2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 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 2 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, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
25 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
50 # include <netdb.h> /* for h_errno */
64 # include "gen_sslfunc.h"
68 extern char *version_string;
79 static int cookies_loaded_p;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
209 hdr += skip_lws (hdr);
215 for (num = 0; ISDIGIT (*hdr); hdr++)
216 num = 10 * num + (*hdr - '0');
217 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
219 closure->first_byte_pos = num;
221 for (num = 0; ISDIGIT (*hdr); hdr++)
222 num = 10 * num + (*hdr - '0');
223 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
225 closure->last_byte_pos = num;
227 for (num = 0; ISDIGIT (*hdr); hdr++)
228 num = 10 * num + (*hdr - '0');
229 closure->entity_length = num;
233 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
234 Used for `Accept-Ranges'. */
236 http_process_none (const char *hdr, void *arg)
238 int *where = (int *)arg;
240 if (strstr (hdr, "none"))
247 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
249 http_process_type (const char *hdr, void *arg)
251 char **result = (char **)arg;
252 /* Locate P on `;' or the terminating zero, whichever comes first. */
253 const char *p = strchr (hdr, ';');
255 p = hdr + strlen (hdr);
256 while (p > hdr && ISSPACE (*(p - 1)))
258 *result = strdupdelim (hdr, p);
262 /* Check whether the `Connection' header is set to "keep-alive". */
264 http_process_connection (const char *hdr, void *arg)
266 int *flag = (int *)arg;
267 if (!strcasecmp (hdr, "Keep-Alive"))
272 /* Persistent connections. Currently, we cache the most recently used
273 connection as persistent, provided that the HTTP server agrees to
274 make it such. The persistence data is stored in the variables
275 below. Ideally, it would be in a structure, and it should be
276 possible to cache an arbitrary fixed number of these connections.
278 I think the code is quite easy to extend in that direction. */
280 /* Whether a persistent connection is active. */
281 static int pc_active_p;
282 /* Host and port of currently active persistent connection. */
283 static struct address_list *pc_last_host_ip;
284 static unsigned short pc_last_port;
286 /* File descriptor of the currently active persistent connection. */
287 static int pc_last_fd;
290 /* Whether a ssl handshake has occoured on this connection */
291 static int pc_active_ssl;
292 /* SSL connection of the currently active persistent connection. */
293 static SSL *pc_last_ssl;
294 #endif /* HAVE_SSL */
296 /* Mark the persistent connection as invalid. This is used by the
297 CLOSE_* macros after they forcefully close a registered persistent
298 connection. This does not close the file descriptor -- it is left
299 to the caller to do that. (Maybe it should, though.) */
302 invalidate_persistent (void)
307 #endif /* HAVE_SSL */
308 if (pc_last_host_ip != NULL)
310 address_list_release (pc_last_host_ip);
311 pc_last_host_ip = NULL;
313 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
316 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
317 persistent. This will enable someone to use the same connection
318 later. In the context of HTTP, this must be called only AFTER the
319 response has been received and the server has promised that the
320 connection will remain alive.
322 If a previous connection was persistent, it is closed. */
325 register_persistent (const char *host, unsigned short port, int fd
333 if (pc_last_fd == fd)
335 /* The connection FD is already registered. Nothing to
341 /* The old persistent connection is still active; let's
342 close it first. This situation arises whenever a
343 persistent connection exists, but we then connect to a
344 different host, and try to register a persistent
345 connection to that one. */
347 /* The ssl disconnect has to take place before the closing
350 shutdown_ssl(pc_last_ssl);
353 invalidate_persistent ();
357 assert (pc_last_host_ip == NULL);
359 /* This lookup_host cannot fail, because it has the results in the
361 pc_last_host_ip = lookup_host (host, 1);
362 assert (pc_last_host_ip != NULL);
369 pc_active_ssl = ssl ? 1 : 0;
371 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
374 /* Return non-zero if a persistent connection is available for
375 connecting to HOST:PORT. */
378 persistent_available_p (const char *host, unsigned short port
385 struct address_list *this_host_ip;
387 /* First, check whether a persistent connection is active at all. */
390 /* Second, check if the active connection pertains to the correct
391 (HOST, PORT) ordered pair. */
392 if (port != pc_last_port)
396 /* Second, a): check if current connection is (not) ssl, too. This
397 test is unlikely to fail because HTTP and HTTPS typicaly use
398 different ports. Yet it is possible, or so I [Christian
399 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
401 if (ssl != pc_active_ssl)
403 #endif /* HAVE_SSL */
405 this_host_ip = lookup_host (host, 1);
409 /* To equate the two host names for the purposes of persistent
410 connections, they need to share all the IP addresses in the
412 success = address_list_match_all (pc_last_host_ip, this_host_ip);
413 address_list_release (this_host_ip);
417 /* Third: check whether the connection is still open. This is
418 important because most server implement a liberal (short) timeout
419 on persistent connections. Wget can of course always reconnect
420 if the connection doesn't work out, but it's nicer to know in
421 advance. This test is a logical followup of the first test, but
422 is "expensive" and therefore placed at the end of the list. */
423 if (!test_socket_open (pc_last_fd))
425 /* Oops, the socket is no longer open. Now that we know that,
426 let's invalidate the persistent connection before returning
429 invalidate_persistent ();
436 # define SHUTDOWN_SSL(ssl) do { \
438 shutdown_ssl (ssl); \
441 # define SHUTDOWN_SSL(ssl)
444 /* The idea behind these two CLOSE macros is to distinguish between
445 two cases: one when the job we've been doing is finished, and we
446 want to close the connection and leave, and two when something is
447 seriously wrong and we're closing the connection as part of
450 In case of keep_alive, CLOSE_FINISH should leave the connection
451 open, while CLOSE_INVALIDATE should still close it.
