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 */
67 # include "gen_sslfunc.h"
71 extern char *version_string;
82 static int cookies_loaded_p;
84 #define TEXTHTML_S "text/html"
85 #define HTTP_ACCEPT "*/*"
87 /* Some status code validation macros: */
88 #define H_20X(x) (((x) >= 200) && ((x) < 300))
89 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
90 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
91 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
93 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
95 #define HTTP_STATUS_OK 200
96 #define HTTP_STATUS_CREATED 201
97 #define HTTP_STATUS_ACCEPTED 202
98 #define HTTP_STATUS_NO_CONTENT 204
99 #define HTTP_STATUS_PARTIAL_CONTENTS 206
101 /* Redirection 3xx. */
102 #define HTTP_STATUS_MULTIPLE_CHOICES 300
103 #define HTTP_STATUS_MOVED_PERMANENTLY 301
104 #define HTTP_STATUS_MOVED_TEMPORARILY 302
105 #define HTTP_STATUS_NOT_MODIFIED 304
107 /* Client error 4xx. */
108 #define HTTP_STATUS_BAD_REQUEST 400
109 #define HTTP_STATUS_UNAUTHORIZED 401
110 #define HTTP_STATUS_FORBIDDEN 403
111 #define HTTP_STATUS_NOT_FOUND 404
113 /* Server errors 5xx. */
114 #define HTTP_STATUS_INTERNAL 500
115 #define HTTP_STATUS_NOT_IMPLEMENTED 501
116 #define HTTP_STATUS_BAD_GATEWAY 502
117 #define HTTP_STATUS_UNAVAILABLE 503
120 /* Parse the HTTP status line, which is of format:
122 HTTP-Version SP Status-Code SP Reason-Phrase
124 The function returns the status-code, or -1 if the status line is
125 malformed. The pointer to reason-phrase is returned in RP. */
127 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
129 /* (the variables must not be named `major' and `minor', because
130 that breaks compilation with SunOS4 cc.) */
131 int mjr, mnr, statcode;
134 *reason_phrase_ptr = NULL;
136 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
137 major version, and Y is minor version. */
138 if (strncmp (line, "HTTP/", 5) != 0)
142 /* Calculate major HTTP version. */
144 for (mjr = 0; ISDIGIT (*line); line++)
145 mjr = 10 * mjr + (*line - '0');
146 if (*line != '.' || p == line)
150 /* Calculate minor HTTP version. */
152 for (mnr = 0; ISDIGIT (*line); line++)
153 mnr = 10 * mnr + (*line - '0');
154 if (*line != ' ' || p == line)
156 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
157 minor version can be safely ignored. */
162 /* Calculate status code. */
163 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
165 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
167 /* Set up the reason phrase pointer. */
169 /* RFC2068 requires SPC here, but we allow the string to finish
170 here, in case no reason-phrase is present. */
174 *reason_phrase_ptr = line;
179 *reason_phrase_ptr = line + 1;
184 /* Functions to be used as arguments to header_process(): */
186 struct http_process_range_closure {
192 /* Parse the `Content-Range' header and extract the information it
193 contains. Returns 1 if successful, -1 otherwise. */
195 http_process_range (const char *hdr, void *arg)
197 struct http_process_range_closure *closure
198 = (struct http_process_range_closure *)arg;
201 /* Certain versions of Nutscape proxy server send out
202 `Content-Length' without "bytes" specifier, which is a breach of
203 RFC2068 (as well as the HTTP/1.1 draft which was current at the
204 time). But hell, I must support it... */
205 if (!strncasecmp (hdr, "bytes", 5))
208 hdr += skip_lws (hdr);
214 for (num = 0; ISDIGIT (*hdr); hdr++)
215 num = 10 * num + (*hdr - '0');
216 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
218 closure->first_byte_pos = num;
220 for (num = 0; ISDIGIT (*hdr); hdr++)
221 num = 10 * num + (*hdr - '0');
222 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
224 closure->last_byte_pos = num;
226 for (num = 0; ISDIGIT (*hdr); hdr++)
227 num = 10 * num + (*hdr - '0');
228 closure->entity_length = num;
232 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
233 Used for `Accept-Ranges'. */
235 http_process_none (const char *hdr, void *arg)
237 int *where = (int *)arg;
239 if (strstr (hdr, "none"))
246 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
248 http_process_type (const char *hdr, void *arg)
250 char **result = (char **)arg;
251 /* Locate P on `;' or the terminating zero, whichever comes first. */
252 const char *p = strchr (hdr, ';');
254 p = hdr + strlen (hdr);
255 while (p > hdr && ISSPACE (*(p - 1)))
257 *result = strdupdelim (hdr, p);
261 /* Check whether the `Connection' header is set to "keep-alive". */
263 http_process_connection (const char *hdr, void *arg)
265 int *flag = (int *)arg;
266 if (!strcasecmp (hdr, "Keep-Alive"))
271 /* Persistent connections. Currently, we cache the most recently used
272 connection as persistent, provided that the HTTP server agrees to
273 make it such. The persistence data is stored in the variables
274 below. Ideally, it would be in a structure, and it should be
275 possible to cache an arbitrary fixed number of these connections.
277 I think the code is quite easy to extend in that direction. */
279 /* Whether a persistent connection is active. */
280 static int pc_active_p;
281 /* Host and port of currently active persistent connection. */
282 static unsigned char pc_last_host[4];
283 static unsigned short pc_last_port;
285 /* File descriptor of the currently active persistent connection. */
286 static int pc_last_fd;
289 /* Whether a ssl handshake has occoured on this connection */
290 static int pc_active_ssl;
291 /* SSL connection of the currently active persistent connection. */
292 static SSL *pc_last_ssl;
293 #endif /* HAVE_SSL */
295 /* Mark the persistent connection as invalid. This is used by the
296 CLOSE_* macros after they forcefully close a registered persistent
297 connection. This does not close the file descriptor -- it is left
298 to the caller to do that. (Maybe it should, though.) */
301 invalidate_persistent (void)
306 #endif /* HAVE_SSL */
307 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
310 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
311 persistent. This will enable someone to use the same connection
312 later. In the context of HTTP, this must be called only AFTER the
313 response has been received and the server has promised that the
314 connection will remain alive.
