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>
46 # else /* def __VMS */
48 # endif /* def __VMS [else] */
49 # define SET_H_ERRNO(err) ((void)(h_errno = (err)))
51 # define SET_H_ERRNO(err) WSASetLastError (err)
62 # define NO_ADDRESS NO_DATA
65 /* Lists of IP addresses that result from running DNS queries. See
66 lookup_host for details. */
69 int count; /* number of adrresses */
70 ip_address *addresses; /* pointer to the string of addresses */
72 int faulty; /* number of addresses known not to work. */
73 bool connected; /* whether we were able to connect to
74 one of the addresses in the list,
77 int refcount; /* reference count; when it drops to
78 0, the entry is freed. */
81 /* Get the bounds of the address list. */
84 address_list_get_bounds (const struct address_list *al, int *start, int *end)
90 /* Return a pointer to the address at position POS. */
93 address_list_address_at (const struct address_list *al, int pos)
95 assert (pos >= al->faulty && pos < al->count);
96 return al->addresses + pos;
99 /* Return true if AL contains IP, false otherwise. */
102 address_list_contains (const struct address_list *al, const ip_address *ip)
108 for (i = 0; i < al->count; i++)
110 ip_address *cur = al->addresses + i;
111 if (cur->family == AF_INET
112 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
118 for (i = 0; i < al->count; i++)
120 ip_address *cur = al->addresses + i;
121 if (cur->family == AF_INET6
122 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
123 && cur->ipv6_scope == ip->ipv6_scope
125 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
129 #endif /* ENABLE_IPV6 */
135 /* Mark the INDEXth element of AL as faulty, so that the next time
136 this address list is used, the faulty element will be skipped. */
139 address_list_set_faulty (struct address_list *al, int index)
141 /* We assume that the address list is traversed in order, so that a
142 "faulty" attempt is always preceded with all-faulty addresses,
143 and this is how Wget uses it. */
144 assert (index == al->faulty);
147 if (al->faulty >= al->count)
148 /* All addresses have been proven faulty. Since there's not much
149 sense in returning the user an empty address list the next
150 time, we'll rather make them all clean, so that they can be
155 /* Set the "connected" flag to true. This flag used by connect.c to
156 see if the host perhaps needs to be resolved again. */
159 address_list_set_connected (struct address_list *al)
161 al->connected = true;
164 /* Return the value of the "connected" flag. */
167 address_list_connected_p (const struct address_list *al)
169 return al->connected;
174 /* Create an address_list from the addresses in the given struct
177 static struct address_list *
178 address_list_from_addrinfo (const struct addrinfo *ai)
180 struct address_list *al;
181 const struct addrinfo *ptr;
186 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
187 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
192 al = xnew0 (struct address_list);
193 al->addresses = xnew_array (ip_address, cnt);
198 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
199 if (ptr->ai_family == AF_INET6)
201 const struct sockaddr_in6 *sin6 =
202 (const struct sockaddr_in6 *)ptr->ai_addr;
203 ip->family = AF_INET6;
204 ip->data.d6 = sin6->sin6_addr;
205 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
206 ip->ipv6_scope = sin6->sin6_scope_id;
210 else if (ptr->ai_family == AF_INET)
212 const struct sockaddr_in *sin =
213 (const struct sockaddr_in *)ptr->ai_addr;
214 ip->family = AF_INET;
215 ip->data.d4 = sin->sin_addr;
218 assert (ip - al->addresses == cnt);
222 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
224 /* Compare two IP addresses by family, giving preference to the IPv4
225 address (sorting it first). In other words, return -1 if ADDR1 is
226 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
229 This is intended to be used as the comparator arg to a qsort-like
230 sorting function, which is why it accepts generic pointers. */
233 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
235 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
238 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
240 /* Like the above, but give preference to the IPv6 address. */
243 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
245 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
248 #else /* not ENABLE_IPV6 */
250 /* Create an address_list from a NULL-terminated vector of IPv4
251 addresses. This kind of vector is returned by gethostbyname. */
253 static struct address_list *
254 address_list_from_ipv4_addresses (char **vec)
257 struct address_list *al = xnew0 (struct address_list);
264 al->addresses = xnew_array (ip_address, count);
268 for (i = 0; i < count; i++)
270 ip_address *ip = &al->addresses[i];
271 ip->family = AF_INET;
272 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
278 #endif /* not ENABLE_IPV6 */
281 address_list_delete (struct address_list *al)
283 xfree (al->addresses);
287 /* Mark the address list as being no longer in use. This will reduce
288 its reference count which will cause the list to be freed when the
292 address_list_release (struct address_list *al)
295 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
297 if (al->refcount <= 0)
299 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
300 address_list_delete (al);
304 /* Versions of gethostbyname and getaddrinfo that support timeout. */
