]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
afc1d536 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "socket-util.h" | |
19 | #include <arpa/inet.h> | |
20 | #include <errno.h> | |
21 | #include <fcntl.h> | |
f89ffb0e | 22 | #include <net/if.h> |
064af421 | 23 | #include <netdb.h> |
b7cefbf7 | 24 | #include <netinet/tcp.h> |
064af421 BP |
25 | #include <poll.h> |
26 | #include <stddef.h> | |
27 | #include <stdio.h> | |
78ff0270 | 28 | #include <stdlib.h> |
064af421 | 29 | #include <string.h> |
3762274e BP |
30 | #include <sys/socket.h> |
31 | #include <sys/stat.h> | |
a7a27d07 | 32 | #include <sys/uio.h> |
064af421 BP |
33 | #include <sys/un.h> |
34 | #include <unistd.h> | |
3e8a2ad1 | 35 | #include "openvswitch/dynamic-string.h" |
b26f46a4 | 36 | #include "ovs-thread.h" |
f89ffb0e | 37 | #include "packets.h" |
fd94a42c | 38 | #include "poll-loop.h" |
064af421 | 39 | #include "util.h" |
e6211adc | 40 | #include "openvswitch/vlog.h" |
2f51a7eb | 41 | #ifdef __linux__ |
f89ffb0e BP |
42 | #include <linux/if_packet.h> |
43 | #endif | |
44 | #ifdef HAVE_NETLINK | |
45 | #include "netlink-protocol.h" | |
46 | #include "netlink-socket.h" | |
47 | #endif | |
5136ce49 | 48 | |
d98e6007 | 49 | VLOG_DEFINE_THIS_MODULE(socket_util); |
064af421 | 50 | |
c1c19657 BP |
51 | static int getsockopt_int(int fd, int level, int option, const char *optname, |
52 | int *valuep); | |
53 | ||
064af421 BP |
54 | /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a |
55 | * positive errno value. */ | |
56 | int | |
57 | set_nonblocking(int fd) | |
58 | { | |
de8bd597 | 59 | #ifndef _WIN32 |
064af421 BP |
60 | int flags = fcntl(fd, F_GETFL, 0); |
61 | if (flags != -1) { | |
62 | if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { | |
63 | return 0; | |
64 | } else { | |
10a89ef0 | 65 | VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno)); |
064af421 BP |
66 | return errno; |
67 | } | |
68 | } else { | |
10a89ef0 | 69 | VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno)); |
064af421 BP |
70 | return errno; |
71 | } | |
de8bd597 GS |
72 | #else |
73 | unsigned long arg = 1; | |
74 | if (ioctlsocket(fd, FIONBIO, &arg)) { | |
75 | int error = sock_errno(); | |
76 | VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error)); | |
77 | return error; | |
78 | } | |
79 | return 0; | |
80 | #endif | |
064af421 BP |
81 | } |
82 | ||
a0505c49 BP |
83 | void |
84 | xset_nonblocking(int fd) | |
85 | { | |
86 | if (set_nonblocking(fd)) { | |
87 | exit(EXIT_FAILURE); | |
88 | } | |
89 | } | |
90 | ||
b7cefbf7 GS |
91 | void |
92 | setsockopt_tcp_nodelay(int fd) | |
93 | { | |
94 | int on = 1; | |
95 | int retval; | |
96 | ||
97 | retval = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof on); | |
98 | if (retval) { | |
99 | retval = sock_errno(); | |
100 | VLOG_ERR("setsockopt(TCP_NODELAY): %s", sock_strerror(retval)); | |
101 | } | |
102 | } | |
103 | ||
afc1d536 BP |
104 | /* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less. |
105 | * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to | |
106 | * do anything useful). */ | |
6b9c1eab | 107 | int |
afc1d536 | 108 | set_dscp(int fd, int family, uint8_t dscp) |
cea15768 | 109 | { |
afc1d536 | 110 | int retval; |
eb0cb316 EM |
111 | int val; |
112 | ||
5349904d GS |
113 | #ifdef _WIN32 |
114 | /* XXX: Consider using QoS2 APIs for Windows to set dscp. */ | |
115 | return 0; | |
116 | #endif | |
117 | ||
cea15768 EJ |
118 | if (dscp > 63) { |
119 | return EINVAL; | |
120 | } | |
eb0cb316 | 121 | val = dscp << 2; |
afc1d536 BP |
122 | |
123 | switch (family) { | |
124 | case AF_INET: | |
125 | retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val); | |
126 | break; | |
127 | ||
128 | case AF_INET6: | |
129 | retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val); | |
130 | break; | |
131 | ||
132 | default: | |
92ae5930 | 133 | return ENOPROTOOPT; |
cea15768 EJ |
134 | } |
135 | ||
afc1d536 | 136 | return retval ? sock_errno() : 0; |
cea15768 EJ |
137 | } |
138 | ||
9835576b JB |
139 | /* Checks whether 'host_name' is an IPv4 or IPv6 address. It is assumed |
140 | * that 'host_name' is valid. Returns false if it is IPv4 address, true if | |
141 | * it is IPv6 address. */ | |
142 | bool | |
143 | addr_is_ipv6(const char *host_name) | |
144 | { | |
145 | return strchr(host_name, ':') != NULL; | |
146 | } | |
147 | ||
2b35e147 BP |
148 | /* Translates 'host_name', which must be a string representation of an IP |