316 If a previous connection was persistent, it is closed. */
319 register_persistent (const char *host, unsigned short port, int fd
329 if (pc_last_fd == fd)
331 /* The connection FD is already registered. Nothing to
337 /* The old persistent connection is still active; let's
338 close it first. This situation arises whenever a
339 persistent connection exists, but we then connect to a
340 different host, and try to register a persistent
341 connection to that one. */
343 /* The ssl disconnect has to take place before the closing
346 shutdown_ssl(pc_last_ssl);
349 invalidate_persistent ();
353 /* This store_hostaddress may not fail, because it has the results
355 success = store_hostaddress (pc_last_host, host);
362 pc_active_ssl = ssl ? 1 : 0;
364 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
367 /* Return non-zero if a persistent connection is available for
368 connecting to HOST:PORT. */
371 persistent_available_p (const char *host, unsigned short port
377 unsigned char this_host[4];
378 /* First, check whether a persistent connection is active at all. */
381 /* Second, check if the active connection pertains to the correct
382 (HOST, PORT) ordered pair. */
383 if (port != pc_last_port)
386 /* Second, a): check if current connection is (not) ssl, too. This
387 test is unlikely to fail because HTTP and HTTPS typicaly use
388 different ports. Yet it is possible, or so I [Christian
389 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
391 if (ssl != pc_active_ssl)
393 #endif /* HAVE_SSL */
394 if (!store_hostaddress (this_host, host))
396 if (memcmp (pc_last_host, this_host, 4))
398 /* Third: check whether the connection is still open. This is
399 important because most server implement a liberal (short) timeout
400 on persistent connections. Wget can of course always reconnect
401 if the connection doesn't work out, but it's nicer to know in
402 advance. This test is a logical followup of the first test, but
403 is "expensive" and therefore placed at the end of the list. */
404 if (!test_socket_open (pc_last_fd))
406 /* Oops, the socket is no longer open. Now that we know that,
407 let's invalidate the persistent connection before returning
410 invalidate_persistent ();
417 # define SHUTDOWN_SSL(ssl) do { \
419 shutdown_ssl (ssl); \
422 # define SHUTDOWN_SSL(ssl)
425 /* The idea behind these two CLOSE macros is to distinguish between
426 two cases: one when the job we've been doing is finished, and we
427 want to close the connection and leave, and two when something is
428 seriously wrong and we're closing the connection as part of
431 In case of keep_alive, CLOSE_FINISH should leave the connection
432 open, while CLOSE_INVALIDATE should still close it.
434 Note that the semantics of the flag `keep_alive' is "this
435 connection *will* be reused (the server has promised not to close
436 the connection once we're done)", while the semantics of
437 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
438 active, registered connection". */
440 #define CLOSE_FINISH(fd) do { \
443 SHUTDOWN_SSL (ssl); \
445 if (pc_active_p && (fd) == pc_last_fd) \
446 invalidate_persistent (); \
450 #define CLOSE_INVALIDATE(fd) do { \
451 SHUTDOWN_SSL (ssl); \
453 if (pc_active_p && (fd) == pc_last_fd) \
454 invalidate_persistent (); \
459 long len; /* received length */
460 long contlen; /* expected length */
461 long restval; /* the restart value */
462 int res; /* the result of last read */
463 char *newloc; /* new location (redirection) */
464 char *remote_time; /* remote time-stamp string */
465 char *error; /* textual HTTP error */
466 int statcode; /* status code */
467 long dltime; /* time of the download */
468 int no_truncate; /* whether truncating the file is
472 /* Free the elements of hstat X. */
473 #define FREEHSTAT(x) do \
475 FREE_MAYBE ((x).newloc); \
476 FREE_MAYBE ((x).remote_time); \
477 FREE_MAYBE ((x).error); \
478 (x).newloc = (x).remote_time = (x).error = NULL; \
481 static char *create_authorization_line PARAMS ((const char *, const char *,
482 const char *, const char *,
484 static char *basic_authentication_encode PARAMS ((const char *, const char *,
486 static int known_authentication_scheme_p PARAMS ((const char *));
488 time_t http_atotm PARAMS ((char *));
490 #define BEGINS_WITH(line, string_constant) \
491 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
492 && (ISSPACE (line[sizeof (string_constant) - 1]) \
493 || !line[sizeof (string_constant) - 1]))
495 /* Retrieve a document through HTTP protocol. It recognizes status
496 code, and correctly handles redirections. It closes the network
497 socket. If it receives an error from the functions below it, it
498 will print it if there is enough information to do so (almost
499 always), returning the error to the caller (i.e. http_loop).
501 Various HTTP parameters are stored to hs. Although it parses the
502 response code correctly, it is not used in a sane way. The caller
505 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
506 and u->proxy->url will be given to the proxy server (bad naming,
509 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
511 char *request, *type, *command, *path;
513 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
514 char *authenticate_h;
518 char *request_keep_alive;
519 int sock, hcount, num_written, all_length, remport, statcode;
520 long contlen, contrange;
524 int auth_tried_already;
527 static SSL_CTX *ssl_ctx = NULL;
529 #endif /* HAVE_SSL */
530 struct wget_timer *timer;
531 char *cookies = NULL;
533 /* Whether this connection will be kept alive after the HTTP request
537 /* Flags that detect the two ways of specifying HTTP keep-alive
539 int http_keep_alive_1, http_keep_alive_2;
541 /* Whether keep-alive should be inhibited. */
542 int inhibit_keep_alive;
545 /* initialize ssl_ctx on first run */
548 err=init_ssl (&ssl_ctx);
553 case SSLERRCTXCREATE:
555 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
559 /* try without certfile */
560 logprintf (LOG_NOTQUIET,
561 _("Failed to load certificates from %s\n"),
564 logprintf (LOG_NOTQUIET,
565 _("Trying without the specified certificate\n"));
568 logprintf (LOG_NOTQUIET,
569 _("Failed to get certificate key from %s\n"),
572 logprintf (LOG_NOTQUIET,
573 _("Trying without the specified certificate\n"));
580 #endif /* HAVE_SSL */
582 if (!(*dt & HEAD_ONLY))
583 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
584 know the local filename so we can save to it. */
585 assert (u->local != NULL);
588 auth_tried_already = 0;
590 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
593 /* We need to come back here when the initial attempt to retrieve
594 without authorization header fails. (Expected to happen at least
595 for the Digest authorization scheme.) */
598 http_keep_alive_1 = http_keep_alive_2 = 0;
600 /* Initialize certain elements of struct http_stat. */
605 hs->remote_time = NULL;
608 /* Which structure to use to retrieve the original URL data. */
614 /* First: establish the connection. */
615 if (inhibit_keep_alive
618 !persistent_available_p (u->host, u->port)
620 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
621 #endif /* HAVE_SSL */
624 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
625 err = make_connection (&sock, u->host, u->port);
629 logputs (LOG_VERBOSE, "\n");
630 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
634 logputs (LOG_VERBOSE, "\n");
635 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
639 logputs (LOG_VERBOSE, "\n");
640 logprintf (LOG_NOTQUIET,
641 _("Connection to %s:%hu refused.\n"), u->host, u->port);
645 logputs (LOG_VERBOSE, "\n");
646 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
651 /* Everything is fine! */
652 logputs (LOG_VERBOSE, _("connected!\n"));
659 if (u->proto == URLHTTPS)
660 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
662 logputs (LOG_VERBOSE, "\n");
663 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
668 #endif /* HAVE_SSL */
672 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
673 /* #### pc_last_fd should be accessed through an accessor
678 #endif /* HAVE_SSL */
679 DEBUGP (("Reusing fd %d.\n", sock));
683 path = u->proxy->url;
687 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
691 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
692 sprintf (referer, "Referer: %s\r\n", ou->referer);
694 if (*dt & SEND_NOCACHE)
695 pragma_h = "Pragma: no-cache\r\n";
700 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
701 /* Gag me! Some servers (e.g. WebSitePro) have been known to
702 respond to the following `Range' format by generating a
703 multipart/x-byte-ranges MIME document! This MIME type was
704 present in an old draft of the byteranges specification.