453 Note that the semantics of the flag `keep_alive' is "this
454 connection *will* be reused (the server has promised not to close
455 the connection once we're done)", while the semantics of
456 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
457 active, registered connection". */
459 #define CLOSE_FINISH(fd) do { \
462 SHUTDOWN_SSL (ssl); \
464 if (pc_active_p && (fd) == pc_last_fd) \
465 invalidate_persistent (); \
469 #define CLOSE_INVALIDATE(fd) do { \
470 SHUTDOWN_SSL (ssl); \
472 if (pc_active_p && (fd) == pc_last_fd) \
473 invalidate_persistent (); \
478 long len; /* received length */
479 long contlen; /* expected length */
480 long restval; /* the restart value */
481 int res; /* the result of last read */
482 char *newloc; /* new location (redirection) */
483 char *remote_time; /* remote time-stamp string */
484 char *error; /* textual HTTP error */
485 int statcode; /* status code */
486 long dltime; /* time of the download */
487 int no_truncate; /* whether truncating the file is
489 const char *referer; /* value of the referer header. */
490 char **local_file; /* local file. */
494 free_hstat (struct http_stat *hs)
496 FREE_MAYBE (hs->newloc);
497 FREE_MAYBE (hs->remote_time);
498 FREE_MAYBE (hs->error);
500 /* Guard against being called twice. */
502 hs->remote_time = NULL;
506 static char *create_authorization_line PARAMS ((const char *, const char *,
507 const char *, const char *,
509 static char *basic_authentication_encode PARAMS ((const char *, const char *,
511 static int known_authentication_scheme_p PARAMS ((const char *));
513 time_t http_atotm PARAMS ((char *));
515 #define BEGINS_WITH(line, string_constant) \
516 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
517 && (ISSPACE (line[sizeof (string_constant) - 1]) \
518 || !line[sizeof (string_constant) - 1]))
520 /* Retrieve a document through HTTP protocol. It recognizes status
521 code, and correctly handles redirections. It closes the network
522 socket. If it receives an error from the functions below it, it
523 will print it if there is enough information to do so (almost
524 always), returning the error to the caller (i.e. http_loop).
526 Various HTTP parameters are stored to hs. Although it parses the
527 response code correctly, it is not used in a sane way. The caller
530 If PROXY is non-NULL, the connection will be made to the proxy
531 server, and u->url will be requested. */
533 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
535 char *request, *type, *command, *full_path;
537 char *pragma_h, *referer, *useragent, *range, *wwwauth;
538 char *authenticate_h;
542 char *request_keep_alive;
543 int sock, hcount, num_written, all_length, statcode;
544 long contlen, contrange;
547 int auth_tried_already;
550 static SSL_CTX *ssl_ctx = NULL;
552 #endif /* HAVE_SSL */
553 char *cookies = NULL;
555 /* Whether this connection will be kept alive after the HTTP request
559 /* Flags that detect the two ways of specifying HTTP keep-alive
561 int http_keep_alive_1, http_keep_alive_2;
563 /* Whether keep-alive should be inhibited. */
564 int inhibit_keep_alive;
567 /* initialize ssl_ctx on first run */
570 uerr_t err = init_ssl (&ssl_ctx);
575 case SSLERRCTXCREATE:
577 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
581 /* try without certfile */
582 logprintf (LOG_NOTQUIET,
583 _("Failed to load certificates from %s\n"),
586 logprintf (LOG_NOTQUIET,
587 _("Trying without the specified certificate\n"));
590 logprintf (LOG_NOTQUIET,
591 _("Failed to get certificate key from %s\n"),
594 logprintf (LOG_NOTQUIET,
595 _("Trying without the specified certificate\n"));
602 #endif /* HAVE_SSL */
604 if (!(*dt & HEAD_ONLY))
605 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
606 know the local filename so we can save to it. */
607 assert (*hs->local_file != NULL);
610 auth_tried_already = 0;
612 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
615 /* We need to come back here when the initial attempt to retrieve
616 without authorization header fails. (Expected to happen at least
617 for the Digest authorization scheme.) */
620 http_keep_alive_1 = http_keep_alive_2 = 0;
622 /* Initialize certain elements of struct http_stat. */
627 hs->remote_time = NULL;
630 /* If we're using a proxy, we will be connecting to the proxy
632 conn = proxy ? proxy : u;
634 /* First: establish the connection. */
635 if (inhibit_keep_alive
638 !persistent_available_p (conn->host, conn->port)
640 !persistent_available_p (conn->host, conn->port,
641 u->scheme == SCHEME_HTTPS)
642 #endif /* HAVE_SSL */
645 struct address_list *al = lookup_host (conn->host, 0);
648 set_connection_host_name (conn->host);
649 sock = connect_to_many (al, conn->port, 0);
650 set_connection_host_name (NULL);
651 address_list_release (al);
654 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
657 if (conn->scheme == SCHEME_HTTPS)
658 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
660 logputs (LOG_VERBOSE, "\n");
661 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
666 #endif /* HAVE_SSL */
670 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
671 conn->host, conn->port);
672 /* #### pc_last_fd should be accessed through an accessor
677 #endif /* HAVE_SSL */
678 DEBUGP (("Reusing fd %d.\n", sock));
681 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
686 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
687 sprintf (referer, "Referer: %s\r\n", hs->referer);
690 if (*dt & SEND_NOCACHE)
691 pragma_h = "Pragma: no-cache\r\n";
697 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
698 /* Gag me! Some servers (e.g. WebSitePro) have been known to
699 respond to the following `Range' format by generating a
700 multipart/x-byte-ranges MIME document! This MIME type was
701 present in an old draft of the byteranges specification.
702 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
703 only if multiple non-overlapping ranges are requested --
704 which Wget never does. */
705 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
710 STRDUP_ALLOCA (useragent, opt.useragent);
713 useragent = (char *)alloca (10 + strlen (version_string));
714 sprintf (useragent, "Wget/%s", version_string);
716 /* Construct the authentication, if userid is present. */
719 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
720 user = user ? user : opt.http_user;
721 passwd = passwd ? passwd : opt.http_passwd;
728 /* We have the username and the password, but haven't tried
729 any authorization yet. Let's see if the "Basic" method
730 works. If not, we'll come back here and construct a
731 proper authorization method with the right challenges.
733 If we didn't employ this kind of logic, every URL that
734 requires authorization would have to be processed twice,
735 which is very suboptimal and generates a bunch of false
736 "unauthorized" errors in the server log.
738 #### But this logic also has a serious problem when used
739 with stronger authentications: we *first* transmit the
740 username and the password in clear text, and *then*
741 attempt a stronger authentication scheme. That cannot be
742 right! We are only fortunate that almost everyone still
743 uses the `Basic' scheme anyway.