149 | * address, into a numeric IP address in '*addr'. Returns 0 if successful, | |
150 | * otherwise a positive errno value. */ | |
064af421 | 151 | int |
d295e8e9 | 152 | lookup_ip(const char *host_name, struct in_addr *addr) |
064af421 | 153 | { |
e7695092 | 154 | if (!ip_parse(host_name, &addr->s_addr)) { |
db5ce514 | 155 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2b35e147 BP |
156 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name); |
157 | return ENOENT; | |
064af421 BP |
158 | } |
159 | return 0; | |
160 | } | |
161 | ||
d31f1109 JP |
162 | /* Translates 'host_name', which must be a string representation of an IPv6 |
163 | * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, | |
164 | * otherwise a positive errno value. */ | |
165 | int | |
166 | lookup_ipv6(const char *host_name, struct in6_addr *addr) | |
167 | { | |
e7695092 | 168 | if (!ipv6_parse(host_name, addr)) { |
d31f1109 JP |
169 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
170 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name); | |
171 | return ENOENT; | |
172 | } | |
173 | return 0; | |
174 | } | |
175 | ||
13f2ef97 BP |
176 | /* Translates 'host_name', which must be a host name or a string representation |
177 | * of an IP address, into a numeric IP address in '*addr'. Returns 0 if | |
178 | * successful, otherwise a positive errno value. | |
179 | * | |
180 | * Most Open vSwitch code should not use this because it causes deadlocks: | |
3cbb5dc7 | 181 | * getaddrinfo() sends out a DNS request but that starts a new flow for which |
13f2ef97 | 182 | * OVS must set up a flow, but it can't because it's waiting for a DNS reply. |
293d49bd | 183 | * The synchronous lookup also delays other activity. (Of course we can solve |
13f2ef97 BP |
184 | * this but it doesn't seem worthwhile quite yet.) */ |
185 | int | |
186 | lookup_hostname(const char *host_name, struct in_addr *addr) | |
187 | { | |
3cbb5dc7 BP |
188 | struct addrinfo *result; |
189 | struct addrinfo hints; | |
13f2ef97 | 190 | |
e7695092 | 191 | if (ip_parse(host_name, &addr->s_addr)) { |
13f2ef97 BP |
192 | return 0; |
193 | } | |
194 | ||
3cbb5dc7 BP |
195 | memset(&hints, 0, sizeof hints); |
196 | hints.ai_family = AF_INET; | |
197 | ||
198 | switch (getaddrinfo(host_name, NULL, &hints, &result)) { | |
199 | case 0: | |
db5a1019 AW |
200 | *addr = ALIGNED_CAST(struct sockaddr_in *, |
201 | result->ai_addr)->sin_addr; | |
3cbb5dc7 | 202 | freeaddrinfo(result); |
13f2ef97 | 203 | return 0; |
13f2ef97 | 204 | |
44f645a5 | 205 | #ifdef EAI_ADDRFAMILY |
3cbb5dc7 | 206 | case EAI_ADDRFAMILY: |
44f645a5 | 207 | #endif |
3cbb5dc7 BP |
208 | case EAI_NONAME: |
209 | case EAI_SERVICE: | |
210 | return ENOENT; | |
211 | ||
212 | case EAI_AGAIN: | |
213 | return EAGAIN; | |
214 | ||
215 | case EAI_BADFLAGS: | |
216 | case EAI_FAMILY: | |
217 | case EAI_SOCKTYPE: | |
218 | return EINVAL; | |
219 | ||
220 | case EAI_FAIL: | |
221 | return EIO; | |
222 | ||
223 | case EAI_MEMORY: | |
224 | return ENOMEM; | |
225 | ||
99ad8ba8 | 226 | #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME |
3cbb5dc7 BP |
227 | case EAI_NODATA: |
228 | return ENXIO; | |
44f645a5 | 229 | #endif |
3cbb5dc7 | 230 | |
99ad8ba8 | 231 | #ifdef EAI_SYSTEM |
3cbb5dc7 | 232 | case EAI_SYSTEM: |
0f0b5401 | 233 | return sock_errno(); |
99ad8ba8 | 234 | #endif |
3cbb5dc7 BP |
235 | |
236 | default: | |
237 | return EPROTO; | |
238 | } | |
13f2ef97 BP |
239 | } |
240 | ||
064af421 | 241 | int |
d295e8e9 | 242 | check_connection_completion(int fd) |
064af421 | 243 | { |
d6cedfd9 | 244 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
064af421 BP |
245 | struct pollfd pfd; |
246 | int retval; | |
247 | ||
248 | pfd.fd = fd; | |
249 | pfd.events = POLLOUT; | |
1bada7ab GS |
250 | |
251 | #ifndef _WIN32 | |
064af421 BP |
252 | do { |
253 | retval = poll(&pfd, 1, 0); | |
254 | } while (retval < 0 && errno == EINTR); | |
1bada7ab | 255 | #else |
ff792c6a AS |
256 | fd_set wrset, exset; |
257 | FD_ZERO(&wrset); | |
258 | FD_ZERO(&exset); | |
259 | FD_SET(fd, &exset); | |
260 | FD_SET(fd, &wrset); | |
261 | pfd.revents = 0; | |
262 | struct timeval tv = { 0, 0 }; | |
263 | /* WSAPoll is broken on Windows, instead do a select */ | |
264 | retval = select(0, NULL, &wrset, &exset, &tv); | |
265 | if (retval == 1) { | |
266 | if (FD_ISSET(fd, &wrset)) { | |
267 | pfd.revents |= pfd.events; | |
268 | } | |
269 | if (FD_ISSET(fd, &exset)) { | |
270 | pfd.revents |= POLLERR; | |
271 | } | |
272 | } | |
1bada7ab | 273 | #endif |
