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>
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"
70 extern char *version_string;
82 #define TEXTHTML_S "text/html"
83 #define HTTP_ACCEPT "*/*"
85 /* Some status code validation macros: */
86 #define H_20X(x) (((x) >= 200) && ((x) < 300))
87 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
88 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
89 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
91 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
93 #define HTTP_STATUS_OK 200
94 #define HTTP_STATUS_CREATED 201
95 #define HTTP_STATUS_ACCEPTED 202
96 #define HTTP_STATUS_NO_CONTENT 204
97 #define HTTP_STATUS_PARTIAL_CONTENTS 206
99 /* Redirection 3xx. */
100 #define HTTP_STATUS_MULTIPLE_CHOICES 300
101 #define HTTP_STATUS_MOVED_PERMANENTLY 301
102 #define HTTP_STATUS_MOVED_TEMPORARILY 302
103 #define HTTP_STATUS_NOT_MODIFIED 304
105 /* Client error 4xx. */
106 #define HTTP_STATUS_BAD_REQUEST 400
107 #define HTTP_STATUS_UNAUTHORIZED 401
108 #define HTTP_STATUS_FORBIDDEN 403
109 #define HTTP_STATUS_NOT_FOUND 404
111 /* Server errors 5xx. */
112 #define HTTP_STATUS_INTERNAL 500
113 #define HTTP_STATUS_NOT_IMPLEMENTED 501
114 #define HTTP_STATUS_BAD_GATEWAY 502
115 #define HTTP_STATUS_UNAVAILABLE 503
118 /* Parse the HTTP status line, which is of format:
120 HTTP-Version SP Status-Code SP Reason-Phrase
122 The function returns the status-code, or -1 if the status line is
123 malformed. The pointer to reason-phrase is returned in RP. */
125 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
127 /* (the variables must not be named `major' and `minor', because
128 that breaks compilation with SunOS4 cc.) */
129 int mjr, mnr, statcode;
132 *reason_phrase_ptr = NULL;
134 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
135 major version, and Y is minor version. */
136 if (strncmp (line, "HTTP/", 5) != 0)
140 /* Calculate major HTTP version. */
142 for (mjr = 0; ISDIGIT (*line); line++)
143 mjr = 10 * mjr + (*line - '0');
144 if (*line != '.' || p == line)
148 /* Calculate minor HTTP version. */
150 for (mnr = 0; ISDIGIT (*line); line++)
151 mnr = 10 * mnr + (*line - '0');
152 if (*line != ' ' || p == line)
154 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
155 minor version can be safely ignored. */
160 /* Calculate status code. */
161 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
163 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
165 /* Set up the reason phrase pointer. */
167 /* RFC2068 requires SPC here, but we allow the string to finish
168 here, in case no reason-phrase is present. */
172 *reason_phrase_ptr = line;
177 *reason_phrase_ptr = line + 1;
182 /* Functions to be used as arguments to header_process(): */
184 struct http_process_range_closure {
190 /* Parse the `Content-Range' header and extract the information it
191 contains. Returns 1 if successful, -1 otherwise. */
193 http_process_range (const char *hdr, void *arg)
195 struct http_process_range_closure *closure
196 = (struct http_process_range_closure *)arg;
199 /* Certain versions of Nutscape proxy server send out
200 `Content-Length' without "bytes" specifier, which is a breach of
201 RFC2068 (as well as the HTTP/1.1 draft which was current at the
202 time). But hell, I must support it... */
203 if (!strncasecmp (hdr, "bytes", 5))
206 hdr += skip_lws (hdr);
212 for (num = 0; ISDIGIT (*hdr); hdr++)
213 num = 10 * num + (*hdr - '0');
214 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
216 closure->first_byte_pos = num;
218 for (num = 0; ISDIGIT (*hdr); hdr++)
219 num = 10 * num + (*hdr - '0');
220 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
222 closure->last_byte_pos = num;
224 for (num = 0; ISDIGIT (*hdr); hdr++)
225 num = 10 * num + (*hdr - '0');
226 closure->entity_length = num;
230 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
231 Used for `Accept-Ranges'. */
233 http_process_none (const char *hdr, void *arg)
235 int *where = (int *)arg;
237 if (strstr (hdr, "none"))
244 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
246 http_process_type (const char *hdr, void *arg)
248 char **result = (char **)arg;
249 /* Locate P on `;' or the terminating zero, whichever comes first. */
250 const char *p = strchr (hdr, ';');
252 p = hdr + strlen (hdr);
253 while (p > hdr && ISSPACE (*(p - 1)))
255 *result = strdupdelim (hdr, p);
259 /* Check whether the `Connection' header is set to "keep-alive". */
261 http_process_connection (const char *hdr, void *arg)
263 int *flag = (int *)arg;
264 if (!strcasecmp (hdr, "Keep-Alive"))
269 /* Persistent connections. Currently, we cache the most recently used
270 connection as persistent, provided that the HTTP server agrees to
271 make it such. The persistence data is stored in the variables
272 below. Ideally, it would be in a structure, and it should be
273 possible to cache an arbitrary fixed number of these connections.
275 I think the code is quite easy to extend in that direction. */
277 /* Whether a persistent connection is active. */
278 static int pc_active_p;
279 /* Host and port of currently active persistent connection. */
280 static unsigned char pc_last_host[4];
281 static unsigned short pc_last_port;
283 /* File descriptor of the currently active persistent connection. */
284 static int pc_last_fd;
287 /* Whether a ssl handshake has occoured on this connection */
288 static int pc_active_ssl;
289 /* SSL connection of the currently active persistent connection. */
290 static SSL *pc_last_ssl;
291 #endif /* HAVE_SSL */
293 /* Mark the persistent connection as invalid. This is used by the
294 CLOSE_* macros after they forcefully close a registered persistent
295 connection. This does not close the file descriptor -- it is left
296 to the caller to do that. (Maybe it should, though.) */
299 invalidate_persistent (void)
304 #endif /* HAVE_SSL */
305 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
308 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
309 persistent. This will enable someone to use the same connection
310 later. In the context of HTTP, this must be called only AFTER the
311 response has been received and the server has promised that the
312 connection will remain alive.
