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 /* Lists of IP addresses that result from running DNS queries. See
62 lookup_host for details. */
65 int count; /* number of adrresses */
66 ip_address *addresses; /* pointer to the string of addresses */
68 int faulty; /* number of addresses known not to work. */
69 bool connected; /* whether we were able to connect to
70 one of the addresses in the list,
73 int refcount; /* reference count; when it drops to
74 0, the entry is freed. */
77 /* Get the bounds of the address list. */
80 address_list_get_bounds (const struct address_list *al, int *start, int *end)
86 /* Return a pointer to the address at position POS. */
89 address_list_address_at (const struct address_list *al, int pos)
91 assert (pos >= al->faulty && pos < al->count);
92 return al->addresses + pos;
95 /* Return true if AL contains IP, false otherwise. */
98 address_list_contains (const struct address_list *al, const ip_address *ip)
104 for (i = 0; i < al->count; i++)
106 ip_address *cur = al->addresses + i;
107 if (cur->family == AF_INET
108 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
114 for (i = 0; i < al->count; i++)
116 ip_address *cur = al->addresses + i;
117 if (cur->family == AF_INET6
118 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
119 && cur->ipv6_scope == ip->ipv6_scope
121 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
125 #endif /* ENABLE_IPV6 */
131 /* Mark the INDEXth element of AL as faulty, so that the next time
132 this address list is used, the faulty element will be skipped. */
135 address_list_set_faulty (struct address_list *al, int index)
137 /* We assume that the address list is traversed in order, so that a
138 "faulty" attempt is always preceded with all-faulty addresses,
139 and this is how Wget uses it. */
140 assert (index == al->faulty);
143 if (al->faulty >= al->count)
144 /* All addresses have been proven faulty. Since there's not much
145 sense in returning the user an empty address list the next
146 time, we'll rather make them all clean, so that they can be
151 /* Set the "connected" flag to true. This flag used by connect.c to
152 see if the host perhaps needs to be resolved again. */
155 address_list_set_connected (struct address_list *al)
157 al->connected = true;
160 /* Return the value of the "connected" flag. */
163 address_list_connected_p (const struct address_list *al)
165 return al->connected;
170 /* Create an address_list from the addresses in the given struct
173 static struct address_list *
174 address_list_from_addrinfo (const struct addrinfo *ai)
176 struct address_list *al;
177 const struct addrinfo *ptr;
182 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
183 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
188 al = xnew0 (struct address_list);
189 al->addresses = xnew_array (ip_address, cnt);
194 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
195 if (ptr->ai_family == AF_INET6)
197 const struct sockaddr_in6 *sin6 =
198 (const struct sockaddr_in6 *)ptr->ai_addr;
199 ip->family = AF_INET6;
200 ip->data.d6 = sin6->sin6_addr;
201 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
202 ip->ipv6_scope = sin6->sin6_scope_id;
206 else if (ptr->ai_family == AF_INET)
208 const struct sockaddr_in *sin =
209 (const struct sockaddr_in *)ptr->ai_addr;
210 ip->family = AF_INET;
211 ip->data.d4 = sin->sin_addr;
214 assert (ip - al->addresses == cnt);
218 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
220 /* Compare two IP addresses by family, giving preference to the IPv4
221 address (sorting it first). In other words, return -1 if ADDR1 is
222 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
225 This is intended to be used as the comparator arg to a qsort-like
226 sorting function, which is why it accepts generic pointers. */
229 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
231 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
234 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
236 /* Like the above, but give preference to the IPv6 address. */
239 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
241 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
244 #else /* not ENABLE_IPV6 */
246 /* Create an address_list from a NULL-terminated vector of IPv4
247 addresses. This kind of vector is returned by gethostbyname. */
249 static struct address_list *
250 address_list_from_ipv4_addresses (char **vec)
253 struct address_list *al = xnew0 (struct address_list);
260 al->addresses = xnew_array (ip_address, count);
264 for (i = 0; i < count; i++)
266 ip_address *ip = &al->addresses[i];
267 ip->family = AF_INET;
268 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
274 #endif /* not ENABLE_IPV6 */
277 address_list_delete (struct address_list *al)
279 xfree (al->addresses);
283 /* Mark the address list as being no longer in use. This will reduce
284 its reference count which will cause the list to be freed when the
288 address_list_release (struct address_list *al)
291 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
293 if (al->refcount <= 0)
295 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
296 address_list_delete (al);
300 /* Versions of gethostbyname and getaddrinfo that support timeout. */
304 struct ghbnwt_context {
305 const char *host_name;
306 struct hostent *hptr;
310 gethostbyname_with_timeout_callback (void *arg)
312 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
313 ctx->hptr = gethostbyname (ctx->host_name);
316 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
317 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
318 The function makes sure that when NULL is returned for reasons
319 other than timeout, errno is reset. */
321 static struct hostent *
322 gethostbyname_with_timeout (const char *host_name, double timeout)
324 struct ghbnwt_context ctx;
325 ctx.host_name = host_name;
326 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
328 SET_H_ERRNO (HOST_NOT_FOUND);
337 /* Print error messages for host errors. */
339 host_errstr (int error)
341 /* Can't use switch since some of these constants can be equal,
