2 Copyright (C) 1995, 1996, 1997, 1998, 2000 Free Software Foundation, Inc.
4 This file is part of Wget.
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
24 #include <sys/types.h>
35 #if TIME_WITH_SYS_TIME
36 # include <sys/time.h>
40 # include <sys/time.h>
49 # include <netdb.h> /* for h_errno */
66 # include "gen_sslfunc.h"
69 extern char *version_string;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 hdr += skip_lws (hdr);
211 for (num = 0; ISDIGIT (*hdr); hdr++)
212 num = 10 * num + (*hdr - '0');
213 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
215 closure->first_byte_pos = num;
217 for (num = 0; ISDIGIT (*hdr); hdr++)
218 num = 10 * num + (*hdr - '0');
219 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
221 closure->last_byte_pos = num;
223 for (num = 0; ISDIGIT (*hdr); hdr++)
224 num = 10 * num + (*hdr - '0');
225 closure->entity_length = num;
229 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
230 Used for `Accept-Ranges'. */
232 http_process_none (const char *hdr, void *arg)
234 int *where = (int *)arg;
236 if (strstr (hdr, "none"))
243 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
245 http_process_type (const char *hdr, void *arg)
247 char **result = (char **)arg;
248 /* Locate P on `;' or the terminating zero, whichever comes first. */
249 const char *p = strchr (hdr, ';');
251 p = hdr + strlen (hdr);
252 while (p > hdr && ISSPACE (*(p - 1)))
254 *result = strdupdelim (hdr, p);
258 /* Check whether the `Connection' header is set to "keep-alive". */
260 http_process_connection (const char *hdr, void *arg)
262 int *flag = (int *)arg;
263 if (!strcasecmp (hdr, "Keep-Alive"))
268 /* Persistent connections. Currently, we cache the most recently used
269 connection as persistent, provided that the HTTP server agrees to
270 make it such. The persistence data is stored in the variables
271 below. Ideally, it would be in a structure, and it should be
272 possible to cache an arbitrary fixed number of these connections.
274 I think the code is quite easy to extend in that direction. */
276 /* Whether a persistent connection is active. */
277 static int pc_active_p;
278 /* Host and port of currently active persistent connection. */
279 static unsigned char pc_last_host[4];
280 static unsigned short pc_last_port;
282 /* File descriptor of the currently active persistent connection. */
283 static int pc_last_fd;
286 /* Whether a ssl handshake has occoured on this connection */
287 static int pc_active_ssl;
288 /* SSL connection of the currently active persistent connection. */
289 static SSL *pc_last_ssl;
290 #endif /* HAVE_SSL */
292 /* Mark the persistent connection as invalid. This is used by the
293 CLOSE_* macros after they forcefully close a registered persistent
294 connection. This does not close the file descriptor -- it is left
295 to the caller to do that. (Maybe it should, though.) */
298 invalidate_persistent (void)
303 #endif /* HAVE_SSL */
304 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
307 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
308 persistent. This will enable someone to use the same connection
309 later. In the context of HTTP, this must be called only AFTER the
310 response has been received and the server has promised that the
311 connection will remain alive.
313 If a previous connection was persistent, it is closed. */
316 register_persistent (const char *host, unsigned short port, int fd
326 if (pc_last_fd == fd)
328 /* The connection FD is already registered. Nothing to
334 /* The old persistent connection is still active; let's
335 close it first. This situation arises whenever a
336 persistent connection exists, but we then connect to a
337 different host, and try to register a persistent
338 connection to that one. */
340 /* The ssl disconnect has to take place before the closing
343 shutdown_ssl(pc_last_ssl);
346 invalidate_persistent ();
350 /* This store_hostaddress may not fail, because it has the results
352 success = store_hostaddress (pc_last_host, host);
359 pc_active_ssl = ssl ? 1 : 0;
361 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
364 /* Return non-zero if a persistent connection is available for
365 connecting to HOST:PORT. */
368 persistent_available_p (const char *host, unsigned short port
374 unsigned char this_host[4];
375 /* First, check whether a persistent connection is active at all. */
378 /* Second, check if the active connection pertains to the correct
379 (HOST, PORT) ordered pair. */
380 if (port != pc_last_port)
383 /* Second, a): check if current connection is (not) ssl, too. This
384 test is unlikely to fail because HTTP and HTTPS typicaly use
385 different ports. Yet it is possible, or so I [Christian
386 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
388 if (ssl != pc_active_ssl)
390 #endif /* HAVE_SSL */
391 if (!store_hostaddress (this_host, host))
393 if (memcmp (pc_last_host, this_host, 4))
395 /* Third: check whether the connection is still open. This is
396 important because most server implement a liberal (short) timeout
397 on persistent connections. Wget can of course always reconnect
398 if the connection doesn't work out, but it's nicer to know in
399 advance. This test is a logical followup of the first test, but
400 is "expensive" and therefore placed at the end of the list. */
401 if (!test_socket_open (pc_last_fd))
403 /* Oops, the socket is no longer open. Now that we know that,
404 let's invalidate the persistent connection before returning
407 invalidate_persistent ();
414 # define SHUTDOWN_SSL(ssl) do { \
416 shutdown_ssl (ssl); \
419 # define SHUTDOWN_SSL(ssl)
422 /* The idea behind these two CLOSE macros is to distinguish between
423 two cases: one when the job we've been doing is finished, and we
424 want to close the connection and leave, and two when something is
425 seriously wrong and we're closing the connection as part of
428 In case of keep_alive, CLOSE_FINISH should leave the connection
429 open, while CLOSE_INVALIDATE should still close it.
