]> git.proxmox.com Git - ovs.git/blob - lib/socket-util.c
Replace inet_aton() with inet_pton().
[ovs.git] / lib / socket-util.c
1 /*
2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
3 *
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:
7 *
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.
15 */
16
17 #include <config.h>
18 #include "socket-util.h"
19 #include <arpa/inet.h>
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <net/if.h>
23 #include <netdb.h>
24 #include <poll.h>
25 #include <stddef.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/ioctl.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/uio.h>
34 #include <sys/un.h>
35 #include <unistd.h>
36 #include "dynamic-string.h"
37 #include "fatal-signal.h"
38 #include "ovs-thread.h"
39 #include "packets.h"
40 #include "poll-loop.h"
41 #include "util.h"
42 #include "vlog.h"
43 #if AF_PACKET && LINUX_DATAPATH
44 #include <linux/if_packet.h>
45 #endif
46 #ifdef HAVE_NETLINK
47 #include "netlink-protocol.h"
48 #include "netlink-socket.h"
49 #endif
50
51 VLOG_DEFINE_THIS_MODULE(socket_util);
52
53 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
54 * Thus, this file compiles all of the code regardless of the target, by
55 * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
56 #ifndef LINUX_DATAPATH
57 #define LINUX_DATAPATH 0
58 #endif
59
60 #ifndef O_DIRECTORY
61 #define O_DIRECTORY 0
62 #endif
63
64 /* Maximum length of the sun_path member in a struct sockaddr_un, excluding
65 * space for a null terminator. */
66 #define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
67
68 static int getsockopt_int(int fd, int level, int option, const char *optname,
69 int *valuep);
70
71 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
72 * positive errno value. */
73 int
74 set_nonblocking(int fd)
75 {
76 #ifndef _WIN32
77 int flags = fcntl(fd, F_GETFL, 0);
78 if (flags != -1) {
79 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
80 return 0;
81 } else {
82 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
83 return errno;
84 }
85 } else {
86 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
87 return errno;
88 }
89 #else
90 unsigned long arg = 1;
91 if (ioctlsocket(fd, FIONBIO, &arg)) {
92 int error = sock_errno();
93 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
94 return error;
95 }
96 return 0;
97 #endif
98 }
99
100 void
101 xset_nonblocking(int fd)
102 {
103 if (set_nonblocking(fd)) {
104 exit(EXIT_FAILURE);
105 }
106 }
107
108 int
109 set_dscp(int fd, uint8_t dscp)
110 {
111 int val;
112
113 if (dscp > 63) {
114 return EINVAL;
115 }
116
117 val = dscp << 2;
118 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
119 return sock_errno();
120 }
121
122 return 0;
123 }
124
125 /* Translates 'host_name', which must be a string representation of an IP
126 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
127 * otherwise a positive errno value. */
128 int
129 lookup_ip(const char *host_name, struct in_addr *addr)
130 {
131 if (!inet_pton(AF_INET, host_name, addr)) {
132 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
133 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
134 return ENOENT;
135 }
136 return 0;
137 }
138
139 /* Translates 'host_name', which must be a string representation of an IPv6
140 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
141 * otherwise a positive errno value. */
142 int
143 lookup_ipv6(const char *host_name, struct in6_addr *addr)
144 {
145 if (inet_pton(AF_INET6, host_name, addr) != 1) {
146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
147 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
148 return ENOENT;
149 }
150 return 0;
151 }
152
153 /* Translates 'host_name', which must be a host name or a string representation
154 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
155 * successful, otherwise a positive errno value.