745 There should be an option to prevent this from happening,
746 for those who use strong authentication schemes and value
748 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
752 wwwauth = create_authorization_line (authenticate_h, user, passwd,
760 char *proxy_user, *proxy_passwd;
761 /* For normal username and password, URL components override
762 command-line/wgetrc parameters. With proxy authentication,
763 it's the reverse, because proxy URLs are normally the
764 "permanent" ones, so command-line args should take
766 if (opt.proxy_user && opt.proxy_passwd)
768 proxy_user = opt.proxy_user;
769 proxy_passwd = opt.proxy_passwd;
773 proxy_user = proxy->user;
774 proxy_passwd = proxy->passwd;
776 /* #### This does not appear right. Can't the proxy request,
777 say, `Digest' authentication? */
778 if (proxy_user && proxy_passwd)
779 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
780 "Proxy-Authorization");
783 /* String of the form :PORT. Used only for non-standard ports. */
785 if (u->port != scheme_default_port (u->scheme))
787 port_maybe = (char *)alloca (numdigit (u->port) + 2);
788 sprintf (port_maybe, ":%d", u->port);
791 if (!inhibit_keep_alive)
792 request_keep_alive = "Connection: Keep-Alive\r\n";
794 request_keep_alive = NULL;
797 cookies = build_cookies_request (u->host, u->port, u->path,
799 u->scheme == SCHEME_HTTPS
806 full_path = xstrdup (u->url);
808 full_path = url_full_path (u);
810 /* Allocate the memory for the request. */
811 request = (char *)alloca (strlen (command)
815 + (port_maybe ? strlen (port_maybe) : 0)
816 + strlen (HTTP_ACCEPT)
817 + (request_keep_alive
818 ? strlen (request_keep_alive) : 0)
819 + (referer ? strlen (referer) : 0)
820 + (cookies ? strlen (cookies) : 0)
821 + (wwwauth ? strlen (wwwauth) : 0)
822 + (proxyauth ? strlen (proxyauth) : 0)
823 + (range ? strlen (range) : 0)
825 + (opt.user_header ? strlen (opt.user_header) : 0)
827 /* Construct the request. */
833 %s%s%s%s%s%s%s%s\r\n",
836 port_maybe ? port_maybe : "",
838 request_keep_alive ? request_keep_alive : "",
839 referer ? referer : "",
840 cookies ? cookies : "",
841 wwwauth ? wwwauth : "",
842 proxyauth ? proxyauth : "",
845 opt.user_header ? opt.user_header : "");
846 DEBUGP (("---request begin---\n%s---request end---\n", request));
848 /* Free the temporary memory. */
849 FREE_MAYBE (wwwauth);
850 FREE_MAYBE (proxyauth);
851 FREE_MAYBE (cookies);
854 /* Send the request to server. */
856 if (u->scheme == SCHEME_HTTPS)
857 num_written = ssl_iwrite (ssl, request, strlen (request));
859 #endif /* HAVE_SSL */
860 num_written = iwrite (sock, request, strlen (request));
864 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
866 CLOSE_INVALIDATE (sock);
869 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
870 proxy ? "Proxy" : "HTTP");
871 contlen = contrange = -1;
876 /* Before reading anything, initialize the rbuf. */
877 rbuf_initialize (&rbuf, sock);
879 if (u->scheme == SCHEME_HTTPS)
883 #endif /* HAVE_SSL */
886 /* Header-fetching loop. */
894 /* Get the header. */
895 status = header_get (&rbuf, &hdr,
896 /* Disallow continuations for status line. */
897 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
899 /* Check for errors. */
900 if (status == HG_EOF && *hdr)
902 /* This used to be an unconditional error, but that was
903 somewhat controversial, because of a large number of
904 broken CGI's that happily "forget" to send the second EOL
905 before closing the connection of a HEAD request.
907 So, the deal is to check whether the header is empty
908 (*hdr is zero if it is); if yes, it means that the
909 previous header was fully retrieved, and that -- most
910 probably -- the request is complete. "...be liberal in
911 what you accept." Oh boy. */
912 logputs (LOG_VERBOSE, "\n");
913 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
916 FREE_MAYBE (all_headers);
917 CLOSE_INVALIDATE (sock);
920 else if (status == HG_ERROR)
922 logputs (LOG_VERBOSE, "\n");
923 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
927 FREE_MAYBE (all_headers);
928 CLOSE_INVALIDATE (sock);
932 /* If the headers are to be saved to a file later, save them to
934 if (opt.save_headers)
936 int lh = strlen (hdr);
937 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
938 memcpy (all_headers + all_length, hdr, lh);
940 all_headers[all_length++] = '\n';
941 all_headers[all_length] = '\0';
944 /* Check for status line. */
948 /* Parse the first line of server response. */
949 statcode = parse_http_status_line (hdr, &error);
950 hs->statcode = statcode;
951 /* Store the descriptive response. */
952 if (statcode == -1) /* malformed response */
954 /* A common reason for "malformed response" error is the
955 case when no data was actually received. Handle this
958 hs->error = xstrdup (_("No data received"));
960 hs->error = xstrdup (_("Malformed status line"));
965 hs->error = xstrdup (_("(no description)"));
967 hs->error = xstrdup (error);
975 if (opt.server_response)
976 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
978 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
984 /* Exit on empty header. */
991 /* Print the header if requested. */
992 if (opt.server_response && hcount != 1)
993 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
995 /* Try getting content-length. */
996 if (contlen == -1 && !opt.ignore_length)
997 if (header_process (hdr, "Content-Length", header_extract_number,
1000 /* Try getting content-type. */
1002 if (header_process (hdr, "Content-Type", http_process_type, &type))
1004 /* Try getting location. */
1006 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1008 /* Try getting last-modified. */
1009 if (!hs->remote_time)
1010 if (header_process (hdr, "Last-Modified", header_strdup,
1013 /* Try getting cookies. */
1015 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1017 /* Try getting www-authentication. */
1018 if (!authenticate_h)
1019 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1022 /* Check for accept-ranges header. If it contains the word
1023 `none', disable the ranges. */
1024 if (*dt & ACCEPTRANGES)
1027 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1030 *dt &= ~ACCEPTRANGES;
1034 /* Try getting content-range. */
1035 if (contrange == -1)
1037 struct http_process_range_closure closure;
1038 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1040 contrange = closure.first_byte_pos;
1044 /* Check for keep-alive related responses. */
1045 if (!inhibit_keep_alive)
1047 /* Check for the `Keep-Alive' header. */
1048 if (!http_keep_alive_1)
1050 if (header_process (hdr, "Keep-Alive", header_exists,