064af421 | 274 | if (retval == 1) { |
d6cedfd9 | 275 | if (pfd.revents & POLLERR) { |
1bada7ab | 276 | ssize_t n = send(fd, "", 1, 0); |
d6cedfd9 | 277 | if (n < 0) { |
1bada7ab | 278 | return sock_errno(); |
d6cedfd9 BP |
279 | } else { |
280 | VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded"); | |
281 | return EPROTO; | |
282 | } | |
283 | } | |
284 | return 0; | |
064af421 | 285 | } else if (retval < 0) { |
1bada7ab | 286 | VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno())); |
064af421 BP |
287 | return errno; |
288 | } else { | |
289 | return EAGAIN; | |
290 | } | |
291 | } | |
292 | ||
c1c19657 BP |
293 | /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a |
294 | * negative errno value if an error occurs. */ | |
295 | int | |
296 | get_socket_rcvbuf(int sock) | |
297 | { | |
298 | int rcvbuf; | |
299 | int error; | |
300 | ||
301 | error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf); | |
302 | return error ? -error : rcvbuf; | |
303 | } | |
304 | ||
064af421 BP |
305 | /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no |
306 | * more data can be immediately read. ('fd' should therefore be in | |
307 | * non-blocking mode.)*/ | |
308 | void | |
309 | drain_fd(int fd, size_t n_packets) | |
310 | { | |
311 | for (; n_packets > 0; n_packets--) { | |
312 | /* 'buffer' only needs to be 1 byte long in most circumstances. This | |
313 | * size is defensive against the possibility that we someday want to | |
314 | * use a Linux tap device without TUN_NO_PI, in which case a buffer | |
315 | * smaller than sizeof(struct tun_pi) will give EINVAL on read. */ | |
316 | char buffer[128]; | |
317 | if (read(fd, buffer, sizeof buffer) <= 0) { | |
318 | break; | |
319 | } | |
320 | } | |
321 | } | |
322 | ||
dbba996b BP |
323 | ovs_be32 |
324 | guess_netmask(ovs_be32 ip_) | |
064af421 | 325 | { |
dbba996b | 326 | uint32_t ip = ntohl(ip_); |
064af421 BP |
327 | return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ |
328 | : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ | |
329 | : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ | |
330 | : htonl(0)); /* ??? */ | |
331 | } | |
332 | ||
e731d71b | 333 | /* This is like strsep() except: |
4f2eb9a7 | 334 | * |
e731d71b AS |
335 | * - The separator string is ":". |
336 | * | |
337 | * - Square brackets [] quote ":" separators and are removed from the | |
338 | * tokens. */ | |
339 | static char * | |
340 | parse_bracketed_token(char **pp) | |
78ff0270 | 341 | { |
e731d71b AS |
342 | char *p = *pp; |
343 | ||
344 | if (p == NULL) { | |
345 | return NULL; | |
346 | } else if (*p == '\0') { | |
347 | *pp = NULL; | |
348 | return p; | |
349 | } else if (*p == '[') { | |
350 | char *start = p + 1; | |
351 | char *end = start + strcspn(start, "]"); | |
352 | *pp = (*end == '\0' ? NULL | |
353 | : end[1] == ':' ? end + 2 | |
354 | : end + 1); | |
355 | *end = '\0'; | |
356 | return start; | |
357 | } else { | |
358 | char *start = p; | |
359 | char *end = start + strcspn(start, ":"); | |
360 | *pp = *end == '\0' ? NULL : end + 1; | |
361 | *end = '\0'; | |
362 | return start; | |
78ff0270 | 363 | } |
e731d71b | 364 | } |
78ff0270 | 365 | |
e731d71b AS |
366 | static bool |
367 | parse_sockaddr_components(struct sockaddr_storage *ss, | |
368 | const char *host_s, | |
369 | const char *port_s, uint16_t default_port, | |
370 | const char *s) | |
371 | { | |
372 | struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss); | |
373 | int port; | |
374 | ||
375 | if (port_s && port_s[0]) { | |
376 | if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) { | |
377 | VLOG_ERR("%s: bad port number \"%s\"", s, port_s); | |
12c5f227 | 378 | goto exit; |
e731d71b AS |
379 | } |
380 | } else { | |
381 | port = default_port; | |
78ff0270 | 382 | } |
e731d71b AS |
383 | |
384 | memset(ss, 0, sizeof *ss); | |
385 | if (strchr(host_s, ':')) { | |
386 | struct sockaddr_in6 *sin6 | |
387 | = ALIGNED_CAST(struct sockaddr_in6 *, ss); | |
388 | ||
389 | sin6->sin6_family = AF_INET6; | |
390 | sin6->sin6_port = htons(port); | |
e7695092 | 391 | if (!ipv6_parse(host_s, &sin6->sin6_addr)) { |
e731d71b AS |
392 | VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s); |
393 | goto exit; | |
394 | } | |
395 | } else { | |
396 | sin->sin_family = AF_INET; | |
397 | sin->sin_port = htons(port); | |
e7695092 | 398 | if (!ip_parse(host_s, &sin->sin_addr.s_addr)) { |
e731d71b AS |
399 | VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s); |
400 | goto exit; | |
401 | } | |