314 If a previous connection was persistent, it is closed. */
317 register_persistent (const char *host, unsigned short port, int fd
327 if (pc_last_fd == fd)
329 /* The connection FD is already registered. Nothing to
335 /* The old persistent connection is still active; let's
336 close it first. This situation arises whenever a
337 persistent connection exists, but we then connect to a
338 different host, and try to register a persistent
339 connection to that one. */
341 /* The ssl disconnect has to take place before the closing
344 shutdown_ssl(pc_last_ssl);
347 invalidate_persistent ();
351 /* This store_hostaddress may not fail, because it has the results
353 success = store_hostaddress (pc_last_host, host);
360 pc_active_ssl = ssl ? 1 : 0;
362 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
365 /* Return non-zero if a persistent connection is available for
366 connecting to HOST:PORT. */
369 persistent_available_p (const char *host, unsigned short port
375 unsigned char this_host[4];
376 /* First, check whether a persistent connection is active at all. */
379 /* Second, check if the active connection pertains to the correct
380 (HOST, PORT) ordered pair. */
381 if (port != pc_last_port)
384 /* Second, a): check if current connection is (not) ssl, too. This
385 test is unlikely to fail because HTTP and HTTPS typicaly use
386 different ports. Yet it is possible, or so I [Christian
387 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
389 if (ssl != pc_active_ssl)
391 #endif /* HAVE_SSL */
392 if (!store_hostaddress (this_host, host))
394 if (memcmp (pc_last_host, this_host, 4))
396 /* Third: check whether the connection is still open. This is
397 important because most server implement a liberal (short) timeout
398 on persistent connections. Wget can of course always reconnect
399 if the connection doesn't work out, but it's nicer to know in
400 advance. This test is a logical followup of the first test, but
401 is "expensive" and therefore placed at the end of the list. */
402 if (!test_socket_open (pc_last_fd))
404 /* Oops, the socket is no longer open. Now that we know that,
405 let's invalidate the persistent connection before returning
408 invalidate_persistent ();
415 # define SHUTDOWN_SSL(ssl) do { \
417 shutdown_ssl (ssl); \
420 # define SHUTDOWN_SSL(ssl)
423 /* The idea behind these two CLOSE macros is to distinguish between
424 two cases: one when the job we've been doing is finished, and we
425 want to close the connection and leave, and two when something is
426 seriously wrong and we're closing the connection as part of
429 In case of keep_alive, CLOSE_FINISH should leave the connection
430 open, while CLOSE_INVALIDATE should still close it.
432 Note that the semantics of the flag `keep_alive' is "this
433 connection *will* be reused (the server has promised not to close
434 the connection once we're done)", while the semantics of
435 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
436 active, registered connection". */
438 #define CLOSE_FINISH(fd) do { \
441 SHUTDOWN_SSL (ssl); \
443 if (pc_active_p && (fd) == pc_last_fd) \
444 invalidate_persistent (); \
448 #define CLOSE_INVALIDATE(fd) do { \
449 SHUTDOWN_SSL (ssl); \
451 if (pc_active_p && (fd) == pc_last_fd) \
452 invalidate_persistent (); \
457 long len; /* received length */
458 long contlen; /* expected length */
459 long restval; /* the restart value */
460 int res; /* the result of last read */
461 char *newloc; /* new location (redirection) */
462 char *remote_time; /* remote time-stamp string */
463 char *error; /* textual HTTP error */
464 int statcode; /* status code */
465 long dltime; /* time of the download */
468 /* Free the elements of hstat X. */
469 #define FREEHSTAT(x) do \
471 FREE_MAYBE ((x).newloc); \
472 FREE_MAYBE ((x).remote_time); \
473 FREE_MAYBE ((x).error); \
474 (x).newloc = (x).remote_time = (x).error = NULL; \
477 static char *create_authorization_line PARAMS ((const char *, const char *,
478 const char *, const char *,
480 static char *basic_authentication_encode PARAMS ((const char *, const char *,
482 static int known_authentication_scheme_p PARAMS ((const char *));
484 static time_t http_atotm PARAMS ((char *));
486 #define BEGINS_WITH(line, string_constant) \
487 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
488 && (ISSPACE (line[sizeof (string_constant) - 1]) \
489 || !line[sizeof (string_constant) - 1]))
491 /* Retrieve a document through HTTP protocol. It recognizes status
492 code, and correctly handles redirections. It closes the network
493 socket. If it receives an error from the functions below it, it
494 will print it if there is enough information to do so (almost
495 always), returning the error to the caller (i.e. http_loop).