1051 &http_keep_alive_1))
1054 /* Check for `Connection: Keep-Alive'. */
1055 if (!http_keep_alive_2)
1057 if (header_process (hdr, "Connection", http_process_connection,
1058 &http_keep_alive_2))
1066 logputs (LOG_VERBOSE, "\n");
1069 && (http_keep_alive_1 || http_keep_alive_2))
1071 assert (inhibit_keep_alive == 0);
1075 /* The server has promised that it will not close the connection
1076 when we're done. This means that we can register it. */
1078 register_persistent (conn->host, conn->port, sock);
1080 register_persistent (conn->host, conn->port, sock, ssl);
1081 #endif /* HAVE_SSL */
1083 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1086 /* Authorization is required. */
1090 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1091 might be more bytes in the body. */
1092 if (auth_tried_already)
1094 /* If we have tried it already, then there is not point
1097 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1098 xfree (authenticate_h);
1101 else if (!known_authentication_scheme_p (authenticate_h))
1103 xfree (authenticate_h);
1104 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1107 else if (BEGINS_WITH (authenticate_h, "Basic"))
1109 /* The authentication scheme is basic, the one we try by
1110 default, and it failed. There's no sense in trying
1116 auth_tried_already = 1;
1120 /* We do not need this anymore. */
1123 xfree (authenticate_h);
1124 authenticate_h = NULL;
1127 /* 20x responses are counted among successful by default. */
1128 if (H_20X (statcode))
1131 /* Return if redirected. */
1132 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1134 /* RFC2068 says that in case of the 300 (multiple choices)
1135 response, the server can output a preferred URL through
1136 `Location' header; otherwise, the request should be treated
1137 like GET. So, if the location is set, it will be a
1138 redirection; otherwise, just proceed normally. */
1139 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1143 logprintf (LOG_VERBOSE,
1144 _("Location: %s%s\n"),
1145 hs->newloc ? hs->newloc : _("unspecified"),
1146 hs->newloc ? _(" [following]") : "");
1147 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1148 might be more bytes in the body. */
1150 FREE_MAYBE (all_headers);
1155 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1158 /* We don't assume text/html by default. */
1161 if (opt.html_extension && (*dt & TEXTHTML))
1162 /* -E / --html-extension / html_extension = on was specified, and this is a
1163 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1164 already the file's suffix, tack on ".html". */
1166 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1168 if (last_period_in_local_filename == NULL ||
1169 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1170 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1172 size_t local_filename_len = strlen(*hs->local_file);
1174 *hs->local_file = xrealloc(*hs->local_file,
1175 local_filename_len + sizeof(".html"));
1176 strcpy(*hs->local_file + local_filename_len, ".html");
1178 *dt |= ADDED_HTML_EXTENSION;
1182 if (contrange == -1)
1184 /* We did not get a content-range header. This means that the
1185 server did not honor our `Range' request. Normally, this
1186 means we should reset hs->restval and continue normally. */
1188 /* However, if `-c' is used, we need to be a bit more careful:
1190 1. If `-c' is specified and the file already existed when
1191 Wget was started, it would be a bad idea for us to start
1192 downloading it from scratch, effectively truncating it. I
1193 believe this cannot happen unless `-c' was specified.
1195 2. If `-c' is used on a file that is already fully
1196 downloaded, we're requesting bytes after the end of file,
1197 which can result in server not honoring `Range'. If this is
1198 the case, `Content-Length' will be equal to the length of the
1200 if (opt.always_rest)
1202 /* Check for condition #2. */
1203 if (hs->restval > 0 /* restart was requested. */
1204 && contlen != -1 /* we got content-length. */
1205 && hs->restval >= contlen /* file fully downloaded
1209 logputs (LOG_VERBOSE, _("\
1210 \n The file is already fully retrieved; nothing to do.\n\n"));
1211 /* In case the caller inspects. */
1215 FREE_MAYBE (all_headers);
1216 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1217 might be more bytes in the body. */
1218 return RETRUNNEEDED;
1221 /* Check for condition #1. */
1222 if (hs->no_truncate)
1224 logprintf (LOG_NOTQUIET,
1227 Continued download failed on this file, which conflicts with `-c'.\n\
1228 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1230 FREE_MAYBE (all_headers);
1231 CLOSE_INVALIDATE (sock);
1232 return CONTNOTSUPPORTED;
1240 else if (contrange != hs->restval ||
1241 (H_PARTIAL (statcode) && contrange == -1))
1243 /* This means the whole request was somehow misunderstood by the
1244 server. Bail out. */
1246 FREE_MAYBE (all_headers);
1247 CLOSE_INVALIDATE (sock);
1254 contlen += contrange;
1256 contrange = -1; /* If conent-length was not sent,
1257 content-range will be ignored. */
1259 hs->contlen = contlen;
1263 if ((*dt & RETROKF) && !opt.server_response)
1265 /* No need to print this output if the body won't be
1266 downloaded at all, or if the original server response is
1268 logputs (LOG_VERBOSE, _("Length: "));
1271 logputs (LOG_VERBOSE, legible (contlen));
1272 if (contrange != -1)
1273 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1274 legible (contlen - contrange));
1277 logputs (LOG_VERBOSE,
1278 opt.ignore_length ? _("ignored") : _("unspecified"));
1280 logprintf (LOG_VERBOSE, " [%s]\n", type);
1282 logputs (LOG_VERBOSE, "\n");
1286 type = NULL; /* We don't need it any more. */
1288 /* Return if we have no intention of further downloading. */
1289 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1291 /* In case the caller cares to look... */
1295 FREE_MAYBE (all_headers);
1296 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1297 might be more bytes in the body. */
1298 return RETRFINISHED;
1301 /* Open the local file. */
1304 mkalldirs (*hs->local_file);
1306 rotate_backups (*hs->local_file);
1307 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1310 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1311 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1312 might be more bytes in the body. */
1313 FREE_MAYBE (all_headers);
1319 extern int global_download_count;
1321 /* To ensure that repeated "from scratch" downloads work for -O
1322 files, we rewind the file pointer, unless restval is
1323 non-zero. (This works only when -O is used on regular files,
1324 but it's still a valuable feature.)