342 which makes the compiler complain about duplicate case
344 if (error == HOST_NOT_FOUND
345 || error == NO_RECOVERY
347 || error == NO_ADDRESS)
348 return _("Unknown host");
349 else if (error == TRY_AGAIN)
350 /* Message modeled after what gai_strerror returns in similar
352 return _("Temporary failure in name resolution");
354 return _("Unknown error");
357 #else /* ENABLE_IPV6 */
359 struct gaiwt_context {
362 const struct addrinfo *hints;
363 struct addrinfo **res;
368 getaddrinfo_with_timeout_callback (void *arg)
370 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
371 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
374 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
375 In case of timeout, the EAI_SYSTEM error code is returned and errno
376 is set to ETIMEDOUT. */
379 getaddrinfo_with_timeout (const char *node, const char *service,
380 const struct addrinfo *hints, struct addrinfo **res,
383 struct gaiwt_context ctx;
385 ctx.service = service;
389 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
394 return ctx.exit_code;
397 #endif /* ENABLE_IPV6 */
399 /* Return a textual representation of ADDR, i.e. the dotted quad for
400 IPv4 addresses, and the colon-separated list of hex words (with all
401 zeros omitted, etc.) for IPv6 addresses. */
404 print_address (const ip_address *addr)
408 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
409 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
412 return inet_ntoa (addr->data.d4);
416 /* The following two functions were adapted from glibc's
417 implementation of inet_pton, written by Paul Vixie. */
420 is_valid_ipv4_address (const char *str, const char *end)
422 bool saw_digit = false;
430 if (ch >= '0' && ch <= '9')
432 val = val * 10 + (ch - '0');
443 else if (ch == '.' && saw_digit)
460 is_valid_ipv6_address (const char *str, const char *end)
462 /* Use lower-case for these to avoid clash with system headers. */
481 /* Leading :: requires some special handling. */
485 if (str == end || *str != ':')
497 /* if ch is a number, add it to val. */
501 val |= XDIGIT_TO_NUM (ch);
508 /* if ch is a colon ... */
521 if (tp > ns_in6addrsz - ns_int16sz)
529 /* if ch is a dot ... */
530 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
531 && is_valid_ipv4_address (curtok, end) == 1)
543 if (tp > ns_in6addrsz - ns_int16sz)
550 if (tp == ns_in6addrsz)
555 if (tp != ns_in6addrsz)
561 /* Simple host cache, used by lookup_host to speed up resolving. The
562 cache doesn't handle TTL because Wget is a fairly short-lived
563 application. Refreshing is attempted when connect fails, though --
564 see connect_to_host. */
566 /* Mapping between known hosts and to lists of their addresses. */
567 static struct hash_table *host_name_addresses_map;
570 /* Return the host's resolved addresses from the cache, if
573 static struct address_list *
574 cache_query (const char *host)
576 struct address_list *al;
577 if (!host_name_addresses_map)
579 al = hash_table_get (host_name_addresses_map, host);
582 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
589 /* Cache the DNS lookup of HOST. Subsequent invocations of
590 lookup_host will return the cached value. */
593 cache_store (const char *host, struct address_list *al)
595 if (!host_name_addresses_map)
596 host_name_addresses_map = make_nocase_string_hash_table (0);
599 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
604 debug_logprintf ("Caching %s =>", host);
605 for (i = 0; i < al->count; i++)
606 debug_logprintf (" %s", print_address (al->addresses + i));
607 debug_logprintf ("\n");
611 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
615 cache_remove (const char *host)
617 struct address_list *al;
618 if (!host_name_addresses_map)
620 al = hash_table_get (host_name_addresses_map, host);
623 address_list_release (al);
624 hash_table_remove (host_name_addresses_map, host);
628 /* Look up HOST in DNS and return a list of IP addresses.
630 This function caches its result so that, if the same host is passed
631 the second time, the addresses are returned without DNS lookup.
632 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
633 globally disable caching.)
635 The order of the returned addresses is affected by the setting of
636 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
637 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
638 at the beginning; otherwise, the order is left intact. The
639 relative order of addresses with the same family is left
640 undisturbed in either case.
642 FLAGS can be a combination of:
643 LH_SILENT - don't print the "resolving ... done" messages.
644 LH_BIND - resolve addresses for use with bind, which under
645 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
646 Passive lookups are not cached under IPv6.
647 LH_REFRESH - if HOST is cached, remove the entry from the cache
648 and resolve it anew. */
650 struct address_list *
651 lookup_host (const char *host, int flags)
653 struct address_list *al;
654 bool silent = !!(flags & LH_SILENT);
656 bool numeric_address = false;
657 double timeout = opt.dns_timeout;
660 /* If we're not using getaddrinfo, first check if HOST specifies a
661 numeric IPv4 address. Some implementations of gethostbyname
662 (e.g. the Ultrix one and possibly Winsock) don't accept
663 dotted-decimal IPv4 addresses. */
665 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
666 if (addr_ipv4 != (uint32_t) -1)
668 /* No need to cache host->addr relation, just return the
671 vec[0] = (char *)&addr_ipv4;
673 return address_list_from_ipv4_addresses (vec);
676 #else /* ENABLE_IPV6 */
677 /* If we're using getaddrinfo, at least check whether the address is
678 already numeric, in which case there is no need to print the
679 "Resolving..." output. (This comes at no additional cost since
680 the is_valid_ipv*_address are already required for
683 const char *end = host + strlen (host);
684 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
685 numeric_address = true;
689 /* Cache is normally on, but can be turned off with --no-dns-cache.