156 *
157 * Most Open vSwitch code should not use this because it causes deadlocks:
158 * getaddrinfo() sends out a DNS request but that starts a new flow for which
159 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
160 * The synchronous lookup also delays other activity. (Of course we can solve
161 * this but it doesn't seem worthwhile quite yet.) */
162 int
163 lookup_hostname(const char *host_name, struct in_addr *addr)
164 {
165 struct addrinfo *result;
166 struct addrinfo hints;
167
168 if (inet_pton(AF_INET, host_name, addr)) {
169 return 0;
170 }
171
172 memset(&hints, 0, sizeof hints);
173 hints.ai_family = AF_INET;
174
175 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
176 case 0:
177 *addr = ALIGNED_CAST(struct sockaddr_in *,
178 result->ai_addr)->sin_addr;
179 freeaddrinfo(result);
180 return 0;
181
182 #ifdef EAI_ADDRFAMILY
183 case EAI_ADDRFAMILY:
184 #endif
185 case EAI_NONAME:
186 case EAI_SERVICE:
187 return ENOENT;
188
189 case EAI_AGAIN:
190 return EAGAIN;
191
192 case EAI_BADFLAGS:
193 case EAI_FAMILY:
194 case EAI_SOCKTYPE:
195 return EINVAL;
196
197 case EAI_FAIL:
198 return EIO;
199
200 case EAI_MEMORY:
201 return ENOMEM;
202
203 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
204 case EAI_NODATA:
205 return ENXIO;
206 #endif
207
208 #ifdef EAI_SYSTEM
209 case EAI_SYSTEM:
210 return sock_errno();
211 #endif
212
213 default:
214 return EPROTO;
215 }
216 }
217
218 int
219 check_connection_completion(int fd)
220 {
221 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
222 struct pollfd pfd;
223 int retval;
224
225 pfd.fd = fd;
226 pfd.events = POLLOUT;
227 do {
228 retval = poll(&pfd, 1, 0);
229 } while (retval < 0 && errno == EINTR);
230 if (retval == 1) {
231 if (pfd.revents & POLLERR) {
232 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
233 if (n < 0) {
234 return errno;
235 } else {
236 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
237 return EPROTO;
238 }
239 }
240 return 0;
241 } else if (retval < 0) {
242 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
243 return errno;
244 } else {
245 return EAGAIN;
246 }
247 }
248
249 /* Drain all the data currently in the receive queue of a datagram socket (and
250 * possibly additional data). There is no way to know how many packets are in
251 * the receive queue, but we do know that the total number of bytes queued does
252 * not exceed the receive buffer size, so we pull packets until none are left
253 * or we've read that many bytes. */
254 int
255 drain_rcvbuf(int fd)
256 {
257 int rcvbuf;
258
259 rcvbuf = get_socket_rcvbuf(fd);
260 if (rcvbuf < 0) {
261 return -rcvbuf;
262 }
263
264 while (rcvbuf > 0) {
265 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
266 * datagram length to be returned, even if that is longer than the
267 * buffer provided. Thus, we can use a 1-byte buffer to discard the
268 * incoming datagram and still be able to account how many bytes were
269 * removed from the receive buffer.
270 *
271 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
272 * argument. */
273 char buffer[LINUX_DATAPATH ? 1 : 2048];
274 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
275 MSG_TRUNC | MSG_DONTWAIT);
276 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
277 break;
278 }
279 rcvbuf -= n_bytes;
280 }
281 return 0;
282 }
283
284 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
285 * negative errno value if an error occurs. */
286 int
287 get_socket_rcvbuf(int sock)
288 {
289 int rcvbuf;
290 int error;
291
292 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
293 return error ? -error : rcvbuf;
294 }
295
296 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
297 * more data can be immediately read. ('fd' should therefore be in
298 * non-blocking mode.)*/
299 void
300 drain_fd(int fd, size_t n_packets)
301 {
302 for (; n_packets > 0; n_packets--) {
303 /* 'buffer' only needs to be 1 byte long in most circumstances. This
304 * size is defensive against the possibility that we someday want to
305 * use a Linux tap device without TUN_NO_PI, in which case a buffer
306 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
307 char buffer[128];
308 if (read(fd, buffer, sizeof buffer) <= 0) {
309 break;
310 }
311 }
312 }
313
314 #ifndef _WIN32
315 /* Attempts to shorten 'name' by opening a file descriptor for the directory
316 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
317 * On systems with Linux-like /proc, this works as long as <basename> isn't too
318 * long.
319 *
320 * On success, returns 0 and stores the short name in 'short_name' and a
321 * directory file descriptor to eventually be closed in '*dirfpd'. */
322 static int
323 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
324 int *dirfdp)
325 {
326 char *dir, *base;
327 int dirfd;
328 int len;
329
330 if (!LINUX_DATAPATH) {
331 return ENAMETOOLONG;
332 }
333
334 dir = dir_name(name);
335 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
336 if (dirfd < 0) {
337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
338 int error = errno;
339
340 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
341 free(dir);
342
343 return error;
344 }
345 free(dir);
346
347 base = base_name(name);
348 len = snprintf(short_name, MAX_UN_LEN + 1,
349 "/proc/self/fd/%d/%s", dirfd, base);
350 free(base);
351
352 if (len >= 0 && len <= MAX_UN_LEN) {
353 *dirfdp = dirfd;
354 return 0;
355 } else {
356 close(dirfd);
357 return ENAMETOOLONG;
358 }
359 }
360
361 /* Attempts to shorten 'name' by creating a symlink for the directory part of
362 * the name and indirecting through <symlink>/<basename>. This works on
363 * systems that support symlinks, as long as <basename> isn't too long.