1326 However, this loses when more than one URL is specified on
1327 the command line the second rewinds eradicates the contents
1328 of the first download. Thus we disable the above trick for
1329 all the downloads except the very first one.
1331 #### A possible solution to this would be to remember the
1332 file position in the output document and to seek to that
1333 position, instead of rewinding. */
1334 if (!hs->restval && global_download_count == 0)
1336 /* This will silently fail for streams that don't correspond
1337 to regular files, but that's OK. */
1339 /* ftruncate is needed because opt.dfp is opened in append
1340 mode if opt.always_rest is set. */
1341 ftruncate (fileno (fp), 0);
1346 /* #### This confuses the code that checks for file size. There
1347 should be some overhead information. */
1348 if (opt.save_headers)
1349 fwrite (all_headers, 1, all_length, fp);
1351 /* Get the contents of the document. */
1352 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1353 (contlen != -1 ? contlen : 0),
1354 &rbuf, keep_alive, &hs->dltime);
1357 /* Close or flush the file. We have to be careful to check for
1358 error here. Checking the result of fwrite() is not enough --
1359 errors could go unnoticed! */
1362 flush_res = fclose (fp);
1364 flush_res = fflush (fp);
1365 if (flush_res == EOF)
1368 FREE_MAYBE (all_headers);
1369 CLOSE_FINISH (sock);
1372 return RETRFINISHED;
1375 /* The genuine HTTP loop! This is the part where the retrieval is
1376 retried, and retried, and retried, and... */
1378 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1379 int *dt, struct url *proxy)
1382 int use_ts, got_head = 0; /* time-stamping info */
1383 char *filename_plus_orig_suffix;
1384 char *local_filename = NULL;
1385 char *tms, *suf, *locf, *tmrate;
1387 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1388 long local_size = 0; /* the size of the local file */
1389 size_t filename_len;
1390 struct http_stat hstat; /* HTTP status */
1394 /* This used to be done in main(), but it's a better idea to do it
1395 here so that we don't go through the hoops if we're just using
1397 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1399 load_cookies (opt.cookies_input);
1400 cookies_loaded_p = 1;
1405 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1406 has_wildcards_p because it would also warn on `?', and we know that
1407 shows up in CGI paths a *lot*. */
1408 if (strchr (u->url, '*'))
1409 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1411 /* Determine the local filename. */
1412 if (local_file && *local_file)
1413 hstat.local_file = local_file;
1414 else if (local_file)
1416 *local_file = url_filename (u);
1417 hstat.local_file = local_file;
1421 dummy = url_filename (u);
1422 hstat.local_file = &dummy;
1425 if (!opt.output_document)
1426 locf = *hstat.local_file;
1428 locf = opt.output_document;
1430 hstat.referer = referer;
1432 filename_len = strlen (*hstat.local_file);
1433 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1435 if (opt.noclobber && file_exists_p (*hstat.local_file))
1437 /* If opt.noclobber is turned on and file already exists, do not
1438 retrieve the file */
1439 logprintf (LOG_VERBOSE, _("\
1440 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1441 /* If the file is there, we suppose it's retrieved OK. */
1444 /* #### Bogusness alert. */
1445 /* If its suffix is "html" or "htm", assume text/html. */
1446 if (((suf = suffix (*hstat.local_file)) != NULL)
1447 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1455 if (opt.timestamping)
1457 boolean local_dot_orig_file_exists = FALSE;
1459 if (opt.backup_converted)
1460 /* If -K is specified, we'll act on the assumption that it was specified
1461 last time these files were downloaded as well, and instead of just
1462 comparing local file X against server file X, we'll compare local
1463 file X.orig (if extant, else X) against server file X. If -K
1464 _wasn't_ specified last time, or the server contains files called
1465 *.orig, -N will be back to not operating correctly with -k. */
1467 /* Would a single s[n]printf() call be faster? --dan
1469 Definitely not. sprintf() is horribly slow. It's a
1470 different question whether the difference between the two
1471 affects a program. Usually I'd say "no", but at one
1472 point I profiled Wget, and found that a measurable and
1473 non-negligible amount of time was lost calling sprintf()
1474 in url.c. Replacing sprintf with inline calls to
1475 strcpy() and long_to_string() made a difference.
1477 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1478 memcpy (filename_plus_orig_suffix + filename_len,
1479 ".orig", sizeof (".orig"));
1481 /* Try to stat() the .orig file. */
1482 if (stat (filename_plus_orig_suffix, &st) == 0)
1484 local_dot_orig_file_exists = TRUE;
1485 local_filename = filename_plus_orig_suffix;
1489 if (!local_dot_orig_file_exists)
1490 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1491 if (stat (*hstat.local_file, &st) == 0)
1492 local_filename = *hstat.local_file;
1494 if (local_filename != NULL)
1495 /* There was a local file, so we'll check later to see if the version
1496 the server has is the same version we already have, allowing us to
1501 local_size = st.st_size;
1505 /* Reset the counter. */
1507 *dt = 0 | ACCEPTRANGES;
1511 /* Increment the pass counter. */
1513 sleep_between_retrievals (count);
1514 /* Get the current time string. */
1515 tms = time_str (NULL);
1516 /* Print fetch message, if opt.verbose. */
1519 char *hurl = url_string (u, 1);
1523 sprintf (tmp, _("(try:%2d)"), count);
1524 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1525 tms, hurl, tmp, locf);
1527 ws_changetitle (hurl, 1);
1532 /* Default document type is empty. However, if spider mode is
1533 on or time-stamping is employed, HEAD_ONLY commands is
1534 encoded within *dt. */
1535 if (opt.spider || (use_ts && !got_head))
1539 /* Assume no restarting. */
1541 /* Decide whether or not to restart. */
1542 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1543 /* #### this calls access() and then stat(); could be optimized. */
1544 && file_exists_p (locf))
1545 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1546 hstat.restval = st.st_size;
1548 /* In `-c' is used and the file is existing and non-empty,
1549 refuse to truncate it if the server doesn't support continued
1551 hstat.no_truncate = 0;
1552 if (opt.always_rest && hstat.restval)
1553 hstat.no_truncate = 1;
1555 /* Decide whether to send the no-cache directive. We send it in
1557 a) we're using a proxy, and we're past our first retrieval.
