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"));
667 #endif /* HAVE_SSL */
671 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
672 /* #### pc_last_fd should be accessed through an accessor
677 #endif /* HAVE_SSL */
678 DEBUGP (("Reusing fd %d.\n", sock));
682 path = u->proxy->url;
686 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
690 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
691 sprintf (referer, "Referer: %s\r\n", ou->referer);
693 if (*dt & SEND_NOCACHE)
694 pragma_h = "Pragma: no-cache\r\n";
699 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
700 /* Gag me! Some servers (e.g. WebSitePro) have been known to
701 respond to the following `Range' format by generating a
702 multipart/x-byte-ranges MIME document! This MIME type was
703 present in an old draft of the byteranges specification.
704 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
705 only if multiple non-overlapping ranges are requested --
706 which Wget never does. */
707 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
712 STRDUP_ALLOCA (useragent, opt.useragent);
715 useragent = (char *)alloca (10 + strlen (version_string));
716 sprintf (useragent, "Wget/%s", version_string);
718 /* Construct the authentication, if userid is present. */
721 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
722 user = user ? user : opt.http_user;
723 passwd = passwd ? passwd : opt.http_passwd;
730 /* We have the username and the password, but haven't tried
731 any authorization yet. Let's see if the "Basic" method
732 works. If not, we'll come back here and construct a
733 proper authorization method with the right challenges.
735 If we didn't employ this kind of logic, every URL that
736 requires authorization would have to be processed twice,
737 which is very suboptimal and generates a bunch of false
738 "unauthorized" errors in the server log.
740 #### But this logic also has a serious problem when used
741 with stronger authentications: we *first* transmit the
742 username and the password in clear text, and *then*
743 attempt a stronger authentication scheme. That cannot be
744 right! We are only fortunate that almost everyone still
745 uses the `Basic' scheme anyway.
747 There should be an option to prevent this from happening,
748 for those who use strong authentication schemes and value
750 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
754 wwwauth = create_authorization_line (authenticate_h, user, passwd,
762 char *proxy_user, *proxy_passwd;
763 /* For normal username and password, URL components override
764 command-line/wgetrc parameters. With proxy authentication,
765 it's the reverse, because proxy URLs are normally the
766 "permanent" ones, so command-line args should take
768 if (opt.proxy_user && opt.proxy_passwd)
770 proxy_user = opt.proxy_user;
771 proxy_passwd = opt.proxy_passwd;
775 proxy_user = u->user;
776 proxy_passwd = u->passwd;
778 /* #### This is junky. Can't the proxy request, say, `Digest'
780 if (proxy_user && proxy_passwd)
781 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
782 "Proxy-Authorization");
787 /* String of the form :PORT. Used only for non-standard ports. */
791 && remport != (u->proto == URLHTTPS
792 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
794 && remport != DEFAULT_HTTP_PORT
798 port_maybe = (char *)alloca (numdigit (remport) + 2);
799 sprintf (port_maybe, ":%d", remport);
802 if (!inhibit_keep_alive)
803 request_keep_alive = "Connection: Keep-Alive\r\n";
805 request_keep_alive = NULL;
808 cookies = build_cookies_request (ou->host, ou->port, ou->path,
809 ou->proto == URLHTTPS);
811 /* Allocate the memory for the request. */
812 request = (char *)alloca (strlen (command) + strlen (path)
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",
834 command, path, useragent, remhost,
835 port_maybe ? port_maybe : "",
837 request_keep_alive ? request_keep_alive : "",
838 referer ? referer : "",
839 cookies ? cookies : "",
840 wwwauth ? wwwauth : "",
841 proxyauth ? proxyauth : "",
844 opt.user_header ? opt.user_header : "");
845 DEBUGP (("---request begin---\n%s---request end---\n", request));
846 /* Free the temporary memory. */
847 FREE_MAYBE (wwwauth);
848 FREE_MAYBE (proxyauth);
849 FREE_MAYBE (cookies);
851 /* Send the request to server. */
853 if (u->proto == URLHTTPS)
854 num_written = ssl_iwrite (ssl, request, strlen (request));
856 #endif /* HAVE_SSL */
857 num_written = iwrite (sock, request, strlen (request));
861 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
863 CLOSE_INVALIDATE (sock);
866 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
867 u->proxy ? "Proxy" : "HTTP");
868 contlen = contrange = -1;
873 /* Before reading anything, initialize the rbuf. */
874 rbuf_initialize (&rbuf, sock);
876 if (u->proto == URLHTTPS)
880 #endif /* HAVE_SSL */
883 /* Header-fetching loop. */
891 /* Get the header. */
892 status = header_get (&rbuf, &hdr,
893 /* Disallow continuations for status line. */
894 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
896 /* Check for errors. */
897 if (status == HG_EOF && *hdr)
899 /* This used to be an unconditional error, but that was
900 somewhat controversial, because of a large number of
901 broken CGI's that happily "forget" to send the second EOL
902 before closing the connection of a HEAD request.