364 *
365 * On success, returns 0 and stores the short name in 'short_name' and the
366 * symbolic link to eventually delete in 'linkname'. */
367 static int
368 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
369 char linkname[MAX_UN_LEN + 1])
370 {
371 char *abs, *dir, *base;
372 const char *tmpdir;
373 int error;
374 int i;
375
376 abs = abs_file_name(NULL, name);
377 dir = dir_name(abs);
378 base = base_name(abs);
379 free(abs);
380
381 tmpdir = getenv("TMPDIR");
382 if (tmpdir == NULL) {
383 tmpdir = "/tmp";
384 }
385
386 for (i = 0; i < 1000; i++) {
387 int len;
388
389 len = snprintf(linkname, MAX_UN_LEN + 1,
390 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
391 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
392 : symlink(dir, linkname) ? errno
393 : 0);
394 if (error != EEXIST) {
395 break;
396 }
397 }
398
399 if (!error) {
400 int len;
401
402 fatal_signal_add_file_to_unlink(linkname);
403
404 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
405 if (len < 0 || len > MAX_UN_LEN) {
406 fatal_signal_unlink_file_now(linkname);
407 error = ENAMETOOLONG;
408 }
409 }
410
411 if (error) {
412 linkname[0] = '\0';
413 }
414 free(dir);
415 free(base);
416
417 return error;
418 }
419
420 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
421 * '*un_len' the size of the sockaddr_un.
422 *
423 * Returns 0 on success, otherwise a positive errno value.
424 *
425 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
426 * longer needs to use 'un'. On success, freeing these references with
427 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
428 * unnecessary but harmless. */
429 static int
430 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
431 int *dirfdp, char linkname[MAX_UN_LEN + 1])
432 {
433 char short_name[MAX_UN_LEN + 1];
434
435 *dirfdp = -1;
436 linkname[0] = '\0';
437 if (strlen(name) > MAX_UN_LEN) {
438 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
439 int error = shorten_name_via_proc(name, short_name, dirfdp);
440 if (error == ENAMETOOLONG) {
441 error = shorten_name_via_symlink(name, short_name, linkname);
442 }
443 if (error) {
444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
445
446 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
447 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
448 return error;
449 }
450
451 name = short_name;
452 }
453
454 un->sun_family = AF_UNIX;
455 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
456 *un_len = (offsetof(struct sockaddr_un, sun_path)
457 + strlen (un->sun_path) + 1);
458 return 0;
459 }
460
461 /* Clean up after make_sockaddr_un(). */
462 static void
463 free_sockaddr_un(int dirfd, const char *linkname)
464 {
465 if (dirfd >= 0) {
466 close(dirfd);
467 }
468 if (linkname[0]) {
469 fatal_signal_unlink_file_now(linkname);
470 }
471 }
472
473 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
474 static int
475 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
476 {
477 /* According to _Unix Network Programming_, umask should affect bind(). */
478 mode_t old_umask = umask(0077);
479 int error = bind(fd, sun, sun_len) ? errno : 0;
480 umask(old_umask);
481 return error;
482 }
483
484 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
485 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
486 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
487 * is true, the socket is made non-blocking.