52f8a75e BP |
402 | } |
403 | ||
e731d71b | 404 | return true; |
52f8a75e BP |
405 | |
406 | exit: | |
e731d71b AS |
407 | memset(ss, 0, sizeof *ss); |
408 | return false; | |
409 | } | |
410 | ||
411 | /* Parses 'target', which should be a string in the format "<host>[:<port>]". | |
412 | * <host>, which is required, may be an IPv4 address or an IPv6 address | |
413 | * enclosed in square brackets. If 'default_port' is nonzero then <port> is | |
414 | * optional and defaults to 'default_port'. | |
415 | * | |
416 | * On success, returns true and stores the parsed remote address into '*ss'. | |
417 | * On failure, logs an error, stores zeros into '*ss', and returns false. */ | |
418 | bool | |
419 | inet_parse_active(const char *target_, uint16_t default_port, | |
420 | struct sockaddr_storage *ss) | |
421 | { | |
422 | char *target = xstrdup(target_); | |
423 | const char *port; | |
424 | const char *host; | |
425 | char *p; | |
426 | bool ok; | |
427 | ||
428 | p = target; | |
429 | host = parse_bracketed_token(&p); | |
430 | port = parse_bracketed_token(&p); | |
431 | if (!host) { | |
432 | VLOG_ERR("%s: host must be specified", target_); | |
433 | ok = false; | |
434 | } else if (!port && !default_port) { | |
435 | VLOG_ERR("%s: port must be specified", target_); | |
436 | ok = false; | |
437 | } else { | |
438 | ok = parse_sockaddr_components(ss, host, port, default_port, target_); | |
439 | } | |
52f8a75e | 440 | if (!ok) { |
e731d71b | 441 | memset(ss, 0, sizeof *ss); |
52f8a75e BP |
442 | } |
443 | free(target); | |
444 | return ok; | |
445 | } | |
446 | ||
e731d71b AS |
447 | |
448 | /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and | |
449 | * connects to 'target', which should be a string in the format | |
450 | * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an | |
451 | * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then | |
452 | * <port> is optional and defaults to 'default_port'. | |
52f8a75e BP |
453 | * |
454 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
455 | * | |
456 | * On success, returns 0 (indicating connection complete) or EAGAIN (indicating | |
457 | * connection in progress), in which case the new file descriptor is stored | |
458 | * into '*fdp'. On failure, returns a positive errno value other than EAGAIN | |
459 | * and stores -1 into '*fdp'. | |
460 | * | |
e731d71b | 461 | * If 'ss' is non-null, then on success stores the target address into '*ss'. |
f125905c | 462 | * |
cea15768 EJ |
463 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
464 | * should be in the range [0, 63] and will automatically be shifted to the | |
465 | * appropriately place in the IP tos field. */ | |
52f8a75e BP |
466 | int |
467 | inet_open_active(int style, const char *target, uint16_t default_port, | |
e731d71b | 468 | struct sockaddr_storage *ssp, int *fdp, uint8_t dscp) |
52f8a75e | 469 | { |
e731d71b | 470 | struct sockaddr_storage ss; |
52f8a75e BP |
471 | int fd = -1; |
472 | int error; | |
473 | ||
474 | /* Parse. */ | |
e731d71b | 475 | if (!inet_parse_active(target, default_port, &ss)) { |
1901968e BP |
476 | error = EAFNOSUPPORT; |
477 | goto exit; | |
78ff0270 BP |
478 | } |
479 | ||
480 | /* Create non-blocking socket. */ | |
e731d71b | 481 | fd = socket(ss.ss_family, style, 0); |
78ff0270 | 482 | if (fd < 0) { |
0f0b5401 GS |
483 | error = sock_errno(); |
484 | VLOG_ERR("%s: socket: %s", target, sock_strerror(error)); | |
78ff0270 BP |
485 | goto exit; |
486 | } | |
487 | error = set_nonblocking(fd); | |
488 | if (error) { | |
a4efa3fc | 489 | goto exit; |
78ff0270 BP |
490 | } |
491 | ||
e731d71b AS |
492 | /* The dscp bits must be configured before connect() to ensure that the |
493 | * TOS field is set during the connection establishment. If set after | |
cea15768 | 494 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ |
afc1d536 | 495 | error = set_dscp(fd, ss.ss_family, dscp); |
cea15768 | 496 | if (error) { |
909dad2b | 497 | VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error)); |
ef8a3d14 | 498 | goto exit; |
f125905c MM |
499 | } |
500 | ||
78ff0270 | 501 | /* Connect. */ |
e731d71b AS |
502 | error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0 |
503 | ? 0 | |
0f0b5401 GS |
504 | : sock_errno(); |
505 | if (error == EINPROGRESS | |
506 | #ifdef _WIN32 | |
507 | || error == WSAEALREADY || error == WSAEWOULDBLOCK | |
508 | #endif | |
509 | ) { | |
78ff0270 | 510 | error = EAGAIN; |
78ff0270 BP |
511 | } |
512 | ||
78ff0270 | 513 | exit: |