1558 Some proxies are notorious for caching incomplete data, so
1559 we require a fresh get.
1560 b) caching is explicitly inhibited. */
1561 if ((proxy && count > 1) /* a */
1562 || !opt.allow_cache /* b */
1564 *dt |= SEND_NOCACHE;
1566 *dt &= ~SEND_NOCACHE;
1568 /* Try fetching the document, or at least its head. */
1569 err = gethttp (u, &hstat, dt, proxy);
1571 /* It's unfortunate that wget determines the local filename before finding
1572 out the Content-Type of the file. Barring a major restructuring of the
1573 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1574 *hstat.local_file to tack on ".html". */
1575 if (!opt.output_document)
1576 locf = *hstat.local_file;
1578 locf = opt.output_document;
1581 tms = time_str (NULL);
1582 /* Get the new location (with or without the redirection). */
1584 *newloc = xstrdup (hstat.newloc);
1587 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1588 case CONERROR: case READERR: case WRITEFAILED:
1590 /* Non-fatal errors continue executing the loop, which will
1591 bring them to "while" statement at the end, to judge
1592 whether the number of tries was exceeded. */
1593 free_hstat (&hstat);
1594 printwhat (count, opt.ntry);
1597 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1598 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1599 /* Fatal errors just return from the function. */
1600 free_hstat (&hstat);
1604 case FWRITEERR: case FOPENERR:
1605 /* Another fatal error. */
1606 logputs (LOG_VERBOSE, "\n");
1607 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1608 *hstat.local_file, strerror (errno));
1609 free_hstat (&hstat);
1614 /* Another fatal error. */
1615 logputs (LOG_VERBOSE, "\n");
1616 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1617 free_hstat (&hstat);
1622 /* Return the new location to the caller. */
1625 logprintf (LOG_NOTQUIET,
1626 _("ERROR: Redirection (%d) without location.\n"),
1628 free_hstat (&hstat);
1632 free_hstat (&hstat);
1637 /* The file was already fully retrieved. */
1638 free_hstat (&hstat);
1643 /* Deal with you later. */
1646 /* All possibilities should have been exhausted. */
1649 if (!(*dt & RETROKF))
1653 /* #### Ugly ugly ugly! */
1654 char *hurl = url_string (u, 1);
1655 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1658 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1659 tms, hstat.statcode, hstat.error);
1660 logputs (LOG_VERBOSE, "\n");
1661 free_hstat (&hstat);
1666 /* Did we get the time-stamp? */
1669 if (opt.timestamping && !hstat.remote_time)
1671 logputs (LOG_NOTQUIET, _("\
1672 Last-modified header missing -- time-stamps turned off.\n"));
1674 else if (hstat.remote_time)
1676 /* Convert the date-string into struct tm. */
1677 tmr = http_atotm (hstat.remote_time);
1678 if (tmr == (time_t) (-1))
1679 logputs (LOG_VERBOSE, _("\
1680 Last-modified header invalid -- time-stamp ignored.\n"));
1684 /* The time-stamping section. */
1689 use_ts = 0; /* no more time-stamping */
1690 count = 0; /* the retrieve count for HEAD is
1692 if (hstat.remote_time && tmr != (time_t) (-1))
1694 /* Now time-stamping can be used validly. Time-stamping
1695 means that if the sizes of the local and remote file
1696 match, and local file is newer than the remote file,
1697 it will not be retrieved. Otherwise, the normal
1698 download procedure is resumed. */
1700 (hstat.contlen == -1 || local_size == hstat.contlen))
1702 logprintf (LOG_VERBOSE, _("\
1703 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1705 free_hstat (&hstat);
1709 else if (tml >= tmr)
1710 logprintf (LOG_VERBOSE, _("\
1711 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1713 logputs (LOG_VERBOSE,
1714 _("Remote file is newer, retrieving.\n"));
1716 free_hstat (&hstat);
1719 if ((tmr != (time_t) (-1))
1721 && ((hstat.len == hstat.contlen) ||
1722 ((hstat.res == 0) &&
1723 ((hstat.contlen == -1) ||
1724 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1726 /* #### This code repeats in http.c and ftp.c. Move it to a
1728 const char *fl = NULL;
1729 if (opt.output_document)
1731 if (opt.od_known_regular)
1732 fl = opt.output_document;
1735 fl = *hstat.local_file;
1739 /* End of time-stamping section. */
1743 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1748 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1750 if (hstat.len == hstat.contlen)
1754 logprintf (LOG_VERBOSE,
1755 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1756 tms, tmrate, locf, hstat.len, hstat.contlen);
1757 logprintf (LOG_NONVERBOSE,
1758 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1759 tms, u->url, hstat.len, hstat.contlen, locf, count);
1762 downloaded_increase (hstat.len);
1764 /* Remember that we downloaded the file for later ".orig" code. */
1765 if (*dt & ADDED_HTML_EXTENSION)
1766 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1768 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1770 free_hstat (&hstat);
1774 else if (hstat.res == 0) /* No read error */
1776 if (hstat.contlen == -1) /* We don't know how much we were supposed
1777 to get, so assume we succeeded. */
1781 logprintf (LOG_VERBOSE,
1782 _("%s (%s) - `%s' saved [%ld]\n\n"),
1783 tms, tmrate, locf, hstat.len);
1784 logprintf (LOG_NONVERBOSE,
1785 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1786 tms, u->url, hstat.len, locf, count);
1789 downloaded_increase (hstat.len);
1791 /* Remember that we downloaded the file for later ".orig" code. */
1792 if (*dt & ADDED_HTML_EXTENSION)
1793 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1795 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1797 free_hstat (&hstat);
1801 else if (hstat.len < hstat.contlen) /* meaning we lost the
1802 connection too soon */
1804 logprintf (LOG_VERBOSE,
1805 _("%s (%s) - Connection closed at byte %ld. "),
1806 tms, tmrate, hstat.len);
1807 printwhat (count, opt.ntry);
1808 free_hstat (&hstat);
1811 else if (!opt.kill_longer) /* meaning we got more than expected */
1813 logprintf (LOG_VERBOSE,
1814 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1815 tms, tmrate, locf, hstat.len, hstat.contlen);
1816 logprintf (LOG_NONVERBOSE,
1817 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1818 tms, u->url, hstat.len, hstat.contlen, locf, count);
1820 downloaded_increase (hstat.len);
1822 /* Remember that we downloaded the file for later ".orig" code. */
1823 if (*dt & ADDED_HTML_EXTENSION)
1824 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1826 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1828 free_hstat (&hstat);
1832 else /* the same, but not accepted */
1834 logprintf (LOG_VERBOSE,
1835 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1836 tms, tmrate, hstat.len, hstat.contlen);
1837 printwhat (count, opt.ntry);
1838 free_hstat (&hstat);
1842 else /* now hstat.res can only be -1 */
1844 if (hstat.contlen == -1)
1846 logprintf (LOG_VERBOSE,
1847 _("%s (%s) - Read error at byte %ld (%s)."),
1848 tms, tmrate, hstat.len, strerror (errno));
1849 printwhat (count, opt.ntry);
1850 free_hstat (&hstat);
1853 else /* hstat.res == -1 and contlen is given */
1855 logprintf (LOG_VERBOSE,
1856 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1857 tms, tmrate, hstat.len, hstat.contlen,
1859 printwhat (count, opt.ntry);
1860 free_hstat (&hstat);
1867 while (!opt.ntry || (count < opt.ntry));
1871 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1872 than local timezone.