488 *
489 * Returns the socket's fd if successful, otherwise a negative errno value. */
490 int
491 make_unix_socket(int style, bool nonblock,
492 const char *bind_path, const char *connect_path)
493 {
494 int error;
495 int fd;
496
497 fd = socket(PF_UNIX, style, 0);
498 if (fd < 0) {
499 return -errno;
500 }
501
502 /* Set nonblocking mode right away, if we want it. This prevents blocking
503 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
504 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
505 * if a backlog of un-accepted connections has built up in the kernel.) */
506 if (nonblock) {
507 error = set_nonblocking(fd);
508 if (error) {
509 goto error;
510 }
511 }
512
513 if (bind_path) {
514 char linkname[MAX_UN_LEN + 1];
515 struct sockaddr_un un;
516 socklen_t un_len;
517 int dirfd;
518
519 if (unlink(bind_path) && errno != ENOENT) {
520 VLOG_WARN("unlinking \"%s\": %s\n",
521 bind_path, ovs_strerror(errno));
522 }
523 fatal_signal_add_file_to_unlink(bind_path);
524
525 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
526 if (!error) {
527 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
528 }
529 free_sockaddr_un(dirfd, linkname);
530
531 if (error) {
532 goto error;
533 }
534 }
535
536 if (connect_path) {
537 char linkname[MAX_UN_LEN + 1];
538 struct sockaddr_un un;
539 socklen_t un_len;
540 int dirfd;
541
542 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
543 if (!error
544 && connect(fd, (struct sockaddr*) &un, un_len)
545 && errno != EINPROGRESS) {
546 error = errno;
547 }
548 free_sockaddr_un(dirfd, linkname);
549
550 if (error) {
551 goto error;
552 }
553 }
554
555 return fd;
556
557 error:
558 if (error == EAGAIN) {
559 error = EPROTO;
560 }
561 if (bind_path) {
562 fatal_signal_unlink_file_now(bind_path);
563 }
564 close(fd);
565 return -error;
566 }
567
568 int
569 get_unix_name_len(socklen_t sun_len)
570 {
571 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
572 ? sun_len - offsetof(struct sockaddr_un, sun_path)
573 : 0);
574 }
575 #endif /* _WIN32 */
576
577 ovs_be32
578 guess_netmask(ovs_be32 ip_)
579 {
580 uint32_t ip = ntohl(ip_);
581 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
582 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
583 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
584 : htonl(0)); /* ??? */
585 }
586
587 /* This is like strsep() except:
588 *
589 * - The separator string is ":".
590 *
591 * - Square brackets [] quote ":" separators and are removed from the
592 * tokens. */
593 static char *
594 parse_bracketed_token(char **pp)
595 {
596 char *p = *pp;
597
598 if (p == NULL) {
599 return NULL;
600 } else if (*p == '\0') {
601 *pp = NULL;
602 return p;
603 } else if (*p == '[') {
604 char *start = p + 1;
605 char *end = start + strcspn(start, "]");
606 *pp = (*end == '\0' ? NULL
607 : end[1] == ':' ? end + 2
608 : end + 1);
609 *end = '\0';
610 return start;
611 } else {
612 char *start = p;
613 char *end = start + strcspn(start, ":");
614 *pp = *end == '\0' ? NULL : end + 1;
615 *end = '\0';
616 return start;
617 }
618 }
619
620 static bool
621 parse_sockaddr_components(struct sockaddr_storage *ss,
622 const char *host_s,
623 const char *port_s, uint16_t default_port,
624 const char *s)
625 {
626 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
627 int port;
628
629 if (port_s && port_s[0]) {
630 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
631 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
632 }
633 } else {
634 port = default_port;
635 }
636
637 memset(ss, 0, sizeof *ss);
638 if (strchr(host_s, ':')) {
639 struct sockaddr_in6 *sin6
640 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
641
642 sin6->sin6_family = AF_INET6;
643 sin6->sin6_port = htons(port);
644 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
645 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
646 goto exit;
647 }
648 } else {
649 sin->sin_family = AF_INET;
650 sin->sin_port = htons(port);
651 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
652 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
653 goto exit;
654 }
655 }
656
657 return true;
658
659 exit:
660 memset(ss, 0, sizeof *ss);
661 return false;
662 }
663
664 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
665 * <host>, which is required, may be an IPv4 address or an IPv6 address
666 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
667 * optional and defaults to 'default_port'.
668 *
669 * On success, returns true and stores the parsed remote address into '*ss'.
670 * On failure, logs an error, stores zeros into '*ss', and returns false. */
671 bool
672 inet_parse_active(const char *target_, uint16_t default_port,
673 struct sockaddr_storage *ss)
674 {
675 char *target = xstrdup(target_);
676 const char *port;
677 const char *host;
678 char *p;
679 bool ok;
680
681 p = target;
682 host = parse_bracketed_token(&p);
683 port = parse_bracketed_token(&p);
684 if (!host) {
685 VLOG_ERR("%s: host must be specified", target_);
686 ok = false;
687 } else if (!port && !default_port) {
688 VLOG_ERR("%s: port must be specified", target_);
689 ok = false;
690 } else {
691 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
692 }
693 if (!ok) {
694 memset(ss, 0, sizeof *ss);
695 }
696 free(target);
697 return ok;
698 }
699
700
701 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
702 * connects to 'target', which should be a string in the format
703 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
704 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
705 * <port> is optional and defaults to 'default_port'.