705 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
706 only if multiple non-overlapping ranges are requested --
707 which Wget never does. */
708 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
713 STRDUP_ALLOCA (useragent, opt.useragent);
716 useragent = (char *)alloca (10 + strlen (version_string));
717 sprintf (useragent, "Wget/%s", version_string);
719 /* Construct the authentication, if userid is present. */
722 search_netrc (ou->host, (const char **)&user, (const char **)&passwd, 0);
723 user = user ? user : opt.http_user;
724 passwd = passwd ? passwd : opt.http_passwd;
731 /* We have the username and the password, but haven't tried
732 any authorization yet. Let's see if the "Basic" method
733 works. If not, we'll come back here and construct a
734 proper authorization method with the right challenges.
736 If we didn't employ this kind of logic, every URL that
737 requires authorization would have to be processed twice,
738 which is very suboptimal and generates a bunch of false
739 "unauthorized" errors in the server log.
741 #### But this logic also has a serious problem when used
742 with stronger authentications: we *first* transmit the
743 username and the password in clear text, and *then*
744 attempt a stronger authentication scheme. That cannot be
745 right! We are only fortunate that almost everyone still
746 uses the `Basic' scheme anyway.
748 There should be an option to prevent this from happening,
749 for those who use strong authentication schemes and value
751 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
755 wwwauth = create_authorization_line (authenticate_h, user, passwd,
763 char *proxy_user, *proxy_passwd;
764 /* For normal username and password, URL components override
765 command-line/wgetrc parameters. With proxy authentication,
766 it's the reverse, because proxy URLs are normally the
767 "permanent" ones, so command-line args should take
769 if (opt.proxy_user && opt.proxy_passwd)
771 proxy_user = opt.proxy_user;
772 proxy_passwd = opt.proxy_passwd;
776 proxy_user = u->user;
777 proxy_passwd = u->passwd;
779 /* #### This is junky. Can't the proxy request, say, `Digest'
781 if (proxy_user && proxy_passwd)
782 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
783 "Proxy-Authorization");
788 /* String of the form :PORT. Used only for non-standard ports. */
792 && remport != (u->proto == URLHTTPS
793 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
795 && remport != DEFAULT_HTTP_PORT
799 port_maybe = (char *)alloca (numdigit (remport) + 2);
800 sprintf (port_maybe, ":%d", remport);
803 if (!inhibit_keep_alive)
804 request_keep_alive = "Connection: Keep-Alive\r\n";
806 request_keep_alive = NULL;
809 cookies = build_cookies_request (ou->host, ou->port, ou->path,
810 ou->proto == URLHTTPS);
812 /* Allocate the memory for the request. */
813 request = (char *)alloca (strlen (command) + strlen (path)
816 + (port_maybe ? strlen (port_maybe) : 0)
817 + strlen (HTTP_ACCEPT)
818 + (request_keep_alive
819 ? strlen (request_keep_alive) : 0)
820 + (referer ? strlen (referer) : 0)
821 + (cookies ? strlen (cookies) : 0)
822 + (wwwauth ? strlen (wwwauth) : 0)
823 + (proxyauth ? strlen (proxyauth) : 0)
824 + (range ? strlen (range) : 0)
826 + (opt.user_header ? strlen (opt.user_header) : 0)
828 /* Construct the request. */
834 %s%s%s%s%s%s%s%s\r\n",
835 command, path, useragent, remhost,
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));
847 /* Free the temporary memory. */
848 FREE_MAYBE (wwwauth);
849 FREE_MAYBE (proxyauth);
850 FREE_MAYBE (cookies);
852 /* Send the request to server. */
854 if (u->proto == URLHTTPS)
855 num_written = ssl_iwrite (ssl, request, strlen (request));
857 #endif /* HAVE_SSL */
858 num_written = iwrite (sock, request, strlen (request));
862 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
864 CLOSE_INVALIDATE (sock);
867 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
868 u->proxy ? "Proxy" : "HTTP");
869 contlen = contrange = -1;
874 /* Before reading anything, initialize the rbuf. */
875 rbuf_initialize (&rbuf, sock);
877 if (u->proto == URLHTTPS)
881 #endif /* HAVE_SSL */
884 /* Header-fetching loop. */
892 /* Get the header. */
893 status = header_get (&rbuf, &hdr,
894 /* Disallow continuations for status line. */
895 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
897 /* Check for errors. */
898 if (status == HG_EOF && *hdr)
900 /* This used to be an unconditional error, but that was
901 somewhat controversial, because of a large number of
902 broken CGI's that happily "forget" to send the second EOL
903 before closing the connection of a HEAD request.
