1 /* Host name resolution and matching.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
39 # include <sys/socket.h>
40 # include <netinet/in.h>
42 # include <arpa/inet.h>
45 # define SET_H_ERRNO(err) ((void)(h_errno = (err)))
47 # define SET_H_ERRNO(err) WSASetLastError (err)
58 # define NO_ADDRESS NO_DATA
61 #if !HAVE_DECL_H_ERRNO
66 /* Lists of IP addresses that result from running DNS queries. See
67 lookup_host for details. */
70 int count; /* number of adrresses */
71 ip_address *addresses; /* pointer to the string of addresses */
73 int faulty; /* number of addresses known not to work. */
74 bool connected; /* whether we were able to connect to
75 one of the addresses in the list,
78 int refcount; /* reference count; when it drops to
79 0, the entry is freed. */
82 /* Get the bounds of the address list. */
85 address_list_get_bounds (const struct address_list *al, int *start, int *end)
91 /* Return a pointer to the address at position POS. */
94 address_list_address_at (const struct address_list *al, int pos)
96 assert (pos >= al->faulty && pos < al->count);
97 return al->addresses + pos;
100 /* Return true if AL contains IP, false otherwise. */
103 address_list_contains (const struct address_list *al, const ip_address *ip)
109 for (i = 0; i < al->count; i++)
111 ip_address *cur = al->addresses + i;
112 if (cur->family == AF_INET
113 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
119 for (i = 0; i < al->count; i++)
121 ip_address *cur = al->addresses + i;
122 if (cur->family == AF_INET6
123 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
124 && cur->ipv6_scope == ip->ipv6_scope
126 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
130 #endif /* ENABLE_IPV6 */
136 /* Mark the INDEXth element of AL as faulty, so that the next time
137 this address list is used, the faulty element will be skipped. */
140 address_list_set_faulty (struct address_list *al, int index)
142 /* We assume that the address list is traversed in order, so that a
143 "faulty" attempt is always preceded with all-faulty addresses,
144 and this is how Wget uses it. */
145 assert (index == al->faulty);
148 if (al->faulty >= al->count)
149 /* All addresses have been proven faulty. Since there's not much
150 sense in returning the user an empty address list the next
151 time, we'll rather make them all clean, so that they can be
156 /* Set the "connected" flag to true. This flag used by connect.c to
157 see if the host perhaps needs to be resolved again. */
160 address_list_set_connected (struct address_list *al)
162 al->connected = true;
165 /* Return the value of the "connected" flag. */
168 address_list_connected_p (const struct address_list *al)
170 return al->connected;
175 /* Create an address_list from the addresses in the given struct
178 static struct address_list *
179 address_list_from_addrinfo (const struct addrinfo *ai)
181 struct address_list *al;
182 const struct addrinfo *ptr;
187 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
188 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
193 al = xnew0 (struct address_list);
194 al->addresses = xnew_array (ip_address, cnt);
199 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
200 if (ptr->ai_family == AF_INET6)
202 const struct sockaddr_in6 *sin6 =
203 (const struct sockaddr_in6 *)ptr->ai_addr;
204 ip->family = AF_INET6;
205 ip->data.d6 = sin6->sin6_addr;
206 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
207 ip->ipv6_scope = sin6->sin6_scope_id;
211 else if (ptr->ai_family == AF_INET)
213 const struct sockaddr_in *sin =
214 (const struct sockaddr_in *)ptr->ai_addr;
215 ip->family = AF_INET;
216 ip->data.d4 = sin->sin_addr;
219 assert (ip - al->addresses == cnt);
223 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
225 /* Compare two IP addresses by family, giving preference to the IPv4
226 address (sorting it first). In other words, return -1 if ADDR1 is
227 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
230 This is intended to be used as the comparator arg to a qsort-like
231 sorting function, which is why it accepts generic pointers. */
234 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
236 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
239 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
241 /* Like the above, but give preference to the IPv6 address. */
244 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
246 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
249 #else /* not ENABLE_IPV6 */
251 /* Create an address_list from a NULL-terminated vector of IPv4
252 addresses. This kind of vector is returned by gethostbyname. */
254 static struct address_list *
255 address_list_from_ipv4_addresses (char **vec)
258 struct address_list *al = xnew0 (struct address_list);
265 al->addresses = xnew_array (ip_address, count);
269 for (i = 0; i < count; i++)
271 ip_address *ip = &al->addresses[i];
272 ip->family = AF_INET;
273 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
279 #endif /* not ENABLE_IPV6 */
282 address_list_delete (struct address_list *al)
284 xfree (al->addresses);
288 /* Mark the address list as being no longer in use. This will reduce
289 its reference count which will cause the list to be freed when the
293 address_list_release (struct address_list *al)
296 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
298 if (al->refcount <= 0)
300 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
301 address_list_delete (al);