706 *
707 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
708 *
709 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
710 * connection in progress), in which case the new file descriptor is stored
711 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
712 * and stores -1 into '*fdp'.
713 *
714 * If 'ss' is non-null, then on success stores the target address into '*ss'.
715 *
716 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
717 * should be in the range [0, 63] and will automatically be shifted to the
718 * appropriately place in the IP tos field. */
719 int
720 inet_open_active(int style, const char *target, uint16_t default_port,
721 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
722 {
723 struct sockaddr_storage ss;
724 int fd = -1;
725 int error;
726
727 /* Parse. */
728 if (!inet_parse_active(target, default_port, &ss)) {
729 error = EAFNOSUPPORT;
730 goto exit;
731 }
732
733 /* Create non-blocking socket. */
734 fd = socket(ss.ss_family, style, 0);
735 if (fd < 0) {
736 error = sock_errno();
737 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
738 goto exit;
739 }
740 error = set_nonblocking(fd);
741 if (error) {
742 goto exit;
743 }
744
745 /* The dscp bits must be configured before connect() to ensure that the
746 * TOS field is set during the connection establishment. If set after
747 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
748 error = set_dscp(fd, dscp);
749 if (error) {
750 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
751 goto exit;
752 }
753
754 /* Connect. */
755 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
756 ? 0
757 : sock_errno();
758 if (error == EINPROGRESS
759 #ifdef _WIN32
760 || error == WSAEALREADY || error == WSAEWOULDBLOCK
761 #endif
762 ) {
763 error = EAGAIN;
764 }
765
766 exit:
767 if (error && error != EAGAIN) {
768 if (ssp) {
769 memset(ssp, 0, sizeof *ssp);
770 }
771 if (fd >= 0) {
772 closesocket(fd);
773 fd = -1;
774 }
775 } else {
776 if (ssp) {
777 *ssp = ss;
778 }
779 }
780 *fdp = fd;
781 return error;
782 }
783
784 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
785 *
786 * - If 'default_port' is -1, then <port> is required. Otherwise, if
787 * <port> is omitted, then 'default_port' is used instead.
788 *
789 * - If <port> (or 'default_port', if used) is 0, then no port is bound
790 * and the TCP/IP stack will select a port.
791 *
792 * - <host> is optional. If supplied, it may be an IPv4 address or an
793 * IPv6 address enclosed in square brackets. If omitted, the IP address
794 * is wildcarded.
795 *
796 * If successful, stores the address into '*ss' and returns true; otherwise
797 * zeros '*ss' and returns false. */
798 bool
799 inet_parse_passive(const char *target_, int default_port,
800 struct sockaddr_storage *ss)
801 {
802 char *target = xstrdup(target_);
803 const char *port;
804 const char *host;
805 char *p;
806 bool ok;
807
808 p = target;
809 port = parse_bracketed_token(&p);
810 host = parse_bracketed_token(&p);
811 if (!port && default_port < 0) {
812 VLOG_ERR("%s: port must be specified", target_);
813 ok = false;
814 } else {
815 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
816 port, default_port, target_);
817 }
818 if (!ok) {
819 memset(ss, 0, sizeof *ss);
820 }
821 free(target);
822 return ok;
823 }
824
825
826 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
827 * 'target', and listens for incoming connections. Parses 'target' in the same
828 * way was inet_parse_passive().
829 *
830 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
831 *
832 * For TCP, the socket will have SO_REUSEADDR turned on.
833 *
834 * On success, returns a non-negative file descriptor. On failure, returns a
835 * negative errno value.
