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/types.h>
40 # include <sys/socket.h>
41 # include <netinet/in.h>
43 # include <arpa/inet.h>
46 # define SET_H_ERRNO(err) ((void)(h_errno = (err)))
48 # define SET_H_ERRNO(err) WSASetLastError (err)
59 # define NO_ADDRESS NO_DATA
62 #if !HAVE_DECL_H_ERRNO
67 /* Lists of IP addresses that result from running DNS queries. See
68 lookup_host for details. */
71 int count; /* number of adrresses */
72 ip_address *addresses; /* pointer to the string of addresses */
74 int faulty; /* number of addresses known not to work. */
75 bool connected; /* whether we were able to connect to
76 one of the addresses in the list,
79 int refcount; /* reference count; when it drops to
80 0, the entry is freed. */
83 /* Get the bounds of the address list. */
86 address_list_get_bounds (const struct address_list *al, int *start, int *end)
92 /* Return a pointer to the address at position POS. */
95 address_list_address_at (const struct address_list *al, int pos)
97 assert (pos >= al->faulty && pos < al->count);
98 return al->addresses + pos;
101 /* Return true if AL contains IP, false otherwise. */
104 address_list_contains (const struct address_list *al, const ip_address *ip)
110 for (i = 0; i < al->count; i++)
112 ip_address *cur = al->addresses + i;
113 if (cur->family == AF_INET
114 && (cur->data.d4.s_addr == ip->data.d4.s_addr))
120 for (i = 0; i < al->count; i++)
122 ip_address *cur = al->addresses + i;
123 if (cur->family == AF_INET6
124 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
125 && cur->ipv6_scope == ip->ipv6_scope
127 && IN6_ARE_ADDR_EQUAL (&cur->data.d6, &ip->data.d6))
131 #endif /* ENABLE_IPV6 */
137 /* Mark the INDEXth element of AL as faulty, so that the next time
138 this address list is used, the faulty element will be skipped. */
141 address_list_set_faulty (struct address_list *al, int index)
143 /* We assume that the address list is traversed in order, so that a
144 "faulty" attempt is always preceded with all-faulty addresses,
145 and this is how Wget uses it. */
146 assert (index == al->faulty);
149 if (al->faulty >= al->count)
150 /* All addresses have been proven faulty. Since there's not much
151 sense in returning the user an empty address list the next
152 time, we'll rather make them all clean, so that they can be
157 /* Set the "connected" flag to true. This flag used by connect.c to
158 see if the host perhaps needs to be resolved again. */
161 address_list_set_connected (struct address_list *al)
163 al->connected = true;
166 /* Return the value of the "connected" flag. */
169 address_list_connected_p (const struct address_list *al)
171 return al->connected;
176 /* Create an address_list from the addresses in the given struct
179 static struct address_list *
180 address_list_from_addrinfo (const struct addrinfo *ai)
182 struct address_list *al;
183 const struct addrinfo *ptr;
188 for (ptr = ai; ptr != NULL ; ptr = ptr->ai_next)
189 if (ptr->ai_family == AF_INET || ptr->ai_family == AF_INET6)
194 al = xnew0 (struct address_list);
195 al->addresses = xnew_array (ip_address, cnt);
200 for (ptr = ai; ptr != NULL; ptr = ptr->ai_next)
201 if (ptr->ai_family == AF_INET6)
203 const struct sockaddr_in6 *sin6 =
204 (const struct sockaddr_in6 *)ptr->ai_addr;
205 ip->family = AF_INET6;
206 ip->data.d6 = sin6->sin6_addr;
207 #ifdef HAVE_SOCKADDR_IN6_SCOPE_ID
208 ip->ipv6_scope = sin6->sin6_scope_id;
212 else if (ptr->ai_family == AF_INET)
214 const struct sockaddr_in *sin =
215 (const struct sockaddr_in *)ptr->ai_addr;
216 ip->family = AF_INET;
217 ip->data.d4 = sin->sin_addr;
220 assert (ip - al->addresses == cnt);
224 #define IS_IPV4(addr) (((const ip_address *) addr)->family == AF_INET)
226 /* Compare two IP addresses by family, giving preference to the IPv4
227 address (sorting it first). In other words, return -1 if ADDR1 is
228 IPv4 and ADDR2 is IPv6, +1 if ADDR1 is IPv6 and ADDR2 is IPv4, and
231 This is intended to be used as the comparator arg to a qsort-like
232 sorting function, which is why it accepts generic pointers. */
235 cmp_prefer_ipv4 (const void *addr1, const void *addr2)
237 return !IS_IPV4 (addr1) - !IS_IPV4 (addr2);