905 So, the deal is to check whether the header is empty
906 (*hdr is zero if it is); if yes, it means that the
907 previous header was fully retrieved, and that -- most
908 probably -- the request is complete. "...be liberal in
909 what you accept." Oh boy. */
910 logputs (LOG_VERBOSE, "\n");
911 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
914 FREE_MAYBE (all_headers);
915 CLOSE_INVALIDATE (sock);
918 else if (status == HG_ERROR)
920 logputs (LOG_VERBOSE, "\n");
921 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
925 FREE_MAYBE (all_headers);
926 CLOSE_INVALIDATE (sock);
930 /* If the headers are to be saved to a file later, save them to
932 if (opt.save_headers)
934 int lh = strlen (hdr);
935 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
936 memcpy (all_headers + all_length, hdr, lh);
938 all_headers[all_length++] = '\n';
939 all_headers[all_length] = '\0';
942 /* Print the header if requested. */
943 if (opt.server_response && hcount != 1)
944 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
946 /* Check for status line. */
950 /* Parse the first line of server response. */
951 statcode = parse_http_status_line (hdr, &error);
952 hs->statcode = statcode;
953 /* Store the descriptive response. */
954 if (statcode == -1) /* malformed response */
956 /* A common reason for "malformed response" error is the
957 case when no data was actually received. Handle this
960 hs->error = xstrdup (_("No data received"));
962 hs->error = xstrdup (_("Malformed status line"));
967 hs->error = xstrdup (_("(no description)"));
969 hs->error = xstrdup (error);
976 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
981 /* Exit on empty header. */
988 /* Try getting content-length. */
989 if (contlen == -1 && !opt.ignore_length)
990 if (header_process (hdr, "Content-Length", header_extract_number,
993 /* Try getting content-type. */
995 if (header_process (hdr, "Content-Type", http_process_type, &type))
997 /* Try getting location. */
999 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1001 /* Try getting last-modified. */
1002 if (!hs->remote_time)
1003 if (header_process (hdr, "Last-Modified", header_strdup,
1006 /* Try getting cookies. */
1008 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1010 /* Try getting www-authentication. */
1011 if (!authenticate_h)
1012 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1015 /* Check for accept-ranges header. If it contains the word
1016 `none', disable the ranges. */
1017 if (*dt & ACCEPTRANGES)
1020 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1023 *dt &= ~ACCEPTRANGES;
1027 /* Try getting content-range. */
1028 if (contrange == -1)
1030 struct http_process_range_closure closure;
1031 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1033 contrange = closure.first_byte_pos;
1037 /* Check for keep-alive related responses. */
1038 if (!inhibit_keep_alive)
1040 /* Check for the `Keep-Alive' header. */
1041 if (!http_keep_alive_1)
1043 if (header_process (hdr, "Keep-Alive", header_exists,
1044 &http_keep_alive_1))
1047 /* Check for `Connection: Keep-Alive'. */
1048 if (!http_keep_alive_2)
1050 if (header_process (hdr, "Connection", http_process_connection,
1051 &http_keep_alive_2))
1059 logputs (LOG_VERBOSE, "\n");
1062 && (http_keep_alive_1 || http_keep_alive_2))
1064 assert (inhibit_keep_alive == 0);
1068 /* The server has promised that it will not close the connection
1069 when we're done. This means that we can register it. */
1071 register_persistent (u->host, u->port, sock);
1073 register_persistent (u->host, u->port, sock, ssl);
1074 #endif /* HAVE_SSL */
1076 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1079 /* Authorization is required. */
1083 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1084 might be more bytes in the body. */
1085 if (auth_tried_already)
1087 /* If we have tried it already, then there is not point
1090 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1091 xfree (authenticate_h);
1094 else if (!known_authentication_scheme_p (authenticate_h))
1096 xfree (authenticate_h);
1097 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1100 else if (BEGINS_WITH (authenticate_h, "Basic"))
1102 /* The authentication scheme is basic, the one we try by
1103 default, and it failed. There's no sense in trying
1109 auth_tried_already = 1;
1113 /* We do not need this anymore. */
1116 xfree (authenticate_h);
1117 authenticate_h = NULL;
1120 /* 20x responses are counted among successful by default. */
1121 if (H_20X (statcode))
1124 /* Return if redirected. */
1125 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1127 /* RFC2068 says that in case of the 300 (multiple choices)
1128 response, the server can output a preferred URL through
1129 `Location' header; otherwise, the request should be treated
1130 like GET. So, if the location is set, it will be a
1131 redirection; otherwise, just proceed normally. */
1132 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1136 logprintf (LOG_VERBOSE,
1137 _("Location: %s%s\n"),
1138 hs->newloc ? hs->newloc : _("unspecified"),
1139 hs->newloc ? _(" [following]") : "");
1140 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1141 might be more bytes in the body. */
1143 FREE_MAYBE (all_headers);
1148 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1151 /* We don't assume text/html by default. */
1154 if (opt.html_extension && (*dt & TEXTHTML))
1155 /* -E / --html-extension / html_extension = on was specified, and this is a
1156 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1157 already the file's suffix, tack on ".html". */
1159 char* last_period_in_local_filename = strrchr(u->local, '.');
1161 if (last_period_in_local_filename == NULL ||
1162 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1163 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1165 size_t local_filename_len = strlen(u->local);
1167 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1168 strcpy(u->local + local_filename_len, ".html");
1170 *dt |= ADDED_HTML_EXTENSION;
1174 if (contrange == -1)
1176 /* We did not get a content-range header. This means that the
1177 server did not honor our `Range' request. Normally, this
1178 means we should reset hs->restval and continue normally. */
1180 /* However, if `-c' is used, we need to be a bit more careful:
1182 1. If `-c' is specified and the file already existed when
1183 Wget was started, it would be a bad idea for us to start
1184 downloading it from scratch, effectively truncating it. I
1185 believe this cannot happen unless `-c' was specified.
