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)
59 # define NO_ADDRESS NO_DATA
62 /* Lists of IP addresses that result from running DNS queries. See
63 lookup_host for details. */
66 int count; /* number of adrresses */
67 ip_address *addresses; /* pointer to the string of addresses */
69 int faulty; /* number of addresses known not to work. */
70 bool connected; /* whether we were able to connect to
71 one of the addresses in the list,
74 int refcount; /* reference count; when it drops to
75 0, the entry is freed. */
78 /* Get the bounds of the address list. */
81 address_list_get_bounds (const struct address_list *al, int *start, int *end)
87 /* Return a pointer to the address at position POS. */
90 address_list_address_at (const struct address_list *al, int pos)
92 assert (pos >= al->faulty && pos < al->count);
93 return al->addresses + pos;
96 /* Return true if AL contains IP, false otherwise. */
99 address_list_contains (const struct address_list *al, const ip_address *ip)
105 for (i = 0; i < al->count; i++)
107 ip_address *cur = al->addresses + i;
108 if (cur->family == AF_INET
109 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
115 for (i = 0; i < al->count; i++)
117 ip_address *cur = al->addresses + i;
118 if (cur->family == AF_INET6
119 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
120 && cur->ipv6_scope == ip->ipv6_scope
122 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
126 #endif /* ENABLE_IPV6 */
132 /* Mark the INDEXth element of AL as faulty, so that the next time
133 this address list is used, the faulty element will be skipped. */
136 address_list_set_faulty (struct address_list *al, int index)
138 /* We assume that the address list is traversed in order, so that a
139 "faulty" attempt is always preceded with all-faulty addresses,
140 and this is how Wget uses it. */
141 assert (index == al->faulty);
144 if (al->faulty >= al->count)
145 /* All addresses have been proven faulty. Since there's not much
146 sense in returning the user an empty address list the next
147 time, we'll rather make them all clean, so that they can be
152 /* Set the "connected" flag to true. This flag used by connect.c to
153 see if the host perhaps needs to be resolved again. */
156 address_list_set_connected (struct address_list *al)
158 al->connected = true;
161 /* Return the value of the "connected" flag. */
164 address_list_connected_p (const struct address_list *al)
166 return al->connected;
171 /* Create an address_list from the addresses in the given struct
174 static struct address_list *
175 address_list_from_addrinfo (const struct addrinfo *ai)
177 struct address_list *al;
178 const struct addrinfo *ptr;
183 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
184 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
189 al = xnew0 (struct address_list);
190 al->addresses = xnew_array (ip_address, cnt);
195 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
196 if (ptr->ai_family == AF_INET6)
198 const struct sockaddr_in6 *sin6 =
199 (const struct sockaddr_in6 *)ptr->ai_addr;
200 ip->family = AF_INET6;
201 ip->data.d6 = sin6->sin6_addr;
202 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
203 ip->ipv6_scope = sin6->sin6_scope_id;
207 else if (ptr->ai_family == AF_INET)
209 const struct sockaddr_in *sin =
210 (const struct sockaddr_in *)ptr->ai_addr;
211 ip->family = AF_INET;
212 ip->data.d4 = sin->sin_addr;
215 assert (ip - al->addresses == cnt);
219 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
221 /* Compare two IP addresses by family, giving preference to the IPv4
222 address (sorting it first). In other words, return -1 if ADDR1 is
223 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
226 This is intended to be used as the comparator arg to a qsort-like
227 sorting function, which is why it accepts generic pointers. */
230 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
232 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
235 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
237 /* Like the above, but give preference to the IPv6 address. */
240 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
242 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
245 #else /* not ENABLE_IPV6 */
247 /* Create an address_list from a NULL-terminated vector of IPv4
248 addresses. This kind of vector is returned by gethostbyname. */
250 static struct address_list *
251 address_list_from_ipv4_addresses (char **vec)
254 struct address_list *al = xnew0 (struct address_list);
261 al->addresses = xnew_array (ip_address, count);
265 for (i = 0; i < count; i++)
267 ip_address *ip = &al->addresses[i];
268 ip->family = AF_INET;
269 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
275 #endif /* not ENABLE_IPV6 */
278 address_list_delete (struct address_list *al)
280 xfree (al->addresses);
284 /* Mark the address list as being no longer in use. This will reduce
285 its reference count which will cause the list to be freed when the
289 address_list_release (struct address_list *al)
292 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
294 if (al->refcount <= 0)
296 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
297 address_list_delete (al);
301 /* Versions of gethostbyname and getaddrinfo that support timeout. */
305 struct ghbnwt_context {
306 const char *host_name;