904 So, the deal is to check whether the header is empty
905 (*hdr is zero if it is); if yes, it means that the
906 previous header was fully retrieved, and that -- most
907 probably -- the request is complete. "...be liberal in
908 what you accept." Oh boy. */
909 logputs (LOG_VERBOSE, "\n");
910 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
913 FREE_MAYBE (all_headers);
914 CLOSE_INVALIDATE (sock);
917 else if (status == HG_ERROR)
919 logputs (LOG_VERBOSE, "\n");
920 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
924 FREE_MAYBE (all_headers);
925 CLOSE_INVALIDATE (sock);
929 /* If the headers are to be saved to a file later, save them to
931 if (opt.save_headers)
933 int lh = strlen (hdr);
934 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
935 memcpy (all_headers + all_length, hdr, lh);
937 all_headers[all_length++] = '\n';
938 all_headers[all_length] = '\0';
941 /* Print the header if requested. */
942 if (opt.server_response && hcount != 1)
943 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
945 /* Check for status line. */
949 /* Parse the first line of server response. */
950 statcode = parse_http_status_line (hdr, &error);
951 hs->statcode = statcode;
952 /* Store the descriptive response. */
953 if (statcode == -1) /* malformed response */
955 /* A common reason for "malformed response" error is the
956 case when no data was actually received. Handle this
959 hs->error = xstrdup (_("No data received"));
961 hs->error = xstrdup (_("Malformed status line"));
966 hs->error = xstrdup (_("(no description)"));
968 hs->error = xstrdup (error);
975 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
980 /* Exit on empty header. */
987 /* Try getting content-length. */
988 if (contlen == -1 && !opt.ignore_length)
989 if (header_process (hdr, "Content-Length", header_extract_number,
992 /* Try getting content-type. */
994 if (header_process (hdr, "Content-Type", http_process_type, &type))
996 /* Try getting location. */
998 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1000 /* Try getting last-modified. */
1001 if (!hs->remote_time)
1002 if (header_process (hdr, "Last-Modified", header_strdup,
1005 /* Try getting cookies. */
1007 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1009 /* Try getting www-authentication. */
1010 if (!authenticate_h)
1011 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1014 /* Check for accept-ranges header. If it contains the word
1015 `none', disable the ranges. */
1016 if (*dt & ACCEPTRANGES)
1019 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1022 *dt &= ~ACCEPTRANGES;
1026 /* Try getting content-range. */
1027 if (contrange == -1)
1029 struct http_process_range_closure closure;
1030 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1032 contrange = closure.first_byte_pos;
1036 /* Check for keep-alive related responses. */
1037 if (!inhibit_keep_alive)
1039 /* Check for the `Keep-Alive' header. */
1040 if (!http_keep_alive_1)
1042 if (header_process (hdr, "Keep-Alive", header_exists,
1043 &http_keep_alive_1))
1046 /* Check for `Connection: Keep-Alive'. */
1047 if (!http_keep_alive_2)
1049 if (header_process (hdr, "Connection", http_process_connection,
1050 &http_keep_alive_2))
1058 logputs (LOG_VERBOSE, "\n");
1061 && (http_keep_alive_1 || http_keep_alive_2))
1063 assert (inhibit_keep_alive == 0);
1067 /* The server has promised that it will not close the connection
1068 when we're done. This means that we can register it. */
1070 register_persistent (u->host, u->port, sock);
1072 register_persistent (u->host, u->port, sock, ssl);
1073 #endif /* HAVE_SSL */
1075 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1078 /* Authorization is required. */
1082 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1083 might be more bytes in the body. */
1084 if (auth_tried_already)
1086 /* If we have tried it already, then there is not point
1089 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1090 xfree (authenticate_h);
1093 else if (!known_authentication_scheme_p (authenticate_h))
1095 xfree (authenticate_h);
1096 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1099 else if (BEGINS_WITH (authenticate_h, "Basic"))
1101 /* The authentication scheme is basic, the one we try by
1102 default, and it failed. There's no sense in trying
1108 auth_tried_already = 1;
1112 /* We do not need this anymore. */
1115 xfree (authenticate_h);
1116 authenticate_h = NULL;
1119 /* 20x responses are counted among successful by default. */
1120 if (H_20X (statcode))
1123 /* Return if redirected. */
1124 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1126 /* RFC2068 says that in case of the 300 (multiple choices)
1127 response, the server can output a preferred URL through
1128 `Location' header; otherwise, the request should be treated
1129 like GET. So, if the location is set, it will be a
1130 redirection; otherwise, just proceed normally. */
1131 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1135 logprintf (LOG_VERBOSE,
1136 _("Location: %s%s\n"),
1137 hs->newloc ? hs->newloc : _("unspecified"),
1138 hs->newloc ? _(" [following]") : "");
1139 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1140 might be more bytes in the body. */
1142 FREE_MAYBE (all_headers);
1147 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1150 /* We don't assume text/html by default. */
1153 if (opt.html_extension && (*dt & TEXTHTML))
1154 /* -E / --html-extension / html_extension = on was specified, and this is a
1155 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1156 already the file's suffix, tack on ".html". */
1158 char* last_period_in_local_filename = strrchr(u->local, '.');
1160 if (last_period_in_local_filename == NULL ||
1161 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1162 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1164 size_t local_filename_len = strlen(u->local);
1166 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1167 strcpy(u->local + local_filename_len, ".html");
1169 *dt |= ADDED_HTML_EXTENSION;
1173 if (contrange == -1)
1175 /* We did not get a content-range header. This means that the
1176 server did not honor our `Range' request. Normally, this
1177 means we should reset hs->restval and continue normally. */
1179 /* However, if `-c' is used, we need to be a bit more careful:
1181 1. If `-c' is specified and the file already existed when
1182 Wget was started, it would be a bad idea for us to start
1183 downloading it from scratch, effectively truncating it. I
1184 believe this cannot happen unless `-c' was specified.
