1 /* Host name resolution and matching.
2 Copyright (C) 1996-2006 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget. If not, see <http://www.gnu.org/licenses/>.
19 In addition, as a special exception, the Free Software Foundation
20 gives permission to link the code of its release of Wget with the
21 OpenSSL project's "OpenSSL" library (or with modified versions of it
22 that use the same license as the "OpenSSL" library), and distribute
23 the linked executables. You must obey the GNU General Public License
24 in all respects for all of the code used other than "OpenSSL". If you
25 modify this file, you may extend this exception to your version of the
26 file, but you are not obligated to do so. If you do not wish to do
27 so, delete this exception statement from your version. */
37 # include <sys/socket.h>
38 # include <netinet/in.h>
40 # include <arpa/inet.h>
43 # define SET_H_ERRNO(err) ((void)(h_errno = (err)))
45 # define SET_H_ERRNO(err) WSASetLastError (err)
57 # define NO_ADDRESS NO_DATA
60 /* Lists of IP addresses that result from running DNS queries. See
61 lookup_host for details. */
64 int count; /* number of adrresses */
65 ip_address *addresses; /* pointer to the string of addresses */
67 int faulty; /* number of addresses known not to work. */
68 bool connected; /* whether we were able to connect to
69 one of the addresses in the list,
72 int refcount; /* reference count; when it drops to
73 0, the entry is freed. */
76 /* Get the bounds of the address list. */
79 address_list_get_bounds (const struct address_list *al, int *start, int *end)
85 /* Return a pointer to the address at position POS. */
88 address_list_address_at (const struct address_list *al, int pos)
90 assert (pos >= al->faulty && pos < al->count);
91 return al->addresses + pos;
94 /* Return true if AL contains IP, false otherwise. */
97 address_list_contains (const struct address_list *al, const ip_address *ip)
103 for (i = 0; i < al->count; i++)
105 ip_address *cur = al->addresses + i;
106 if (cur->family == AF_INET
107 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
113 for (i = 0; i < al->count; i++)
115 ip_address *cur = al->addresses + i;
116 if (cur->family == AF_INET6
117 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
118 && cur->ipv6_scope == ip->ipv6_scope
120 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
124 #endif /* ENABLE_IPV6 */
130 /* Mark the INDEXth element of AL as faulty, so that the next time
131 this address list is used, the faulty element will be skipped. */
134 address_list_set_faulty (struct address_list *al, int index)
136 /* We assume that the address list is traversed in order, so that a
137 "faulty" attempt is always preceded with all-faulty addresses,
138 and this is how Wget uses it. */
139 assert (index == al->faulty);
142 if (al->faulty >= al->count)
143 /* All addresses have been proven faulty. Since there's not much
144 sense in returning the user an empty address list the next
145 time, we'll rather make them all clean, so that they can be
150 /* Set the "connected" flag to true. This flag used by connect.c to
151 see if the host perhaps needs to be resolved again. */
154 address_list_set_connected (struct address_list *al)
156 al->connected = true;
159 /* Return the value of the "connected" flag. */
162 address_list_connected_p (const struct address_list *al)
164 return al->connected;
169 /* Create an address_list from the addresses in the given struct
172 static struct address_list *
173 address_list_from_addrinfo (const struct addrinfo *ai)
175 struct address_list *al;
176 const struct addrinfo *ptr;
181 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
182 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
187 al = xnew0 (struct address_list);
188 al->addresses = xnew_array (ip_address, cnt);
193 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
194 if (ptr->ai_family == AF_INET6)
196 const struct sockaddr_in6 *sin6 =
197 (const struct sockaddr_in6 *)ptr->ai_addr;
198 ip->family = AF_INET6;
199 ip->data.d6 = sin6->sin6_addr;
200 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
201 ip->ipv6_scope = sin6->sin6_scope_id;
205 else if (ptr->ai_family == AF_INET)
207 const struct sockaddr_in *sin =
208 (const struct sockaddr_in *)ptr->ai_addr;
209 ip->family = AF_INET;
210 ip->data.d4 = sin->sin_addr;
213 assert (ip - al->addresses == cnt);