308 struct ghbnwt_context {
309 const char *host_name;
310 struct hostent *hptr;
314 gethostbyname_with_timeout_callback (void *arg)
316 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
317 ctx->hptr = gethostbyname (ctx->host_name);
320 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
321 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
322 The function makes sure that when NULL is returned for reasons
323 other than timeout, errno is reset. */
325 static struct hostent *
326 gethostbyname_with_timeout (const char *host_name, double timeout)
328 struct ghbnwt_context ctx;
329 ctx.host_name = host_name;
330 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
332 SET_H_ERRNO (HOST_NOT_FOUND);
341 /* Print error messages for host errors. */
343 host_errstr (int error)
345 /* Can't use switch since some of these constants can be equal,
346 which makes the compiler complain about duplicate case
348 if (error == HOST_NOT_FOUND
349 || error == NO_RECOVERY
351 || error == NO_ADDRESS)
352 return _("Unknown host");
353 else if (error == TRY_AGAIN)
354 /* Message modeled after what gai_strerror returns in similar
356 return _("Temporary failure in name resolution");
358 return _("Unknown error");
361 #else /* ENABLE_IPV6 */
363 struct gaiwt_context {
366 const struct addrinfo *hints;
367 struct addrinfo **res;
372 getaddrinfo_with_timeout_callback (void *arg)
374 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
375 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
378 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
379 In case of timeout, the EAI_SYSTEM error code is returned and errno
380 is set to ETIMEDOUT. */
383 getaddrinfo_with_timeout (const char *node, const char *service,
384 const struct addrinfo *hints, struct addrinfo **res,
387 struct gaiwt_context ctx;
389 ctx.service = service;
393 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
398 return ctx.exit_code;
401 #endif /* ENABLE_IPV6 */
403 /* Return a textual representation of ADDR, i.e. the dotted quad for
404 IPv4 addresses, and the colon-separated list of hex words (with all
405 zeros omitted, etc.) for IPv6 addresses. */
408 print_address (const ip_address *addr)
412 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
413 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
416 return inet_ntoa (addr->data.d4);
420 /* The following two functions were adapted from glibc's
421 implementation of inet_pton, written by Paul Vixie. */
424 is_valid_ipv4_address (const char *str, const char *end)
426 bool saw_digit = false;
434 if (ch >= '0' && ch <= '9')
436 val = val * 10 + (ch - '0');
447 else if (ch == '.' && saw_digit)
464 is_valid_ipv6_address (const char *str, const char *end)
466 /* Use lower-case for these to avoid clash with system headers. */
485 /* Leading :: requires some special handling. */
489 if (str == end || *str != ':')
501 /* if ch is a number, add it to val. */
505 val |= XDIGIT_TO_NUM (ch);
512 /* if ch is a colon ... */
525 if (tp > ns_in6addrsz - ns_int16sz)
533 /* if ch is a dot ... */
534 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
535 && is_valid_ipv4_address (curtok, end) == 1)
547 if (tp > ns_in6addrsz - ns_int16sz)
554 if (tp == ns_in6addrsz)
559 if (tp != ns_in6addrsz)
565 /* Simple host cache, used by lookup_host to speed up resolving. The
566 cache doesn't handle TTL because Wget is a fairly short-lived
567 application. Refreshing is attempted when connect fails, though --
568 see connect_to_host. */
570 /* Mapping between known hosts and to lists of their addresses. */
571 static struct hash_table *host_name_addresses_map;
574 /* Return the host's resolved addresses from the cache, if
577 static struct address_list *
578 cache_query (const char *host)
580 struct address_list *al;
581 if (!host_name_addresses_map)
583 al = hash_table_get (host_name_addresses_map, host);
586 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
593 /* Cache the DNS lookup of HOST. Subsequent invocations of
594 lookup_host will return the cached value. */
597 cache_store (const char *host, struct address_list *al)
599 if (!host_name_addresses_map)
600 host_name_addresses_map = make_nocase_string_hash_table (0);
603 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
608 debug_logprintf ("Caching %s =>", host);
609 for (i = 0; i < al->count; i++)
610 debug_logprintf (" %s", print_address (al->addresses + i));
611 debug_logprintf ("\n");
615 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
619 cache_remove (const char *host)
621 struct address_list *al;
622 if (!host_name_addresses_map)
624 al = hash_table_get (host_name_addresses_map, host);
627 address_list_release (al);
628 hash_table_remove (host_name_addresses_map, host);
632 /* Look up HOST in DNS and return a list of IP addresses.
634 This function caches its result so that, if the same host is passed
635 the second time, the addresses are returned without DNS lookup.
636 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
637 globally disable caching.)
639 The order of the returned addresses is affected by the setting of
640 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
641 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
642 at the beginning; otherwise, the order is left intact. The
643 relative order of addresses with the same family is left
644 undisturbed in either case.
