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. */
31 #define USE_GNULIB_ALLOC
41 # include <sys/socket.h>
42 # include <netinet/in.h>
44 # include <arpa/inet.h>
47 # define SET_H_ERRNO(err) ((void)(h_errno = (err)))
49 # define SET_H_ERRNO(err) WSASetLastError (err)
60 # define NO_ADDRESS NO_DATA
63 /* Lists of IP addresses that result from running DNS queries. See
64 lookup_host for details. */
67 int count; /* number of adrresses */
68 ip_address *addresses; /* pointer to the string of addresses */
70 int faulty; /* number of addresses known not to work. */
71 bool connected; /* whether we were able to connect to
72 one of the addresses in the list,
75 int refcount; /* reference count; when it drops to
76 0, the entry is freed. */
79 /* Get the bounds of the address list. */
82 address_list_get_bounds (const struct address_list *al, int *start, int *end)
88 /* Return a pointer to the address at position POS. */
91 address_list_address_at (const struct address_list *al, int pos)
93 assert (pos >= al->faulty && pos < al->count);
94 return al->addresses + pos;
97 /* Return true if AL contains IP, false otherwise. */
100 address_list_contains (const struct address_list *al, const ip_address *ip)
106 for (i = 0; i < al->count; i++)
108 ip_address *cur = al->addresses + i;
109 if (cur->family == AF_INET
110 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
116 for (i = 0; i < al->count; i++)
118 ip_address *cur = al->addresses + i;
119 if (cur->family == AF_INET6
120 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
121 && cur->ipv6_scope == ip->ipv6_scope
123 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
127 #endif /* ENABLE_IPV6 */
133 /* Mark the INDEXth element of AL as faulty, so that the next time
134 this address list is used, the faulty element will be skipped. */
137 address_list_set_faulty (struct address_list *al, int index)
139 /* We assume that the address list is traversed in order, so that a
140 "faulty" attempt is always preceded with all-faulty addresses,
141 and this is how Wget uses it. */
142 assert (index == al->faulty);
145 if (al->faulty >= al->count)
146 /* All addresses have been proven faulty. Since there's not much
147 sense in returning the user an empty address list the next
148 time, we'll rather make them all clean, so that they can be
153 /* Set the "connected" flag to true. This flag used by connect.c to
154 see if the host perhaps needs to be resolved again. */
157 address_list_set_connected (struct address_list *al)
159 al->connected = true;
162 /* Return the value of the "connected" flag. */
165 address_list_connected_p (const struct address_list *al)
167 return al->connected;
172 /* Create an address_list from the addresses in the given struct
175 static struct address_list *
176 address_list_from_addrinfo (const struct addrinfo *ai)
178 struct address_list *al;
179 const struct addrinfo *ptr;
184 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
185 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
190 al = xnew0 (struct address_list);
191 al->addresses = xnew_array (ip_address, cnt);
196 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
197 if (ptr->ai_family == AF_INET6)
199 const struct sockaddr_in6 *sin6 =
200 (const struct sockaddr_in6 *)ptr->ai_addr;
201 ip->family = AF_INET6;
202 ip->data.d6 = sin6->sin6_addr;
203 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
204 ip->ipv6_scope = sin6->sin6_scope_id;
208 else if (ptr->ai_family == AF_INET)
210 const struct sockaddr_in *sin =
211 (const struct sockaddr_in *)ptr->ai_addr;
212 ip->family = AF_INET;
213 ip->data.d4 = sin->sin_addr;
216 assert (ip - al->addresses == cnt);
220 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
222 /* Compare two IP addresses by family, giving preference to the IPv4
223 address (sorting it first). In other words, return -1 if ADDR1 is
224 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
227 This is intended to be used as the comparator arg to a qsort-like
228 sorting function, which is why it accepts generic pointers. */
231 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
233 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
236 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
238 /* Like the above, but give preference to the IPv6 address. */
241 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
243 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
246 #else /* not ENABLE_IPV6 */
248 /* Create an address_list from a NULL-terminated vector of IPv4
249 addresses. This kind of vector is returned by gethostbyname. */
251 static struct address_list *
252 address_list_from_ipv4_addresses (char **vec)
255 struct address_list *al = xnew0 (struct address_list);
262 al->addresses = xnew_array (ip_address, count);
266 for (i = 0; i < count; i++)
268 ip_address *ip = &al->addresses[i];
269 ip->family = AF_INET;
270 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
276 #endif /* not ENABLE_IPV6 */
279 address_list_delete (struct address_list *al)
281 xfree (al->addresses);
285 /* Mark the address list as being no longer in use. This will reduce
286 its reference count which will cause the list to be freed when the
290 address_list_release (struct address_list *al)
293 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
295 if (al->refcount <= 0)
297 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
298 address_list_delete (al);