497 Various HTTP parameters are stored to hs. Although it parses the
498 response code correctly, it is not used in a sane way. The caller
501 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
502 and u->proxy->url will be given to the proxy server (bad naming,
505 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
507 char *request, *type, *command, *path;
509 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
510 char *authenticate_h;
514 char *request_keep_alive;
515 int sock, hcount, num_written, all_length, remport, statcode;
516 long contlen, contrange;
520 int auth_tried_already;
523 static SSL_CTX *ssl_ctx = NULL;
525 #endif /* HAVE_SSL */
527 /* Whether this connection will be kept alive after the HTTP request
531 /* Flags that detect the two ways of specifying HTTP keep-alive
533 int http_keep_alive_1, http_keep_alive_2;
535 /* Whether keep-alive should be inhibited. */
536 int inhibit_keep_alive;
539 /* initialize ssl_ctx on first run */
542 err=init_ssl (&ssl_ctx);
547 case SSLERRCTXCREATE:
549 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
553 /* try without certfile */
554 logprintf (LOG_NOTQUIET,
555 _("Failed to load certificates from %s\n"),
558 logprintf (LOG_NOTQUIET,
559 _("Trying without the specified certificate\n"));
562 logprintf (LOG_NOTQUIET,
563 _("Failed to get certificate key from %s\n"),
566 logprintf (LOG_NOTQUIET,
567 _("Trying without the specified certificate\n"));
574 #endif /* HAVE_SSL */
576 if (!(*dt & HEAD_ONLY))
577 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
578 know the local filename so we can save to it. */
579 assert (u->local != NULL);
582 auth_tried_already = 0;
584 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
587 /* We need to come back here when the initial attempt to retrieve
588 without authorization header fails. (Expected to happen at least
589 for the Digest authorization scheme.) */
592 http_keep_alive_1 = http_keep_alive_2 = 0;
594 /* Initialize certain elements of struct http_stat. */
599 hs->remote_time = NULL;
602 /* Which structure to use to retrieve the original URL data. */
608 /* First: establish the connection. */
609 if (inhibit_keep_alive
612 !persistent_available_p (u->host, u->port)
614 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
615 #endif /* HAVE_SSL */
618 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
619 err = make_connection (&sock, u->host, u->port);
623 logputs (LOG_VERBOSE, "\n");
624 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
628 logputs (LOG_VERBOSE, "\n");
629 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
633 logputs (LOG_VERBOSE, "\n");
634 logprintf (LOG_NOTQUIET,
635 _("Connection to %s:%hu refused.\n"), u->host, u->port);
639 logputs (LOG_VERBOSE, "\n");
640 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
645 /* Everything is fine! */
646 logputs (LOG_VERBOSE, _("connected!\n"));
653 if (u->proto == URLHTTPS)
654 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
656 logputs (LOG_VERBOSE, "\n");
657 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
661 #endif /* HAVE_SSL */
665 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
666 /* #### pc_last_fd should be accessed through an accessor
671 #endif /* HAVE_SSL */
672 DEBUGP (("Reusing fd %d.\n", sock));
676 path = u->proxy->url;
680 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
684 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
685 sprintf (referer, "Referer: %s\r\n", ou->referer);
687 if (*dt & SEND_NOCACHE)
688 pragma_h = "Pragma: no-cache\r\n";
693 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
694 /* Gag me! Some servers (e.g. WebSitePro) have been known to
695 respond to the following `Range' format by generating a
696 multipart/x-byte-ranges MIME document! This MIME type was
697 present in an old draft of the byteranges specification.
698 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
699 only if multiple non-overlapping ranges are requested --
700 which Wget never does. */
701 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
706 STRDUP_ALLOCA (useragent, opt.useragent);
709 useragent = (char *)alloca (10 + strlen (version_string));
710 sprintf (useragent, "Wget/%s", version_string);
712 /* Construct the authentication, if userid is present. */
715 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
716 user = user ? user : opt.http_user;
717 passwd = passwd ? passwd : opt.http_passwd;
724 /* We have the username and the password, but haven't tried
725 any authorization yet. Let's see if the "Basic" method
726 works. If not, we'll come back here and construct a
727 proper authorization method with the right challenges.
729 If we didn't employ this kind of logic, every URL that
730 requires authorization would have to be processed twice,
731 which is very suboptimal and generates a bunch of false
732 "unauthorized" errors in the server log.
734 #### But this logic also has a serious problem when used
735 with stronger authentications: we *first* transmit the
736 username and the password in clear text, and *then*
737 attempt a stronger authentication scheme. That cannot be
738 right! We are only fortunate that almost everyone still
739 uses the `Basic' scheme anyway.
741 There should be an option to prevent this from happening,
742 for those who use strong authentication schemes and value
744 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
748 wwwauth = create_authorization_line (authenticate_h, user, passwd,
756 char *proxy_user, *proxy_passwd;
757 /* For normal username and password, URL components override
758 command-line/wgetrc parameters. With proxy authentication,
759 it's the reverse, because proxy URLs are normally the
760 "permanent" ones, so command-line args should take
762 if (opt.proxy_user && opt.proxy_passwd)
764 proxy_user = opt.proxy_user;
765 proxy_passwd = opt.proxy_passwd;
769 proxy_user = u->user;
770 proxy_passwd = u->passwd;
772 /* #### This is junky. Can't the proxy request, say, `Digest'
774 if (proxy_user && proxy_passwd)
775 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
776 "Proxy-Authorization");
781 /* String of the form :PORT. Used only for non-standard ports. */
784 if (remport != (u->proto == URLHTTPS ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT) )
786 if (remport != DEFAULT_HTTP_PORT)
789 port_maybe = (char *)alloca (numdigit (remport) + 2);
790 sprintf (port_maybe, ":%d", remport);
793 if (!inhibit_keep_alive)
794 request_keep_alive = "Connection: Keep-Alive\r\n";
796 request_keep_alive = NULL;
798 /* Allocate the memory for the request. */
799 request = (char *)alloca (strlen (command) + strlen (path)
802 + (port_maybe ? strlen (port_maybe) : 0)
803 + strlen (HTTP_ACCEPT)
804 + (request_keep_alive
805 ? strlen (request_keep_alive) : 0)
806 + (referer ? strlen (referer) : 0)
807 + (wwwauth ? strlen (wwwauth) : 0)
808 + (proxyauth ? strlen (proxyauth) : 0)
809 + (range ? strlen (range) : 0)
811 + (opt.user_header ? strlen (opt.user_header) : 0)
813 /* Construct the request. */
820 command, path, useragent, remhost,
821 port_maybe ? port_maybe : "",
823 request_keep_alive ? request_keep_alive : "",
824 referer ? referer : "",
825 wwwauth ? wwwauth : "",
826 proxyauth ? proxyauth : "",
829 opt.user_header ? opt.user_header : "");
830 DEBUGP (("---request begin---\n%s---request end---\n", request));
831 /* Free the temporary memory. */
832 FREE_MAYBE (wwwauth);
833 FREE_MAYBE (proxyauth);
835 /* Send the request to server. */
837 if (u->proto == URLHTTPS)
838 num_written = ssl_iwrite (ssl, request, strlen (request));
840 #endif /* HAVE_SSL */
841 num_written = iwrite (sock, request, strlen (request));
845 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
847 CLOSE_INVALIDATE (sock);
850 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
851 u->proxy ? "Proxy" : "HTTP");
852 contlen = contrange = -1;
857 /* Before reading anything, initialize the rbuf. */
858 rbuf_initialize (&rbuf, sock);
860 if (u->proto == URLHTTPS)
864 #endif /* HAVE_SSL */
867 /* Header-fetching loop. */
875 /* Get the header. */
876 status = header_get (&rbuf, &hdr,
877 /* Disallow continuations for status line. */
878 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
880 /* Check for errors. */
881 if (status == HG_EOF && *hdr)
883 /* This used to be an unconditional error, but that was
884 somewhat controversial, because of a large number of
885 broken CGI's that happily "forget" to send the second EOL
886 before closing the connection of a HEAD request.
