | |
852 | error = errno; | |
853 | VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); | |
854 | } | |
855 | free(dir); | |
856 | ||
857 | return error; | |
858 | } | |
26efd256 BP |
859 | |
860 | /* Obtains the modification time of the file named 'file_name' to the greatest | |
861 | * supported precision. If successful, stores the mtime in '*mtime' and | |
862 | * returns 0. On error, returns a positive errno value and stores zeros in | |
863 | * '*mtime'. */ | |
864 | int | |
865 | get_mtime(const char *file_name, struct timespec *mtime) | |
866 | { | |
867 | struct stat s; | |
868 | ||
869 | if (!stat(file_name, &s)) { | |
870 | mtime->tv_sec = s.st_mtime; | |
871 | ||
872 | #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC | |
873 | mtime->tv_nsec = s.st_mtim.tv_nsec; | |
874 | #elif HAVE_STRUCT_STAT_ST_MTIMENSEC | |
875 | mtime->tv_nsec = s.st_mtimensec; | |
876 | #else | |
877 | mtime->tv_nsec = 0; | |
878 | #endif | |
879 | ||
880 | return 0; | |
881 | } else { | |
882 | mtime->tv_sec = mtime->tv_nsec = 0; | |
883 | return errno; | |
884 | } | |
885 | } | |
886 | ||
279c9e03 BP |
887 | void |
888 | xpipe(int fds[2]) | |
889 | { | |
890 | if (pipe(fds)) { | |
891 | VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); | |
892 | } | |
893 | } | |
f89ffb0e BP |
894 | |
895 | static int | |
c1c19657 | 896 | getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) |
f89ffb0e | 897 | { |
c1c19657 BP |
898 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
899 | socklen_t len; | |
900 | int value; | |
901 | int error; | |
f89ffb0e | 902 | |
c1c19657 BP |
903 | len = sizeof value; |
904 | if (getsockopt(fd, level, option, &value, &len)) { | |
905 | error = errno; | |
906 | VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error)); | |
907 | } else if (len != sizeof value) { | |
908 | error = EINVAL; | |
909 | VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)", | |
910 | optname, (unsigned int) len, sizeof value); | |
911 | } else { | |
912 | error = 0; | |
913 | } | |
914 | ||
915 | *valuep = error ? 0 : value; | |
916 | return error; | |
f89ffb0e BP |
917 | } |
918 | ||
919 | static void | |
920 | describe_sockaddr(struct ds *string, int fd, | |
921 | int (*getaddr)(int, struct sockaddr *, socklen_t *)) | |
922 | { | |
923 | struct sockaddr_storage ss; | |
924 | socklen_t len = sizeof ss; | |
925 | ||
926 | if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { | |
927 | if (ss.ss_family == AF_INET) { | |
928 | struct sockaddr_in sin; | |
929 | ||
930 | memcpy(&sin, &ss, sizeof sin); | |
931 | ds_put_format(string, IP_FMT":%"PRIu16, | |
932 | IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port)); | |
933 | } else if (ss.ss_family == AF_UNIX) { | |
934 | struct sockaddr_un sun; | |
935 | const char *null; | |
936 | size_t maxlen; | |
937 | ||
938 | memcpy(&sun, &ss, sizeof sun); | |
939 | maxlen = len - offsetof(struct sockaddr_un, sun_path); | |
940 | null = memchr(sun.sun_path, '\0', maxlen); | |
941 | ds_put_buffer(string, sun.sun_path, | |
942 | null ? null - sun.sun_path : maxlen); | |
943 | } | |
944 | #ifdef HAVE_NETLINK | |
945 | else if (ss.ss_family == AF_NETLINK) { | |
946 | int protocol; | |
947 | ||
948 | /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version | |