431 Note that the semantics of the flag `keep_alive' is "this
432 connection *will* be reused (the server has promised not to close
433 the connection once we're done)", while the semantics of
434 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
435 active, registered connection". */
437 #define CLOSE_FINISH(fd) do { \
440 SHUTDOWN_SSL (ssl); \
442 if (pc_active_p && (fd) == pc_last_fd) \
443 invalidate_persistent (); \
447 #define CLOSE_INVALIDATE(fd) do { \
448 SHUTDOWN_SSL (ssl); \
450 if (pc_active_p && (fd) == pc_last_fd) \
451 invalidate_persistent (); \
456 long len; /* received length */
457 long contlen; /* expected length */
458 long restval; /* the restart value */
459 int res; /* the result of last read */
460 char *newloc; /* new location (redirection) */
461 char *remote_time; /* remote time-stamp string */
462 char *error; /* textual HTTP error */
463 int statcode; /* status code */
464 long dltime; /* time of the download */
465 int no_truncate; /* whether truncating the file is
469 /* Free the elements of hstat X. */
470 #define FREEHSTAT(x) do \
472 FREE_MAYBE ((x).newloc); \
473 FREE_MAYBE ((x).remote_time); \
474 FREE_MAYBE ((x).error); \
475 (x).newloc = (x).remote_time = (x).error = NULL; \
478 static char *create_authorization_line PARAMS ((const char *, const char *,
479 const char *, const char *,
481 static char *basic_authentication_encode PARAMS ((const char *, const char *,
483 static int known_authentication_scheme_p PARAMS ((const char *));
485 static time_t http_atotm PARAMS ((char *));
487 #define BEGINS_WITH(line, string_constant) \
488 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
489 && (ISSPACE (line[sizeof (string_constant) - 1]) \
490 || !line[sizeof (string_constant) - 1]))
492 /* Retrieve a document through HTTP protocol. It recognizes status
493 code, and correctly handles redirections. It closes the network
494 socket. If it receives an error from the functions below it, it
495 will print it if there is enough information to do so (almost
496 always), returning the error to the caller (i.e. http_loop).
498 Various HTTP parameters are stored to hs. Although it parses the
499 response code correctly, it is not used in a sane way. The caller
502 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
503 and u->proxy->url will be given to the proxy server (bad naming,
506 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
508 char *request, *type, *command, *path;
510 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
511 char *authenticate_h;
515 char *request_keep_alive;
516 int sock, hcount, num_written, all_length, remport, statcode;
517 long contlen, contrange;
521 int auth_tried_already;
524 static SSL_CTX *ssl_ctx = NULL;
526 #endif /* HAVE_SSL */
528 /* Whether this connection will be kept alive after the HTTP request
532 /* Flags that detect the two ways of specifying HTTP keep-alive
534 int http_keep_alive_1, http_keep_alive_2;
536 /* Whether keep-alive should be inhibited. */
537 int inhibit_keep_alive;
540 /* initialize ssl_ctx on first run */
543 err=init_ssl (&ssl_ctx);
548 case SSLERRCTXCREATE:
550 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
554 /* try without certfile */
555 logprintf (LOG_NOTQUIET,
556 _("Failed to load certificates from %s\n"),
559 logprintf (LOG_NOTQUIET,
560 _("Trying without the specified certificate\n"));
563 logprintf (LOG_NOTQUIET,
564 _("Failed to get certificate key from %s\n"),
567 logprintf (LOG_NOTQUIET,
568 _("Trying without the specified certificate\n"));
575 #endif /* HAVE_SSL */
577 if (!(*dt & HEAD_ONLY))
578 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
579 know the local filename so we can save to it. */
580 assert (u->local != NULL);
583 auth_tried_already = 0;
585 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
588 /* We need to come back here when the initial attempt to retrieve
589 without authorization header fails. (Expected to happen at least
590 for the Digest authorization scheme.) */
593 http_keep_alive_1 = http_keep_alive_2 = 0;
595 /* Initialize certain elements of struct http_stat. */
600 hs->remote_time = NULL;
603 /* Which structure to use to retrieve the original URL data. */
609 /* First: establish the connection. */
610 if (inhibit_keep_alive
613 !persistent_available_p (u->host, u->port)
615 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
616 #endif /* HAVE_SSL */
619 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
620 err = make_connection (&sock, u->host, u->port);
624 logputs (LOG_VERBOSE, "\n");
625 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
629 logputs (LOG_VERBOSE, "\n");
630 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
634 logputs (LOG_VERBOSE, "\n");
635 logprintf (LOG_NOTQUIET,
636 _("Connection to %s:%hu refused.\n"), u->host, u->port);
640 logputs (LOG_VERBOSE, "\n");
641 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
646 /* Everything is fine! */
647 logputs (LOG_VERBOSE, _("connected!\n"));
654 if (u->proto == URLHTTPS)
655 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
657 logputs (LOG_VERBOSE, "\n");
658 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
662 #endif /* HAVE_SSL */
666 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
667 /* #### pc_last_fd should be accessed through an accessor
672 #endif /* HAVE_SSL */
673 DEBUGP (("Reusing fd %d.\n", sock));
677 path = u->proxy->url;
681 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
685 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
686 sprintf (referer, "Referer: %s\r\n", ou->referer);
688 if (*dt & SEND_NOCACHE)
689 pragma_h = "Pragma: no-cache\r\n";
694 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
695 /* Gag me! Some servers (e.g. WebSitePro) have been known to
696 respond to the following `Range' format by generating a
697 multipart/x-byte-ranges MIME document! This MIME type was
698 present in an old draft of the byteranges specification.
699 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
700 only if multiple non-overlapping ranges are requested --
701 which Wget never does. */
702 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
707 STRDUP_ALLOCA (useragent, opt.useragent);
710 useragent = (char *)alloca (10 + strlen (version_string));
711 sprintf (useragent, "Wget/%s", version_string);
713 /* Construct the authentication, if userid is present. */
716 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
717 user = user ? user : opt.http_user;
718 passwd = passwd ? passwd : opt.http_passwd;
725 /* We have the username and the password, but haven't tried
726 any authorization yet. Let's see if the "Basic" method
727 works. If not, we'll come back here and construct a
728 proper authorization method with the right challenges.