888 So, the deal is to check whether the header is empty
889 (*hdr is zero if it is); if yes, it means that the
890 previous header was fully retrieved, and that -- most
891 probably -- the request is complete. "...be liberal in
892 what you accept." Oh boy. */
893 logputs (LOG_VERBOSE, "\n");
894 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
897 FREE_MAYBE (hs->newloc);
898 FREE_MAYBE (all_headers);
899 CLOSE_INVALIDATE (sock);
902 else if (status == HG_ERROR)
904 logputs (LOG_VERBOSE, "\n");
905 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
909 FREE_MAYBE (hs->newloc);
910 FREE_MAYBE (all_headers);
911 CLOSE_INVALIDATE (sock);
915 /* If the headers are to be saved to a file later, save them to
917 if (opt.save_headers)
919 int lh = strlen (hdr);
920 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
921 memcpy (all_headers + all_length, hdr, lh);
923 all_headers[all_length++] = '\n';
924 all_headers[all_length] = '\0';
927 /* Print the header if requested. */
928 if (opt.server_response && hcount != 1)
929 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
931 /* Check for status line. */
935 /* Parse the first line of server response. */
936 statcode = parse_http_status_line (hdr, &error);
937 hs->statcode = statcode;
938 /* Store the descriptive response. */
939 if (statcode == -1) /* malformed response */
941 /* A common reason for "malformed response" error is the
942 case when no data was actually received. Handle this
945 hs->error = xstrdup (_("No data received"));
947 hs->error = xstrdup (_("Malformed status line"));
952 hs->error = xstrdup (_("(no description)"));
954 hs->error = xstrdup (error);
961 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
966 /* Exit on empty header. */
973 /* Try getting content-length. */
974 if (contlen == -1 && !opt.ignore_length)
975 if (header_process (hdr, "Content-Length", header_extract_number,
978 /* Try getting content-type. */
980 if (header_process (hdr, "Content-Type", http_process_type, &type))
982 /* Try getting location. */
984 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
986 /* Try getting last-modified. */
987 if (!hs->remote_time)
988 if (header_process (hdr, "Last-Modified", header_strdup,
991 /* Try getting www-authentication. */
993 if (header_process (hdr, "WWW-Authenticate", header_strdup,
996 /* Check for accept-ranges header. If it contains the word
997 `none', disable the ranges. */
998 if (*dt & ACCEPTRANGES)
1001 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1004 *dt &= ~ACCEPTRANGES;
1008 /* Try getting content-range. */
1009 if (contrange == -1)
1011 struct http_process_range_closure closure;
1012 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1014 contrange = closure.first_byte_pos;
1018 /* Check for keep-alive related responses. */
1019 if (!inhibit_keep_alive)
1021 /* Check for the `Keep-Alive' header. */
1022 if (!http_keep_alive_1)
1024 if (header_process (hdr, "Keep-Alive", header_exists,
1025 &http_keep_alive_1))
1028 /* Check for `Connection: Keep-Alive'. */
1029 if (!http_keep_alive_2)
1031 if (header_process (hdr, "Connection", http_process_connection,
1032 &http_keep_alive_2))
1040 logputs (LOG_VERBOSE, "\n");
1043 && (http_keep_alive_1 || http_keep_alive_2))
1045 assert (inhibit_keep_alive == 0);
1049 /* The server has promised that it will not close the connection
1050 when we're done. This means that we can register it. */
1052 register_persistent (u->host, u->port, sock);
1054 register_persistent (u->host, u->port, sock, ssl);
1055 #endif /* HAVE_SSL */
1057 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1060 /* Authorization is required. */
1064 CLOSE_FINISH (sock);
1065 if (auth_tried_already)
1067 /* If we have tried it already, then there is not point
1070 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1071 xfree (authenticate_h);
1074 else if (!known_authentication_scheme_p (authenticate_h))
1076 xfree (authenticate_h);
1077 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1080 else if (BEGINS_WITH (authenticate_h, "Basic"))
1082 /* The authentication scheme is basic, the one we try by
1083 default, and it failed. There's no sense in trying
1089 auth_tried_already = 1;
1093 /* We do not need this anymore. */
1096 xfree (authenticate_h);
1097 authenticate_h = NULL;
1100 /* 20x responses are counted among successful by default. */
1101 if (H_20X (statcode))
1104 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1107 /* We don't assume text/html by default. */
1110 if (opt.html_extension && (*dt & TEXTHTML))
1111 /* -E / --html-extension / html_extension = on was specified, and this is a
1112 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1113 already the file's suffix, tack on ".html". */
1115 char* last_period_in_local_filename = strrchr(u->local, '.');
1117 if (last_period_in_local_filename == NULL ||
1118 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1119 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1121 size_t local_filename_len = strlen(u->local);
1123 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1124 strcpy(u->local + local_filename_len, ".html");
1126 *dt |= ADDED_HTML_EXTENSION;
1130 if (contrange == -1)
1132 else if (contrange != hs->restval ||
1133 (H_PARTIAL (statcode) && contrange == -1))
1135 /* This means the whole request was somehow misunderstood by the
1136 server. Bail out. */
1138 FREE_MAYBE (hs->newloc);
1139 FREE_MAYBE (all_headers);
1140 CLOSE_INVALIDATE (sock);
1147 contlen += contrange;