1186 2. If `-c' is used on a file that is already fully
1187 downloaded, we're requesting bytes after the end of file,
1188 which can result in server not honoring `Range'. If this is
1189 the case, `Content-Length' will be equal to the length of the
1191 if (opt.always_rest)
1193 /* Check for condition #2. */
1194 if (hs->restval > 0 /* restart was requested. */
1195 && contlen != -1 /* we got content-length. */
1196 && hs->restval >= contlen /* file fully downloaded
1200 logputs (LOG_VERBOSE, _("\
1201 \n The file is already fully retrieved; nothing to do.\n\n"));
1202 /* In case the caller inspects. */
1206 FREE_MAYBE (all_headers);
1207 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1208 might be more bytes in the body. */
1209 return RETRUNNEEDED;
1212 /* Check for condition #1. */
1213 if (hs->no_truncate)
1215 logprintf (LOG_NOTQUIET,
1218 Continued download failed on this file, which conflicts with `-c'.\n\
1219 Refusing to truncate existing file `%s'.\n\n"), u->local);
1221 FREE_MAYBE (all_headers);
1222 CLOSE_INVALIDATE (sock);
1223 return CONTNOTSUPPORTED;
1231 else if (contrange != hs->restval ||
1232 (H_PARTIAL (statcode) && contrange == -1))
1234 /* This means the whole request was somehow misunderstood by the
1235 server. Bail out. */
1237 FREE_MAYBE (all_headers);
1238 CLOSE_INVALIDATE (sock);
1245 contlen += contrange;
1247 contrange = -1; /* If conent-length was not sent,
1248 content-range will be ignored. */
1250 hs->contlen = contlen;
1254 if ((*dt & RETROKF) && !opt.server_response)
1256 /* No need to print this output if the body won't be
1257 downloaded at all, or if the original server response is
1259 logputs (LOG_VERBOSE, _("Length: "));
1262 logputs (LOG_VERBOSE, legible (contlen));
1263 if (contrange != -1)
1264 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1265 legible (contlen - contrange));
1268 logputs (LOG_VERBOSE,
1269 opt.ignore_length ? _("ignored") : _("unspecified"));
1271 logprintf (LOG_VERBOSE, " [%s]\n", type);
1273 logputs (LOG_VERBOSE, "\n");
1277 type = NULL; /* We don't need it any more. */
1279 /* Return if we have no intention of further downloading. */
1280 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1282 /* In case the caller cares to look... */
1286 FREE_MAYBE (all_headers);
1287 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1288 might be more bytes in the body. */
1289 return RETRFINISHED;
1292 /* Open the local file. */
1295 mkalldirs (u->local);
1297 rotate_backups (u->local);
1298 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1301 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1302 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1303 might be more bytes in the body. */
1304 FREE_MAYBE (all_headers);
1310 extern int global_download_count;
1312 /* To ensure that repeated "from scratch" downloads work for -O
1313 files, we rewind the file pointer, unless restval is
1314 non-zero. (This works only when -O is used on regular files,
1315 but it's still a valuable feature.)
1317 However, this loses when more than one URL is specified on
1318 the command line the second rewinds eradicates the contents
1319 of the first download. Thus we disable the above trick for
1320 all the downloads except the very first one.