836 *
837 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
838 *
839 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
840 * should be in the range [0, 63] and will automatically be shifted to the
841 * appropriately place in the IP tos field. */
842 int
843 inet_open_passive(int style, const char *target, int default_port,
844 struct sockaddr_storage *ssp, uint8_t dscp)
845 {
846 bool kernel_chooses_port;
847 struct sockaddr_storage ss;
848 int fd = 0, error;
849 unsigned int yes = 1;
850
851 if (!inet_parse_passive(target, default_port, &ss)) {
852 return -EAFNOSUPPORT;
853 }
854 kernel_chooses_port = ss_get_port(&ss) == 0;
855
856 /* Create non-blocking socket, set SO_REUSEADDR. */
857 fd = socket(ss.ss_family, style, 0);
858 if (fd < 0) {
859 error = sock_errno();
860 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
861 return -error;
862 }
863 error = set_nonblocking(fd);
864 if (error) {
865 goto error;
866 }
867 if (style == SOCK_STREAM
868 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
869 error = sock_errno();
870 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
871 target, sock_strerror(error));
872 goto error;
873 }
874
875 /* Bind. */
876 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
877 error = sock_errno();
878 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
879 goto error;
880 }
881
882 /* The dscp bits must be configured before connect() to ensure that the TOS
883 * field is set during the connection establishment. If set after
884 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
885 error = set_dscp(fd, dscp);
886 if (error) {
887 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
888 goto error;
889 }
890
891 /* Listen. */
892 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
893 error = sock_errno();
894 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
895 goto error;
896 }
897
898 if (ssp || kernel_chooses_port) {
899 socklen_t ss_len = sizeof ss;
900 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
901 error = sock_errno();
902 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
903 goto error;
904 }
905 if (kernel_chooses_port) {
906 VLOG_INFO("%s: listening on port %"PRIu16,
907 target, ss_get_port(&ss));
908 }
909 if (ssp) {
910 *ssp = ss;
911 }
912 }
913
914 return fd;
915
916 error:
917 if (ssp) {
918 memset(ssp, 0, sizeof *ssp);
919 }
920 closesocket(fd);
921 return -error;
922 }
923
924 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
925 * a negative errno value. The caller must not close the returned fd (because
926 * the same fd will be handed out to subsequent callers). */
927 int
928 get_null_fd(void)
929 {
930 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
931 static int null_fd;
932
933 if (ovsthread_once_start(&once)) {
934 null_fd = open("/dev/null", O_RDWR);
935 if (null_fd < 0) {
936 int error = errno;
937 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
938 null_fd = -error;
939 }
940 ovsthread_once_done(&once);
941 }
942
943 return null_fd;
944 }
945
946 int
947 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
948 {
949 uint8_t *p = p_;
950
951 *bytes_read = 0;
952 while (size > 0) {
953 ssize_t retval = read(fd, p, size);
954 if (retval > 0) {
955 *bytes_read += retval;
956 size -= retval;
957 p += retval;
958 } else if (retval == 0) {
959 return EOF;
960 } else if (errno != EINTR) {
961 return errno;
962 }
963 }
964 return 0;
965 }
966
967 int
968 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
969 {
970 const uint8_t *p = p_;
971
972 *bytes_written = 0;
973 while (size > 0) {
974 ssize_t retval = write(fd, p, size);
975 if (retval > 0) {
976 *bytes_written += retval;
977 size -= retval;
978 p += retval;
979 } else if (retval == 0) {
980 VLOG_WARN("write returned 0");
981 return EPROTO;
982 } else if (errno != EINTR) {
983 return errno;
984 }
985 }
986 return 0;
987 }
988
989 /* Given file name 'file_name', fsyncs the directory in which it is contained.
990 * Returns 0 if successful, otherwise a positive errno value. */
991 int
992 fsync_parent_dir(const char *file_name)
993 {
994 int error = 0;
995 #ifndef _WIN32
996 char *dir;
997 int fd;
998
999 dir = dir_name(file_name);
1000 fd = open(dir, O_RDONLY);
1001 if (fd >= 0) {
1002 if (fsync(fd)) {
1003 if (errno == EINVAL || errno == EROFS) {
1004 /* This directory does not support synchronization. Not
1005 * really an error. */
1006 } else {
1007 error = errno;
1008 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1009 }
1010 }
1011 close(fd);