730 If we didn't employ this kind of logic, every URL that
731 requires authorization would have to be processed twice,
732 which is very suboptimal and generates a bunch of false
733 "unauthorized" errors in the server log.
735 #### But this logic also has a serious problem when used
736 with stronger authentications: we *first* transmit the
737 username and the password in clear text, and *then*
738 attempt a stronger authentication scheme. That cannot be
739 right! We are only fortunate that almost everyone still
740 uses the `Basic' scheme anyway.
742 There should be an option to prevent this from happening,
743 for those who use strong authentication schemes and value
745 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
749 wwwauth = create_authorization_line (authenticate_h, user, passwd,
757 char *proxy_user, *proxy_passwd;
758 /* For normal username and password, URL components override
759 command-line/wgetrc parameters. With proxy authentication,
760 it's the reverse, because proxy URLs are normally the
761 "permanent" ones, so command-line args should take
763 if (opt.proxy_user && opt.proxy_passwd)
765 proxy_user = opt.proxy_user;
766 proxy_passwd = opt.proxy_passwd;
770 proxy_user = u->user;
771 proxy_passwd = u->passwd;
773 /* #### This is junky. Can't the proxy request, say, `Digest'
775 if (proxy_user && proxy_passwd)
776 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
777 "Proxy-Authorization");
782 /* String of the form :PORT. Used only for non-standard ports. */
785 if (remport != (u->proto == URLHTTPS ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT) )
787 if (remport != DEFAULT_HTTP_PORT)
790 port_maybe = (char *)alloca (numdigit (remport) + 2);
791 sprintf (port_maybe, ":%d", remport);
794 if (!inhibit_keep_alive)
795 request_keep_alive = "Connection: Keep-Alive\r\n";
797 request_keep_alive = NULL;
799 /* Allocate the memory for the request. */
800 request = (char *)alloca (strlen (command) + strlen (path)
803 + (port_maybe ? strlen (port_maybe) : 0)
804 + strlen (HTTP_ACCEPT)
805 + (request_keep_alive
806 ? strlen (request_keep_alive) : 0)
807 + (referer ? strlen (referer) : 0)
808 + (wwwauth ? strlen (wwwauth) : 0)
809 + (proxyauth ? strlen (proxyauth) : 0)
810 + (range ? strlen (range) : 0)
812 + (opt.user_header ? strlen (opt.user_header) : 0)
814 /* Construct the request. */
821 command, path, useragent, remhost,
822 port_maybe ? port_maybe : "",
824 request_keep_alive ? request_keep_alive : "",
825 referer ? referer : "",
826 wwwauth ? wwwauth : "",
827 proxyauth ? proxyauth : "",
830 opt.user_header ? opt.user_header : "");
831 DEBUGP (("---request begin---\n%s---request end---\n", request));
832 /* Free the temporary memory. */
833 FREE_MAYBE (wwwauth);
834 FREE_MAYBE (proxyauth);
836 /* Send the request to server. */
838 if (u->proto == URLHTTPS)
839 num_written = ssl_iwrite (ssl, request, strlen (request));
841 #endif /* HAVE_SSL */
842 num_written = iwrite (sock, request, strlen (request));
846 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
848 CLOSE_INVALIDATE (sock);
851 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
852 u->proxy ? "Proxy" : "HTTP");
853 contlen = contrange = -1;
858 /* Before reading anything, initialize the rbuf. */
859 rbuf_initialize (&rbuf, sock);
861 if (u->proto == URLHTTPS)
865 #endif /* HAVE_SSL */
868 /* Header-fetching loop. */
876 /* Get the header. */
877 status = header_get (&rbuf, &hdr,
878 /* Disallow continuations for status line. */
879 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
881 /* Check for errors. */
882 if (status == HG_EOF && *hdr)
884 /* This used to be an unconditional error, but that was
885 somewhat controversial, because of a large number of
886 broken CGI's that happily "forget" to send the second EOL
887 before closing the connection of a HEAD request.
889 So, the deal is to check whether the header is empty
890 (*hdr is zero if it is); if yes, it means that the
891 previous header was fully retrieved, and that -- most
892 probably -- the request is complete. "...be liberal in
893 what you accept." Oh boy. */
894 logputs (LOG_VERBOSE, "\n");
895 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
898 FREE_MAYBE (hs->newloc);
899 FREE_MAYBE (all_headers);
900 CLOSE_INVALIDATE (sock);
903 else if (status == HG_ERROR)
905 logputs (LOG_VERBOSE, "\n");
906 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
910 FREE_MAYBE (hs->newloc);
911 FREE_MAYBE (all_headers);
912 CLOSE_INVALIDATE (sock);
916 /* If the headers are to be saved to a file later, save them to
918 if (opt.save_headers)
920 int lh = strlen (hdr);
921 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
922 memcpy (all_headers + all_length, hdr, lh);
924 all_headers[all_length++] = '\n';
925 all_headers[all_length] = '\0';
928 /* Print the header if requested. */
929 if (opt.server_response && hcount != 1)
930 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
932 /* Check for status line. */
936 /* Parse the first line of server response. */
937 statcode = parse_http_status_line (hdr, &error);
938 hs->statcode = statcode;
939 /* Store the descriptive response. */
940 if (statcode == -1) /* malformed response */
942 /* A common reason for "malformed response" error is the
943 case when no data was actually received. Handle this
946 hs->error = xstrdup (_("No data received"));
948 hs->error = xstrdup (_("Malformed status line"));
953 hs->error = xstrdup (_("(no description)"));
955 hs->error = xstrdup (error);
962 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