307 struct hostent *hptr;
311 gethostbyname_with_timeout_callback (void *arg)
313 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
314 ctx->hptr = gethostbyname (ctx->host_name);
317 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
318 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
319 The function makes sure that when NULL is returned for reasons
320 other than timeout, errno is reset. */
322 static struct hostent *
323 gethostbyname_with_timeout (const char *host_name, double timeout)
325 struct ghbnwt_context ctx;
326 ctx.host_name = host_name;
327 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
329 SET_H_ERRNO (HOST_NOT_FOUND);
338 /* Print error messages for host errors. */
340 host_errstr (int error)
342 /* Can't use switch since some of these constants can be equal,
343 which makes the compiler complain about duplicate case
345 if (error == HOST_NOT_FOUND
346 || error == NO_RECOVERY
348 || error == NO_ADDRESS)
349 return _("Unknown host");
350 else if (error == TRY_AGAIN)
351 /* Message modeled after what gai_strerror returns in similar
353 return _("Temporary failure in name resolution");
355 return _("Unknown error");
358 #else /* ENABLE_IPV6 */
360 struct gaiwt_context {
363 const struct addrinfo *hints;
364 struct addrinfo **res;
369 getaddrinfo_with_timeout_callback (void *arg)
371 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
372 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
375 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
376 In case of timeout, the EAI_SYSTEM error code is returned and errno
377 is set to ETIMEDOUT. */
380 getaddrinfo_with_timeout (const char *node, const char *service,
381 const struct addrinfo *hints, struct addrinfo **res,
384 struct gaiwt_context ctx;
386 ctx.service = service;
390 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
395 return ctx.exit_code;
398 #endif /* ENABLE_IPV6 */
400 /* Return a textual representation of ADDR, i.e. the dotted quad for
401 IPv4 addresses, and the colon-separated list of hex words (with all
402 zeros omitted, etc.) for IPv6 addresses. */
405 print_address (const ip_address *addr)
409 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
410 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
413 return inet_ntoa (addr->data.d4);
417 /* The following two functions were adapted from glibc's
418 implementation of inet_pton, written by Paul Vixie. */
421 is_valid_ipv4_address (const char *str, const char *end)
423 bool saw_digit = false;
431 if (ch >= '0' && ch <= '9')
433 val = val * 10 + (ch - '0');
444 else if (ch == '.' && saw_digit)
461 is_valid_ipv6_address (const char *str, const char *end)
463 /* Use lower-case for these to avoid clash with system headers. */
482 /* Leading :: requires some special handling. */
486 if (str == end || *str != ':')
498 /* if ch is a number, add it to val. */
502 val |= XDIGIT_TO_NUM (ch);
509 /* if ch is a colon ... */
522 if (tp > ns_in6addrsz - ns_int16sz)
530 /* if ch is a dot ... */
531 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
532 && is_valid_ipv4_address (curtok, end) == 1)
544 if (tp > ns_in6addrsz - ns_int16sz)
551 if (tp == ns_in6addrsz)
556 if (tp != ns_in6addrsz)
562 /* Simple host cache, used by lookup_host to speed up resolving. The
563 cache doesn't handle TTL because Wget is a fairly short-lived
564 application. Refreshing is attempted when connect fails, though --
565 see connect_to_host. */
567 /* Mapping between known hosts and to lists of their addresses. */
568 static struct hash_table *host_name_addresses_map;
571 /* Return the host's resolved addresses from the cache, if
574 static struct address_list *
575 cache_query (const char *host)
577 struct address_list *al;
578 if (!host_name_addresses_map)
580 al = hash_table_get (host_name_addresses_map, host);
583 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
590 /* Cache the DNS lookup of HOST. Subsequent invocations of
591 lookup_host will return the cached value. */
594 cache_store (const char *host, struct address_list *al)
596 if (!host_name_addresses_map)
597 host_name_addresses_map = make_nocase_string_hash_table (0);
600 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
605 debug_logprintf ("Caching %s =>", host);
606 for (i = 0; i < al->count; i++)
607 debug_logprintf (" %s", print_address (al->addresses + i));
608 debug_logprintf ("\n");
612 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
616 cache_remove (const char *host)
618 struct address_list *al;
619 if (!host_name_addresses_map)
621 al = hash_table_get (host_name_addresses_map, host);
624 address_list_release (al);
625 hash_table_remove (host_name_addresses_map, host);
629 /* Look up HOST in DNS and return a list of IP addresses.
631 This function caches its result so that, if the same host is passed
632 the second time, the addresses are returned without DNS lookup.
633 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
634 globally disable caching.)
636 The order of the returned addresses is affected by the setting of
637 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
638 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
639 at the beginning; otherwise, the order is left intact. The
640 relative order of addresses with the same family is left
641 undisturbed in either case.