e731d71b AS |
514 | if (error && error != EAGAIN) { |
515 | if (ssp) { | |
516 | memset(ssp, 0, sizeof *ssp); | |
517 | } | |
518 | if (fd >= 0) { | |
7009a594 | 519 | closesocket(fd); |
e731d71b AS |
520 | fd = -1; |
521 | } | |
522 | } else { | |
523 | if (ssp) { | |
524 | *ssp = ss; | |
78ff0270 | 525 | } |
78ff0270 | 526 | } |
a4efa3fc | 527 | *fdp = fd; |
78ff0270 BP |
528 | return error; |
529 | } | |
530 | ||
e731d71b | 531 | /* Parses 'target', which should be a string in the format "[<port>][:<host>]": |
e1bd3bee BP |
532 | * |
533 | * - If 'default_port' is -1, then <port> is required. Otherwise, if | |
534 | * <port> is omitted, then 'default_port' is used instead. | |
535 | * | |
536 | * - If <port> (or 'default_port', if used) is 0, then no port is bound | |
537 | * and the TCP/IP stack will select a port. | |
538 | * | |
e731d71b AS |
539 | * - <host> is optional. If supplied, it may be an IPv4 address or an |
540 | * IPv6 address enclosed in square brackets. If omitted, the IP address | |
541 | * is wildcarded. | |
4f2eb9a7 | 542 | * |
e731d71b AS |
543 | * If successful, stores the address into '*ss' and returns true; otherwise |
544 | * zeros '*ss' and returns false. */ | |
d98fa503 | 545 | bool |
995337c4 | 546 | inet_parse_passive(const char *target_, int default_port, |
e731d71b | 547 | struct sockaddr_storage *ss) |
78ff0270 BP |
548 | { |
549 | char *target = xstrdup(target_); | |
e731d71b AS |
550 | const char *port; |
551 | const char *host; | |
552 | char *p; | |
553 | bool ok; | |
554 | ||
555 | p = target; | |
556 | port = parse_bracketed_token(&p); | |
557 | host = parse_bracketed_token(&p); | |
558 | if (!port && default_port < 0) { | |
559 | VLOG_ERR("%s: port must be specified", target_); | |
560 | ok = false; | |
561 | } else { | |
562 | ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0", | |
563 | port, default_port, target_); | |
d98fa503 | 564 | } |
d98fa503 | 565 | if (!ok) { |
e731d71b | 566 | memset(ss, 0, sizeof *ss); |
d98fa503 BP |
567 | } |
568 | free(target); | |
569 | return ok; | |
570 | } | |
571 | ||
572 | ||
e731d71b | 573 | /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to |
d98fa503 BP |
574 | * 'target', and listens for incoming connections. Parses 'target' in the same |
575 | * way was inet_parse_passive(). | |
576 | * | |
577 | * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). | |
578 | * | |
579 | * For TCP, the socket will have SO_REUSEADDR turned on. | |
580 | * | |
581 | * On success, returns a non-negative file descriptor. On failure, returns a | |
582 | * negative errno value. | |
583 | * | |
e731d71b | 584 | * If 'ss' is non-null, then on success stores the bound address into '*ss'. |
f125905c | 585 | * |
cea15768 EJ |
586 | * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It |
587 | * should be in the range [0, 63] and will automatically be shifted to the | |
b52ecd96 GS |
588 | * appropriately place in the IP tos field. |
589 | * | |
590 | * If 'kernel_print_port' is true and the port is dynamically assigned by | |
591 | * the kernel, print the chosen port. */ | |
d98fa503 BP |
592 | int |
593 | inet_open_passive(int style, const char *target, int default_port, | |
b52ecd96 GS |
594 | struct sockaddr_storage *ssp, uint8_t dscp, |
595 | bool kernel_print_port) | |
d98fa503 | 596 | { |
df451457 | 597 | bool kernel_chooses_port; |
e731d71b | 598 | struct sockaddr_storage ss; |
d98fa503 BP |
599 | int fd = 0, error; |
600 | unsigned int yes = 1; | |
601 | ||
e731d71b | 602 | if (!inet_parse_passive(target, default_port, &ss)) { |
ca286ba9 | 603 | return -EAFNOSUPPORT; |
78ff0270 | 604 | } |
e731d71b | 605 | kernel_chooses_port = ss_get_port(&ss) == 0; |
78ff0270 BP |
606 | |
607 | /* Create non-blocking socket, set SO_REUSEADDR. */ | |
e731d71b | 608 | fd = socket(ss.ss_family, style, 0); |
78ff0270 | 609 | if (fd < 0) { |
0f0b5401 GS |
610 | error = sock_errno(); |
611 | VLOG_ERR("%s: socket: %s", target, sock_strerror(error)); | |
ca286ba9 | 612 | return -error; |
78ff0270 BP |
613 | } |
614 | error = set_nonblocking(fd); | |
615 | if (error) { | |
d98fa503 | 616 | goto error; |
78ff0270 | 617 | } |
4f2eb9a7 BP |
618 | if (style == SOCK_STREAM |
619 | && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) { | |
0f0b5401 | 620 | error = sock_errno(); |
10a89ef0 | 621 | VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", |
0f0b5401 | 622 | target, sock_strerror(error)); |
d98fa503 | 623 | goto error; |
78ff0270 BP |
624 | } |