1149 contrange = -1; /* If conent-length was not sent,
1150 content-range will be ignored. */
1152 hs->contlen = contlen;
1154 /* Return if redirected. */
1155 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1157 /* RFC2068 says that in case of the 300 (multiple choices)
1158 response, the server can output a preferred URL through
1159 `Location' header; otherwise, the request should be treated
1160 like GET. So, if the location is set, it will be a
1161 redirection; otherwise, just proceed normally. */
1162 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1166 logprintf (LOG_VERBOSE,
1167 _("Location: %s%s\n"),
1168 hs->newloc ? hs->newloc : _("unspecified"),
1169 hs->newloc ? _(" [following]") : "");
1170 CLOSE_FINISH (sock);
1172 FREE_MAYBE (all_headers);
1178 if ((*dt & RETROKF) && !opt.server_response)
1180 /* No need to print this output if the body won't be
1181 downloaded at all, or if the original server response is
1183 logputs (LOG_VERBOSE, _("Length: "));
1186 logputs (LOG_VERBOSE, legible (contlen));
1187 if (contrange != -1)
1188 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1189 legible (contlen - contrange));
1192 logputs (LOG_VERBOSE,
1193 opt.ignore_length ? _("ignored") : _("unspecified"));
1195 logprintf (LOG_VERBOSE, " [%s]\n", type);
1197 logputs (LOG_VERBOSE, "\n");
1201 type = NULL; /* We don't need it any more. */
1203 /* Return if we have no intention of further downloading. */
1204 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1206 /* In case someone cares to look... */
1210 FREE_MAYBE (all_headers);
1211 CLOSE_FINISH (sock);
1212 return RETRFINISHED;
1215 /* Open the local file. */
1218 mkalldirs (u->local);
1220 rotate_backups (u->local);
1221 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1224 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1225 CLOSE_FINISH (sock);
1226 FREE_MAYBE (all_headers);
1235 /* This will silently fail for streams that don't correspond
1236 to regular files, but that's OK. */
1242 /* #### This confuses the code that checks for file size. There
1243 should be some overhead information. */
1244 if (opt.save_headers)
1245 fwrite (all_headers, 1, all_length, fp);
1247 /* Get the contents of the document. */
1248 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1249 (contlen != -1 ? contlen : 0),
1251 hs->dltime = elapsed_time ();
1253 /* Close or flush the file. We have to be careful to check for
1254 error here. Checking the result of fwrite() is not enough --
1255 errors could go unnoticed! */
1258 flush_res = fclose (fp);
1260 flush_res = fflush (fp);
1261 if (flush_res == EOF)
1264 FREE_MAYBE (all_headers);
1265 CLOSE_FINISH (sock);
1268 return RETRFINISHED;
1271 /* The genuine HTTP loop! This is the part where the retrieval is
1272 retried, and retried, and retried, and... */
1274 http_loop (struct urlinfo *u, char **newloc, int *dt)
1277 int use_ts, got_head = 0; /* time-stamping info */
1278 char *filename_plus_orig_suffix;
1279 char *local_filename = NULL;
1280 char *tms, *suf, *locf, *tmrate;
1282 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1283 long local_size = 0; /* the size of the local file */
1284 size_t filename_len;
1285 struct http_stat hstat; /* HTTP status */
1290 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1291 has_wildcards_p because it would also warn on `?', and we know that
1292 shows up in CGI paths a *lot*. */
1293 if (strchr (u->url, '*'))
1294 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1296 /* Determine the local filename. */
1298 u->local = url_filename (u->proxy ? u->proxy : u);
1300 if (!opt.output_document)
1303 locf = opt.output_document;
1305 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1306 xmalloc() here before EACH return. This is one reason it's better to set
1307 flags that influence flow control and then return once at the end. */
1308 filename_len = strlen(u->local);
1309 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1311 if (opt.noclobber && file_exists_p (u->local))
1313 /* If opt.noclobber is turned on and file already exists, do not
1314 retrieve the file */
1315 logprintf (LOG_VERBOSE, _("\
1316 File `%s' already there, will not retrieve.\n"), u->local);
1317 /* If the file is there, we suppose it's retrieved OK. */
1320 /* #### Bogusness alert. */
1321 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1322 text/html, a harmless lie. */
1323 if (((suf = suffix (u->local)) != NULL)
1324 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1327 xfree (filename_plus_orig_suffix); /* must precede every return! */
1328 /* Another harmless lie: */
1333 if (opt.timestamping)
1335 boolean local_dot_orig_file_exists = FALSE;
1337 if (opt.backup_converted)
1338 /* If -K is specified, we'll act on the assumption that it was specified
1339 last time these files were downloaded as well, and instead of just
1340 comparing local file X against server file X, we'll compare local
1341 file X.orig (if extant, else X) against server file X. If -K
1342 _wasn't_ specified last time, or the server contains files called
1343 *.orig, -N will be back to not operating correctly with -k. */
1345 /* Would a single s[n]printf() call be faster? --dan
1347 It wouldn't. sprintf() is horribly slow. At one point I
1348 profiled Wget, and found that a measurable and
1349 non-negligible amount of time was lost calling sprintf()
1350 in url.c. Replacing sprintf with inline calls to
1351 strcpy() and long_to_string() made a difference.