1187 2. If `-c' is used on a file that is already fully
1188 downloaded, we're requesting bytes after the end of file,
1189 which can result in server not honoring `Range'. If this is
1190 the case, `Content-Length' will be equal to the length of the
1192 if (opt.always_rest)
1194 /* Check for condition #2. */
1195 if (hs->restval > 0 /* restart was requested. */
1196 && contlen != -1 /* we got content-length. */
1197 && hs->restval >= contlen /* file fully downloaded
1201 logputs (LOG_VERBOSE, _("\
1202 \n The file is already fully retrieved; nothing to do.\n\n"));
1203 /* In case the caller inspects. */
1207 FREE_MAYBE (all_headers);
1208 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1209 might be more bytes in the body. */
1210 return RETRUNNEEDED;
1213 /* Check for condition #1. */
1214 if (hs->no_truncate)
1216 logprintf (LOG_NOTQUIET,
1219 Continued download failed on this file, which conflicts with `-c'.\n\
1220 Refusing to truncate existing file `%s'.\n\n"), u->local);
1222 FREE_MAYBE (all_headers);
1223 CLOSE_INVALIDATE (sock);
1224 return CONTNOTSUPPORTED;
1232 else if (contrange != hs->restval ||
1233 (H_PARTIAL (statcode) && contrange == -1))
1235 /* This means the whole request was somehow misunderstood by the
1236 server. Bail out. */
1238 FREE_MAYBE (all_headers);
1239 CLOSE_INVALIDATE (sock);
1246 contlen += contrange;
1248 contrange = -1; /* If conent-length was not sent,
1249 content-range will be ignored. */
1251 hs->contlen = contlen;
1255 if ((*dt & RETROKF) && !opt.server_response)
1257 /* No need to print this output if the body won't be
1258 downloaded at all, or if the original server response is
1260 logputs (LOG_VERBOSE, _("Length: "));
1263 logputs (LOG_VERBOSE, legible (contlen));
1264 if (contrange != -1)
1265 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1266 legible (contlen - contrange));
1269 logputs (LOG_VERBOSE,
1270 opt.ignore_length ? _("ignored") : _("unspecified"));
1272 logprintf (LOG_VERBOSE, " [%s]\n", type);
1274 logputs (LOG_VERBOSE, "\n");
1278 type = NULL; /* We don't need it any more. */
1280 /* Return if we have no intention of further downloading. */
1281 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1283 /* In case the caller cares to look... */
1287 FREE_MAYBE (all_headers);
1288 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1289 might be more bytes in the body. */
1290 return RETRFINISHED;
1293 /* Open the local file. */
1296 mkalldirs (u->local);
1298 rotate_backups (u->local);
1299 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1302 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1303 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1304 might be more bytes in the body. */
1305 FREE_MAYBE (all_headers);
1311 extern int global_download_count;
1313 /* To ensure that repeated "from scratch" downloads work for -O
1314 files, we rewind the file pointer, unless restval is
1315 non-zero. (This works only when -O is used on regular files,
1316 but it's still a valuable feature.)
1318 However, this loses when more than one URL is specified on
1319 the command line the second rewinds eradicates the contents
1320 of the first download. Thus we disable the above trick for
1321 all the downloads except the very first one.
1323 #### A possible solution to this would be to remember the
1324 file position in the output document and to seek to that
1325 position, instead of rewinding. */
1326 if (!hs->restval && global_download_count == 0)
1328 /* This will silently fail for streams that don't correspond
1329 to regular files, but that's OK. */
1331 /* ftruncate is needed because opt.dfp is opened in append
1332 mode if opt.always_rest is set. */
1333 ftruncate (fileno (fp), 0);
1338 /* #### This confuses the code that checks for file size. There
1339 should be some overhead information. */
1340 if (opt.save_headers)
1341 fwrite (all_headers, 1, all_length, fp);
1342 timer = wtimer_new ();
1343 /* Get the contents of the document. */
1344 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1345 (contlen != -1 ? contlen : 0),
1347 hs->dltime = wtimer_elapsed (timer);
1348 wtimer_delete (timer);
1350 /* Close or flush the file. We have to be careful to check for
1351 error here. Checking the result of fwrite() is not enough --
1352 errors could go unnoticed! */
1355 flush_res = fclose (fp);
1357 flush_res = fflush (fp);
1358 if (flush_res == EOF)
1361 FREE_MAYBE (all_headers);
1362 CLOSE_FINISH (sock);
1365 return RETRFINISHED;
1368 /* The genuine HTTP loop! This is the part where the retrieval is
1369 retried, and retried, and retried, and... */
1371 http_loop (struct urlinfo *u, char **newloc, int *dt)
1374 int use_ts, got_head = 0; /* time-stamping info */
1375 char *filename_plus_orig_suffix;
1376 char *local_filename = NULL;
1377 char *tms, *suf, *locf, *tmrate;
1379 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1380 long local_size = 0; /* the size of the local file */
1381 size_t filename_len;
1382 struct http_stat hstat; /* HTTP status */
1385 /* This used to be done in main(), but it's a better idea to do it
1386 here so that we don't go through the hoops if we're just using
1388 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1390 load_cookies (opt.cookies_input);
1391 cookies_loaded_p = 1;
1396 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1397 has_wildcards_p because it would also warn on `?', and we know that
1398 shows up in CGI paths a *lot*. */
1399 if (strchr (u->url, '*'))
1400 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1402 /* Determine the local filename. */
1404 u->local = url_filename (u->proxy ? u->proxy : u);
1406 if (!opt.output_document)
1409 locf = opt.output_document;
1411 filename_len = strlen (u->local);
1412 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1414 if (opt.noclobber && file_exists_p (u->local))
1416 /* If opt.noclobber is turned on and file already exists, do not
1417 retrieve the file */
1418 logprintf (LOG_VERBOSE, _("\
1419 File `%s' already there, will not retrieve.\n"), u->local);
1420 /* If the file is there, we suppose it's retrieved OK. */
1423 /* #### Bogusness alert. */
1424 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1425 text/html, a harmless lie. */
1426 if (((suf = suffix (u->local)) != NULL)
1427 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1430 /* Another harmless lie: */
1435 if (opt.timestamping)
1437 boolean local_dot_orig_file_exists = FALSE;
1439 if (opt.backup_converted)
1440 /* If -K is specified, we'll act on the assumption that it was specified
1441 last time these files were downloaded as well, and instead of just
1442 comparing local file X against server file X, we'll compare local
1443 file X.orig (if extant, else X) against server file X. If -K
1444 _wasn't_ specified last time, or the server contains files called
1445 *.orig, -N will be back to not operating correctly with -k. */
1447 /* Would a single s[n]printf() call be faster? --dan
1449 Definitely not. sprintf() is horribly slow. It's a
1450 different question whether the difference between the two
1451 affects a program. Usually I'd say "no", but at one
1452 point I profiled Wget, and found that a measurable and
1453 non-negligible amount of time was lost calling sprintf()
1454 in url.c. Replacing sprintf with inline calls to
1455 strcpy() and long_to_string() made a difference.