625 | ||
626 | /* Bind. */ | |
e731d71b | 627 | if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) { |
0f0b5401 GS |
628 | error = sock_errno(); |
629 | VLOG_ERR("%s: bind: %s", target, sock_strerror(error)); | |
d98fa503 | 630 | goto error; |
78ff0270 BP |
631 | } |
632 | ||
cea15768 EJ |
633 | /* The dscp bits must be configured before connect() to ensure that the TOS |
634 | * field is set during the connection establishment. If set after | |
635 | * connect(), the handshake SYN frames will be sent with a TOS of 0. */ | |
afc1d536 | 636 | error = set_dscp(fd, ss.ss_family, dscp); |
cea15768 | 637 | if (error) { |
909dad2b | 638 | VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error)); |
ef8a3d14 | 639 | goto error; |
f125905c MM |
640 | } |
641 | ||
78ff0270 | 642 | /* Listen. */ |
041dc07f | 643 | if (style == SOCK_STREAM && listen(fd, 10) < 0) { |
0f0b5401 GS |
644 | error = sock_errno(); |
645 | VLOG_ERR("%s: listen: %s", target, sock_strerror(error)); | |
d98fa503 | 646 | goto error; |
78ff0270 | 647 | } |
36775dad | 648 | |
e731d71b AS |
649 | if (ssp || kernel_chooses_port) { |
650 | socklen_t ss_len = sizeof ss; | |
651 | if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) { | |
0f0b5401 GS |
652 | error = sock_errno(); |
653 | VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error)); | |
d98fa503 | 654 | goto error; |
36775dad | 655 | } |
b52ecd96 | 656 | if (kernel_chooses_port && kernel_print_port) { |
df451457 | 657 | VLOG_INFO("%s: listening on port %"PRIu16, |
e731d71b AS |
658 | target, ss_get_port(&ss)); |
659 | } | |
660 | if (ssp) { | |
661 | *ssp = ss; | |
df451457 | 662 | } |
36775dad BP |
663 | } |
664 | ||
d98fa503 | 665 | return fd; |
78ff0270 | 666 | |
d98fa503 | 667 | error: |
e731d71b AS |
668 | if (ssp) { |
669 | memset(ssp, 0, sizeof *ssp); | |
670 | } | |
7009a594 | 671 | closesocket(fd); |
ca286ba9 | 672 | return -error; |
78ff0270 BP |
673 | } |
674 | ||
064af421 BP |
675 | int |
676 | read_fully(int fd, void *p_, size_t size, size_t *bytes_read) | |
677 | { | |
678 | uint8_t *p = p_; | |
679 | ||
680 | *bytes_read = 0; | |
681 | while (size > 0) { | |
682 | ssize_t retval = read(fd, p, size); | |
683 | if (retval > 0) { | |
684 | *bytes_read += retval; | |
685 | size -= retval; | |
686 | p += retval; | |
687 | } else if (retval == 0) { | |
688 | return EOF; | |
689 | } else if (errno != EINTR) { | |
690 | return errno; | |
691 | } | |
692 | } | |
693 | return 0; | |
694 | } | |
695 | ||
696 | int | |
697 | write_fully(int fd, const void *p_, size_t size, size_t *bytes_written) | |
698 | { | |
699 | const uint8_t *p = p_; | |
700 | ||
701 | *bytes_written = 0; | |
702 | while (size > 0) { | |
703 | ssize_t retval = write(fd, p, size); | |
704 | if (retval > 0) { | |
705 | *bytes_written += retval; | |
706 | size -= retval; | |
707 | p += retval; | |
708 | } else if (retval == 0) { | |
709 | VLOG_WARN("write returned 0"); | |
710 | return EPROTO; | |
711 | } else if (errno != EINTR) { | |
712 | return errno; | |
713 | } | |
714 | } | |
715 | return 0; | |
716 | } | |
8e71cf88 BP |
717 | |
718 | /* Given file name 'file_name', fsyncs the directory in which it is contained. | |
719 | * Returns 0 if successful, otherwise a positive errno value. */ | |
720 | int | |
721 | fsync_parent_dir(const char *file_name) | |
722 | { | |
723 | int error = 0; | |
cbf414e5 | 724 | #ifndef _WIN32 |
8e71cf88 BP |
725 | char *dir; |
726 | int fd; | |
727 | ||
728 | dir = dir_name(file_name); | |
729 | fd = open(dir, O_RDONLY); | |
730 | if (fd >= 0) { | |
731 | if (fsync(fd)) { | |
732 | if (errno == EINVAL || errno == EROFS) { | |
733 | /* This directory does not support synchronization. Not | |
734 | * really an error. */ | |
735 | } else { | |
736 | error = errno; | |
10a89ef0 | 737 | VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error)); |
8e71cf88 BP |
738 | } |
739 | } | |
740 | close(fd); | |
741 | } else { | |
742 | error = errno; | |
10a89ef0 | 743 | VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error)); |
8e71cf88 BP |
744 | } |
745 | free(dir); | |
cbf414e5 | 746 | #endif |
8e71cf88 BP |
747 | |
748 | return error; | |
749 | } | |
26efd256 BP |
750 | |
751 | /* Obtains the modification time of the file named 'file_name' to the greatest | |
752 | * supported precision. If successful, stores the mtime in '*mtime' and | |
753 | * returns 0. On error, returns a positive errno value and stores zeros in | |
754 | * '*mtime'. */ | |
755 | int | |