1353 strcpy(filename_plus_orig_suffix, u->local);
1354 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1356 /* Try to stat() the .orig file. */
1357 if (stat(filename_plus_orig_suffix, &st) == 0)
1359 local_dot_orig_file_exists = TRUE;
1360 local_filename = filename_plus_orig_suffix;
1364 if (!local_dot_orig_file_exists)
1365 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1366 if (stat (u->local, &st) == 0)
1367 local_filename = u->local;
1369 if (local_filename != NULL)
1370 /* There was a local file, so we'll check later to see if the version
1371 the server has is the same version we already have, allowing us to
1376 local_size = st.st_size;
1380 /* Reset the counter. */
1382 *dt = 0 | ACCEPTRANGES;
1386 /* Increment the pass counter. */
1388 sleep_between_retrievals (count);
1389 /* Get the current time string. */
1390 tms = time_str (NULL);
1391 /* Print fetch message, if opt.verbose. */
1394 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1398 sprintf (tmp, _("(try:%2d)"), count);
1399 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1400 tms, hurl, tmp, locf);
1402 ws_changetitle (hurl, 1);
1407 /* Default document type is empty. However, if spider mode is
1408 on or time-stamping is employed, HEAD_ONLY commands is
1409 encoded within *dt. */
1410 if (opt.spider || (use_ts && !got_head))
1414 /* Assume no restarting. */
1416 /* Decide whether or not to restart. */
1417 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1418 && file_exists_p (u->local))
1419 if (stat (u->local, &st) == 0)
1420 hstat.restval = st.st_size;
1421 /* Decide whether to send the no-cache directive. */
1422 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1423 *dt |= SEND_NOCACHE;
1425 *dt &= ~SEND_NOCACHE;
1427 /* Try fetching the document, or at least its head. :-) */
1428 err = gethttp (u, &hstat, dt);
1430 /* It's unfortunate that wget determines the local filename before finding
1431 out the Content-Type of the file. Barring a major restructuring of the
1432 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1433 u->local to tack on ".html". */
1434 if (!opt.output_document)
1437 locf = opt.output_document;
1440 tms = time_str (NULL);
1441 /* Get the new location (with or without the redirection). */
1443 *newloc = xstrdup (hstat.newloc);
1446 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1447 case CONERROR: case READERR: case WRITEFAILED:
1449 /* Non-fatal errors continue executing the loop, which will
1450 bring them to "while" statement at the end, to judge
1451 whether the number of tries was exceeded. */
1453 printwhat (count, opt.ntry);
1456 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1457 case SSLERRCTXCREATE:
1458 /* Fatal errors just return from the function. */
1460 xfree (filename_plus_orig_suffix); /* must precede every return! */
1463 case FWRITEERR: case FOPENERR:
1464 /* Another fatal error. */
1465 logputs (LOG_VERBOSE, "\n");
1466 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1467 u->local, strerror (errno));
1472 /* Another fatal error. */
1473 logputs (LOG_VERBOSE, "\n");
1474 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1476 xfree (filename_plus_orig_suffix); /* must precede every return! */
1480 /* Return the new location to the caller. */
1483 logprintf (LOG_NOTQUIET,
1484 _("ERROR: Redirection (%d) without location.\n"),
1486 xfree (filename_plus_orig_suffix); /* must precede every return! */
1490 xfree (filename_plus_orig_suffix); /* must precede every return! */
1494 /* Deal with you later. */
1497 /* All possibilities should have been exhausted. */
1500 if (!(*dt & RETROKF))
1504 /* #### Ugly ugly ugly! */
1505 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1506 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1509 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1510 tms, hstat.statcode, hstat.error);
1511 logputs (LOG_VERBOSE, "\n");
1513 xfree (filename_plus_orig_suffix); /* must precede every return! */
1517 /* Did we get the time-stamp? */
1520 if (opt.timestamping && !hstat.remote_time)
1522 logputs (LOG_NOTQUIET, _("\
1523 Last-modified header missing -- time-stamps turned off.\n"));
1525 else if (hstat.remote_time)
1527 /* Convert the date-string into struct tm. */
1528 tmr = http_atotm (hstat.remote_time);
1529 if (tmr == (time_t) (-1))
1530 logputs (LOG_VERBOSE, _("\
1531 Last-modified header invalid -- time-stamp ignored.\n"));
1535 /* The time-stamping section. */
1540 use_ts = 0; /* no more time-stamping */
1541 count = 0; /* the retrieve count for HEAD is
1543 if (hstat.remote_time && tmr != (time_t) (-1))
1545 /* Now time-stamping can be used validly. Time-stamping
1546 means that if the sizes of the local and remote file
1547 match, and local file is newer than the remote file,
1548 it will not be retrieved. Otherwise, the normal
1549 download procedure is resumed. */
1551 (hstat.contlen == -1 || local_size == hstat.contlen))
1553 logprintf (LOG_VERBOSE, _("\
1554 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1557 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1560 else if (tml >= tmr)
1561 logprintf (LOG_VERBOSE, _("\
1562 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1564 logputs (LOG_VERBOSE,
1565 _("Remote file is newer, retrieving.\n"));
1570 if ((tmr != (time_t) (-1))
1572 && ((hstat.len == hstat.contlen) ||
1573 ((hstat.res == 0) &&
1574 ((hstat.contlen == -1) ||
1575 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1577 /* #### This code repeats in http.c and ftp.c. Move it to a
1579 const char *fl = NULL;
1580 if (opt.output_document)
1582 if (opt.od_known_regular)
1583 fl = opt.output_document;
1590 /* End of time-stamping section. */
1594 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1595 xfree (filename_plus_orig_suffix); /* must precede every return! */
1599 /* It is now safe to free the remainder of hstat, since the
1600 strings within it will no longer be used. */
1603 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1605 if (hstat.len == hstat.contlen)
1609 logprintf (LOG_VERBOSE,
1610 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1611 tms, tmrate, locf, hstat.len, hstat.contlen);
1612 logprintf (LOG_NONVERBOSE,
1613 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1614 tms, u->url, hstat.len, hstat.contlen, locf, count);
1617 downloaded_increase (hstat.len);
1619 /* Remember that we downloaded the file for later ".orig" code. */
1620 if (*dt & ADDED_HTML_EXTENSION)
1621 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1623 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1625 xfree(filename_plus_orig_suffix); /* must precede every return! */
1628 else if (hstat.res == 0) /* No read error */
1630 if (hstat.contlen == -1) /* We don't know how much we were supposed
1631 to get, so assume we succeeded. */
1635 logprintf (LOG_VERBOSE,
1636 _("%s (%s) - `%s' saved [%ld]\n\n"),
1637 tms, tmrate, locf, hstat.len);
1638 logprintf (LOG_NONVERBOSE,
1639 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1640 tms, u->url, hstat.len, locf, count);
1643 downloaded_increase (hstat.len);
1645 /* Remember that we downloaded the file for later ".orig" code. */
1646 if (*dt & ADDED_HTML_EXTENSION)
1647 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1649 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1651 xfree (filename_plus_orig_suffix); /* must precede every return! */