1874 mktime is similar but assumes struct tm, also known as the
1875 "broken-down" form of time, is in local time zone. mktime_from_utc
1876 uses mktime to make the conversion understanding that an offset
1877 will be introduced by the local time assumption.
1879 mktime_from_utc then measures the introduced offset by applying
1880 gmtime to the initial result and applying mktime to the resulting
1881 "broken-down" form. The difference between the two mktime results
1882 is the measured offset which is then subtracted from the initial
1883 mktime result to yield a calendar time which is the value returned.
1885 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1886 consistent offset (the non DST offset) since tm and tm+o might be
1887 on opposite sides of a DST change.
1889 Some implementations of mktime return -1 for the nonexistent
1890 localtime hour at the beginning of DST. In this event, use
1891 mktime(tm - 1hr) + 3600.
1895 gmtime(t+o) --> tm+o
1896 mktime(tm+o) --> t+2o
1897 t+o - (t+2o - t+o) = t
1899 Note that glibc contains a function of the same purpose named
1900 `timegm' (reverse of gmtime). But obviously, it is not universally
1901 available, and unfortunately it is not straightforwardly
1902 extractable for use here. Perhaps configure should detect timegm
1903 and use it where available.
1905 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1906 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1907 Further improved by Roger with assistance from Edward J. Sabol
1908 based on input by Jamie Zawinski. */
1911 mktime_from_utc (struct tm *t)
1922 return -1; /* can't deal with output from strptime */
1933 return -1; /* can't deal with output from gmtime */
1936 return (tl - (tb - tl));
1939 /* Check whether the result of strptime() indicates success.
1940 strptime() returns the pointer to how far it got to in the string.
1941 The processing has been successful if the string is at `GMT' or
1942 `+X', or at the end of the string.
1944 In extended regexp parlance, the function returns 1 if P matches
1945 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1946 can return) is considered a failure and 0 is returned. */
1948 check_end (const char *p)
1952 while (ISSPACE (*p))
1955 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1956 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1962 /* Convert the textual specification of time in TIME_STRING to the
1963 number of seconds since the Epoch.
1965 TIME_STRING can be in any of the three formats RFC2068 allows the
1966 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1967 Timezones are ignored, and should be GMT.
1969 Return the computed time_t representation, or -1 if the conversion
1972 This function uses strptime with various string formats for parsing
1973 TIME_STRING. This results in a parser that is not as lenient in
1974 interpreting TIME_STRING as I would like it to be. Being based on
1975 strptime, it always allows shortened months, one-digit days, etc.,
1976 but due to the multitude of formats in which time can be
1977 represented, an ideal HTTP time parser would be even more
1978 forgiving. It should completely ignore things like week days and
1979 concentrate only on the various forms of representing years,
1980 months, days, hours, minutes, and seconds. For example, it would
1981 be nice if it accepted ISO 8601 out of the box.
1983 I've investigated free and PD code for this purpose, but none was
1984 usable. getdate was big and unwieldy, and had potential copyright
1985 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1986 distributed with phttpd, is excellent, but we cannot use it because
1987 it is not assigned to the FSF. So I stuck it with strptime. */
1990 http_atotm (char *time_string)
1992 /* NOTE: Solaris strptime man page claims that %n and %t match white
1993 space, but that's not universally available. Instead, we simply
1994 use ` ' to mean "skip all WS", which works under all strptime
1995 implementations I've tested. */
1997 static const char *time_formats[] = {
1998 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1999 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2000 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2001 (google.com uses this for their cookies.) */
2002 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2008 /* According to Roger Beeman, we need to initialize tm_isdst, since
2009 strptime won't do it. */
2012 /* Note that under foreign locales Solaris strptime() fails to
2013 recognize English dates, which renders this function useless. We
2014 solve this by being careful not to affect LC_TIME when
2015 initializing locale.
2017 Another solution would be to temporarily set locale to C, invoke
2018 strptime(), and restore it back. This is slow and dirty,
2019 however, and locale support other than LC_MESSAGES can mess other
2020 things, so I rather chose to stick with just setting LC_MESSAGES.