1322 #### A possible solution to this would be to remember the
1323 file position in the output document and to seek to that
1324 position, instead of rewinding. */
1325 if (!hs->restval && global_download_count == 0)
1327 /* This will silently fail for streams that don't correspond
1328 to regular files, but that's OK. */
1330 /* ftruncate is needed because opt.dfp is opened in append
1331 mode if opt.always_rest is set. */
1332 ftruncate (fileno (fp), 0);
1337 /* #### This confuses the code that checks for file size. There
1338 should be some overhead information. */
1339 if (opt.save_headers)
1340 fwrite (all_headers, 1, all_length, fp);
1341 timer = wtimer_new ();
1342 /* Get the contents of the document. */
1343 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1344 (contlen != -1 ? contlen : 0),
1346 hs->dltime = wtimer_elapsed (timer);
1347 wtimer_delete (timer);
1349 /* Close or flush the file. We have to be careful to check for
1350 error here. Checking the result of fwrite() is not enough --
1351 errors could go unnoticed! */
1354 flush_res = fclose (fp);
1356 flush_res = fflush (fp);
1357 if (flush_res == EOF)
1360 FREE_MAYBE (all_headers);
1361 CLOSE_FINISH (sock);
1364 return RETRFINISHED;
1367 /* The genuine HTTP loop! This is the part where the retrieval is
1368 retried, and retried, and retried, and... */
1370 http_loop (struct urlinfo *u, char **newloc, int *dt)
1373 int use_ts, got_head = 0; /* time-stamping info */
1374 char *filename_plus_orig_suffix;
1375 char *local_filename = NULL;
1376 char *tms, *suf, *locf, *tmrate;
1378 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1379 long local_size = 0; /* the size of the local file */
1380 size_t filename_len;
1381 struct http_stat hstat; /* HTTP status */
1384 /* This used to be done in main(), but it's a better idea to do it
1385 here so that we don't go through the hoops if we're just using
1387 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1389 load_cookies (opt.cookies_input);
1390 cookies_loaded_p = 1;
1395 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1396 has_wildcards_p because it would also warn on `?', and we know that
1397 shows up in CGI paths a *lot*. */
1398 if (strchr (u->url, '*'))
1399 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1401 /* Determine the local filename. */
1403 u->local = url_filename (u->proxy ? u->proxy : u);
1405 if (!opt.output_document)
1408 locf = opt.output_document;
1410 filename_len = strlen (u->local);
1411 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1413 if (opt.noclobber && file_exists_p (u->local))
1415 /* If opt.noclobber is turned on and file already exists, do not
1416 retrieve the file */
1417 logprintf (LOG_VERBOSE, _("\
1418 File `%s' already there, will not retrieve.\n"), u->local);
1419 /* If the file is there, we suppose it's retrieved OK. */
1422 /* #### Bogusness alert. */
1423 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1424 text/html, a harmless lie. */
1425 if (((suf = suffix (u->local)) != NULL)
1426 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1429 /* Another harmless lie: */
1434 if (opt.timestamping)
1436 boolean local_dot_orig_file_exists = FALSE;
1438 if (opt.backup_converted)
1439 /* If -K is specified, we'll act on the assumption that it was specified
1440 last time these files were downloaded as well, and instead of just
1441 comparing local file X against server file X, we'll compare local
1442 file X.orig (if extant, else X) against server file X. If -K
1443 _wasn't_ specified last time, or the server contains files called
1444 *.orig, -N will be back to not operating correctly with -k. */
1446 /* Would a single s[n]printf() call be faster? --dan
1448 Definitely not. sprintf() is horribly slow. It's a
1449 different question whether the difference between the two
1450 affects a program. Usually I'd say "no", but at one
1451 point I profiled Wget, and found that a measurable and
1452 non-negligible amount of time was lost calling sprintf()
1453 in url.c. Replacing sprintf with inline calls to
1454 strcpy() and long_to_string() made a difference.
1456 memcpy (filename_plus_orig_suffix, u->local, filename_len);
1457 memcpy (filename_plus_orig_suffix + filename_len,
1458 ".orig", sizeof (".orig"));
1460 /* Try to stat() the .orig file. */
1461 if (stat (filename_plus_orig_suffix, &st) == 0)
1463 local_dot_orig_file_exists = TRUE;
1464 local_filename = filename_plus_orig_suffix;
1468 if (!local_dot_orig_file_exists)
1469 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1470 if (stat (u->local, &st) == 0)
1471 local_filename = u->local;
1473 if (local_filename != NULL)
1474 /* There was a local file, so we'll check later to see if the version
1475 the server has is the same version we already have, allowing us to
1480 local_size = st.st_size;
1484 /* Reset the counter. */
1486 *dt = 0 | ACCEPTRANGES;
1490 /* Increment the pass counter. */
1492 sleep_between_retrievals (count);
1493 /* Get the current time string. */
1494 tms = time_str (NULL);
1495 /* Print fetch message, if opt.verbose. */
1498 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1502 sprintf (tmp, _("(try:%2d)"), count);
1503 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1504 tms, hurl, tmp, locf);
1506 ws_changetitle (hurl, 1);
1511 /* Default document type is empty. However, if spider mode is
1512 on or time-stamping is employed, HEAD_ONLY commands is
1513 encoded within *dt. */
1514 if (opt.spider || (use_ts && !got_head))
1518 /* Assume no restarting. */
1520 /* Decide whether or not to restart. */
1521 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1522 /* #### this calls access() and then stat(); could be optimized. */
1523 && file_exists_p (locf))
1524 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1525 hstat.restval = st.st_size;
1527 /* In `-c' is used and the file is existing and non-empty,
1528 refuse to truncate it if the server doesn't support continued
1530 hstat.no_truncate = 0;
1531 if (opt.always_rest && hstat.restval)
1532 hstat.no_truncate = 1;
1534 /* Decide whether to send the no-cache directive. We send it in
1536 a) we're using a proxy, and we're past our first retrieval.