217 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
219 /* Compare two IP addresses by family, giving preference to the IPv4
220 address (sorting it first). In other words, return -1 if ADDR1 is
221 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
224 This is intended to be used as the comparator arg to a qsort-like
225 sorting function, which is why it accepts generic pointers. */
228 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
230 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
233 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
235 /* Like the above, but give preference to the IPv6 address. */
238 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
240 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
243 #else /* not ENABLE_IPV6 */
245 /* Create an address_list from a NULL-terminated vector of IPv4
246 addresses. This kind of vector is returned by gethostbyname. */
248 static struct address_list *
249 address_list_from_ipv4_addresses (char **vec)
252 struct address_list *al = xnew0 (struct address_list);
259 al->addresses = xnew_array (ip_address, count);
263 for (i = 0; i < count; i++)
265 ip_address *ip = &al->addresses[i];
266 ip->family = AF_INET;
267 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
273 #endif /* not ENABLE_IPV6 */
276 address_list_delete (struct address_list *al)
278 xfree (al->addresses);
282 /* Mark the address list as being no longer in use. This will reduce
283 its reference count which will cause the list to be freed when the
287 address_list_release (struct address_list *al)
290 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
292 if (al->refcount <= 0)
294 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
295 address_list_delete (al);
299 /* Versions of gethostbyname and getaddrinfo that support timeout. */
303 struct ghbnwt_context {
304 const char *host_name;
305 struct hostent *hptr;
309 gethostbyname_with_timeout_callback (void *arg)
311 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
312 ctx->hptr = gethostbyname (ctx->host_name);
315 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
316 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
317 The function makes sure that when NULL is returned for reasons
318 other than timeout, errno is reset. */
320 static struct hostent *
321 gethostbyname_with_timeout (const char *host_name, double timeout)
323 struct ghbnwt_context ctx;
324 ctx.host_name = host_name;
325 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
327 SET_H_ERRNO (HOST_NOT_FOUND);
336 /* Print error messages for host errors. */
338 host_errstr (int error)
340 /* Can't use switch since some of these constants can be equal,
341 which makes the compiler complain about duplicate case
343 if (error == HOST_NOT_FOUND
344 || error == NO_RECOVERY
346 || error == NO_ADDRESS)
347 return _("Unknown host");
348 else if (error == TRY_AGAIN)
349 /* Message modeled after what gai_strerror returns in similar
351 return _("Temporary failure in name resolution");
353 return _("Unknown error");
356 #else /* ENABLE_IPV6 */
358 struct gaiwt_context {
361 const struct addrinfo *hints;
362 struct addrinfo **res;
367 getaddrinfo_with_timeout_callback (void *arg)
369 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
370 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
373 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
374 In case of timeout, the EAI_SYSTEM error code is returned and errno
375 is set to ETIMEDOUT. */
378 getaddrinfo_with_timeout (const char *node, const char *service,
379 const struct addrinfo *hints, struct addrinfo **res,
382 struct gaiwt_context ctx;
384 ctx.service = service;
388 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
393 return ctx.exit_code;
396 #endif /* ENABLE_IPV6 */
398 /* Return a textual representation of ADDR, i.e. the dotted quad for
399 IPv4 addresses, and the colon-separated list of hex words (with all
400 zeros omitted, etc.) for IPv6 addresses. */
403 print_address (const ip_address *addr)
407 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
408 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
411 return inet_ntoa (addr->data.d4);
415 /* The following two functions were adapted from glibc's
416 implementation of inet_pton, written by Paul Vixie. */
419 is_valid_ipv4_address (const char *str, const char *end)