967 /* Exit on empty header. */
974 /* Try getting content-length. */
975 if (contlen == -1 && !opt.ignore_length)
976 if (header_process (hdr, "Content-Length", header_extract_number,
979 /* Try getting content-type. */
981 if (header_process (hdr, "Content-Type", http_process_type, &type))
983 /* Try getting location. */
985 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
987 /* Try getting last-modified. */
988 if (!hs->remote_time)
989 if (header_process (hdr, "Last-Modified", header_strdup,
992 /* Try getting www-authentication. */
994 if (header_process (hdr, "WWW-Authenticate", header_strdup,
997 /* Check for accept-ranges header. If it contains the word
998 `none', disable the ranges. */
999 if (*dt & ACCEPTRANGES)
1002 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1005 *dt &= ~ACCEPTRANGES;
1009 /* Try getting content-range. */
1010 if (contrange == -1)
1012 struct http_process_range_closure closure;
1013 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1015 contrange = closure.first_byte_pos;
1019 /* Check for keep-alive related responses. */
1020 if (!inhibit_keep_alive)
1022 /* Check for the `Keep-Alive' header. */
1023 if (!http_keep_alive_1)
1025 if (header_process (hdr, "Keep-Alive", header_exists,
1026 &http_keep_alive_1))
1029 /* Check for `Connection: Keep-Alive'. */
1030 if (!http_keep_alive_2)
1032 if (header_process (hdr, "Connection", http_process_connection,
1033 &http_keep_alive_2))
1041 logputs (LOG_VERBOSE, "\n");
1044 && (http_keep_alive_1 || http_keep_alive_2))
1046 assert (inhibit_keep_alive == 0);
1050 /* The server has promised that it will not close the connection
1051 when we're done. This means that we can register it. */
1053 register_persistent (u->host, u->port, sock);
1055 register_persistent (u->host, u->port, sock, ssl);
1056 #endif /* HAVE_SSL */
1058 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1061 /* Authorization is required. */
1065 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1066 might be more bytes in the body. */
1067 if (auth_tried_already)
1069 /* If we have tried it already, then there is not point
1072 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1073 xfree (authenticate_h);
1076 else if (!known_authentication_scheme_p (authenticate_h))
1078 xfree (authenticate_h);
1079 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1082 else if (BEGINS_WITH (authenticate_h, "Basic"))
1084 /* The authentication scheme is basic, the one we try by
1085 default, and it failed. There's no sense in trying
1091 auth_tried_already = 1;
1095 /* We do not need this anymore. */
1098 xfree (authenticate_h);
1099 authenticate_h = NULL;
1102 /* 20x responses are counted among successful by default. */
1103 if (H_20X (statcode))
1106 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1109 /* We don't assume text/html by default. */
1112 if (opt.html_extension && (*dt & TEXTHTML))
1113 /* -E / --html-extension / html_extension = on was specified, and this is a
1114 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1115 already the file's suffix, tack on ".html". */
1117 char* last_period_in_local_filename = strrchr(u->local, '.');
1119 if (last_period_in_local_filename == NULL ||
1120 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1121 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1123 size_t local_filename_len = strlen(u->local);
1125 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1126 strcpy(u->local + local_filename_len, ".html");
1128 *dt |= ADDED_HTML_EXTENSION;
1132 if (contrange == -1)
1134 /* We did not get a content-range header. This means that the
1135 server did not honor our `Range' request. Normally, this
1136 means we should reset hs->restval and continue normally. */
1138 /* However, if `-c' is used, we need to be a bit more careful:
1140 1. If `-c' is specified and the file already existed when
1141 Wget was started, it would be a bad idea for us to start
1142 downloading it from scratch, effectively truncating it. I
1143 believe this cannot happen unless `-c' was specified.
1145 2. If `-c' is used on a file that is already fully
1146 downloaded, we're requesting bytes after the end of file,
1147 which can result in server not honoring `Range'. If this is
1148 the case, `Content-Length' will be equal to the length of the
1150 if (opt.always_rest)
1152 /* Check for condition #2. */
1153 if (hs->restval == contlen)
1155 logputs (LOG_VERBOSE, _("\
1156 \n The file is already fully retrieved; nothing to do.\n\n"));
1157 /* In case the caller inspects. */
1161 FREE_MAYBE (hs->newloc);
1162 FREE_MAYBE (all_headers);
1163 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1164 might be more bytes in the body. */
1165 return RETRFINISHED;
1168 /* Check for condition #1. */
1169 if (hs->no_truncate)
1171 logprintf (LOG_NOTQUIET,
1174 The server does not support continued download;
1175 refusing to truncate `%s'.\n\n"), u->local);
1176 return CONTNOTSUPPORTED;
1185 else if (contrange != hs->restval ||
1186 (H_PARTIAL (statcode) && contrange == -1))
1188 /* This means the whole request was somehow misunderstood by the
1189 server. Bail out. */
1191 FREE_MAYBE (hs->newloc);
1192 FREE_MAYBE (all_headers);
1193 CLOSE_INVALIDATE (sock);
1200 contlen += contrange;
1202 contrange = -1; /* If conent-length was not sent,
1203 content-range will be ignored. */
1205 hs->contlen = contlen;
1207 /* Return if redirected. */
1208 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1210 /* RFC2068 says that in case of the 300 (multiple choices)