305 /* Versions of gethostbyname and getaddrinfo that support timeout. */
309 struct ghbnwt_context {
310 const char *host_name;
311 struct hostent *hptr;
315 gethostbyname_with_timeout_callback (void *arg)
317 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
318 ctx->hptr = gethostbyname (ctx->host_name);
321 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
322 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
323 The function makes sure that when NULL is returned for reasons
324 other than timeout, errno is reset. */
326 static struct hostent *
327 gethostbyname_with_timeout (const char *host_name, double timeout)
329 struct ghbnwt_context ctx;
330 ctx.host_name = host_name;
331 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
333 SET_H_ERRNO (HOST_NOT_FOUND);
342 /* Print error messages for host errors. */
344 host_errstr (int error)
346 /* Can't use switch since some of these constants can be equal,
347 which makes the compiler complain about duplicate case
349 if (error == HOST_NOT_FOUND
350 || error == NO_RECOVERY
352 || error == NO_ADDRESS)
353 return _("Unknown host");
354 else if (error == TRY_AGAIN)
355 /* Message modeled after what gai_strerror returns in similar
357 return _("Temporary failure in name resolution");
359 return _("Unknown error");
362 #else /* ENABLE_IPV6 */
364 struct gaiwt_context {
367 const struct addrinfo *hints;
368 struct addrinfo **res;
373 getaddrinfo_with_timeout_callback (void *arg)
375 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
376 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
379 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
380 In case of timeout, the EAI_SYSTEM error code is returned and errno
381 is set to ETIMEDOUT. */
384 getaddrinfo_with_timeout (const char *node, const char *service,
385 const struct addrinfo *hints, struct addrinfo **res,
388 struct gaiwt_context ctx;
390 ctx.service = service;
394 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
399 return ctx.exit_code;
402 #endif /* ENABLE_IPV6 */
404 /* Return a textual representation of ADDR, i.e. the dotted quad for
405 IPv4 addresses, and the colon-separated list of hex words (with all
406 zeros omitted, etc.) for IPv6 addresses. */
409 print_address (const ip_address *addr)
413 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
414 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
417 return inet_ntoa (addr->data.d4);
421 /* The following two functions were adapted from glibc's
422 implementation of inet_pton, written by Paul Vixie. */
425 is_valid_ipv4_address (const char *str, const char *end)
427 bool saw_digit = false;
435 if (ch >= '0' && ch <= '9')
437 val = val * 10 + (ch - '0');
448 else if (ch == '.' && saw_digit)
465 is_valid_ipv6_address (const char *str, const char *end)
467 /* Use lower-case for these to avoid clash with system headers. */
486 /* Leading :: requires some special handling. */
490 if (str == end || *str != ':')
502 /* if ch is a number, add it to val. */
506 val |= XDIGIT_TO_NUM (ch);
513 /* if ch is a colon ... */
526 if (tp > ns_in6addrsz - ns_int16sz)
534 /* if ch is a dot ... */
535 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
536 && is_valid_ipv4_address (curtok, end) == 1)
548 if (tp > ns_in6addrsz - ns_int16sz)
555 if (tp == ns_in6addrsz)
560 if (tp != ns_in6addrsz)
566 /* Simple host cache, used by lookup_host to speed up resolving. The
567 cache doesn't handle TTL because Wget is a fairly short-lived
568 application. Refreshing is attempted when connect fails, though --
569 see connect_to_host. */
571 /* Mapping between known hosts and to lists of their addresses. */
572 static struct hash_table *host_name_addresses_map;
575 /* Return the host's resolved addresses from the cache, if
578 static struct address_list *
579 cache_query (const char *host)
581 struct address_list *al;
582 if (!host_name_addresses_map)
584 al = hash_table_get (host_name_addresses_map, host);
587 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
594 /* Cache the DNS lookup of HOST. Subsequent invocations of
595 lookup_host will return the cached value. */
598 cache_store (const char *host, struct address_list *al)
600 if (!host_name_addresses_map)
601 host_name_addresses_map = make_nocase_string_hash_table (0);
604 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
609 debug_logprintf ("Caching %s =>", host);
610 for (i = 0; i < al->count; i++)
611 debug_logprintf (" %s", print_address (al->addresses + i));
612 debug_logprintf ("\n");
616 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
620 cache_remove (const char *host)
622 struct address_list *al;
623 if (!host_name_addresses_map)
625 al = hash_table_get (host_name_addresses_map, host);
628 address_list_release (al);
629 hash_table_remove (host_name_addresses_map, host);
633 /* Look up HOST in DNS and return a list of IP addresses.
635 This function caches its result so that, if the same host is passed
636 the second time, the addresses are returned without DNS lookup.
637 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
638 globally disable caching.)
640 The order of the returned addresses is affected by the setting of
641 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
642 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
643 at the beginning; otherwise, the order is left intact. The
644 relative order of addresses with the same family is left
645 undisturbed in either case.