421 bool saw_digit = false;
429 if (ch >= '0' && ch <= '9')
431 val = val * 10 + (ch - '0');
442 else if (ch == '.' && saw_digit)
459 is_valid_ipv6_address (const char *str, const char *end)
461 /* Use lower-case for these to avoid clash with system headers. */
480 /* Leading :: requires some special handling. */
484 if (str == end || *str != ':')
496 /* if ch is a number, add it to val. */
500 val |= XDIGIT_TO_NUM (ch);
507 /* if ch is a colon ... */
520 if (tp > ns_in6addrsz - ns_int16sz)
528 /* if ch is a dot ... */
529 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
530 && is_valid_ipv4_address (curtok, end) == 1)
542 if (tp > ns_in6addrsz - ns_int16sz)
549 if (tp == ns_in6addrsz)
554 if (tp != ns_in6addrsz)
560 /* Simple host cache, used by lookup_host to speed up resolving. The
561 cache doesn't handle TTL because Wget is a fairly short-lived
562 application. Refreshing is attempted when connect fails, though --
563 see connect_to_host. */
565 /* Mapping between known hosts and to lists of their addresses. */
566 static struct hash_table *host_name_addresses_map;
569 /* Return the host's resolved addresses from the cache, if
572 static struct address_list *
573 cache_query (const char *host)
575 struct address_list *al;
576 if (!host_name_addresses_map)
578 al = hash_table_get (host_name_addresses_map, host);
581 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
588 /* Cache the DNS lookup of HOST. Subsequent invocations of
589 lookup_host will return the cached value. */
592 cache_store (const char *host, struct address_list *al)
594 if (!host_name_addresses_map)
595 host_name_addresses_map = make_nocase_string_hash_table (0);
598 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
603 debug_logprintf ("Caching %s =>", host);
604 for (i = 0; i < al->count; i++)
605 debug_logprintf (" %s", print_address (al->addresses + i));
606 debug_logprintf ("\n");
610 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
614 cache_remove (const char *host)
616 struct address_list *al;
617 if (!host_name_addresses_map)
619 al = hash_table_get (host_name_addresses_map, host);
622 address_list_release (al);
623 hash_table_remove (host_name_addresses_map, host);
627 /* Look up HOST in DNS and return a list of IP addresses.
629 This function caches its result so that, if the same host is passed
630 the second time, the addresses are returned without DNS lookup.
631 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
632 globally disable caching.)
634 The order of the returned addresses is affected by the setting of
635 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
636 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
637 at the beginning; otherwise, the order is left intact. The
638 relative order of addresses with the same family is left
639 undisturbed in either case.
641 FLAGS can be a combination of:
642 LH_SILENT - don't print the "resolving ... done" messages.
643 LH_BIND - resolve addresses for use with bind, which under
644 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
645 Passive lookups are not cached under IPv6.
646 LH_REFRESH - if HOST is cached, remove the entry from the cache
647 and resolve it anew. */
649 struct address_list *
650 lookup_host (const char *host, int flags)
652 struct address_list *al;
653 bool silent = !!(flags & LH_SILENT);
655 bool numeric_address = false;
656 double timeout = opt.dns_timeout;
659 /* If we're not using getaddrinfo, first check if HOST specifies a
660 numeric IPv4 address. Some implementations of gethostbyname
661 (e.g. the Ultrix one and possibly Winsock) don't accept
662 dotted-decimal IPv4 addresses. */
664 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
665 if (addr_ipv4 != (uint32_t) -1)
667 /* No need to cache host->addr relation, just return the
670 vec[0] = (char *)&addr_ipv4;
672 return address_list_from_ipv4_addresses (vec);
675 #else /* ENABLE_IPV6 */
676 /* If we're using getaddrinfo, at least check whether the address is
677 already numeric, in which case there is no need to print the
678 "Resolving..." output. (This comes at no additional cost since
679 the is_valid_ipv*_address are already required for
682 const char *end = host + strlen (host);
683 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
684 numeric_address = true;
688 /* Cache is normally on, but can be turned off with --no-dns-cache.