1537 Some proxies are notorious for caching incomplete data, so
1538 we require a fresh get.
1539 b) caching is explicitly inhibited. */
1540 if ((u->proxy && count > 1) /* a */
1541 || !opt.allow_cache /* b */
1543 *dt |= SEND_NOCACHE;
1545 *dt &= ~SEND_NOCACHE;
1547 /* Try fetching the document, or at least its head. :-) */
1548 err = gethttp (u, &hstat, dt);
1550 /* It's unfortunate that wget determines the local filename before finding
1551 out the Content-Type of the file. Barring a major restructuring of the
1552 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1553 u->local to tack on ".html". */
1554 if (!opt.output_document)
1557 locf = opt.output_document;
1560 tms = time_str (NULL);
1561 /* Get the new location (with or without the redirection). */
1563 *newloc = xstrdup (hstat.newloc);
1566 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1567 case CONERROR: case READERR: case WRITEFAILED:
1569 /* Non-fatal errors continue executing the loop, which will
1570 bring them to "while" statement at the end, to judge
1571 whether the number of tries was exceeded. */
1573 printwhat (count, opt.ntry);
1576 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1577 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1578 /* Fatal errors just return from the function. */
1582 case FWRITEERR: case FOPENERR:
1583 /* Another fatal error. */
1584 logputs (LOG_VERBOSE, "\n");
1585 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1586 u->local, strerror (errno));
1591 /* Another fatal error. */
1592 logputs (LOG_VERBOSE, "\n");
1593 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1598 /* Return the new location to the caller. */
1601 logprintf (LOG_NOTQUIET,
1602 _("ERROR: Redirection (%d) without location.\n"),
1610 /* The file was already fully retrieved. */
1615 /* Deal with you later. */
1618 /* All possibilities should have been exhausted. */
1621 if (!(*dt & RETROKF))
1625 /* #### Ugly ugly ugly! */
1626 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1627 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1630 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1631 tms, hstat.statcode, hstat.error);
1632 logputs (LOG_VERBOSE, "\n");
1637 /* Did we get the time-stamp? */
1640 if (opt.timestamping && !hstat.remote_time)
1642 logputs (LOG_NOTQUIET, _("\
1643 Last-modified header missing -- time-stamps turned off.\n"));
1645 else if (hstat.remote_time)
1647 /* Convert the date-string into struct tm. */
1648 tmr = http_atotm (hstat.remote_time);
1649 if (tmr == (time_t) (-1))
1650 logputs (LOG_VERBOSE, _("\
1651 Last-modified header invalid -- time-stamp ignored.\n"));
1655 /* The time-stamping section. */
1660 use_ts = 0; /* no more time-stamping */
1661 count = 0; /* the retrieve count for HEAD is
1663 if (hstat.remote_time && tmr != (time_t) (-1))
1665 /* Now time-stamping can be used validly. Time-stamping
1666 means that if the sizes of the local and remote file
1667 match, and local file is newer than the remote file,
1668 it will not be retrieved. Otherwise, the normal
1669 download procedure is resumed. */
1671 (hstat.contlen == -1 || local_size == hstat.contlen))
1673 logprintf (LOG_VERBOSE, _("\
1674 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1679 else if (tml >= tmr)
1680 logprintf (LOG_VERBOSE, _("\
1681 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1683 logputs (LOG_VERBOSE,
1684 _("Remote file is newer, retrieving.\n"));
1689 if ((tmr != (time_t) (-1))
1691 && ((hstat.len == hstat.contlen) ||
1692 ((hstat.res == 0) &&
1693 ((hstat.contlen == -1) ||
1694 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1696 /* #### This code repeats in http.c and ftp.c. Move it to a
1698 const char *fl = NULL;
1699 if (opt.output_document)
1701 if (opt.od_known_regular)
1702 fl = opt.output_document;
1709 /* End of time-stamping section. */
1713 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1717 /* It is now safe to free the remainder of hstat, since the
1718 strings within it will no longer be used. */
1721 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1723 if (hstat.len == hstat.contlen)
1727 logprintf (LOG_VERBOSE,
1728 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1729 tms, tmrate, locf, hstat.len, hstat.contlen);
1730 logprintf (LOG_NONVERBOSE,
1731 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1732 tms, u->url, hstat.len, hstat.contlen, locf, count);
1735 downloaded_increase (hstat.len);