2022 GNU strptime does not have this problem because it recognizes
2023 both international and local dates. */
2025 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2026 if (check_end (strptime (time_string, time_formats[i], &t)))
2027 return mktime_from_utc (&t);
2029 /* All formats have failed. */
2033 /* Authorization support: We support two authorization schemes:
2035 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2037 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2038 consisting of answering to the server's challenge with the proper
2041 /* How many bytes it will take to store LEN bytes in base64. */
2042 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2044 /* Encode the string S of length LENGTH to base64 format and place it
2045 to STORE. STORE will be 0-terminated, and must point to a writable
2046 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2048 base64_encode (const char *s, char *store, int length)
2050 /* Conversion table. */
2051 static char tbl[64] = {
2052 'A','B','C','D','E','F','G','H',
2053 'I','J','K','L','M','N','O','P',
2054 'Q','R','S','T','U','V','W','X',
2055 'Y','Z','a','b','c','d','e','f',
2056 'g','h','i','j','k','l','m','n',
2057 'o','p','q','r','s','t','u','v',
2058 'w','x','y','z','0','1','2','3',
2059 '4','5','6','7','8','9','+','/'
2062 unsigned char *p = (unsigned char *)store;
2064 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2065 for (i = 0; i < length; i += 3)
2067 *p++ = tbl[s[0] >> 2];
2068 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2069 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2070 *p++ = tbl[s[2] & 0x3f];
2073 /* Pad the result if necessary... */
2074 if (i == length + 1)
2076 else if (i == length + 2)
2077 *(p - 1) = *(p - 2) = '=';
2078 /* ...and zero-terminate it. */
2082 /* Create the authentication header contents for the `Basic' scheme.
2083 This is done by encoding the string `USER:PASS' in base64 and
2084 prepending `HEADER: Basic ' to it. */
2086 basic_authentication_encode (const char *user, const char *passwd,
2089 char *t1, *t2, *res;
2090 int len1 = strlen (user) + 1 + strlen (passwd);
2091 int len2 = BASE64_LENGTH (len1);
2093 t1 = (char *)alloca (len1 + 1);
2094 sprintf (t1, "%s:%s", user, passwd);
2095 t2 = (char *)alloca (1 + len2);
2096 base64_encode (t1, t2, len1);
2097 res = (char *)xmalloc (len2 + 11 + strlen (header));
2098 sprintf (res, "%s: Basic %s\r\n", header, t2);
2104 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2105 of a field in such a header. If the field is the one specified by
2106 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2107 digest authorization code), extract its value in the (char*)
2108 variable pointed by RET. Returns negative on a malformed header,
2109 or number of bytes that have been parsed by this call. */
2111 extract_header_attr (const char *au, const char *attr_name, char **ret)
2113 const char *cp, *ep;
2117 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2119 cp += strlen (attr_name);
2122 cp += skip_lws (cp);
2127 cp += skip_lws (cp);
2132 for (ep = cp; *ep && *ep != '\"'; ep++)
2137 *ret = strdupdelim (cp, ep);
2144 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2145 an array of 16 bytes containing the hash keys, and BUF should be a
2146 buffer of 33 writable characters (32 for hex digits plus one for
2147 zero termination). */
2149 dump_hash (unsigned char *buf, const unsigned char *hash)
2153 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2155 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2156 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2161 /* Take the line apart to find the challenge, and compose a digest
2162 authorization header. See RFC2069 section 2.1.2. */
2164 digest_authentication_encode (const char *au, const char *user,
2165 const char *passwd, const char *method,
2168 static char *realm, *opaque, *nonce;
2173 { "realm", &realm },
2174 { "opaque", &opaque },
2179 realm = opaque = nonce = NULL;
2181 au += 6; /* skip over `Digest' */
2186 au += skip_lws (au);
2187 for (i = 0; i < ARRAY_SIZE (options); i++)
2189 int skip = extract_header_attr (au, options[i].name,
2190 options[i].variable);
2194 FREE_MAYBE (opaque);
2204 if (i == ARRAY_SIZE (options))
2206 while (*au && *au != '=')
2210 au += skip_lws (au);
2214 while (*au && *au != '\"')
2221 while (*au && *au != ',')
2226 if (!realm || !nonce || !user || !passwd || !path || !method)
2229 FREE_MAYBE (opaque);
2234 /* Calculate the digest value. */
2236 MD5_CONTEXT_TYPE ctx;
2237 unsigned char hash[MD5_HASHLEN];
2238 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2239 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2241 /* A1BUF = H(user ":" realm ":" password) */
2243 MD5_UPDATE (user, strlen (user), &ctx);
2244 MD5_UPDATE (":", 1, &ctx);
2245 MD5_UPDATE (realm, strlen (realm), &ctx);
2246 MD5_UPDATE (":", 1, &ctx);
2247 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2248 MD5_FINISH (&ctx, hash);
2249 dump_hash (a1buf, hash);
2251 /* A2BUF = H(method ":" path) */
2253 MD5_UPDATE (method, strlen (method), &ctx);
2254 MD5_UPDATE (":", 1, &ctx);
2255 MD5_UPDATE (path, strlen (path), &ctx);
2256 MD5_FINISH (&ctx, hash);
2257 dump_hash (a2buf, hash);
2259 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2261 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2262 MD5_UPDATE (":", 1, &ctx);
2263 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2264 MD5_UPDATE (":", 1, &ctx);
2265 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2266 MD5_FINISH (&ctx, hash);
2267 dump_hash (response_digest, hash);
2269 res = (char*) xmalloc (strlen (user)
2274 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2275 + (opaque ? strlen (opaque) : 0)
2277 sprintf (res, "Authorization: Digest \
2278 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2279 user, realm, nonce, path, response_digest);
2282 char *p = res + strlen (res);
2283 strcat (p, ", opaque=\"");
2287 strcat (res, "\r\n");
2291 #endif /* USE_DIGEST */
2294 #define BEGINS_WITH(line, string_constant) \
2295 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2296 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2297 || !line[sizeof (string_constant) - 1]))
2300 known_authentication_scheme_p (const char *au)
2302 return BEGINS_WITH (au, "Basic")
2303 || BEGINS_WITH (au, "Digest")
2304 || BEGINS_WITH (au, "NTLM");
2309 /* Create the HTTP authorization request header. When the
2310 `WWW-Authenticate' response header is seen, according to the
2311 authorization scheme specified in that header (`Basic' and `Digest'
2312 are supported by the current implementation), produce an
2313 appropriate HTTP authorization request header. */
2315 create_authorization_line (const char *au, const char *user,
2316 const char *passwd, const char *method,
2319 char *wwwauth = NULL;
2321 if (!strncasecmp (au, "Basic", 5))
2322 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2323 if (!strncasecmp (au, "NTLM", 4))
2324 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2326 else if (!strncasecmp (au, "Digest", 6))
2327 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2328 #endif /* USE_DIGEST */
2335 if (pc_last_host_ip)
2336 address_list_release (pc_last_host_ip);