949 | * of the Linux kernel headers in use at build time. */ | |
950 | #ifndef SO_PROTOCOL | |
951 | #define SO_PROTOCOL 38 | |
952 | #endif | |
953 | ||
c1c19657 BP |
954 | if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", |
955 | &protocol)) { | |
f89ffb0e BP |
956 | switch (protocol) { |
957 | case NETLINK_ROUTE: | |
958 | ds_put_cstr(string, "NETLINK_ROUTE"); | |
959 | break; | |
960 | ||
961 | case NETLINK_GENERIC: | |
962 | ds_put_cstr(string, "NETLINK_GENERIC"); | |
963 | break; | |
964 | ||
965 | default: | |
966 | ds_put_format(string, "AF_NETLINK family %d", protocol); | |
967 | break; | |
968 | } | |
969 | } else { | |
970 | ds_put_cstr(string, "AF_NETLINK"); | |
971 | } | |
972 | } | |
973 | #endif | |
974 | #if AF_PACKET && __linux__ | |
975 | else if (ss.ss_family == AF_PACKET) { | |
976 | struct sockaddr_ll sll; | |
977 | ||
978 | memcpy(&sll, &ss, sizeof sll); | |
979 | ds_put_cstr(string, "AF_PACKET"); | |
980 | if (sll.sll_ifindex) { | |
981 | char name[IFNAMSIZ]; | |
982 | ||
983 | if (if_indextoname(sll.sll_ifindex, name)) { | |
984 | ds_put_format(string, "(%s)", name); | |
985 | } else { | |
986 | ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); | |
987 | } | |
988 | } | |
989 | if (sll.sll_protocol) { | |
990 | ds_put_format(string, "(protocol=0x%"PRIu16")", | |
991 | ntohs(sll.sll_protocol)); | |
992 | } | |
993 | } | |
994 | #endif | |
995 | else if (ss.ss_family == AF_UNSPEC) { | |
996 | ds_put_cstr(string, "AF_UNSPEC"); | |
997 | } else { | |
998 | ds_put_format(string, "AF_%d", (int) ss.ss_family); | |
999 | } | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | ||
1004 | #ifdef __linux__ | |
1005 | static void | |
1006 | put_fd_filename(struct ds *string, int fd) | |
1007 | { | |
1008 | char buf[1024]; | |
1009 | char *linkname; | |
1010 | int n; | |
1011 | ||
1012 | linkname = xasprintf("/proc/self/fd/%d", fd); | |
1013 | n = readlink(linkname, buf, sizeof buf); | |
1014 | if (n > 0) { | |
1015 | ds_put_char(string, ' '); | |
1016 | ds_put_buffer(string, buf, n); | |
1017 | if (n > sizeof buf) { | |
1018 | ds_put_cstr(string, "..."); | |
1019 | } | |
1020 | } | |
1021 | free(linkname); | |
1022 | } | |
1023 | #endif | |
1024 | ||
1025 | /* Returns a malloc()'d string describing 'fd', for use in logging. */ | |
1026 | char * | |
1027 | describe_fd(int fd) | |
1028 | { | |
1029 | struct ds string; | |
1030 | struct stat s; | |
1031 | ||
1032 | ds_init(&string); | |
1033 | if (fstat(fd, &s)) { | |
1034 | ds_put_format(&string, "fstat failed (%s)", strerror(errno)); | |
1035 | } else if (S_ISSOCK(s.st_mode)) { | |
1036 | describe_sockaddr(&string, fd, getsockname); | |
1037 | ds_put_cstr(&string, "<->"); | |
1038 | describe_sockaddr(&string, fd, getpeername); | |
1039 | } else { | |
1040 | ds_put_cstr(&string, (isatty(fd) ? "tty" | |
1041 | : S_ISDIR(s.st_mode) ? "directory" | |
1042 | : S_ISCHR(s.st_mode) ? "character device" | |
1043 | : S_ISBLK(s.st_mode) ? "block device" | |
1044 | : S_ISREG(s.st_mode) ? "file" | |
1045 | : S_ISFIFO(s.st_mode) ? "FIFO" | |
1046 | : S_ISLNK(s.st_mode) ? "symbolic link" | |
1047 | : "unknown")); | |
1048 | #ifdef __linux__ | |
1049 | put_fd_filename(&string, fd); | |
1050 | #endif | |
1051 | } | |
1052 | return ds_steal_cstr(&string); | |
1053 | } |