646 FLAGS can be a combination of:
647 LH_SILENT - don't print the "resolving ... done" messages.
648 LH_BIND - resolve addresses for use with bind, which under
649 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
650 Passive lookups are not cached under IPv6.
651 LH_REFRESH - if HOST is cached, remove the entry from the cache
652 and resolve it anew. */
654 struct address_list *
655 lookup_host (const char *host, int flags)
657 struct address_list *al;
658 bool silent = !!(flags & LH_SILENT);
660 bool numeric_address = false;
661 double timeout = opt.dns_timeout;
664 /* If we're not using getaddrinfo, first check if HOST specifies a
665 numeric IPv4 address. Some implementations of gethostbyname
666 (e.g. the Ultrix one and possibly Winsock) don't accept
667 dotted-decimal IPv4 addresses. */
669 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
670 if (addr_ipv4 != (uint32_t) -1)
672 /* No need to cache host->addr relation, just return the
675 vec[0] = (char *)&addr_ipv4;
677 return address_list_from_ipv4_addresses (vec);
680 #else /* ENABLE_IPV6 */
681 /* If we're using getaddrinfo, at least check whether the address is
682 already numeric, in which case there is no need to print the
683 "Resolving..." output. (This comes at no additional cost since
684 the is_valid_ipv*_address are already required for
687 const char *end = host + strlen (host);
688 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
689 numeric_address = true;
693 /* Cache is normally on, but can be turned off with --no-dns-cache.
694 Don't cache passive lookups under IPv6. */
695 use_cache = opt.dns_cache;
697 if ((flags & LH_BIND) || numeric_address)
701 /* Try to find the host in the cache so we don't need to talk to the
702 resolver. If LH_REFRESH is requested, remove HOST from the cache
706 if (!(flags & LH_REFRESH))
708 al = cache_query (host);
716 /* No luck with the cache; resolve HOST. */
718 if (!silent && !numeric_address)
719 logprintf (LOG_VERBOSE, _("Resolving %s... "), escnonprint (host));
724 struct addrinfo hints, *res;
727 hints.ai_socktype = SOCK_STREAM;
729 hints.ai_family = AF_INET;
730 else if (opt.ipv6_only)
731 hints.ai_family = AF_INET6;
733 /* We tried using AI_ADDRCONFIG, but removed it because: it
734 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
735 it's unneeded since we sort the addresses anyway. */
736 hints.ai_family = AF_UNSPEC;
739 hints.ai_flags |= AI_PASSIVE;
741 #ifdef AI_NUMERICHOST
744 /* Where available, the AI_NUMERICHOST hint can prevent costly
745 access to DNS servers. */
746 hints.ai_flags |= AI_NUMERICHOST;
747 timeout = 0; /* no timeout needed when "resolving"
748 numeric hosts -- avoid setting up
749 signal handlers and such. */
753 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
754 if (err != 0 || res == NULL)
757 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
758 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
761 al = address_list_from_addrinfo (res);
765 logprintf (LOG_VERBOSE,
766 _("failed: No IPv4/IPv6 addresses for host.\n"));
770 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
771 --prefer-family) come first. Sorting is stable so the order of
772 the addresses with the same family is undisturbed. */
773 if (al->count > 1 && opt.prefer_family != prefer_none)
774 stable_sort (al->addresses, al->count, sizeof (ip_address),
775 opt.prefer_family == prefer_ipv4
776 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
778 #else /* not ENABLE_IPV6 */
780 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
785 if (errno != ETIMEDOUT)
786 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
787 host_errstr (h_errno));
789 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
793 /* Do older systems have h_addr_list? */
794 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
796 #endif /* not ENABLE_IPV6 */
798 /* Print the addresses determined by DNS lookup, but no more than
800 if (!silent && !numeric_address)
803 int printmax = al->count <= 3 ? al->count : 3;
804 for (i = 0; i < printmax; i++)
806 logputs (LOG_VERBOSE, print_address (al->addresses + i));
807 if (i < printmax - 1)
808 logputs (LOG_VERBOSE, ", ");
810 if (printmax != al->count)
811 logputs (LOG_VERBOSE, ", ...");
812 logputs (LOG_VERBOSE, "\n");
815 /* Cache the lookup information. */
817 cache_store (host, al);
822 /* Determine whether a URL is acceptable to be followed, according to
823 a list of domains to accept. */
825 accept_domain (struct url *u)
827 assert (u->host != NULL);
830 if (!sufmatch ((const char **)opt.domains, u->host))
833 if (opt.exclude_domains)
835 if (sufmatch ((const char **)opt.exclude_domains, u->host))
841 /* Check whether WHAT is matched in LIST, each element of LIST being a
842 pattern to match WHAT against, using backward matching (see
843 match_backwards() in utils.c).
845 If an element of LIST matched, 1 is returned, 0 otherwise. */
847 sufmatch (const char **list, const char *what)
852 for (i = 0; list[i]; i++)
854 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
855 if (c_tolower (list[i][j]) != c_tolower (what[k]))
857 /* The domain must be first to reach to beginning. */
867 if (host_name_addresses_map)
869 hash_table_iterator iter;
870 for (hash_table_iterate (host_name_addresses_map, &iter);
871 hash_table_iter_next (&iter);
874 char *host = iter.key;
875 struct address_list *al = iter.value;
877 assert (al->refcount == 1);
878 address_list_delete (al);
880 hash_table_destroy (host_name_addresses_map);
881 host_name_addresses_map = NULL;