1457 memcpy (filename_plus_orig_suffix, u->local, filename_len);
1458 memcpy (filename_plus_orig_suffix + filename_len,
1459 ".orig", sizeof (".orig"));
1461 /* Try to stat() the .orig file. */
1462 if (stat (filename_plus_orig_suffix, &st) == 0)
1464 local_dot_orig_file_exists = TRUE;
1465 local_filename = filename_plus_orig_suffix;
1469 if (!local_dot_orig_file_exists)
1470 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1471 if (stat (u->local, &st) == 0)
1472 local_filename = u->local;
1474 if (local_filename != NULL)
1475 /* There was a local file, so we'll check later to see if the version
1476 the server has is the same version we already have, allowing us to
1481 local_size = st.st_size;
1485 /* Reset the counter. */
1487 *dt = 0 | ACCEPTRANGES;
1491 /* Increment the pass counter. */
1493 sleep_between_retrievals (count);
1494 /* Get the current time string. */
1495 tms = time_str (NULL);
1496 /* Print fetch message, if opt.verbose. */
1499 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1503 sprintf (tmp, _("(try:%2d)"), count);
1504 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1505 tms, hurl, tmp, locf);
1507 ws_changetitle (hurl, 1);
1512 /* Default document type is empty. However, if spider mode is
1513 on or time-stamping is employed, HEAD_ONLY commands is
1514 encoded within *dt. */
1515 if (opt.spider || (use_ts && !got_head))
1519 /* Assume no restarting. */
1521 /* Decide whether or not to restart. */
1522 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1523 /* #### this calls access() and then stat(); could be optimized. */
1524 && file_exists_p (locf))
1525 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1526 hstat.restval = st.st_size;
1528 /* In `-c' is used and the file is existing and non-empty,
1529 refuse to truncate it if the server doesn't support continued
1531 hstat.no_truncate = 0;
1532 if (opt.always_rest && hstat.restval)
1533 hstat.no_truncate = 1;
1535 /* Decide whether to send the no-cache directive. We send it in
1537 a) we're using a proxy, and we're past our first retrieval.
1538 Some proxies are notorious for caching incomplete data, so
1539 we require a fresh get.
1540 b) caching is explicitly inhibited. */
1541 if ((u->proxy && count > 1) /* a */
1542 || !opt.allow_cache /* b */
1544 *dt |= SEND_NOCACHE;
1546 *dt &= ~SEND_NOCACHE;
1548 /* Try fetching the document, or at least its head. :-) */
1549 err = gethttp (u, &hstat, dt);
1551 /* It's unfortunate that wget determines the local filename before finding
1552 out the Content-Type of the file. Barring a major restructuring of the
1553 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1554 u->local to tack on ".html". */
1555 if (!opt.output_document)
1558 locf = opt.output_document;
1561 tms = time_str (NULL);
1562 /* Get the new location (with or without the redirection). */
1564 *newloc = xstrdup (hstat.newloc);
1567 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1568 case CONERROR: case READERR: case WRITEFAILED:
1570 /* Non-fatal errors continue executing the loop, which will
1571 bring them to "while" statement at the end, to judge
1572 whether the number of tries was exceeded. */
1574 printwhat (count, opt.ntry);
1577 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1578 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1579 /* Fatal errors just return from the function. */
1583 case FWRITEERR: case FOPENERR:
1584 /* Another fatal error. */
1585 logputs (LOG_VERBOSE, "\n");
1586 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1587 u->local, strerror (errno));
1592 /* Another fatal error. */
1593 logputs (LOG_VERBOSE, "\n");
1594 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1599 /* Return the new location to the caller. */
1602 logprintf (LOG_NOTQUIET,
1603 _("ERROR: Redirection (%d) without location.\n"),
1611 /* The file was already fully retrieved. */
1616 /* Deal with you later. */
1619 /* All possibilities should have been exhausted. */
1622 if (!(*dt & RETROKF))
1626 /* #### Ugly ugly ugly! */
1627 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1628 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1631 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1632 tms, hstat.statcode, hstat.error);
1633 logputs (LOG_VERBOSE, "\n");
1638 /* Did we get the time-stamp? */
1641 if (opt.timestamping && !hstat.remote_time)
1643 logputs (LOG_NOTQUIET, _("\
1644 Last-modified header missing -- time-stamps turned off.\n"));
1646 else if (hstat.remote_time)
1648 /* Convert the date-string into struct tm. */
1649 tmr = http_atotm (hstat.remote_time);
1650 if (tmr == (time_t) (-1))
1651 logputs (LOG_VERBOSE, _("\
1652 Last-modified header invalid -- time-stamp ignored.\n"));
1656 /* The time-stamping section. */
1661 use_ts = 0; /* no more time-stamping */
1662 count = 0; /* the retrieve count for HEAD is
1664 if (hstat.remote_time && tmr != (time_t) (-1))
1666 /* Now time-stamping can be used validly. Time-stamping
1667 means that if the sizes of the local and remote file
1668 match, and local file is newer than the remote file,
1669 it will not be retrieved. Otherwise, the normal
1670 download procedure is resumed. */
1672 (hstat.contlen == -1 || local_size == hstat.contlen))
1674 logprintf (LOG_VERBOSE, _("\
1675 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1680 else if (tml >= tmr)
1681 logprintf (LOG_VERBOSE, _("\
1682 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1684 logputs (LOG_VERBOSE,
1685 _("Remote file is newer, retrieving.\n"));
1690 if ((tmr != (time_t) (-1))
1692 && ((hstat.len == hstat.contlen) ||
1693 ((hstat.res == 0) &&
1694 ((hstat.contlen == -1) ||
1695 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1697 /* #### This code repeats in http.c and ftp.c. Move it to a
1699 const char *fl = NULL;
1700 if (opt.output_document)
1702 if (opt.od_known_regular)
1703 fl = opt.output_document;
1710 /* End of time-stamping section. */
1714 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1718 /* It is now safe to free the remainder of hstat, since the
1719 strings within it will no longer be used. */
1722 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1724 if (hstat.len == hstat.contlen)
1728 logprintf (LOG_VERBOSE,
1729 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1730 tms, tmrate, locf, hstat.len, hstat.contlen);
1731 logprintf (LOG_NONVERBOSE,
1732 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1733 tms, u->url, hstat.len, hstat.contlen, locf, count);
1736 downloaded_increase (hstat.len);
1738 /* Remember that we downloaded the file for later ".orig" code. */
1739 if (*dt & ADDED_HTML_EXTENSION)
1740 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1742 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1746 else if (hstat.res == 0) /* No read error */
1748 if (hstat.contlen == -1) /* We don't know how much we were supposed
1749 to get, so assume we succeeded. */
1753 logprintf (LOG_VERBOSE,
1754 _("%s (%s) - `%s' saved [%ld]\n\n"),
1755 tms, tmrate, locf, hstat.len);
1756 logprintf (LOG_NONVERBOSE,
1757 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1758 tms, u->url, hstat.len, locf, count);
1761 downloaded_increase (hstat.len);
1763 /* Remember that we downloaded the file for later ".orig" code. */
1764 if (*dt & ADDED_HTML_EXTENSION)
1765 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1767 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1771 else if (hstat.len < hstat.contlen) /* meaning we lost the
1772 connection too soon */
1774 logprintf (LOG_VERBOSE,
1775 _("%s (%s) - Connection closed at byte %ld. "),
1776 tms, tmrate, hstat.len);
1777 printwhat (count, opt.ntry);
1780 else if (!opt.kill_longer) /* meaning we got more than expected */
1782 logprintf (LOG_VERBOSE,
1783 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1784 tms, tmrate, locf, hstat.len, hstat.contlen);
1785 logprintf (LOG_NONVERBOSE,
1786 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1787 tms, u->url, hstat.len, hstat.contlen, 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);
1799 else /* the same, but not accepted */
1801 logprintf (LOG_VERBOSE,
1802 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1803 tms, tmrate, hstat.len, hstat.contlen);
1804 printwhat (count, opt.ntry);
1808 else /* now hstat.res can only be -1 */
1810 if (hstat.contlen == -1)
1812 logprintf (LOG_VERBOSE,
1813 _("%s (%s) - Read error at byte %ld (%s)."),
1814 tms, tmrate, hstat.len, strerror (errno));
1815 printwhat (count, opt.ntry);
1818 else /* hstat.res == -1 and contlen is given */
1820 logprintf (LOG_VERBOSE,
1821 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1822 tms, tmrate, hstat.len, hstat.contlen,
1824 printwhat (count, opt.ntry);
1831 while (!opt.ntry || (count < opt.ntry));
1835 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1836 than local timezone.