1737 /* Remember that we downloaded the file for later ".orig" code. */
1738 if (*dt & ADDED_HTML_EXTENSION)
1739 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1741 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1745 else if (hstat.res == 0) /* No read error */
1747 if (hstat.contlen == -1) /* We don't know how much we were supposed
1748 to get, so assume we succeeded. */
1752 logprintf (LOG_VERBOSE,
1753 _("%s (%s) - `%s' saved [%ld]\n\n"),
1754 tms, tmrate, locf, hstat.len);
1755 logprintf (LOG_NONVERBOSE,
1756 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1757 tms, u->url, hstat.len, locf, count);
1760 downloaded_increase (hstat.len);
1762 /* Remember that we downloaded the file for later ".orig" code. */
1763 if (*dt & ADDED_HTML_EXTENSION)
1764 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1766 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1770 else if (hstat.len < hstat.contlen) /* meaning we lost the
1771 connection too soon */
1773 logprintf (LOG_VERBOSE,
1774 _("%s (%s) - Connection closed at byte %ld. "),
1775 tms, tmrate, hstat.len);
1776 printwhat (count, opt.ntry);
1779 else if (!opt.kill_longer) /* meaning we got more than expected */
1781 logprintf (LOG_VERBOSE,
1782 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1783 tms, tmrate, locf, hstat.len, hstat.contlen);
1784 logprintf (LOG_NONVERBOSE,
1785 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1786 tms, u->url, hstat.len, hstat.contlen, locf, count);
1788 downloaded_increase (hstat.len);
1790 /* Remember that we downloaded the file for later ".orig" code. */
1791 if (*dt & ADDED_HTML_EXTENSION)
1792 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1794 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1798 else /* the same, but not accepted */
1800 logprintf (LOG_VERBOSE,
1801 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1802 tms, tmrate, hstat.len, hstat.contlen);
1803 printwhat (count, opt.ntry);
1807 else /* now hstat.res can only be -1 */
1809 if (hstat.contlen == -1)
1811 logprintf (LOG_VERBOSE,
1812 _("%s (%s) - Read error at byte %ld (%s)."),
1813 tms, tmrate, hstat.len, strerror (errno));
1814 printwhat (count, opt.ntry);
1817 else /* hstat.res == -1 and contlen is given */
1819 logprintf (LOG_VERBOSE,
1820 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1821 tms, tmrate, hstat.len, hstat.contlen,
1823 printwhat (count, opt.ntry);
1830 while (!opt.ntry || (count < opt.ntry));
1834 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1835 than local timezone.
1837 mktime is similar but assumes struct tm, also known as the
1838 "broken-down" form of time, is in local time zone. mktime_from_utc
1839 uses mktime to make the conversion understanding that an offset
1840 will be introduced by the local time assumption.
1842 mktime_from_utc then measures the introduced offset by applying
1843 gmtime to the initial result and applying mktime to the resulting
1844 "broken-down" form. The difference between the two mktime results
1845 is the measured offset which is then subtracted from the initial
1846 mktime result to yield a calendar time which is the value returned.
1848 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1849 consistent offset (the non DST offset) since tm and tm+o might be
1850 on opposite sides of a DST change.
1852 Some implementations of mktime return -1 for the nonexistent
1853 localtime hour at the beginning of DST. In this event, use
1854 mktime(tm - 1hr) + 3600.
1858 gmtime(t+o) --> tm+o
1859 mktime(tm+o) --> t+2o
1860 t+o - (t+2o - t+o) = t
1862 Note that glibc contains a function of the same purpose named
1863 `timegm' (reverse of gmtime). But obviously, it is not universally
1864 available, and unfortunately it is not straightforwardly
1865 extractable for use here. Perhaps configure should detect timegm
1866 and use it where available.
1868 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1869 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1870 Further improved by Roger with assistance from Edward J. Sabol
1871 based on input by Jamie Zawinski. */
1874 mktime_from_utc (struct tm *t)
1885 return -1; /* can't deal with output from strptime */
1896 return -1; /* can't deal with output from gmtime */
1899 return (tl - (tb - tl));
1902 /* Check whether the result of strptime() indicates success.
1903 strptime() returns the pointer to how far it got to in the string.