1211 response, the server can output a preferred URL through
1212 `Location' header; otherwise, the request should be treated
1213 like GET. So, if the location is set, it will be a
1214 redirection; otherwise, just proceed normally. */
1215 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1219 logprintf (LOG_VERBOSE,
1220 _("Location: %s%s\n"),
1221 hs->newloc ? hs->newloc : _("unspecified"),
1222 hs->newloc ? _(" [following]") : "");
1223 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1224 might be more bytes in the body. */
1226 FREE_MAYBE (all_headers);
1232 if ((*dt & RETROKF) && !opt.server_response)
1234 /* No need to print this output if the body won't be
1235 downloaded at all, or if the original server response is
1237 logputs (LOG_VERBOSE, _("Length: "));
1240 logputs (LOG_VERBOSE, legible (contlen));
1241 if (contrange != -1)
1242 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1243 legible (contlen - contrange));
1246 logputs (LOG_VERBOSE,
1247 opt.ignore_length ? _("ignored") : _("unspecified"));
1249 logprintf (LOG_VERBOSE, " [%s]\n", type);
1251 logputs (LOG_VERBOSE, "\n");
1255 type = NULL; /* We don't need it any more. */
1257 /* Return if we have no intention of further downloading. */
1258 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1260 /* In case the caller cares to look... */
1264 FREE_MAYBE (all_headers);
1265 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1266 might be more bytes in the body. */
1267 return RETRFINISHED;
1270 /* Open the local file. */
1273 mkalldirs (u->local);
1275 rotate_backups (u->local);
1276 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1279 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1280 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1281 might be more bytes in the body. */
1282 FREE_MAYBE (all_headers);
1291 /* This will silently fail for streams that don't correspond
1292 to regular files, but that's OK. */
1298 /* #### This confuses the code that checks for file size. There
1299 should be some overhead information. */
1300 if (opt.save_headers)
1301 fwrite (all_headers, 1, all_length, fp);
1303 /* Get the contents of the document. */
1304 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1305 (contlen != -1 ? contlen : 0),
1307 hs->dltime = elapsed_time ();
1309 /* Close or flush the file. We have to be careful to check for
1310 error here. Checking the result of fwrite() is not enough --
1311 errors could go unnoticed! */
1314 flush_res = fclose (fp);
1316 flush_res = fflush (fp);
1317 if (flush_res == EOF)
1320 FREE_MAYBE (all_headers);
1321 CLOSE_FINISH (sock);
1324 return RETRFINISHED;
1327 /* The genuine HTTP loop! This is the part where the retrieval is
1328 retried, and retried, and retried, and... */
1330 http_loop (struct urlinfo *u, char **newloc, int *dt)
1333 int use_ts, got_head = 0; /* time-stamping info */
1334 char *filename_plus_orig_suffix;
1335 char *local_filename = NULL;
1336 char *tms, *suf, *locf, *tmrate;
1338 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1339 long local_size = 0; /* the size of the local file */
1340 size_t filename_len;
1341 struct http_stat hstat; /* HTTP status */
1346 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1347 has_wildcards_p because it would also warn on `?', and we know that
1348 shows up in CGI paths a *lot*. */
1349 if (strchr (u->url, '*'))
1350 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1352 /* Determine the local filename. */
1354 u->local = url_filename (u->proxy ? u->proxy : u);
1356 if (!opt.output_document)
1359 locf = opt.output_document;
1361 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1362 xmalloc() here before EACH return. This is one reason it's better to set
1363 flags that influence flow control and then return once at the end. */
1364 filename_len = strlen(u->local);
1365 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1367 if (opt.noclobber && file_exists_p (u->local))
1369 /* If opt.noclobber is turned on and file already exists, do not
1370 retrieve the file */
1371 logprintf (LOG_VERBOSE, _("\
1372 File `%s' already there, will not retrieve.\n"), u->local);
1373 /* If the file is there, we suppose it's retrieved OK. */
1376 /* #### Bogusness alert. */
1377 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1378 text/html, a harmless lie. */
1379 if (((suf = suffix (u->local)) != NULL)
1380 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1383 xfree (filename_plus_orig_suffix); /* must precede every return! */
1384 /* Another harmless lie: */
1389 if (opt.timestamping)
1391 boolean local_dot_orig_file_exists = FALSE;
1393 if (opt.backup_converted)
1394 /* If -K is specified, we'll act on the assumption that it was specified
1395 last time these files were downloaded as well, and instead of just
1396 comparing local file X against server file X, we'll compare local
1397 file X.orig (if extant, else X) against server file X. If -K
1398 _wasn't_ specified last time, or the server contains files called
1399 *.orig, -N will be back to not operating correctly with -k. */
1401 /* Would a single s[n]printf() call be faster? --dan
1403 Definitely not. sprintf() is horribly slow. It's a
1404 different question whether the difference between the two
1405 affects a program. Usually I'd say "no", but at one
1406 point I profiled Wget, and found that a measurable and
1407 non-negligible amount of time was lost calling sprintf()
1408 in url.c. Replacing sprintf with inline calls to
1409 strcpy() and long_to_string() made a difference.