1838 mktime is similar but assumes struct tm, also known as the
1839 "broken-down" form of time, is in local time zone. mktime_from_utc
1840 uses mktime to make the conversion understanding that an offset
1841 will be introduced by the local time assumption.
1843 mktime_from_utc then measures the introduced offset by applying
1844 gmtime to the initial result and applying mktime to the resulting
1845 "broken-down" form. The difference between the two mktime results
1846 is the measured offset which is then subtracted from the initial
1847 mktime result to yield a calendar time which is the value returned.
1849 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1850 consistent offset (the non DST offset) since tm and tm+o might be
1851 on opposite sides of a DST change.
1853 Some implementations of mktime return -1 for the nonexistent
1854 localtime hour at the beginning of DST. In this event, use
1855 mktime(tm - 1hr) + 3600.
1859 gmtime(t+o) --> tm+o
1860 mktime(tm+o) --> t+2o
1861 t+o - (t+2o - t+o) = t
1863 Note that glibc contains a function of the same purpose named
1864 `timegm' (reverse of gmtime). But obviously, it is not universally
1865 available, and unfortunately it is not straightforwardly
1866 extractable for use here. Perhaps configure should detect timegm
1867 and use it where available.
1869 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1870 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1871 Further improved by Roger with assistance from Edward J. Sabol
1872 based on input by Jamie Zawinski. */
1875 mktime_from_utc (struct tm *t)
1886 return -1; /* can't deal with output from strptime */
1897 return -1; /* can't deal with output from gmtime */
1900 return (tl - (tb - tl));
1903 /* Check whether the result of strptime() indicates success.
1904 strptime() returns the pointer to how far it got to in the string.
1905 The processing has been successful if the string is at `GMT' or
1906 `+X', or at the end of the string.
1908 In extended regexp parlance, the function returns 1 if P matches
1909 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1910 can return) is considered a failure and 0 is returned. */
1912 check_end (const char *p)
1916 while (ISSPACE (*p))
1919 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1920 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1926 /* Convert the textual specification of time in TIME_STRING to the
1927 number of seconds since the Epoch.
1929 TIME_STRING can be in any of the three formats RFC2068 allows the
1930 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1931 Timezones are ignored, and should be GMT.
1933 Return the computed time_t representation, or -1 if the conversion
1936 This function uses strptime with various string formats for parsing
1937 TIME_STRING. This results in a parser that is not as lenient in
1938 interpreting TIME_STRING as I would like it to be. Being based on
1939 strptime, it always allows shortened months, one-digit days, etc.,
1940 but due to the multitude of formats in which time can be
1941 represented, an ideal HTTP time parser would be even more
1942 forgiving. It should completely ignore things like week days and
1943 concentrate only on the various forms of representing years,
1944 months, days, hours, minutes, and seconds. For example, it would
1945 be nice if it accepted ISO 8601 out of the box.
1947 I've investigated free and PD code for this purpose, but none was
1948 usable. getdate was big and unwieldy, and had potential copyright
1949 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1950 distributed with phttpd, is excellent, but we cannot use it because
1951 it is not assigned to the FSF. So I stuck it with strptime. */
1954 http_atotm (char *time_string)
1956 /* NOTE: Solaris strptime man page claims that %n and %t match white
1957 space, but that's not universally available. Instead, we simply
1958 use ` ' to mean "skip all WS", which works under all strptime
1959 implementations I've tested. */
1961 static const char *time_formats[] = {
1962 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1963 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1964 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1965 (google.com uses this for their cookies.) */
1966 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1972 /* According to Roger Beeman, we need to initialize tm_isdst, since
1973 strptime won't do it. */
1976 /* Note that under foreign locales Solaris strptime() fails to
1977 recognize English dates, which renders this function useless. We
1978 solve this by being careful not to affect LC_TIME when
1979 initializing locale.
1981 Another solution would be to temporarily set locale to C, invoke
1982 strptime(), and restore it back. This is slow and dirty,
1983 however, and locale support other than LC_MESSAGES can mess other
1984 things, so I rather chose to stick with just setting LC_MESSAGES.