1654 else if (hstat.len < hstat.contlen) /* meaning we lost the
1655 connection too soon */
1657 logprintf (LOG_VERBOSE,
1658 _("%s (%s) - Connection closed at byte %ld. "),
1659 tms, tmrate, hstat.len);
1660 printwhat (count, opt.ntry);
1663 else if (!opt.kill_longer) /* meaning we got more than expected */
1665 logprintf (LOG_VERBOSE,
1666 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1667 tms, tmrate, locf, hstat.len, hstat.contlen);
1668 logprintf (LOG_NONVERBOSE,
1669 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1670 tms, u->url, hstat.len, hstat.contlen, locf, count);
1672 downloaded_increase (hstat.len);
1674 /* Remember that we downloaded the file for later ".orig" code. */
1675 if (*dt & ADDED_HTML_EXTENSION)
1676 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1678 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1680 xfree (filename_plus_orig_suffix); /* must precede every return! */
1683 else /* the same, but not accepted */
1685 logprintf (LOG_VERBOSE,
1686 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1687 tms, tmrate, hstat.len, hstat.contlen);
1688 printwhat (count, opt.ntry);
1692 else /* now hstat.res can only be -1 */
1694 if (hstat.contlen == -1)
1696 logprintf (LOG_VERBOSE,
1697 _("%s (%s) - Read error at byte %ld (%s)."),
1698 tms, tmrate, hstat.len, strerror (errno));
1699 printwhat (count, opt.ntry);
1702 else /* hstat.res == -1 and contlen is given */
1704 logprintf (LOG_VERBOSE,
1705 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1706 tms, tmrate, hstat.len, hstat.contlen,
1708 printwhat (count, opt.ntry);
1715 while (!opt.ntry || (count < opt.ntry));
1716 xfree (filename_plus_orig_suffix); /* must precede every return! */
1720 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1721 than local timezone (mktime assumes the latter).
1723 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1724 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1726 mktime_from_utc (struct tm *t)
1733 tb = mktime (gmtime (&tl));
1734 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1737 /* Check whether the result of strptime() indicates success.
1738 strptime() returns the pointer to how far it got to in the string.
1739 The processing has been successful if the string is at `GMT' or
1740 `+X', or at the end of the string.
1742 In extended regexp parlance, the function returns 1 if P matches
1743 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1744 strptime()) is considered a failure and 0 is returned. */
1746 check_end (const char *p)
1750 while (ISSPACE (*p))
1753 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1754 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1760 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1761 the three formats RFC2068 allows the HTTP servers to emit --
1762 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1765 We use strptime() to recognize various dates, which makes it a
1766 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1767 allows shortened dates and months, one-digit days, etc.). It also
1768 allows more than one space anywhere where the specs require one SP.
1769 The routine should probably be even more forgiving (as recommended
1770 by RFC2068), but I do not have the time to write one.
1772 Return the computed time_t representation, or -1 if all the
1775 Needless to say, what we *really* need here is something like
1776 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1777 and does not use strptime(). atotm() is to be found in the sources
1778 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1780 http_atotm (char *time_string)
1784 /* Roger Beeman says: "This function dynamically allocates struct tm
1785 t, but does no initialization. The only field that actually
1786 needs initialization is tm_isdst, since the others will be set by
1787 strptime. Since strptime does not set tm_isdst, it will return
1788 the data structure with whatever data was in tm_isdst to begin
1789 with. For those of us in timezones where DST can occur, there
1790 can be a one hour shift depending on the previous contents of the
1791 data area where the data structure is allocated." */
1794 /* Note that under foreign locales Solaris strptime() fails to
1795 recognize English dates, which renders this function useless. I
1796 assume that other non-GNU strptime's are plagued by the same
1797 disease. We solve this by setting only LC_MESSAGES in
1798 i18n_initialize(), instead of LC_ALL.
1800 Another solution could be to temporarily set locale to C, invoke
1801 strptime(), and restore it back. This is slow and dirty,
1802 however, and locale support other than LC_MESSAGES can mess other
1803 things, so I rather chose to stick with just setting LC_MESSAGES.
1805 Also note that none of this is necessary under GNU strptime(),
1806 because it recognizes both international and local dates. */
1808 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1809 it to eat all white space up to (and including) a newline, and
1810 the function fails if there is no newline (!).
1812 Let's hope all strptime() implementations use ` ' to skip *all*
1813 whitespace instead of just one (it works that way on all the
1814 systems I've tested it on). */
1816 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1817 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1818 return mktime_from_utc (&t);
1819 /* RFC850: Thu, 29-Jan-98 22:12:57 */
1820 if (check_end (strptime (time_string, "%a, %d-%b-%y %T", &t)))
1821 return mktime_from_utc (&t);
1822 /* asctime: Thu Jan 29 22:12:57 1998 */
1823 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1824 return mktime_from_utc (&t);
1829 /* Authorization support: We support two authorization schemes:
1831 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1833 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1834 consisting of answering to the server's challenge with the proper
1837 /* How many bytes it will take to store LEN bytes in base64. */
1838 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1840 /* Encode the string S of length LENGTH to base64 format and place it
1841 to STORE. STORE will be 0-terminated, and must point to a writable
1842 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1844 base64_encode (const char *s, char *store, int length)
1846 /* Conversion table. */
1847 static char tbl[64] = {
1848 'A','B','C','D','E','F','G','H',
1849 'I','J','K','L','M','N','O','P',
1850 'Q','R','S','T','U','V','W','X',
1851 'Y','Z','a','b','c','d','e','f',
1852 'g','h','i','j','k','l','m','n',
1853 'o','p','q','r','s','t','u','v',
1854 'w','x','y','z','0','1','2','3',
1855 '4','5','6','7','8','9','+','/'
1858 unsigned char *p = (unsigned char *)store;
1860 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1861 for (i = 0; i < length; i += 3)
1863 *p++ = tbl[s[0] >> 2];
1864 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1865 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1866 *p++ = tbl[s[2] & 0x3f];
1869 /* Pad the result if necessary... */
1870 if (i == length + 1)
1872 else if (i == length + 2)
1873 *(p - 1) = *(p - 2) = '=';
1874 /* ...and zero-terminate it. */
1878 /* Create the authentication header contents for the `Basic' scheme.