1411 strcpy(filename_plus_orig_suffix, u->local);
1412 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1414 /* Try to stat() the .orig file. */
1415 if (stat(filename_plus_orig_suffix, &st) == 0)
1417 local_dot_orig_file_exists = TRUE;
1418 local_filename = filename_plus_orig_suffix;
1422 if (!local_dot_orig_file_exists)
1423 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1424 if (stat (u->local, &st) == 0)
1425 local_filename = u->local;
1427 if (local_filename != NULL)
1428 /* There was a local file, so we'll check later to see if the version
1429 the server has is the same version we already have, allowing us to
1434 local_size = st.st_size;
1438 /* Reset the counter. */
1440 *dt = 0 | ACCEPTRANGES;
1444 /* Increment the pass counter. */
1446 sleep_between_retrievals (count);
1447 /* Get the current time string. */
1448 tms = time_str (NULL);
1449 /* Print fetch message, if opt.verbose. */
1452 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1456 sprintf (tmp, _("(try:%2d)"), count);
1457 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1458 tms, hurl, tmp, locf);
1460 ws_changetitle (hurl, 1);
1465 /* Default document type is empty. However, if spider mode is
1466 on or time-stamping is employed, HEAD_ONLY commands is
1467 encoded within *dt. */
1468 if (opt.spider || (use_ts && !got_head))
1472 /* Assume no restarting. */
1474 /* Decide whether or not to restart. */
1475 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1476 && file_exists_p (u->local))
1477 if (stat (u->local, &st) == 0)
1478 hstat.restval = st.st_size;
1479 /* Decide whether to send the no-cache directive. */
1480 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1481 *dt |= SEND_NOCACHE;
1483 *dt &= ~SEND_NOCACHE;
1485 /* Try fetching the document, or at least its head. :-) */
1486 err = gethttp (u, &hstat, dt);
1488 /* It's unfortunate that wget determines the local filename before finding
1489 out the Content-Type of the file. Barring a major restructuring of the
1490 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1491 u->local to tack on ".html". */
1492 if (!opt.output_document)
1495 locf = opt.output_document;
1497 /* In `-c' is used, check whether the file we're writing to
1498 exists before we've done anything. If so, we'll refuse to
1499 truncate it if the server doesn't support continued
1501 if (opt.always_rest)
1502 hstat.no_truncate = file_exists_p (locf);
1505 tms = time_str (NULL);
1506 /* Get the new location (with or without the redirection). */
1508 *newloc = xstrdup (hstat.newloc);
1511 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1512 case CONERROR: case READERR: case WRITEFAILED:
1514 /* Non-fatal errors continue executing the loop, which will
1515 bring them to "while" statement at the end, to judge
1516 whether the number of tries was exceeded. */
1518 printwhat (count, opt.ntry);
1521 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1522 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1523 /* Fatal errors just return from the function. */
1525 xfree (filename_plus_orig_suffix); /* must precede every return! */
1528 case FWRITEERR: case FOPENERR:
1529 /* Another fatal error. */
1530 logputs (LOG_VERBOSE, "\n");
1531 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1532 u->local, strerror (errno));
1537 /* Another fatal error. */
1538 logputs (LOG_VERBOSE, "\n");
1539 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1541 xfree (filename_plus_orig_suffix); /* must precede every return! */
1545 /* Return the new location to the caller. */
1548 logprintf (LOG_NOTQUIET,
1549 _("ERROR: Redirection (%d) without location.\n"),
1551 xfree (filename_plus_orig_suffix); /* must precede every return! */
1555 xfree (filename_plus_orig_suffix); /* must precede every return! */
1559 /* Deal with you later. */
1562 /* All possibilities should have been exhausted. */
1565 if (!(*dt & RETROKF))
1569 /* #### Ugly ugly ugly! */
1570 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1571 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1574 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1575 tms, hstat.statcode, hstat.error);
1576 logputs (LOG_VERBOSE, "\n");
1578 xfree (filename_plus_orig_suffix); /* must precede every return! */
1582 /* Did we get the time-stamp? */
1585 if (opt.timestamping && !hstat.remote_time)
1587 logputs (LOG_NOTQUIET, _("\
1588 Last-modified header missing -- time-stamps turned off.\n"));
1590 else if (hstat.remote_time)
1592 /* Convert the date-string into struct tm. */
1593 tmr = http_atotm (hstat.remote_time);
1594 if (tmr == (time_t) (-1))
1595 logputs (LOG_VERBOSE, _("\
1596 Last-modified header invalid -- time-stamp ignored.\n"));
1600 /* The time-stamping section. */
1605 use_ts = 0; /* no more time-stamping */
1606 count = 0; /* the retrieve count for HEAD is
1608 if (hstat.remote_time && tmr != (time_t) (-1))
1610 /* Now time-stamping can be used validly. Time-stamping
1611 means that if the sizes of the local and remote file
1612 match, and local file is newer than the remote file,
1613 it will not be retrieved. Otherwise, the normal
1614 download procedure is resumed. */
1616 (hstat.contlen == -1 || local_size == hstat.contlen))
1618 logprintf (LOG_VERBOSE, _("\
1619 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1622 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1625 else if (tml >= tmr)
1626 logprintf (LOG_VERBOSE, _("\
1627 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1629 logputs (LOG_VERBOSE,
1630 _("Remote file is newer, retrieving.\n"));
1635 if ((tmr != (time_t) (-1))
1637 && ((hstat.len == hstat.contlen) ||
1638 ((hstat.res == 0) &&
1639 ((hstat.contlen == -1) ||
1640 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1642 /* #### This code repeats in http.c and ftp.c. Move it to a
1644 const char *fl = NULL;
1645 if (opt.output_document)
1647 if (opt.od_known_regular)
1648 fl = opt.output_document;
1655 /* End of time-stamping section. */
1659 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1660 xfree (filename_plus_orig_suffix); /* must precede every return! */
1664 /* It is now safe to free the remainder of hstat, since the
1665 strings within it will no longer be used. */
1668 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1670 if (hstat.len == hstat.contlen)
1674 logprintf (LOG_VERBOSE,
1675 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1676 tms, tmrate, locf, hstat.len, hstat.contlen);
1677 logprintf (LOG_NONVERBOSE,
1678 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1679 tms, u->url, hstat.len, hstat.contlen, locf, count);
1682 downloaded_increase (hstat.len);
1684 /* Remember that we downloaded the file for later ".orig" code. */
1685 if (*dt & ADDED_HTML_EXTENSION)
1686 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1688 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1690 xfree(filename_plus_orig_suffix); /* must precede every return! */
1693 else if (hstat.res == 0) /* No read error */
1695 if (hstat.contlen == -1) /* We don't know how much we were supposed
1696 to get, so assume we succeeded. */
1700 logprintf (LOG_VERBOSE,
1701 _("%s (%s) - `%s' saved [%ld]\n\n"),
1702 tms, tmrate, locf, hstat.len);
1703 logprintf (LOG_NONVERBOSE,
1704 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1705 tms, u->url, hstat.len, locf, count);
1708 downloaded_increase (hstat.len);
1710 /* Remember that we downloaded the file for later ".orig" code. */
1711 if (*dt & ADDED_HTML_EXTENSION)
1712 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1714 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1716 xfree (filename_plus_orig_suffix); /* must precede every return! */
1719 else if (hstat.len < hstat.contlen) /* meaning we lost the
1720 connection too soon */
1722 logprintf (LOG_VERBOSE,
1723 _("%s (%s) - Connection closed at byte %ld. "),
1724 tms, tmrate, hstat.len);
1725 printwhat (count, opt.ntry);
1728 else if (!opt.kill_longer) /* meaning we got more than expected */
1730 logprintf (LOG_VERBOSE,
1731 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1732 tms, tmrate, locf, hstat.len, hstat.contlen);
1733 logprintf (LOG_NONVERBOSE,
1734 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1735 tms, u->url, hstat.len, hstat.contlen, locf, count);
1737 downloaded_increase (hstat.len);
1739 /* Remember that we downloaded the file for later ".orig" code. */
1740 if (*dt & ADDED_HTML_EXTENSION)
1741 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1743 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1745 xfree (filename_plus_orig_suffix); /* must precede every return! */
1748 else /* the same, but not accepted */
1750 logprintf (LOG_VERBOSE,
1751 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1752 tms, tmrate, hstat.len, hstat.contlen);
1753 printwhat (count, opt.ntry);
1757 else /* now hstat.res can only be -1 */
1759 if (hstat.contlen == -1)
1761 logprintf (LOG_VERBOSE,
1762 _("%s (%s) - Read error at byte %ld (%s)."),
1763 tms, tmrate, hstat.len, strerror (errno));
1764 printwhat (count, opt.ntry);
1767 else /* hstat.res == -1 and contlen is given */
1769 logprintf (LOG_VERBOSE,
1770 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1771 tms, tmrate, hstat.len, hstat.contlen,
1773 printwhat (count, opt.ntry);
1780 while (!opt.ntry || (count < opt.ntry));
1781 xfree (filename_plus_orig_suffix); /* must precede every return! */
1785 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1786 than local timezone (mktime assumes the latter).