1012 } else {
1013 error = errno;
1014 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1015 }
1016 free(dir);
1017 #endif
1018
1019 return error;
1020 }
1021
1022 /* Obtains the modification time of the file named 'file_name' to the greatest
1023 * supported precision. If successful, stores the mtime in '*mtime' and
1024 * returns 0. On error, returns a positive errno value and stores zeros in
1025 * '*mtime'. */
1026 int
1027 get_mtime(const char *file_name, struct timespec *mtime)
1028 {
1029 struct stat s;
1030
1031 if (!stat(file_name, &s)) {
1032 mtime->tv_sec = s.st_mtime;
1033
1034 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1035 mtime->tv_nsec = s.st_mtim.tv_nsec;
1036 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1037 mtime->tv_nsec = s.st_mtimensec;
1038 #else
1039 mtime->tv_nsec = 0;
1040 #endif
1041
1042 return 0;
1043 } else {
1044 mtime->tv_sec = mtime->tv_nsec = 0;
1045 return errno;
1046 }
1047 }
1048
1049 void
1050 xpipe(int fds[2])
1051 {
1052 if (pipe(fds)) {
1053 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1054 }
1055 }
1056
1057 void
1058 xpipe_nonblocking(int fds[2])
1059 {
1060 xpipe(fds);
1061 xset_nonblocking(fds[0]);
1062 xset_nonblocking(fds[1]);
1063 }
1064
1065 static int
1066 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1067 {
1068 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1069 socklen_t len;
1070 int value;
1071 int error;
1072
1073 len = sizeof value;
1074 if (getsockopt(fd, level, option, &value, &len)) {
1075 error = sock_errno();
1076 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
1077 } else if (len != sizeof value) {
1078 error = EINVAL;
1079 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1080 optname, (unsigned int) len, sizeof value);
1081 } else {
1082 error = 0;
1083 }
1084
1085 *valuep = error ? 0 : value;
1086 return error;
1087 }
1088
1089 static void
1090 describe_sockaddr(struct ds *string, int fd,
1091 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1092 {
1093 struct sockaddr_storage ss;
1094 socklen_t len = sizeof ss;
1095
1096 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1097 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1098 char addrbuf[SS_NTOP_BUFSIZE];
1099
1100 ds_put_format(string, "%s:%"PRIu16,
1101 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1102 ss_get_port(&ss));
1103 #ifndef _WIN32
1104 } else if (ss.ss_family == AF_UNIX) {
1105 struct sockaddr_un sun;
1106 const char *null;
1107 size_t maxlen;
1108
1109 memcpy(&sun, &ss, sizeof sun);
1110 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1111 null = memchr(sun.sun_path, '\0', maxlen);
1112 ds_put_buffer(string, sun.sun_path,
1113 null ? null - sun.sun_path : maxlen);
1114 #endif
1115 }
1116 #ifdef HAVE_NETLINK
1117 else if (ss.ss_family == AF_NETLINK) {
1118 int protocol;
1119
1120 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1121 * of the Linux kernel headers in use at build time. */
1122 #ifndef SO_PROTOCOL
1123 #define SO_PROTOCOL 38
1124 #endif
1125
1126 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1127 &protocol)) {
1128 switch (protocol) {
1129 case NETLINK_ROUTE:
1130 ds_put_cstr(string, "NETLINK_ROUTE");
1131 break;
1132
1133 case NETLINK_GENERIC:
1134 ds_put_cstr(string, "NETLINK_GENERIC");
1135 break;
1136
1137 default:
1138 ds_put_format(string, "AF_NETLINK family %d", protocol);
1139 break;
1140 }
1141 } else {
1142 ds_put_cstr(string, "AF_NETLINK");
1143 }
1144 }
1145 #endif
1146 #if AF_PACKET && LINUX_DATAPATH
1147 else if (ss.ss_family == AF_PACKET) {
1148 struct sockaddr_ll sll;
1149
1150 memcpy(&sll, &ss, sizeof sll);
1151 ds_put_cstr(string, "AF_PACKET");
1152 if (sll.sll_ifindex) {
1153 char name[IFNAMSIZ];
1154
1155 if (if_indextoname(sll.sll_ifindex, name)) {
1156 ds_put_format(string, "(%s)", name);
1157 } else {
1158 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1159 }
1160 }
1161 if (sll.sll_protocol) {
1162 ds_put_format(string, "(protocol=0x%"PRIu16")",
1163 ntohs(sll.sll_protocol));
1164 }
1165 }
1166 #endif
1167 else if (ss.ss_family == AF_UNSPEC) {
1168 ds_put_cstr(string, "AF_UNSPEC");
1169 } else {
1170 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1171 }
1172 }
1173 }
1174
1175
1176 #ifdef LINUX_DATAPATH
1177 static void
1178 put_fd_filename(struct ds *string, int fd)
1179 {
1180 char buf[1024];
1181 char *linkname;
1182 int n;
1183
1184 linkname = xasprintf("/proc/self/fd/%d", fd);
1185 n = readlink(linkname, buf, sizeof buf);
1186 if (n > 0) {
1187 ds_put_char(string, ' ');