690 Don't cache passive lookups under IPv6. */
691 use_cache = opt.dns_cache;
693 if ((flags & LH_BIND) || numeric_address)
697 /* Try to find the host in the cache so we don't need to talk to the
698 resolver. If LH_REFRESH is requested, remove HOST from the cache
702 if (!(flags & LH_REFRESH))
704 al = cache_query (host);
712 /* No luck with the cache; resolve HOST. */
714 if (!silent && !numeric_address)
716 char *str = NULL, *name;
718 if (opt.enable_iri && (name = idn_decode ((char *) host)) != NULL)
720 int len = strlen (host) + strlen (name) + 4;
722 snprintf (str, len, "%s (%s)", name, host);
727 logprintf (LOG_VERBOSE, _("Resolving %s... "),
728 quotearg_style (escape_quoting_style, str ? str : host));
737 struct addrinfo hints, *res;
740 hints.ai_socktype = SOCK_STREAM;
742 hints.ai_family = AF_INET;
743 else if (opt.ipv6_only)
744 hints.ai_family = AF_INET6;
746 /* We tried using AI_ADDRCONFIG, but removed it because: it
747 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
748 it's unneeded since we sort the addresses anyway. */
749 hints.ai_family = AF_UNSPEC;
752 hints.ai_flags |= AI_PASSIVE;
754 #ifdef AI_NUMERICHOST
757 /* Where available, the AI_NUMERICHOST hint can prevent costly
758 access to DNS servers. */
759 hints.ai_flags |= AI_NUMERICHOST;
760 timeout = 0; /* no timeout needed when "resolving"
761 numeric hosts -- avoid setting up
762 signal handlers and such. */
766 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
767 if (err != 0 || res == NULL)
770 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
771 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
774 al = address_list_from_addrinfo (res);
778 logprintf (LOG_VERBOSE,
779 _("failed: No IPv4/IPv6 addresses for host.\n"));
783 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
784 --prefer-family) come first. Sorting is stable so the order of
785 the addresses with the same family is undisturbed. */
786 if (al->count > 1 && opt.prefer_family != prefer_none)
787 stable_sort (al->addresses, al->count, sizeof (ip_address),
788 opt.prefer_family == prefer_ipv4
789 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
791 #else /* not ENABLE_IPV6 */
793 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
798 if (errno != ETIMEDOUT)
799 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
800 host_errstr (h_errno));
802 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
806 /* Do older systems have h_addr_list? */
807 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
809 #endif /* not ENABLE_IPV6 */
811 /* Print the addresses determined by DNS lookup, but no more than
813 if (!silent && !numeric_address)
816 int printmax = al->count <= 3 ? al->count : 3;
817 for (i = 0; i < printmax; i++)
819 logputs (LOG_VERBOSE, print_address (al->addresses + i));
820 if (i < printmax - 1)
821 logputs (LOG_VERBOSE, ", ");
823 if (printmax != al->count)
824 logputs (LOG_VERBOSE, ", ...");
825 logputs (LOG_VERBOSE, "\n");
828 /* Cache the lookup information. */
830 cache_store (host, al);
835 /* Determine whether a URL is acceptable to be followed, according to
836 a list of domains to accept. */
838 accept_domain (struct url *u)
840 assert (u->host != NULL);
843 if (!sufmatch ((const char **)opt.domains, u->host))
846 if (opt.exclude_domains)
848 if (sufmatch ((const char **)opt.exclude_domains, u->host))
854 /* Check whether WHAT is matched in LIST, each element of LIST being a
855 pattern to match WHAT against, using backward matching (see
856 match_backwards() in utils.c).
858 If an element of LIST matched, 1 is returned, 0 otherwise. */
860 sufmatch (const char **list, const char *what)
865 for (i = 0; list[i]; i++)
867 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
868 if (c_tolower (list[i][j]) != c_tolower (what[k]))
870 /* The domain must be first to reach to beginning. */
880 if (host_name_addresses_map)
882 hash_table_iterator iter;
883 for (hash_table_iterate (host_name_addresses_map, &iter);
884 hash_table_iter_next (&iter);
887 char *host = iter.key;
888 struct address_list *al = iter.value;
890 assert (al->refcount == 1);
891 address_list_delete (al);
893 hash_table_destroy (host_name_addresses_map);
894 host_name_addresses_map = NULL;