1904 The processing has been successful if the string is at `GMT' or
1905 `+X', or at the end of the string.
1907 In extended regexp parlance, the function returns 1 if P matches
1908 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1909 can return) is considered a failure and 0 is returned. */
1911 check_end (const char *p)
1915 while (ISSPACE (*p))
1918 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1919 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1925 /* Convert the textual specification of time in TIME_STRING to the
1926 number of seconds since the Epoch.
1928 TIME_STRING can be in any of the three formats RFC2068 allows the
1929 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1930 Timezones are ignored, and should be GMT.
1932 Return the computed time_t representation, or -1 if the conversion
1935 This function uses strptime with various string formats for parsing
1936 TIME_STRING. This results in a parser that is not as lenient in
1937 interpreting TIME_STRING as I would like it to be. Being based on
1938 strptime, it always allows shortened months, one-digit days, etc.,
1939 but due to the multitude of formats in which time can be
1940 represented, an ideal HTTP time parser would be even more
1941 forgiving. It should completely ignore things like week days and
1942 concentrate only on the various forms of representing years,
1943 months, days, hours, minutes, and seconds. For example, it would
1944 be nice if it accepted ISO 8601 out of the box.
1946 I've investigated free and PD code for this purpose, but none was
1947 usable. getdate was big and unwieldy, and had potential copyright
1948 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1949 distributed with phttpd, is excellent, but we cannot use it because
1950 it is not assigned to the FSF. So I stuck it with strptime. */
1953 http_atotm (char *time_string)
1955 /* NOTE: Solaris strptime man page claims that %n and %t match white
1956 space, but that's not universally available. Instead, we simply
1957 use ` ' to mean "skip all WS", which works under all strptime
1958 implementations I've tested. */
1960 static const char *time_formats[] = {
1961 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1962 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1963 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1964 (google.com uses this for their cookies.) */
1965 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1971 /* According to Roger Beeman, we need to initialize tm_isdst, since
1972 strptime won't do it. */
1975 /* Note that under foreign locales Solaris strptime() fails to
1976 recognize English dates, which renders this function useless. We
1977 solve this by being careful not to affect LC_TIME when
1978 initializing locale.
1980 Another solution would be to temporarily set locale to C, invoke
1981 strptime(), and restore it back. This is slow and dirty,
1982 however, and locale support other than LC_MESSAGES can mess other
1983 things, so I rather chose to stick with just setting LC_MESSAGES.
1985 GNU strptime does not have this problem because it recognizes
1986 both international and local dates. */
1988 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1989 if (check_end (strptime (time_string, time_formats[i], &t)))
1990 return mktime_from_utc (&t);
1992 /* All formats have failed. */
1996 /* Authorization support: We support two authorization schemes:
1998 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2000 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2001 consisting of answering to the server's challenge with the proper
2004 /* How many bytes it will take to store LEN bytes in base64. */
2005 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2007 /* Encode the string S of length LENGTH to base64 format and place it
2008 to STORE. STORE will be 0-terminated, and must point to a writable
2009 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2011 base64_encode (const char *s, char *store, int length)
2013 /* Conversion table. */
2014 static char tbl[64] = {
2015 'A','B','C','D','E','F','G','H',
2016 'I','J','K','L','M','N','O','P',
2017 'Q','R','S','T','U','V','W','X',
2018 'Y','Z','a','b','c','d','e','f',
2019 'g','h','i','j','k','l','m','n',
2020 'o','p','q','r','s','t','u','v',
2021 'w','x','y','z','0','1','2','3',
2022 '4','5','6','7','8','9','+','/'
2025 unsigned char *p = (unsigned char *)store;
2027 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2028 for (i = 0; i < length; i += 3)
2030 *p++ = tbl[s[0] >> 2];
2031 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2032 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2033 *p++ = tbl[s[2] & 0x3f];
2036 /* Pad the result if necessary... */
2037 if (i == length + 1)
2039 else if (i == length + 2)
2040 *(p - 1) = *(p - 2) = '=';
2041 /* ...and zero-terminate it. */
2045 /* Create the authentication header contents for the `Basic' scheme.