240 #define IS_IPV6(addr) (((const ip_address *) addr)->family == AF_INET6)
242 /* Like the above, but give preference to the IPv6 address. */
245 cmp_prefer_ipv6 (const void *addr1, const void *addr2)
247 return !IS_IPV6 (addr1) - !IS_IPV6 (addr2);
250 #else /* not ENABLE_IPV6 */
252 /* Create an address_list from a NULL-terminated vector of IPv4
253 addresses. This kind of vector is returned by gethostbyname. */
255 static struct address_list *
256 address_list_from_ipv4_addresses (char **vec)
259 struct address_list *al = xnew0 (struct address_list);
266 al->addresses = xnew_array (ip_address, count);
270 for (i = 0; i < count; i++)
272 ip_address *ip = &al->addresses[i];
273 ip->family = AF_INET;
274 memcpy (IP_INADDR_DATA (ip), vec[i], 4);
280 #endif /* not ENABLE_IPV6 */
283 address_list_delete (struct address_list *al)
285 xfree (al->addresses);
289 /* Mark the address list as being no longer in use. This will reduce
290 its reference count which will cause the list to be freed when the
294 address_list_release (struct address_list *al)
297 DEBUGP (("Releasing 0x%0*lx (new refcount %d).\n", PTR_FORMAT (al),
299 if (al->refcount <= 0)
301 DEBUGP (("Deleting unused 0x%0*lx.\n", PTR_FORMAT (al)));
302 address_list_delete (al);
306 /* Versions of gethostbyname and getaddrinfo that support timeout. */
310 struct ghbnwt_context {
311 const char *host_name;
312 struct hostent *hptr;
316 gethostbyname_with_timeout_callback (void *arg)
318 struct ghbnwt_context *ctx = (struct ghbnwt_context *)arg;
319 ctx->hptr = gethostbyname (ctx->host_name);
322 /* Just like gethostbyname, except it times out after TIMEOUT seconds.
323 In case of timeout, NULL is returned and errno is set to ETIMEDOUT.
324 The function makes sure that when NULL is returned for reasons
325 other than timeout, errno is reset. */
327 static struct hostent *
328 gethostbyname_with_timeout (const char *host_name, double timeout)
330 struct ghbnwt_context ctx;
331 ctx.host_name = host_name;
332 if (run_with_timeout (timeout, gethostbyname_with_timeout_callback, &ctx))
334 SET_H_ERRNO (HOST_NOT_FOUND);
343 /* Print error messages for host errors. */
345 host_errstr (int error)
347 /* Can't use switch since some of these constants can be equal,
348 which makes the compiler complain about duplicate case
350 if (error == HOST_NOT_FOUND
351 || error == NO_RECOVERY
353 || error == NO_ADDRESS)
354 return _("Unknown host");
355 else if (error == TRY_AGAIN)
356 /* Message modeled after what gai_strerror returns in similar
358 return _("Temporary failure in name resolution");
360 return _("Unknown error");
363 #else /* ENABLE_IPV6 */
365 struct gaiwt_context {
368 const struct addrinfo *hints;
369 struct addrinfo **res;
374 getaddrinfo_with_timeout_callback (void *arg)
376 struct gaiwt_context *ctx = (struct gaiwt_context *)arg;
377 ctx->exit_code = getaddrinfo (ctx->node, ctx->service, ctx->hints, ctx->res);
380 /* Just like getaddrinfo, except it times out after TIMEOUT seconds.
381 In case of timeout, the EAI_SYSTEM error code is returned and errno
382 is set to ETIMEDOUT. */
385 getaddrinfo_with_timeout (const char *node, const char *service,
386 const struct addrinfo *hints, struct addrinfo **res,
389 struct gaiwt_context ctx;
391 ctx.service = service;
395 if (run_with_timeout (timeout, getaddrinfo_with_timeout_callback, &ctx))
400 return ctx.exit_code;
403 #endif /* ENABLE_IPV6 */
405 /* Return a textual representation of ADDR, i.e. the dotted quad for
406 IPv4 addresses, and the colon-separated list of hex words (with all
407 zeros omitted, etc.) for IPv6 addresses. */
410 print_address (const ip_address *addr)
414 if (!inet_ntop (addr->family, IP_INADDR_DATA (addr), buf, sizeof buf))
415 snprintf (buf, sizeof buf, "<error: %s>", strerror (errno));
418 return inet_ntoa (addr->data.d4);
422 /* The following two functions were adapted from glibc's
423 implementation of inet_pton, written by Paul Vixie. */
426 is_valid_ipv4_address (const char *str, const char *end)
428 bool saw_digit = false;
436 if (ch >= '0' && ch <= '9')
438 val = val * 10 + (ch - '0');
449 else if (ch == '.' && saw_digit)
466 is_valid_ipv6_address (const char *str, const char *end)
468 /* Use lower-case for these to avoid clash with system headers. */