1188 ds_put_buffer(string, buf, n);
1189 if (n > sizeof buf) {
1190 ds_put_cstr(string, "...");
1191 }
1192 }
1193 free(linkname);
1194 }
1195 #endif
1196
1197 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1198 char *
1199 describe_fd(int fd)
1200 {
1201 struct ds string;
1202 struct stat s;
1203
1204 ds_init(&string);
1205 #ifndef _WIN32
1206 if (fstat(fd, &s)) {
1207 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1208 } else if (S_ISSOCK(s.st_mode)) {
1209 describe_sockaddr(&string, fd, getsockname);
1210 ds_put_cstr(&string, "<->");
1211 describe_sockaddr(&string, fd, getpeername);
1212 } else {
1213 ds_put_cstr(&string, (isatty(fd) ? "tty"
1214 : S_ISDIR(s.st_mode) ? "directory"
1215 : S_ISCHR(s.st_mode) ? "character device"
1216 : S_ISBLK(s.st_mode) ? "block device"
1217 : S_ISREG(s.st_mode) ? "file"
1218 : S_ISFIFO(s.st_mode) ? "FIFO"
1219 : S_ISLNK(s.st_mode) ? "symbolic link"
1220 : "unknown"));
1221 #ifdef LINUX_DATAPATH
1222 put_fd_filename(&string, fd);
1223 #endif
1224 }
1225 #else
1226 ds_put_format(&string,"file descriptor");
1227 #endif /* _WIN32 */
1228 return ds_steal_cstr(&string);
1229 }
1230
1231 #ifndef _WIN32
1232 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1233 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1234 int
1235 af_inet_ioctl(unsigned long int command, const void *arg)
1236 {
1237 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1238 static int sock;
1239
1240 if (ovsthread_once_start(&once)) {
1241 sock = socket(AF_INET, SOCK_DGRAM, 0);
1242 if (sock < 0) {
1243 int error = sock_errno();
1244 VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
1245 sock = -error;
1246 }
1247 ovsthread_once_done(&once);
1248 }
1249
1250 return (sock < 0 ? -sock
1251 : ioctl(sock, command, arg) == -1 ? errno
1252 : 0);
1253 }
1254
1255 int
1256 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1257 const char *cmd_name)
1258 {
1259 int error;
1260
1261 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1262 error = af_inet_ioctl(cmd, ifr);
1263 if (error) {
1264 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1265 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1266 ovs_strerror(error));
1267 }
1268 return error;
1269 }
1270 #endif
1271 \f
1272 /* sockaddr_storage helpers. */
1273
1274 /* Returns the IPv4 or IPv6 port in 'ss'. */
1275 uint16_t
1276 ss_get_port(const struct sockaddr_storage *ss)
1277 {
1278 if (ss->ss_family == AF_INET) {
1279 const struct sockaddr_in *sin
1280 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1281 return ntohs(sin->sin_port);
1282 } else if (ss->ss_family == AF_INET6) {
1283 const struct sockaddr_in6 *sin6
1284 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1285 return ntohs(sin6->sin6_port);
1286 } else {
1287 OVS_NOT_REACHED();
1288 }
1289 }
1290
1291 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1292 * If 'ss' is an IPv6 address, puts square brackets around the address.
1293 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1294 *
1295 * Returns 'buf'. */
1296 char *
1297 ss_format_address(const struct sockaddr_storage *ss,
1298 char *buf, size_t bufsize)
1299 {
1300 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1301 if (ss->ss_family == AF_INET) {
1302 const struct sockaddr_in *sin
1303 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1304
1305 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1306 } else if (ss->ss_family == AF_INET6) {
1307 const struct sockaddr_in6 *sin6
1308 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1309
1310 buf[0] = '[';
1311 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1312 strcpy(strchr(buf, '\0'), "]");
1313 } else {
1314 OVS_NOT_REACHED();
1315 }
1316
1317 return buf;
1318 }
1319
1320 size_t
1321 ss_length(const struct sockaddr_storage *ss)
1322 {
1323 switch (ss->ss_family) {
1324 case AF_INET:
1325 return sizeof(struct sockaddr_in);
1326
1327 case AF_INET6:
1328 return sizeof(struct sockaddr_in6);
1329
1330 default:
1331 OVS_NOT_REACHED();
1332 }
1333 }
1334
1335 /* For Windows socket calls, 'errno' is not set. One has to call
1336 * WSAGetLastError() to get the error number and then pass it to
1337 * this function to get the correct error string.
1338 *
1339 * ovs_strerror() calls strerror_r() and would not get the correct error
1340 * string for Windows sockets, but is good for POSIX. */
1341 const char *
1342 sock_strerror(int error)
1343 {
1344 #ifdef _WIN32
1345 return ovs_format_message(error);
1346 #else
1347 return ovs_strerror(error);
1348 #endif
1349 }