1879 This is done by encoding the string `USER:PASS' in base64 and
1880 prepending `HEADER: Basic ' to it. */
1882 basic_authentication_encode (const char *user, const char *passwd,
1885 char *t1, *t2, *res;
1886 int len1 = strlen (user) + 1 + strlen (passwd);
1887 int len2 = BASE64_LENGTH (len1);
1889 t1 = (char *)alloca (len1 + 1);
1890 sprintf (t1, "%s:%s", user, passwd);
1891 t2 = (char *)alloca (1 + len2);
1892 base64_encode (t1, t2, len1);
1893 res = (char *)xmalloc (len2 + 11 + strlen (header));
1894 sprintf (res, "%s: Basic %s\r\n", header, t2);
1900 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
1901 of a field in such a header. If the field is the one specified by
1902 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
1903 digest authorization code), extract its value in the (char*)
1904 variable pointed by RET. Returns negative on a malformed header,
1905 or number of bytes that have been parsed by this call. */
1907 extract_header_attr (const char *au, const char *attr_name, char **ret)
1909 const char *cp, *ep;
1913 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
1915 cp += strlen (attr_name);
1918 cp += skip_lws (cp);
1923 cp += skip_lws (cp);
1928 for (ep = cp; *ep && *ep != '\"'; ep++)
1933 *ret = strdupdelim (cp, ep);
1940 /* Response value needs to be in lowercase, so we cannot use HEXD2ASC
1941 from url.h. See RFC 2069 2.1.2 for the syntax of response-digest. */
1942 #define HEXD2asc(x) (((x) < 10) ? ((x) + '0') : ((x) - 10 + 'a'))
1944 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
1945 an array of 16 bytes containing the hash keys, and BUF should be a
1946 buffer of 33 writable characters (32 for hex digits plus one for
1947 zero termination). */
1949 dump_hash (unsigned char *buf, const unsigned char *hash)
1953 for (i = 0; i < MD5_HASHLEN; i++, hash++)
1955 *buf++ = HEXD2asc (*hash >> 4);
1956 *buf++ = HEXD2asc (*hash & 0xf);
1961 /* Take the line apart to find the challenge, and compose a digest
1962 authorization header. See RFC2069 section 2.1.2. */
1964 digest_authentication_encode (const char *au, const char *user,
1965 const char *passwd, const char *method,
1968 static char *realm, *opaque, *nonce;
1973 { "realm", &realm },
1974 { "opaque", &opaque },
1979 realm = opaque = nonce = NULL;
1981 au += 6; /* skip over `Digest' */
1986 au += skip_lws (au);
1987 for (i = 0; i < ARRAY_SIZE (options); i++)
1989 int skip = extract_header_attr (au, options[i].name,
1990 options[i].variable);
1994 FREE_MAYBE (opaque);
2004 if (i == ARRAY_SIZE (options))
2006 while (*au && *au != '=')
2010 au += skip_lws (au);
2014 while (*au && *au != '\"')
2021 while (*au && *au != ',')
2026 if (!realm || !nonce || !user || !passwd || !path || !method)
2029 FREE_MAYBE (opaque);
2034 /* Calculate the digest value. */
2037 unsigned char hash[MD5_HASHLEN];
2038 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2039 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2041 /* A1BUF = H(user ":" realm ":" password) */
2042 md5_init_ctx (&ctx);
2043 md5_process_bytes (user, strlen (user), &ctx);
2044 md5_process_bytes (":", 1, &ctx);
2045 md5_process_bytes (realm, strlen (realm), &ctx);
2046 md5_process_bytes (":", 1, &ctx);
2047 md5_process_bytes (passwd, strlen (passwd), &ctx);
2048 md5_finish_ctx (&ctx, hash);
2049 dump_hash (a1buf, hash);
2051 /* A2BUF = H(method ":" path) */
2052 md5_init_ctx (&ctx);
2053 md5_process_bytes (method, strlen (method), &ctx);
2054 md5_process_bytes (":", 1, &ctx);
2055 md5_process_bytes (path, strlen (path), &ctx);
2056 md5_finish_ctx (&ctx, hash);
2057 dump_hash (a2buf, hash);
2059 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2060 md5_init_ctx (&ctx);
2061 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2062 md5_process_bytes (":", 1, &ctx);
2063 md5_process_bytes (nonce, strlen (nonce), &ctx);
2064 md5_process_bytes (":", 1, &ctx);
2065 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2066 md5_finish_ctx (&ctx, hash);
2067 dump_hash (response_digest, hash);
2069 res = (char*) xmalloc (strlen (user)
2074 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2075 + (opaque ? strlen (opaque) : 0)
2077 sprintf (res, "Authorization: Digest \
2078 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2079 user, realm, nonce, path, response_digest);
2082 char *p = res + strlen (res);
2083 strcat (p, ", opaque=\"");
2087 strcat (res, "\r\n");
2091 #endif /* USE_DIGEST */
2094 #define BEGINS_WITH(line, string_constant) \
2095 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2096 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2097 || !line[sizeof (string_constant) - 1]))
2100 known_authentication_scheme_p (const char *au)
2102 return BEGINS_WITH (au, "Basic")
2103 || BEGINS_WITH (au, "Digest")
2104 || BEGINS_WITH (au, "NTLM");
2109 /* Create the HTTP authorization request header. When the
2110 `WWW-Authenticate' response header is seen, according to the
2111 authorization scheme specified in that header (`Basic' and `Digest'
2112 are supported by the current implementation), produce an
2113 appropriate HTTP authorization request header. */
2115 create_authorization_line (const char *au, const char *user,
2116 const char *passwd, const char *method,
2119 char *wwwauth = NULL;
2121 if (!strncasecmp (au, "Basic", 5))
2122 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2123 if (!strncasecmp (au, "NTLM", 4))
2124 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2126 else if (!strncasecmp (au, "Digest", 6))
2127 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2128 #endif /* USE_DIGEST */