1986 GNU strptime does not have this problem because it recognizes
1987 both international and local dates. */
1989 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1990 if (check_end (strptime (time_string, time_formats[i], &t)))
1991 return mktime_from_utc (&t);
1993 /* All formats have failed. */
1997 /* Authorization support: We support two authorization schemes:
1999 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2001 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2002 consisting of answering to the server's challenge with the proper
2005 /* How many bytes it will take to store LEN bytes in base64. */
2006 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2008 /* Encode the string S of length LENGTH to base64 format and place it
2009 to STORE. STORE will be 0-terminated, and must point to a writable
2010 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2012 base64_encode (const char *s, char *store, int length)
2014 /* Conversion table. */
2015 static char tbl[64] = {
2016 'A','B','C','D','E','F','G','H',
2017 'I','J','K','L','M','N','O','P',
2018 'Q','R','S','T','U','V','W','X',
2019 'Y','Z','a','b','c','d','e','f',
2020 'g','h','i','j','k','l','m','n',
2021 'o','p','q','r','s','t','u','v',
2022 'w','x','y','z','0','1','2','3',
2023 '4','5','6','7','8','9','+','/'
2026 unsigned char *p = (unsigned char *)store;
2028 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2029 for (i = 0; i < length; i += 3)
2031 *p++ = tbl[s[0] >> 2];
2032 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2033 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2034 *p++ = tbl[s[2] & 0x3f];
2037 /* Pad the result if necessary... */
2038 if (i == length + 1)
2040 else if (i == length + 2)
2041 *(p - 1) = *(p - 2) = '=';
2042 /* ...and zero-terminate it. */
2046 /* Create the authentication header contents for the `Basic' scheme.
2047 This is done by encoding the string `USER:PASS' in base64 and
2048 prepending `HEADER: Basic ' to it. */
2050 basic_authentication_encode (const char *user, const char *passwd,
2053 char *t1, *t2, *res;
2054 int len1 = strlen (user) + 1 + strlen (passwd);
2055 int len2 = BASE64_LENGTH (len1);
2057 t1 = (char *)alloca (len1 + 1);
2058 sprintf (t1, "%s:%s", user, passwd);
2059 t2 = (char *)alloca (1 + len2);
2060 base64_encode (t1, t2, len1);
2061 res = (char *)xmalloc (len2 + 11 + strlen (header));
2062 sprintf (res, "%s: Basic %s\r\n", header, t2);
2068 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2069 of a field in such a header. If the field is the one specified by
2070 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2071 digest authorization code), extract its value in the (char*)
2072 variable pointed by RET. Returns negative on a malformed header,
2073 or number of bytes that have been parsed by this call. */
2075 extract_header_attr (const char *au, const char *attr_name, char **ret)
2077 const char *cp, *ep;
2081 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2083 cp += strlen (attr_name);
2086 cp += skip_lws (cp);
2091 cp += skip_lws (cp);
2096 for (ep = cp; *ep && *ep != '\"'; ep++)
2101 *ret = strdupdelim (cp, ep);
2108 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2109 an array of 16 bytes containing the hash keys, and BUF should be a
2110 buffer of 33 writable characters (32 for hex digits plus one for
2111 zero termination). */
2113 dump_hash (unsigned char *buf, const unsigned char *hash)
2117 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2119 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2120 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2125 /* Take the line apart to find the challenge, and compose a digest
2126 authorization header. See RFC2069 section 2.1.2. */
2128 digest_authentication_encode (const char *au, const char *user,
2129 const char *passwd, const char *method,
2132 static char *realm, *opaque, *nonce;
2137 { "realm", &realm },
2138 { "opaque", &opaque },
2143 realm = opaque = nonce = NULL;
2145 au += 6; /* skip over `Digest' */
2150 au += skip_lws (au);
2151 for (i = 0; i < ARRAY_SIZE (options); i++)
2153 int skip = extract_header_attr (au, options[i].name,
2154 options[i].variable);
2158 FREE_MAYBE (opaque);
2168 if (i == ARRAY_SIZE (options))
2170 while (*au && *au != '=')
2174 au += skip_lws (au);
2178 while (*au && *au != '\"')
2185 while (*au && *au != ',')
2190 if (!realm || !nonce || !user || !passwd || !path || !method)
2193 FREE_MAYBE (opaque);
2198 /* Calculate the digest value. */
2201 unsigned char hash[MD5_HASHLEN];
2202 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2203 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2205 /* A1BUF = H(user ":" realm ":" password) */
2206 md5_init_ctx (&ctx);
2207 md5_process_bytes (user, strlen (user), &ctx);
2208 md5_process_bytes (":", 1, &ctx);
2209 md5_process_bytes (realm, strlen (realm), &ctx);
2210 md5_process_bytes (":", 1, &ctx);
2211 md5_process_bytes (passwd, strlen (passwd), &ctx);
2212 md5_finish_ctx (&ctx, hash);
2213 dump_hash (a1buf, hash);
2215 /* A2BUF = H(method ":" path) */
2216 md5_init_ctx (&ctx);
2217 md5_process_bytes (method, strlen (method), &ctx);
2218 md5_process_bytes (":", 1, &ctx);
2219 md5_process_bytes (path, strlen (path), &ctx);
2220 md5_finish_ctx (&ctx, hash);
2221 dump_hash (a2buf, hash);
2223 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2224 md5_init_ctx (&ctx);
2225 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2226 md5_process_bytes (":", 1, &ctx);
2227 md5_process_bytes (nonce, strlen (nonce), &ctx);
2228 md5_process_bytes (":", 1, &ctx);
2229 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2230 md5_finish_ctx (&ctx, hash);
2231 dump_hash (response_digest, hash);
2233 res = (char*) xmalloc (strlen (user)
2238 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2239 + (opaque ? strlen (opaque) : 0)
2241 sprintf (res, "Authorization: Digest \
2242 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2243 user, realm, nonce, path, response_digest);
2246 char *p = res + strlen (res);
2247 strcat (p, ", opaque=\"");
2251 strcat (res, "\r\n");
2255 #endif /* USE_DIGEST */
2258 #define BEGINS_WITH(line, string_constant) \
2259 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2260 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2261 || !line[sizeof (string_constant) - 1]))
2264 known_authentication_scheme_p (const char *au)
2266 return BEGINS_WITH (au, "Basic")
2267 || BEGINS_WITH (au, "Digest")
2268 || BEGINS_WITH (au, "NTLM");
2273 /* Create the HTTP authorization request header. When the
2274 `WWW-Authenticate' response header is seen, according to the
2275 authorization scheme specified in that header (`Basic' and `Digest'
2276 are supported by the current implementation), produce an
2277 appropriate HTTP authorization request header. */
2279 create_authorization_line (const char *au, const char *user,
2280 const char *passwd, const char *method,
2283 char *wwwauth = NULL;
2285 if (!strncasecmp (au, "Basic", 5))
2286 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2287 if (!strncasecmp (au, "NTLM", 4))
2288 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2290 else if (!strncasecmp (au, "Digest", 6))
2291 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2292 #endif /* USE_DIGEST */