756 | get_mtime(const char *file_name, struct timespec *mtime) | |
757 | { | |
758 | struct stat s; | |
759 | ||
760 | if (!stat(file_name, &s)) { | |
761 | mtime->tv_sec = s.st_mtime; | |
762 | ||
763 | #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC | |
764 | mtime->tv_nsec = s.st_mtim.tv_nsec; | |
765 | #elif HAVE_STRUCT_STAT_ST_MTIMENSEC | |
766 | mtime->tv_nsec = s.st_mtimensec; | |
767 | #else | |
768 | mtime->tv_nsec = 0; | |
769 | #endif | |
770 | ||
771 | return 0; | |
772 | } else { | |
773 | mtime->tv_sec = mtime->tv_nsec = 0; | |
774 | return errno; | |
775 | } | |
776 | } | |
777 | ||
f89ffb0e | 778 | static int |
c1c19657 | 779 | getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) |
f89ffb0e | 780 | { |
c1c19657 BP |
781 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); |
782 | socklen_t len; | |
783 | int value; | |
784 | int error; | |
f89ffb0e | 785 | |
c1c19657 BP |
786 | len = sizeof value; |
787 | if (getsockopt(fd, level, option, &value, &len)) { | |
0f0b5401 GS |
788 | error = sock_errno(); |
789 | VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error)); | |
c1c19657 BP |
790 | } else if (len != sizeof value) { |
791 | error = EINVAL; | |
34582733 | 792 | VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")", |
c1c19657 BP |
793 | optname, (unsigned int) len, sizeof value); |
794 | } else { | |
795 | error = 0; | |
796 | } | |
797 | ||
798 | *valuep = error ? 0 : value; | |
799 | return error; | |
f89ffb0e BP |
800 | } |
801 | ||
802 | static void | |
803 | describe_sockaddr(struct ds *string, int fd, | |
804 | int (*getaddr)(int, struct sockaddr *, socklen_t *)) | |
805 | { | |
806 | struct sockaddr_storage ss; | |
807 | socklen_t len = sizeof ss; | |
808 | ||
809 | if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { | |
e731d71b AS |
810 | if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) { |
811 | char addrbuf[SS_NTOP_BUFSIZE]; | |
f89ffb0e | 812 | |
e731d71b AS |
813 | ds_put_format(string, "%s:%"PRIu16, |
814 | ss_format_address(&ss, addrbuf, sizeof addrbuf), | |
815 | ss_get_port(&ss)); | |
7ff04d92 | 816 | #ifndef _WIN32 |
f89ffb0e BP |
817 | } else if (ss.ss_family == AF_UNIX) { |
818 | struct sockaddr_un sun; | |
819 | const char *null; | |
820 | size_t maxlen; | |
821 | ||
822 | memcpy(&sun, &ss, sizeof sun); | |
823 | maxlen = len - offsetof(struct sockaddr_un, sun_path); | |
824 | null = memchr(sun.sun_path, '\0', maxlen); | |
825 | ds_put_buffer(string, sun.sun_path, | |
826 | null ? null - sun.sun_path : maxlen); | |
7ff04d92 | 827 | #endif |
f89ffb0e BP |
828 | } |
829 | #ifdef HAVE_NETLINK | |
830 | else if (ss.ss_family == AF_NETLINK) { | |
831 | int protocol; | |
832 | ||
833 | /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version | |
834 | * of the Linux kernel headers in use at build time. */ | |
835 | #ifndef SO_PROTOCOL | |
836 | #define SO_PROTOCOL 38 | |
837 | #endif | |
838 | ||
c1c19657 BP |
839 | if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", |
840 | &protocol)) { | |
f89ffb0e BP |
841 | switch (protocol) { |
842 | case NETLINK_ROUTE: | |
843 | ds_put_cstr(string, "NETLINK_ROUTE"); | |
844 | break; | |
845 | ||
846 | case NETLINK_GENERIC: | |
847 | ds_put_cstr(string, "NETLINK_GENERIC"); | |
848 | break; | |
849 | ||
850 | default: | |
851 | ds_put_format(string, "AF_NETLINK family %d", protocol); | |
852 | break; | |
853 | } | |
854 | } else { | |
855 | ds_put_cstr(string, "AF_NETLINK"); | |
856 | } | |
857 | } | |
858 | #endif | |
2f51a7eb | 859 | #if __linux__ |
f89ffb0e BP |
860 | else if (ss.ss_family == AF_PACKET) { |
861 | struct sockaddr_ll sll; | |
862 | ||
863 | memcpy(&sll, &ss, sizeof sll); | |
864 | ds_put_cstr(string, "AF_PACKET"); | |
865 | if (sll.sll_ifindex) { | |
866 | char name[IFNAMSIZ]; | |
867 | ||
868 | if (if_indextoname(sll.sll_ifindex, name)) { | |
869 | ds_put_format(string, "(%s)", name); | |
870 | } else { | |
871 | ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); | |
872 | } | |
873 | } | |
874 | if (sll.sll_protocol) { | |
875 | ds_put_format(string, "(protocol=0x%"PRIu16")", | |
876 | ntohs(sll.sll_protocol)); | |
877 | } | |
878 | } | |
879 | #endif | |
880 | else if (ss.ss_family == AF_UNSPEC) { | |
881 | ds_put_cstr(string, "AF_UNSPEC"); | |
882 | } else { | |
883 | ds_put_format(string, "AF_%d", (int) ss.ss_family); | |
884 | } | |
885 | } | |
886 | } | |
887 | ||
888 | ||
2f51a7eb | 889 | #ifdef __linux__ |
f89ffb0e BP |
890 | static void |
891 | put_fd_filename(struct ds *string, int fd) | |