487 /* Leading :: requires some special handling. */
491 if (str == end || *str != ':')
503 /* if ch is a number, add it to val. */
507 val |= XDIGIT_TO_NUM (ch);
514 /* if ch is a colon ... */
527 if (tp > ns_in6addrsz - ns_int16sz)
535 /* if ch is a dot ... */
536 if (ch == '.' && (tp <= ns_in6addrsz - ns_inaddrsz)
537 && is_valid_ipv4_address (curtok, end) == 1)
549 if (tp > ns_in6addrsz - ns_int16sz)
556 if (tp == ns_in6addrsz)
561 if (tp != ns_in6addrsz)
567 /* Simple host cache, used by lookup_host to speed up resolving. The
568 cache doesn't handle TTL because Wget is a fairly short-lived
569 application. Refreshing is attempted when connect fails, though --
570 see connect_to_host. */
572 /* Mapping between known hosts and to lists of their addresses. */
573 static struct hash_table *host_name_addresses_map;
576 /* Return the host's resolved addresses from the cache, if
579 static struct address_list *
580 cache_query (const char *host)
582 struct address_list *al;
583 if (!host_name_addresses_map)
585 al = hash_table_get (host_name_addresses_map, host);
588 DEBUGP (("Found %s in host_name_addresses_map (%p)\n", host, al));
595 /* Cache the DNS lookup of HOST. Subsequent invocations of
596 lookup_host will return the cached value. */
599 cache_store (const char *host, struct address_list *al)
601 if (!host_name_addresses_map)
602 host_name_addresses_map = make_nocase_string_hash_table (0);
605 hash_table_put (host_name_addresses_map, xstrdup_lower (host), al);
610 debug_logprintf ("Caching %s =>", host);
611 for (i = 0; i < al->count; i++)
612 debug_logprintf (" %s", print_address (al->addresses + i));
613 debug_logprintf ("\n");
617 /* Remove HOST from the DNS cache. Does nothing is HOST is not in
621 cache_remove (const char *host)
623 struct address_list *al;
624 if (!host_name_addresses_map)
626 al = hash_table_get (host_name_addresses_map, host);
629 address_list_release (al);
630 hash_table_remove (host_name_addresses_map, host);
634 /* Look up HOST in DNS and return a list of IP addresses.
636 This function caches its result so that, if the same host is passed
637 the second time, the addresses are returned without DNS lookup.
638 (Use LH_REFRESH to force lookup, or set opt.dns_cache to 0 to
639 globally disable caching.)
641 The order of the returned addresses is affected by the setting of
642 opt.prefer_family: if it is set to prefer_ipv4, IPv4 addresses are
643 placed at the beginning; if it is prefer_ipv6, IPv6 ones are placed
644 at the beginning; otherwise, the order is left intact. The
645 relative order of addresses with the same family is left
646 undisturbed in either case.
648 FLAGS can be a combination of:
649 LH_SILENT - don't print the "resolving ... done" messages.
650 LH_BIND - resolve addresses for use with bind, which under
651 IPv6 means to use AI_PASSIVE flag to getaddrinfo.
652 Passive lookups are not cached under IPv6.
653 LH_REFRESH - if HOST is cached, remove the entry from the cache
654 and resolve it anew. */
656 struct address_list *
657 lookup_host (const char *host, int flags)
659 struct address_list *al;
660 bool silent = !!(flags & LH_SILENT);
662 bool numeric_address = false;
663 double timeout = opt.dns_timeout;
666 /* If we're not using getaddrinfo, first check if HOST specifies a
667 numeric IPv4 address. Some implementations of gethostbyname
668 (e.g. the Ultrix one and possibly Winsock) don't accept
669 dotted-decimal IPv4 addresses. */
671 uint32_t addr_ipv4 = (uint32_t)inet_addr (host);
672 if (addr_ipv4 != (uint32_t) -1)
674 /* No need to cache host->addr relation, just return the
677 vec[0] = (char *)&addr_ipv4;
679 return address_list_from_ipv4_addresses (vec);
682 #else /* ENABLE_IPV6 */
683 /* If we're using getaddrinfo, at least check whether the address is
684 already numeric, in which case there is no need to print the
685 "Resolving..." output. (This comes at no additional cost since
686 the is_valid_ipv*_address are already required for
689 const char *end = host + strlen (host);
690 if (is_valid_ipv4_address (host, end) || is_valid_ipv6_address (host, end))
691 numeric_address = true;
695 /* Cache is normally on, but can be turned off with --no-dns-cache.