302 /* Versions of gethostbyname and getaddrinfo that support timeout. */
306 struct ghbnwt_context {
307 const char *host_name;
308 struct hostent *hptr;
312 gethostbyname_with_timeout_callback (void *arg)
314 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
315 ctx->hptr = gethostbyname (ctx->host_name);
318 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
319 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
320 The function makes sure that when NULL is returned for reasons
321 other than timeout, errno is reset. */
323 static struct hostent *
324 gethostbyname_with_timeout (const char *host_name, double timeout)
326 struct ghbnwt_context ctx;
327 ctx.host_name = host_name;
328 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
330 SET_H_ERRNO (HOST_NOT_FOUND);
339 /* Print error messages for host errors. */
341 host_errstr (int error)
343 /* Can't use switch since some of these constants can be equal,
344 which makes the compiler complain about duplicate case
346 if (error == HOST_NOT_FOUND
347 || error == NO_RECOVERY
349 || error == NO_ADDRESS)
350 return _("Unknown host");
351 else if (error == TRY_AGAIN)
352 /* Message modeled after what gai_strerror returns in similar
354 return _("Temporary failure in name resolution");
356 return _("Unknown error");
359 #else /* ENABLE_IPV6 */
361 struct gaiwt_context {
364 const struct addrinfo *hints;
365 struct addrinfo **res;
370 getaddrinfo_with_timeout_callback (void *arg)
372 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
373 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
376 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
377 In case of timeout, the EAI_SYSTEM error code is returned and errno
378 is set to ETIMEDOUT. */
381 getaddrinfo_with_timeout (const char *node, const char *service,
382 const struct addrinfo *hints, struct addrinfo **res,
385 struct gaiwt_context ctx;
387 ctx.service = service;
391 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
396 return ctx.exit_code;
399 #endif /* ENABLE_IPV6 */
401 /* Return a textual representation of ADDR, i.e. the dotted quad for
402 IPv4 addresses, and the colon-separated list of hex words (with all
403 zeros omitted, etc.) for IPv6 addresses. */
406 print_address (const ip_address *addr)
410 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
411 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
414 return inet_ntoa (addr->data.d4);
418 /* The following two functions were adapted from glibc's
419 implementation of inet_pton, written by Paul Vixie. */
422 is_valid_ipv4_address (const char *str, const char *end)
424 bool saw_digit = false;
432 if (ch >= '0' && ch <= '9')
434 val = val * 10 + (ch - '0');
445 else if (ch == '.' && saw_digit)
462 is_valid_ipv6_address (const char *str, const char *end)
464 /* Use lower-case for these to avoid clash with system headers. */
483 /* Leading :: requires some special handling. */
487 if (str == end || *str != ':')
499 /* if ch is a number, add it to val. */
503 val |= XDIGIT_TO_NUM (ch);
510 /* if ch is a colon ... */
523 if (tp > ns_in6addrsz - ns_int16sz)
531 /* if ch is a dot ... */
532 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
533 && is_valid_ipv4_address (curtok, end) == 1)
545 if (tp > ns_in6addrsz - ns_int16sz)
552 if (tp == ns_in6addrsz)
557 if (tp != ns_in6addrsz)
563 /* Simple host cache, used by lookup_host to speed up resolving. The
564 cache doesn't handle TTL because Wget is a fairly short-lived
565 application. Refreshing is attempted when connect fails, though --
566 see connect_to_host. */
568 /* Mapping between known hosts and to lists of their addresses. */
569 static struct hash_table *host_name_addresses_map;
572 /* Return the host's resolved addresses from the cache, if
575 static struct address_list *
576 cache_query (const char *host)
578 struct address_list *al;
579 if (!host_name_addresses_map)
581 al = hash_table_get (host_name_addresses_map, host);
584 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
591 /* Cache the DNS lookup of HOST. Subsequent invocations of
592 lookup_host will return the cached value. */
595 cache_store (const char *host, struct address_list *al)
597 if (!host_name_addresses_map)
598 host_name_addresses_map = make_nocase_string_hash_table (0);
601 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
606 debug_logprintf ("Caching %s =>", host);
607 for (i = 0; i < al->count; i++)
608 debug_logprintf (" %s", print_address (al->addresses + i));
609 debug_logprintf ("\n");
613 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
617 cache_remove (const char *host)
619 struct address_list *al;
620 if (!host_name_addresses_map)
622 al = hash_table_get (host_name_addresses_map, host);
625 address_list_release (al);
626 hash_table_remove (host_name_addresses_map, host);
630 /* Look up HOST in DNS and return a list of IP addresses.
632 This function caches its result so that, if the same host is passed
633 the second time, the addresses are returned without DNS lookup.
634 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
635 globally disable caching.)
637 The order of the returned addresses is affected by the setting of
638 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
639 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
640 at the beginning; otherwise, the order is left intact. The
641 relative order of addresses with the same family is left
642 undisturbed in either case.