892 | { | |
893 | char buf[1024]; | |
894 | char *linkname; | |
895 | int n; | |
896 | ||
897 | linkname = xasprintf("/proc/self/fd/%d", fd); | |
898 | n = readlink(linkname, buf, sizeof buf); | |
899 | if (n > 0) { | |
900 | ds_put_char(string, ' '); | |
901 | ds_put_buffer(string, buf, n); | |
902 | if (n > sizeof buf) { | |
903 | ds_put_cstr(string, "..."); | |
904 | } | |
905 | } | |
906 | free(linkname); | |
907 | } | |
908 | #endif | |
909 | ||
910 | /* Returns a malloc()'d string describing 'fd', for use in logging. */ | |
911 | char * | |
912 | describe_fd(int fd) | |
913 | { | |
914 | struct ds string; | |
915 | struct stat s; | |
916 | ||
917 | ds_init(&string); | |
54a1cfb5 | 918 | #ifndef _WIN32 |
f89ffb0e | 919 | if (fstat(fd, &s)) { |
10a89ef0 | 920 | ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno)); |
f89ffb0e BP |
921 | } else if (S_ISSOCK(s.st_mode)) { |
922 | describe_sockaddr(&string, fd, getsockname); | |
923 | ds_put_cstr(&string, "<->"); | |
924 | describe_sockaddr(&string, fd, getpeername); | |
925 | } else { | |
926 | ds_put_cstr(&string, (isatty(fd) ? "tty" | |
927 | : S_ISDIR(s.st_mode) ? "directory" | |
928 | : S_ISCHR(s.st_mode) ? "character device" | |
929 | : S_ISBLK(s.st_mode) ? "block device" | |
930 | : S_ISREG(s.st_mode) ? "file" | |
931 | : S_ISFIFO(s.st_mode) ? "FIFO" | |
932 | : S_ISLNK(s.st_mode) ? "symbolic link" | |
933 | : "unknown")); | |
2f51a7eb | 934 | #ifdef __linux__ |
f89ffb0e BP |
935 | put_fd_filename(&string, fd); |
936 | #endif | |
937 | } | |
54a1cfb5 GS |
938 | #else |
939 | ds_put_format(&string,"file descriptor"); | |
940 | #endif /* _WIN32 */ | |
f89ffb0e BP |
941 | return ds_steal_cstr(&string); |
942 | } | |
fd94a42c | 943 | |
e731d71b AS |
944 | \f |
945 | /* sockaddr_storage helpers. */ | |
259e0b1a | 946 | |
e731d71b AS |
947 | /* Returns the IPv4 or IPv6 port in 'ss'. */ |
948 | uint16_t | |
949 | ss_get_port(const struct sockaddr_storage *ss) | |
950 | { | |
951 | if (ss->ss_family == AF_INET) { | |
952 | const struct sockaddr_in *sin | |
953 | = ALIGNED_CAST(const struct sockaddr_in *, ss); | |
954 | return ntohs(sin->sin_port); | |
955 | } else if (ss->ss_family == AF_INET6) { | |
956 | const struct sockaddr_in6 *sin6 | |
957 | = ALIGNED_CAST(const struct sockaddr_in6 *, ss); | |
958 | return ntohs(sin6->sin6_port); | |
959 | } else { | |
960 | OVS_NOT_REACHED(); | |
961 | } | |
962 | } | |
963 | ||
964 | /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'. | |
965 | * If 'ss' is an IPv6 address, puts square brackets around the address. | |
966 | * 'bufsize' should be at least SS_NTOP_BUFSIZE. | |
967 | * | |
968 | * Returns 'buf'. */ | |
969 | char * | |
970 | ss_format_address(const struct sockaddr_storage *ss, | |
971 | char *buf, size_t bufsize) | |
972 | { | |
973 | ovs_assert(bufsize >= SS_NTOP_BUFSIZE); | |
974 | if (ss->ss_family == AF_INET) { | |
975 | const struct sockaddr_in *sin | |
976 | = ALIGNED_CAST(const struct sockaddr_in *, ss); | |
977 | ||
978 | snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr)); | |
979 | } else if (ss->ss_family == AF_INET6) { | |
980 | const struct sockaddr_in6 *sin6 | |
981 | = ALIGNED_CAST(const struct sockaddr_in6 *, ss); | |
982 | ||
983 | buf[0] = '['; | |
984 | inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1); | |
985 | strcpy(strchr(buf, '\0'), "]"); | |
986 | } else { | |
987 | OVS_NOT_REACHED(); | |
988 | } | |
989 | ||
990 | return buf; | |
991 | } | |
992 | ||
993 | size_t | |
994 | ss_length(const struct sockaddr_storage *ss) | |
995 | { | |
996 | switch (ss->ss_family) { | |
997 | case AF_INET: | |
998 | return sizeof(struct sockaddr_in); | |
999 | ||
1000 | case AF_INET6: | |
1001 | return sizeof(struct sockaddr_in6); | |
1002 | ||
1003 | default: | |
1004 | OVS_NOT_REACHED(); | |
1005 | } | |
1006 | } | |
b26f46a4 GS |
1007 | |
1008 | /* For Windows socket calls, 'errno' is not set. One has to call | |
1009 | * WSAGetLastError() to get the error number and then pass it to | |
315ea327 GS |
1010 | * this function to get the correct error string. |
1011 | * | |
b26f46a4 GS |
1012 | * ovs_strerror() calls strerror_r() and would not get the correct error |
1013 | * string for Windows sockets, but is good for POSIX. */ | |
1014 | const char * | |
1015 | sock_strerror(int error) | |
1016 | { | |
1017 | #ifdef _WIN32 | |
315ea327 | 1018 | return ovs_format_message(error); |
b26f46a4 GS |
1019 | #else |
1020 | return ovs_strerror(error); | |
1021 | #endif | |
1022 | } |