643 FLAGS can be a combination of:
644 LH_SILENT - don't print the "resolving ... done" messages.
645 LH_BIND - resolve addresses for use with bind, which under
646 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
647 Passive lookups are not cached under IPv6.
648 LH_REFRESH - if HOST is cached, remove the entry from the cache
649 and resolve it anew. */
651 struct address_list *
652 lookup_host (const char *host, int flags)
654 struct address_list *al;
655 bool silent = !!(flags & LH_SILENT);
657 bool numeric_address = false;
658 double timeout = opt.dns_timeout;
661 /* If we're not using getaddrinfo, first check if HOST specifies a
662 numeric IPv4 address. Some implementations of gethostbyname
663 (e.g. the Ultrix one and possibly Winsock) don't accept
664 dotted-decimal IPv4 addresses. */
666 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
667 if (addr_ipv4 != (uint32_t) -1)
669 /* No need to cache host->addr relation, just return the
672 vec[0] = (char *)&addr_ipv4;
674 return address_list_from_ipv4_addresses (vec);
677 #else /* ENABLE_IPV6 */
678 /* If we're using getaddrinfo, at least check whether the address is
679 already numeric, in which case there is no need to print the
680 "Resolving..." output. (This comes at no additional cost since
681 the is_valid_ipv*_address are already required for
684 const char *end = host + strlen (host);
685 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
686 numeric_address = true;
690 /* Cache is normally on, but can be turned off with --no-dns-cache.
691 Don't cache passive lookups under IPv6. */
692 use_cache = opt.dns_cache;
694 if ((flags & LH_BIND) || numeric_address)
698 /* Try to find the host in the cache so we don't need to talk to the
699 resolver. If LH_REFRESH is requested, remove HOST from the cache
703 if (!(flags & LH_REFRESH))
705 al = cache_query (host);
713 /* No luck with the cache; resolve HOST. */
715 if (!silent && !numeric_address)
717 char *str = NULL, *name;
719 if (opt.enable_iri && (name = idn_decode ((char *) host)) != NULL)
721 int len = strlen (host) + strlen (name) + 4;
723 snprintf (str, len, "%s (%s)", name, host);
728 logprintf (LOG_VERBOSE, _("Resolving %s... "),
729 quotearg_style (escape_quoting_style, str ? str : host));
738 struct addrinfo hints, *res;
741 hints.ai_socktype = SOCK_STREAM;
743 hints.ai_family = AF_INET;
744 else if (opt.ipv6_only)
745 hints.ai_family = AF_INET6;
747 /* We tried using AI_ADDRCONFIG, but removed it because: it
748 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
749 it's unneeded since we sort the addresses anyway. */
750 hints.ai_family = AF_UNSPEC;
753 hints.ai_flags |= AI_PASSIVE;
755 #ifdef AI_NUMERICHOST
758 /* Where available, the AI_NUMERICHOST hint can prevent costly
759 access to DNS servers. */
760 hints.ai_flags |= AI_NUMERICHOST;
761 timeout = 0; /* no timeout needed when "resolving"
762 numeric hosts -- avoid setting up
763 signal handlers and such. */
767 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
768 if (err != 0 || res == NULL)
771 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
772 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
775 al = address_list_from_addrinfo (res);
779 logprintf (LOG_VERBOSE,
780 _("failed: No IPv4/IPv6 addresses for host.\n"));
784 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
785 --prefer-family) come first. Sorting is stable so the order of
786 the addresses with the same family is undisturbed. */
787 if (al->count > 1 && opt.prefer_family != prefer_none)
788 stable_sort (al->addresses, al->count, sizeof (ip_address),
789 opt.prefer_family == prefer_ipv4
790 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
792 #else /* not ENABLE_IPV6 */
794 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
799 if (errno != ETIMEDOUT)
800 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
801 host_errstr (h_errno));
803 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
807 /* Do older systems have h_addr_list? */
808 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
810 #endif /* not ENABLE_IPV6 */
812 /* Print the addresses determined by DNS lookup, but no more than
814 if (!silent && !numeric_address)
817 int printmax = al->count <= 3 ? al->count : 3;
818 for (i = 0; i < printmax; i++)
820 logputs (LOG_VERBOSE, print_address (al->addresses + i));
821 if (i < printmax - 1)
822 logputs (LOG_VERBOSE, ", ");
824 if (printmax != al->count)
825 logputs (LOG_VERBOSE, ", ...");
826 logputs (LOG_VERBOSE, "\n");
829 /* Cache the lookup information. */
831 cache_store (host, al);
836 /* Determine whether a URL is acceptable to be followed, according to
837 a list of domains to accept. */
839 accept_domain (struct url *u)
841 assert (u->host != NULL);
844 if (!sufmatch ((const char **)opt.domains, u->host))
847 if (opt.exclude_domains)
849 if (sufmatch ((const char **)opt.exclude_domains, u->host))
855 /* Check whether WHAT is matched in LIST, each element of LIST being a
856 pattern to match WHAT against, using backward matching (see
857 match_backwards() in utils.c).
859 If an element of LIST matched, 1 is returned, 0 otherwise. */
861 sufmatch (const char **list, const char *what)
866 for (i = 0; list[i]; i++)
868 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
869 if (c_tolower (list[i][j]) != c_tolower (what[k]))
871 /* The domain must be first to reach to beginning. */
881 if (host_name_addresses_map)
883 hash_table_iterator iter;
884 for (hash_table_iterate (host_name_addresses_map, &iter);
885 hash_table_iter_next (&iter);
888 char *host = iter.key;
889 struct address_list *al = iter.value;
891 assert (al->refcount == 1);
892 address_list_delete (al);
894 hash_table_destroy (host_name_addresses_map);
895 host_name_addresses_map = NULL;