689 Don't cache passive lookups under IPv6. */
690 use_cache = opt.dns_cache;
692 if ((flags & LH_BIND) || numeric_address)
696 /* Try to find the host in the cache so we don't need to talk to the
697 resolver. If LH_REFRESH is requested, remove HOST from the cache
701 if (!(flags & LH_REFRESH))
703 al = cache_query (host);
711 /* No luck with the cache; resolve HOST. */
713 if (!silent && !numeric_address)
714 logprintf (LOG_VERBOSE, _("Resolving %s... "), escnonprint (host));
719 struct addrinfo hints, *res;
722 hints.ai_socktype = SOCK_STREAM;
724 hints.ai_family = AF_INET;
725 else if (opt.ipv6_only)
726 hints.ai_family = AF_INET6;
728 /* We tried using AI_ADDRCONFIG, but removed it because: it
729 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
730 it's unneeded since we sort the addresses anyway. */
731 hints.ai_family = AF_UNSPEC;
734 hints.ai_flags |= AI_PASSIVE;
736 #ifdef AI_NUMERICHOST
739 /* Where available, the AI_NUMERICHOST hint can prevent costly
740 access to DNS servers. */
741 hints.ai_flags |= AI_NUMERICHOST;
742 timeout = 0; /* no timeout needed when "resolving"
743 numeric hosts -- avoid setting up
744 signal handlers and such. */
748 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
749 if (err != 0 || res == NULL)
752 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
753 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
756 al = address_list_from_addrinfo (res);
760 logprintf (LOG_VERBOSE,
761 _("failed: No IPv4/IPv6 addresses for host.\n"));
765 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
766 --prefer-family) come first. Sorting is stable so the order of
767 the addresses with the same family is undisturbed. */
768 if (al->count > 1 && opt.prefer_family != prefer_none)
769 stable_sort (al->addresses, al->count, sizeof (ip_address),
770 opt.prefer_family == prefer_ipv4
771 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
773 #else /* not ENABLE_IPV6 */
775 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
780 if (errno != ETIMEDOUT)
781 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
782 host_errstr (h_errno));
784 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
788 /* Do older systems have h_addr_list? */
789 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
791 #endif /* not ENABLE_IPV6 */
793 /* Print the addresses determined by DNS lookup, but no more than
795 if (!silent && !numeric_address)
798 int printmax = al->count <= 3 ? al->count : 3;
799 for (i = 0; i < printmax; i++)
801 logputs (LOG_VERBOSE, print_address (al->addresses + i));
802 if (i < printmax - 1)
803 logputs (LOG_VERBOSE, ", ");
805 if (printmax != al->count)
806 logputs (LOG_VERBOSE, ", ...");
807 logputs (LOG_VERBOSE, "\n");
810 /* Cache the lookup information. */
812 cache_store (host, al);
817 /* Determine whether a URL is acceptable to be followed, according to
818 a list of domains to accept. */
820 accept_domain (struct url *u)
822 assert (u->host != NULL);
825 if (!sufmatch ((const char **)opt.domains, u->host))
828 if (opt.exclude_domains)
830 if (sufmatch ((const char **)opt.exclude_domains, u->host))
836 /* Check whether WHAT is matched in LIST, each element of LIST being a
837 pattern to match WHAT against, using backward matching (see
838 match_backwards() in utils.c).
840 If an element of LIST matched, 1 is returned, 0 otherwise. */
842 sufmatch (const char **list, const char *what)
847 for (i = 0; list[i]; i++)
849 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
850 if (TOLOWER (list[i][j]) != TOLOWER (what[k]))
852 /* The domain must be first to reach to beginning. */
862 if (host_name_addresses_map)
864 hash_table_iterator iter;
865 for (hash_table_iterate (host_name_addresses_map, &iter);
866 hash_table_iter_next (&iter);
869 char *host = iter.key;
870 struct address_list *al = iter.value;
872 assert (al->refcount == 1);
873 address_list_delete (al);
875 hash_table_destroy (host_name_addresses_map);
876 host_name_addresses_map = NULL;