2046 This is done by encoding the string `USER:PASS' in base64 and
2047 prepending `HEADER: Basic ' to it. */
2049 basic_authentication_encode (const char *user, const char *passwd,
2052 char *t1, *t2, *res;
2053 int len1 = strlen (user) + 1 + strlen (passwd);
2054 int len2 = BASE64_LENGTH (len1);
2056 t1 = (char *)alloca (len1 + 1);
2057 sprintf (t1, "%s:%s", user, passwd);
2058 t2 = (char *)alloca (1 + len2);
2059 base64_encode (t1, t2, len1);
2060 res = (char *)xmalloc (len2 + 11 + strlen (header));
2061 sprintf (res, "%s: Basic %s\r\n", header, t2);
2067 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2068 of a field in such a header. If the field is the one specified by
2069 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2070 digest authorization code), extract its value in the (char*)
2071 variable pointed by RET. Returns negative on a malformed header,
2072 or number of bytes that have been parsed by this call. */
2074 extract_header_attr (const char *au, const char *attr_name, char **ret)
2076 const char *cp, *ep;
2080 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2082 cp += strlen (attr_name);
2085 cp += skip_lws (cp);
2090 cp += skip_lws (cp);
2095 for (ep = cp; *ep && *ep != '\"'; ep++)
2100 *ret = strdupdelim (cp, ep);
2107 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2108 an array of 16 bytes containing the hash keys, and BUF should be a
2109 buffer of 33 writable characters (32 for hex digits plus one for
2110 zero termination). */
2112 dump_hash (unsigned char *buf, const unsigned char *hash)
2116 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2118 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2119 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2124 /* Take the line apart to find the challenge, and compose a digest
2125 authorization header. See RFC2069 section 2.1.2. */
2127 digest_authentication_encode (const char *au, const char *user,
2128 const char *passwd, const char *method,
2131 static char *realm, *opaque, *nonce;
2136 { "realm", &realm },
2137 { "opaque", &opaque },
2142 realm = opaque = nonce = NULL;
2144 au += 6; /* skip over `Digest' */
2149 au += skip_lws (au);
2150 for (i = 0; i < ARRAY_SIZE (options); i++)
2152 int skip = extract_header_attr (au, options[i].name,
2153 options[i].variable);
2157 FREE_MAYBE (opaque);
2167 if (i == ARRAY_SIZE (options))
2169 while (*au && *au != '=')
2173 au += skip_lws (au);
2177 while (*au && *au != '\"')
2184 while (*au && *au != ',')
2189 if (!realm || !nonce || !user || !passwd || !path || !method)
2192 FREE_MAYBE (opaque);
2197 /* Calculate the digest value. */
2200 unsigned char hash[MD5_HASHLEN];
2201 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2202 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2204 /* A1BUF = H(user ":" realm ":" password) */
2205 md5_init_ctx (&ctx);
2206 md5_process_bytes (user, strlen (user), &ctx);
2207 md5_process_bytes (":", 1, &ctx);
2208 md5_process_bytes (realm, strlen (realm), &ctx);
2209 md5_process_bytes (":", 1, &ctx);
2210 md5_process_bytes (passwd, strlen (passwd), &ctx);
2211 md5_finish_ctx (&ctx, hash);
2212 dump_hash (a1buf, hash);
2214 /* A2BUF = H(method ":" path) */
2215 md5_init_ctx (&ctx);
2216 md5_process_bytes (method, strlen (method), &ctx);
2217 md5_process_bytes (":", 1, &ctx);
2218 md5_process_bytes (path, strlen (path), &ctx);
2219 md5_finish_ctx (&ctx, hash);
2220 dump_hash (a2buf, hash);
2222 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2223 md5_init_ctx (&ctx);
2224 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2225 md5_process_bytes (":", 1, &ctx);
2226 md5_process_bytes (nonce, strlen (nonce), &ctx);
2227 md5_process_bytes (":", 1, &ctx);
2228 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2229 md5_finish_ctx (&ctx, hash);
2230 dump_hash (response_digest, hash);
2232 res = (char*) xmalloc (strlen (user)
2237 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2238 + (opaque ? strlen (opaque) : 0)
2240 sprintf (res, "Authorization: Digest \
2241 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2242 user, realm, nonce, path, response_digest);
2245 char *p = res + strlen (res);
2246 strcat (p, ", opaque=\"");
2250 strcat (res, "\r\n");
2254 #endif /* USE_DIGEST */
2257 #define BEGINS_WITH(line, string_constant) \
2258 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2259 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2260 || !line[sizeof (string_constant) - 1]))
2263 known_authentication_scheme_p (const char *au)
2265 return BEGINS_WITH (au, "Basic")
2266 || BEGINS_WITH (au, "Digest")
2267 || BEGINS_WITH (au, "NTLM");
2272 /* Create the HTTP authorization request header. When the
2273 `WWW-Authenticate' response header is seen, according to the
2274 authorization scheme specified in that header (`Basic' and `Digest'
2275 are supported by the current implementation), produce an
2276 appropriate HTTP authorization request header. */
2278 create_authorization_line (const char *au, const char *user,
2279 const char *passwd, const char *method,
2282 char *wwwauth = NULL;
2284 if (!strncasecmp (au, "Basic", 5))
2285 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2286 if (!strncasecmp (au, "NTLM", 4))
2287 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2289 else if (!strncasecmp (au, "Digest", 6))
2290 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2291 #endif /* USE_DIGEST */