1788 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1789 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1791 mktime_from_utc (struct tm *t)
1798 tb = mktime (gmtime (&tl));
1799 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1802 /* Check whether the result of strptime() indicates success.
1803 strptime() returns the pointer to how far it got to in the string.
1804 The processing has been successful if the string is at `GMT' or
1805 `+X', or at the end of the string.
1807 In extended regexp parlance, the function returns 1 if P matches
1808 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1809 strptime()) is considered a failure and 0 is returned. */
1811 check_end (const char *p)
1815 while (ISSPACE (*p))
1818 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1819 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1825 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1826 the three formats RFC2068 allows the HTTP servers to emit --
1827 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1830 We use strptime() to recognize various dates, which makes it a
1831 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1832 allows shortened dates and months, one-digit days, etc.). It also
1833 allows more than one space anywhere where the specs require one SP.
1834 The routine should probably be even more forgiving (as recommended
1835 by RFC2068), but I do not have the time to write one.
1837 Return the computed time_t representation, or -1 if all the
1840 Needless to say, what we *really* need here is something like
1841 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1842 and does not use strptime(). atotm() is to be found in the sources
1843 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1845 http_atotm (char *time_string)
1849 /* Roger Beeman says: "This function dynamically allocates struct tm
1850 t, but does no initialization. The only field that actually
1851 needs initialization is tm_isdst, since the others will be set by
1852 strptime. Since strptime does not set tm_isdst, it will return
1853 the data structure with whatever data was in tm_isdst to begin
1854 with. For those of us in timezones where DST can occur, there
1855 can be a one hour shift depending on the previous contents of the
1856 data area where the data structure is allocated." */
1859 /* Note that under foreign locales Solaris strptime() fails to
1860 recognize English dates, which renders this function useless. I
1861 assume that other non-GNU strptime's are plagued by the same
1862 disease. We solve this by setting only LC_MESSAGES in
1863 i18n_initialize(), instead of LC_ALL.
1865 Another solution could be to temporarily set locale to C, invoke
1866 strptime(), and restore it back. This is slow and dirty,
1867 however, and locale support other than LC_MESSAGES can mess other
1868 things, so I rather chose to stick with just setting LC_MESSAGES.
1870 Also note that none of this is necessary under GNU strptime(),
1871 because it recognizes both international and local dates. */
1873 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1874 it to eat all white space up to (and including) a newline, and
1875 the function fails if there is no newline (!).
1877 Let's hope all strptime() implementations use ` ' to skip *all*
1878 whitespace instead of just one (it works that way on all the
1879 systems I've tested it on). */
1881 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1882 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1883 return mktime_from_utc (&t);
1884 /* RFC850: Thu, 29-Jan-98 22:12:57 */
1885 if (check_end (strptime (time_string, "%a, %d-%b-%y %T", &t)))
1886 return mktime_from_utc (&t);
1887 /* asctime: Thu Jan 29 22:12:57 1998 */
1888 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1889 return mktime_from_utc (&t);
1894 /* Authorization support: We support two authorization schemes:
1896 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1898 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1899 consisting of answering to the server's challenge with the proper
1902 /* How many bytes it will take to store LEN bytes in base64. */
1903 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1905 /* Encode the string S of length LENGTH to base64 format and place it
1906 to STORE. STORE will be 0-terminated, and must point to a writable
1907 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1909 base64_encode (const char *s, char *store, int length)
1911 /* Conversion table. */
1912 static char tbl[64] = {
1913 'A','B','C','D','E','F','G','H',
1914 'I','J','K','L','M','N','O','P',
1915 'Q','R','S','T','U','V','W','X',
1916 'Y','Z','a','b','c','d','e','f',
1917 'g','h','i','j','k','l','m','n',
1918 'o','p','q','r','s','t','u','v',
1919 'w','x','y','z','0','1','2','3',
1920 '4','5','6','7','8','9','+','/'
1923 unsigned char *p = (unsigned char *)store;
1925 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1926 for (i = 0; i < length; i += 3)
1928 *p++ = tbl[s[0] >> 2];
1929 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1930 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1931 *p++ = tbl[s[2] & 0x3f];
1934 /* Pad the result if necessary... */
1935 if (i == length + 1)
1937 else if (i == length + 2)
1938 *(p - 1) = *(p - 2) = '=';
1939 /* ...and zero-terminate it. */
1943 /* Create the authentication header contents for the `Basic' scheme.