644 FLAGS can be a combination of:
645 LH_SILENT - don't print the "resolving ... done" messages.
646 LH_BIND - resolve addresses for use with bind, which under
647 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
648 Passive lookups are not cached under IPv6.
649 LH_REFRESH - if HOST is cached, remove the entry from the cache
650 and resolve it anew. */
652 struct address_list *
653 lookup_host (const char *host, int flags)
655 struct address_list *al;
656 bool silent = !!(flags & LH_SILENT);
658 bool numeric_address = false;
659 double timeout = opt.dns_timeout;
662 /* If we're not using getaddrinfo, first check if HOST specifies a
663 numeric IPv4 address. Some implementations of gethostbyname
664 (e.g. the Ultrix one and possibly Winsock) don't accept
665 dotted-decimal IPv4 addresses. */
667 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
668 if (addr_ipv4 != (uint32_t) -1)
670 /* No need to cache host->addr relation, just return the
673 vec[0] = (char *)&addr_ipv4;
675 return address_list_from_ipv4_addresses (vec);
678 #else /* ENABLE_IPV6 */
679 /* If we're using getaddrinfo, at least check whether the address is
680 already numeric, in which case there is no need to print the
681 "Resolving..." output. (This comes at no additional cost since
682 the is_valid_ipv*_address are already required for
685 const char *end = host + strlen (host);
686 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
687 numeric_address = true;
691 /* Cache is normally on, but can be turned off with --no-dns-cache.
692 Don't cache passive lookups under IPv6. */
693 use_cache = opt.dns_cache;
695 if ((flags & LH_BIND) || numeric_address)
699 /* Try to find the host in the cache so we don't need to talk to the
700 resolver. If LH_REFRESH is requested, remove HOST from the cache
704 if (!(flags & LH_REFRESH))
706 al = cache_query (host);
714 /* No luck with the cache; resolve HOST. */
716 if (!silent && !numeric_address)
717 logprintf (LOG_VERBOSE, _("Resolving %s... "), escnonprint (host));
722 struct addrinfo hints, *res;
725 hints.ai_socktype = SOCK_STREAM;
727 hints.ai_family = AF_INET;
728 else if (opt.ipv6_only)
729 hints.ai_family = AF_INET6;
731 /* We tried using AI_ADDRCONFIG, but removed it because: it
732 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
733 it's unneeded since we sort the addresses anyway. */
734 hints.ai_family = AF_UNSPEC;
737 hints.ai_flags |= AI_PASSIVE;
739 #ifdef AI_NUMERICHOST
742 /* Where available, the AI_NUMERICHOST hint can prevent costly
743 access to DNS servers. */
744 hints.ai_flags |= AI_NUMERICHOST;
745 timeout = 0; /* no timeout needed when "resolving"
746 numeric hosts -- avoid setting up
747 signal handlers and such. */
751 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
752 if (err != 0 || res == NULL)
755 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
756 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
759 al = address_list_from_addrinfo (res);
763 logprintf (LOG_VERBOSE,
764 _("failed: No IPv4/IPv6 addresses for host.\n"));
768 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
769 --prefer-family) come first. Sorting is stable so the order of
770 the addresses with the same family is undisturbed. */
771 if (al->count > 1 && opt.prefer_family != prefer_none)
772 stable_sort (al->addresses, al->count, sizeof (ip_address),
773 opt.prefer_family == prefer_ipv4
774 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
776 #else /* not ENABLE_IPV6 */
778 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
783 if (errno != ETIMEDOUT)
784 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
785 host_errstr (h_errno));
787 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
791 /* Do older systems have h_addr_list? */
792 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
794 #endif /* not ENABLE_IPV6 */
796 /* Print the addresses determined by DNS lookup, but no more than
798 if (!silent && !numeric_address)
801 int printmax = al->count <= 3 ? al->count : 3;
802 for (i = 0; i < printmax; i++)
804 logputs (LOG_VERBOSE, print_address (al->addresses + i));
805 if (i < printmax - 1)
806 logputs (LOG_VERBOSE, ", ");
808 if (printmax != al->count)
809 logputs (LOG_VERBOSE, ", ...");
810 logputs (LOG_VERBOSE, "\n");
813 /* Cache the lookup information. */
815 cache_store (host, al);
820 /* Determine whether a URL is acceptable to be followed, according to
821 a list of domains to accept. */
823 accept_domain (struct url *u)
825 assert (u->host != NULL);
828 if (!sufmatch ((const char **)opt.domains, u->host))
831 if (opt.exclude_domains)
833 if (sufmatch ((const char **)opt.exclude_domains, u->host))
839 /* Check whether WHAT is matched in LIST, each element of LIST being a
840 pattern to match WHAT against, using backward matching (see
841 match_backwards() in utils.c).
843 If an element of LIST matched, 1 is returned, 0 otherwise. */
845 sufmatch (const char **list, const char *what)
850 for (i = 0; list[i]; i++)
852 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
853 if (c_tolower (list[i][j]) != c_tolower (what[k]))
855 /* The domain must be first to reach to beginning. */
865 if (host_name_addresses_map)
867 hash_table_iterator iter;
868 for (hash_table_iterate (host_name_addresses_map, &iter);
869 hash_table_iter_next (&iter);
872 char *host = iter.key;
873 struct address_list *al = iter.value;
875 assert (al->refcount == 1);
876 address_list_delete (al);
878 hash_table_destroy (host_name_addresses_map);
879 host_name_addresses_map = NULL;