647 FLAGS can be a combination of:
648 LH_SILENT - don't print the "resolving ... done" messages.
649 LH_BIND - resolve addresses for use with bind, which under
650 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
651 Passive lookups are not cached under IPv6.
652 LH_REFRESH - if HOST is cached, remove the entry from the cache
653 and resolve it anew. */
655 struct address_list *
656 lookup_host (const char *host, int flags)
658 struct address_list *al;
659 bool silent = !!(flags & LH_SILENT);
661 bool numeric_address = false;
662 double timeout = opt.dns_timeout;
665 /* If we're not using getaddrinfo, first check if HOST specifies a
666 numeric IPv4 address. Some implementations of gethostbyname
667 (e.g. the Ultrix one and possibly Winsock) don't accept
668 dotted-decimal IPv4 addresses. */
670 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
671 if (addr_ipv4 != (uint32_t) -1)
673 /* No need to cache host->addr relation, just return the
676 vec[0] = (char *)&addr_ipv4;
678 return address_list_from_ipv4_addresses (vec);
681 #else /* ENABLE_IPV6 */
682 /* If we're using getaddrinfo, at least check whether the address is
683 already numeric, in which case there is no need to print the
684 "Resolving..." output. (This comes at no additional cost since
685 the is_valid_ipv*_address are already required for
688 const char *end = host + strlen (host);
689 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
690 numeric_address = true;
694 /* Cache is normally on, but can be turned off with --no-dns-cache.
695 Don't cache passive lookups under IPv6. */
696 use_cache = opt.dns_cache;
698 if ((flags & LH_BIND) || numeric_address)
702 /* Try to find the host in the cache so we don't need to talk to the
703 resolver. If LH_REFRESH is requested, remove HOST from the cache
707 if (!(flags & LH_REFRESH))
709 al = cache_query (host);
717 /* No luck with the cache; resolve HOST. */
719 if (!silent && !numeric_address)
720 logprintf (LOG_VERBOSE, _("Resolving %s... "),
721 quotearg_style (escape_quoting_style, host));
726 struct addrinfo hints, *res;
729 hints.ai_socktype = SOCK_STREAM;
731 hints.ai_family = AF_INET;
732 else if (opt.ipv6_only)
733 hints.ai_family = AF_INET6;
735 /* We tried using AI_ADDRCONFIG, but removed it because: it
736 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
737 it's unneeded since we sort the addresses anyway. */
738 hints.ai_family = AF_UNSPEC;
741 hints.ai_flags |= AI_PASSIVE;
743 #ifdef AI_NUMERICHOST
746 /* Where available, the AI_NUMERICHOST hint can prevent costly
747 access to DNS servers. */
748 hints.ai_flags |= AI_NUMERICHOST;
749 timeout = 0; /* no timeout needed when "resolving"
750 numeric hosts -- avoid setting up
751 signal handlers and such. */
755 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
756 if (err != 0 || res == NULL)
759 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
760 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
763 al = address_list_from_addrinfo (res);
767 logprintf (LOG_VERBOSE,
768 _("failed: No IPv4/IPv6 addresses for host.\n"));
772 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
773 --prefer-family) come first. Sorting is stable so the order of
774 the addresses with the same family is undisturbed. */
775 if (al->count > 1 && opt.prefer_family != prefer_none)
776 stable_sort (al->addresses, al->count, sizeof (ip_address),
777 opt.prefer_family == prefer_ipv4
778 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
780 #else /* not ENABLE_IPV6 */
782 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
787 if (errno != ETIMEDOUT)
788 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
789 host_errstr (h_errno));
791 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
795 /* Do older systems have h_addr_list? */
796 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
798 #endif /* not ENABLE_IPV6 */
800 /* Print the addresses determined by DNS lookup, but no more than
802 if (!silent && !numeric_address)
805 int printmax = al->count <= 3 ? al->count : 3;
806 for (i = 0; i < printmax; i++)
808 logputs (LOG_VERBOSE, print_address (al->addresses + i));
809 if (i < printmax - 1)
810 logputs (LOG_VERBOSE, ", ");
812 if (printmax != al->count)
813 logputs (LOG_VERBOSE, ", ...");
814 logputs (LOG_VERBOSE, "\n");
817 /* Cache the lookup information. */
819 cache_store (host, al);
824 /* Determine whether a URL is acceptable to be followed, according to
825 a list of domains to accept. */
827 accept_domain (struct url *u)
829 assert (u->host != NULL);
832 if (!sufmatch ((const char **)opt.domains, u->host))
835 if (opt.exclude_domains)
837 if (sufmatch ((const char **)opt.exclude_domains, u->host))
843 /* Check whether WHAT is matched in LIST, each element of LIST being a
844 pattern to match WHAT against, using backward matching (see
845 match_backwards() in utils.c).
847 If an element of LIST matched, 1 is returned, 0 otherwise. */
849 sufmatch (const char **list, const char *what)
854 for (i = 0; list[i]; i++)
856 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
857 if (c_tolower (list[i][j]) != c_tolower (what[k]))
859 /* The domain must be first to reach to beginning. */
869 if (host_name_addresses_map)
871 hash_table_iterator iter;
872 for (hash_table_iterate (host_name_addresses_map, &iter);
873 hash_table_iter_next (&iter);
876 char *host = iter.key;
877 struct address_list *al = iter.value;
879 assert (al->refcount == 1);
880 address_list_delete (al);
882 hash_table_destroy (host_name_addresses_map);
883 host_name_addresses_map = NULL;