696 Don't cache passive lookups under IPv6. */
697 use_cache = opt.dns_cache;
699 if ((flags & LH_BIND) || numeric_address)
703 /* Try to find the host in the cache so we don't need to talk to the
704 resolver. If LH_REFRESH is requested, remove HOST from the cache
708 if (!(flags & LH_REFRESH))
710 al = cache_query (host);
718 /* No luck with the cache; resolve HOST. */
720 if (!silent && !numeric_address)
722 char *str = NULL, *name;
724 if (opt.enable_iri && (name = idn_decode ((char *) host)) != NULL)
726 int len = strlen (host) + strlen (name) + 4;
728 snprintf (str, len, "%s (%s)", name, host);
733 logprintf (LOG_VERBOSE, _("Resolving %s... "),
734 quotearg_style (escape_quoting_style, str ? str : host));
743 struct addrinfo hints, *res;
746 hints.ai_socktype = SOCK_STREAM;
748 hints.ai_family = AF_INET;
749 else if (opt.ipv6_only)
750 hints.ai_family = AF_INET6;
752 /* We tried using AI_ADDRCONFIG, but removed it because: it
753 misinterprets IPv6 loopbacks, it is broken on AIX 5.1, and
754 it's unneeded since we sort the addresses anyway. */
755 hints.ai_family = AF_UNSPEC;
758 hints.ai_flags |= AI_PASSIVE;
760 #ifdef AI_NUMERICHOST
763 /* Where available, the AI_NUMERICHOST hint can prevent costly
764 access to DNS servers. */
765 hints.ai_flags |= AI_NUMERICHOST;
766 timeout = 0; /* no timeout needed when "resolving"
767 numeric hosts -- avoid setting up
768 signal handlers and such. */
772 err = getaddrinfo_with_timeout (host, NULL, &hints, &res, timeout);
773 if (err != 0 || res == NULL)
776 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
777 err != EAI_SYSTEM ? gai_strerror (err) : strerror (errno));
780 al = address_list_from_addrinfo (res);
784 logprintf (LOG_VERBOSE,
785 _("failed: No IPv4/IPv6 addresses for host.\n"));
789 /* Reorder addresses so that IPv4 ones (or IPv6 ones, as per
790 --prefer-family) come first. Sorting is stable so the order of
791 the addresses with the same family is undisturbed. */
792 if (al->count > 1 && opt.prefer_family != prefer_none)
793 stable_sort (al->addresses, al->count, sizeof (ip_address),
794 opt.prefer_family == prefer_ipv4
795 ? cmp_prefer_ipv4 : cmp_prefer_ipv6);
797 #else /* not ENABLE_IPV6 */
799 struct hostent *hptr = gethostbyname_with_timeout (host, timeout);
804 if (errno != ETIMEDOUT)
805 logprintf (LOG_VERBOSE, _("failed: %s.\n"),
806 host_errstr (h_errno));
808 logputs (LOG_VERBOSE, _("failed: timed out.\n"));
812 /* Do older systems have h_addr_list? */
813 al = address_list_from_ipv4_addresses (hptr->h_addr_list);
815 #endif /* not ENABLE_IPV6 */
817 /* Print the addresses determined by DNS lookup, but no more than
819 if (!silent && !numeric_address)
822 int printmax = al->count <= 3 ? al->count : 3;
823 for (i = 0; i < printmax; i++)
825 logputs (LOG_VERBOSE, print_address (al->addresses + i));
826 if (i < printmax - 1)
827 logputs (LOG_VERBOSE, ", ");
829 if (printmax != al->count)
830 logputs (LOG_VERBOSE, ", ...");
831 logputs (LOG_VERBOSE, "\n");
834 /* Cache the lookup information. */
836 cache_store (host, al);
841 /* Determine whether a URL is acceptable to be followed, according to
842 a list of domains to accept. */
844 accept_domain (struct url *u)
846 assert (u->host != NULL);
849 if (!sufmatch ((const char **)opt.domains, u->host))
852 if (opt.exclude_domains)
854 if (sufmatch ((const char **)opt.exclude_domains, u->host))
860 /* Check whether WHAT is matched in LIST, each element of LIST being a
861 pattern to match WHAT against, using backward matching (see
862 match_backwards() in utils.c).
864 If an element of LIST matched, 1 is returned, 0 otherwise. */
866 sufmatch (const char **list, const char *what)
871 for (i = 0; list[i]; i++)
873 for (j = strlen (list[i]), k = lw; j >= 0 && k >= 0; j--, k--)
874 if (c_tolower (list[i][j]) != c_tolower (what[k]))
876 /* The domain must be first to reach to beginning. */
886 if (host_name_addresses_map)
888 hash_table_iterator iter;
889 for (hash_table_iterate (host_name_addresses_map, &iter);
890 hash_table_iter_next (&iter);
893 char *host = iter.key;
894 struct address_list *al = iter.value;
896 assert (al->refcount == 1);
897 address_list_delete (al);
899 hash_table_destroy (host_name_addresses_map);
900 host_name_addresses_map = NULL;