1944 This is done by encoding the string `USER:PASS' in base64 and
1945 prepending `HEADER: Basic ' to it. */
1947 basic_authentication_encode (const char *user, const char *passwd,
1950 char *t1, *t2, *res;
1951 int len1 = strlen (user) + 1 + strlen (passwd);
1952 int len2 = BASE64_LENGTH (len1);
1954 t1 = (char *)alloca (len1 + 1);
1955 sprintf (t1, "%s:%s", user, passwd);
1956 t2 = (char *)alloca (1 + len2);
1957 base64_encode (t1, t2, len1);
1958 res = (char *)xmalloc (len2 + 11 + strlen (header));
1959 sprintf (res, "%s: Basic %s\r\n", header, t2);
1965 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
1966 of a field in such a header. If the field is the one specified by
1967 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
1968 digest authorization code), extract its value in the (char*)
1969 variable pointed by RET. Returns negative on a malformed header,
1970 or number of bytes that have been parsed by this call. */
1972 extract_header_attr (const char *au, const char *attr_name, char **ret)
1974 const char *cp, *ep;
1978 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
1980 cp += strlen (attr_name);
1983 cp += skip_lws (cp);
1988 cp += skip_lws (cp);
1993 for (ep = cp; *ep && *ep != '\"'; ep++)
1998 *ret = strdupdelim (cp, ep);
2005 /* Response value needs to be in lowercase, so we cannot use HEXD2ASC
2006 from url.h. See RFC 2069 2.1.2 for the syntax of response-digest. */
2007 #define HEXD2asc(x) (((x) < 10) ? ((x) + '0') : ((x) - 10 + 'a'))
2009 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2010 an array of 16 bytes containing the hash keys, and BUF should be a
2011 buffer of 33 writable characters (32 for hex digits plus one for
2012 zero termination). */
2014 dump_hash (unsigned char *buf, const unsigned char *hash)
2018 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2020 *buf++ = HEXD2asc (*hash >> 4);
2021 *buf++ = HEXD2asc (*hash & 0xf);
2026 /* Take the line apart to find the challenge, and compose a digest
2027 authorization header. See RFC2069 section 2.1.2. */
2029 digest_authentication_encode (const char *au, const char *user,
2030 const char *passwd, const char *method,
2033 static char *realm, *opaque, *nonce;
2038 { "realm", &realm },
2039 { "opaque", &opaque },
2044 realm = opaque = nonce = NULL;
2046 au += 6; /* skip over `Digest' */
2051 au += skip_lws (au);
2052 for (i = 0; i < ARRAY_SIZE (options); i++)
2054 int skip = extract_header_attr (au, options[i].name,
2055 options[i].variable);
2059 FREE_MAYBE (opaque);
2069 if (i == ARRAY_SIZE (options))
2071 while (*au && *au != '=')
2075 au += skip_lws (au);
2079 while (*au && *au != '\"')
2086 while (*au && *au != ',')
2091 if (!realm || !nonce || !user || !passwd || !path || !method)
2094 FREE_MAYBE (opaque);
2099 /* Calculate the digest value. */
2102 unsigned char hash[MD5_HASHLEN];
2103 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2104 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2106 /* A1BUF = H(user ":" realm ":" password) */
2107 md5_init_ctx (&ctx);
2108 md5_process_bytes (user, strlen (user), &ctx);
2109 md5_process_bytes (":", 1, &ctx);
2110 md5_process_bytes (realm, strlen (realm), &ctx);
2111 md5_process_bytes (":", 1, &ctx);
2112 md5_process_bytes (passwd, strlen (passwd), &ctx);
2113 md5_finish_ctx (&ctx, hash);
2114 dump_hash (a1buf, hash);
2116 /* A2BUF = H(method ":" path) */
2117 md5_init_ctx (&ctx);
2118 md5_process_bytes (method, strlen (method), &ctx);
2119 md5_process_bytes (":", 1, &ctx);
2120 md5_process_bytes (path, strlen (path), &ctx);
2121 md5_finish_ctx (&ctx, hash);
2122 dump_hash (a2buf, hash);
2124 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2125 md5_init_ctx (&ctx);
2126 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2127 md5_process_bytes (":", 1, &ctx);
2128 md5_process_bytes (nonce, strlen (nonce), &ctx);
2129 md5_process_bytes (":", 1, &ctx);
2130 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2131 md5_finish_ctx (&ctx, hash);
2132 dump_hash (response_digest, hash);
2134 res = (char*) xmalloc (strlen (user)
2139 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2140 + (opaque ? strlen (opaque) : 0)
2142 sprintf (res, "Authorization: Digest \
2143 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2144 user, realm, nonce, path, response_digest);
2147 char *p = res + strlen (res);
2148 strcat (p, ", opaque=\"");
2152 strcat (res, "\r\n");
2156 #endif /* USE_DIGEST */
2159 #define BEGINS_WITH(line, string_constant) \
2160 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2161 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2162 || !line[sizeof (string_constant) - 1]))
2165 known_authentication_scheme_p (const char *au)
2167 return BEGINS_WITH (au, "Basic")
2168 || BEGINS_WITH (au, "Digest")
2169 || BEGINS_WITH (au, "NTLM");
2174 /* Create the HTTP authorization request header. When the
2175 `WWW-Authenticate' response header is seen, according to the
2176 authorization scheme specified in that header (`Basic' and `Digest'
2177 are supported by the current implementation), produce an
2178 appropriate HTTP authorization request header. */
2180 create_authorization_line (const char *au, const char *user,
2181 const char *passwd, const char *method,
2184 char *wwwauth = NULL;
2186 if (!strncasecmp (au, "Basic", 5))
2187 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2188 if (!strncasecmp (au, "NTLM", 4))
2189 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2191 else if (!strncasecmp (au, "Digest", 6))
2192 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2193 #endif /* USE_DIGEST */