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 */
467 /* Free the elements of hstat X. */
468 #define FREEHSTAT(x) do \
470 FREE_MAYBE ((x).newloc); \
471 FREE_MAYBE ((x).remote_time); \
472 FREE_MAYBE ((x).error); \
473 (x).newloc = (x).remote_time = (x).error = NULL; \
476 static char *create_authorization_line PARAMS ((const char *, const char *,
477 const char *, const char *,
479 static char *basic_authentication_encode PARAMS ((const char *, const char *,
481 static int known_authentication_scheme_p PARAMS ((const char *));
483 static time_t http_atotm PARAMS ((char *));
485 #define BEGINS_WITH(line, string_constant) \
486 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
487 && (ISSPACE (line[sizeof (string_constant) - 1]) \
488 || !line[sizeof (string_constant) - 1]))
490 /* Retrieve a document through HTTP protocol. It recognizes status
491 code, and correctly handles redirections. It closes the network
492 socket. If it receives an error from the functions below it, it
493 will print it if there is enough information to do so (almost
494 always), returning the error to the caller (i.e. http_loop).
496 Various HTTP parameters are stored to hs. Although it parses the
497 response code correctly, it is not used in a sane way. The caller
500 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
501 and u->proxy->url will be given to the proxy server (bad naming,
504 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
506 char *request, *type, *command, *path;
508 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
509 char *authenticate_h;
513 char *request_keep_alive;
514 int sock, hcount, num_written, all_length, remport, statcode;
515 long contlen, contrange;
519 int auth_tried_already;
522 static SSL_CTX *ssl_ctx = NULL;
524 #endif /* HAVE_SSL */
526 /* Whether this connection will be kept alive after the HTTP request
530 /* Flags that detect the two ways of specifying HTTP keep-alive
532 int http_keep_alive_1, http_keep_alive_2;
534 /* Whether keep-alive should be inhibited. */
535 int inhibit_keep_alive;
538 /* initialize ssl_ctx on first run */
541 err=init_ssl (&ssl_ctx);
546 case SSLERRCTXCREATE:
548 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
552 /* try without certfile */
553 logprintf (LOG_NOTQUIET,
554 _("Failed to load certificates from %s\n"),
557 logprintf (LOG_NOTQUIET,
558 _("Trying without the specified certificate\n"));
561 logprintf (LOG_NOTQUIET,
562 _("Failed to get certificate key from %s\n"),
565 logprintf (LOG_NOTQUIET,
566 _("Trying without the specified certificate\n"));
573 #endif /* HAVE_SSL */
575 if (!(*dt & HEAD_ONLY))
576 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
577 know the local filename so we can save to it. */
578 assert (u->local != NULL);
581 auth_tried_already = 0;
583 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
586 /* We need to come back here when the initial attempt to retrieve
587 without authorization header fails. (Expected to happen at least
588 for the Digest authorization scheme.) */
591 http_keep_alive_1 = http_keep_alive_2 = 0;
593 /* Initialize certain elements of struct http_stat. */
598 hs->remote_time = NULL;
601 /* Which structure to use to retrieve the original URL data. */
607 /* First: establish the connection. */
608 if (inhibit_keep_alive
611 !persistent_available_p (u->host, u->port)
613 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
614 #endif /* HAVE_SSL */
617 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
618 err = make_connection (&sock, u->host, u->port);
622 logputs (LOG_VERBOSE, "\n");
623 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
627 logputs (LOG_VERBOSE, "\n");
628 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
632 logputs (LOG_VERBOSE, "\n");
633 logprintf (LOG_NOTQUIET,
634 _("Connection to %s:%hu refused.\n"), u->host, u->port);
638 logputs (LOG_VERBOSE, "\n");
639 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
644 /* Everything is fine! */
645 logputs (LOG_VERBOSE, _("connected!\n"));
652 if (u->proto == URLHTTPS)
653 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
655 logputs (LOG_VERBOSE, "\n");
656 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
660 #endif /* HAVE_SSL */
664 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
665 /* #### pc_last_fd should be accessed through an accessor
670 #endif /* HAVE_SSL */
671 DEBUGP (("Reusing fd %d.\n", sock));
675 path = u->proxy->url;
679 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
683 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
684 sprintf (referer, "Referer: %s\r\n", ou->referer);
686 if (*dt & SEND_NOCACHE)
687 pragma_h = "Pragma: no-cache\r\n";
692 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
693 /* Gag me! Some servers (e.g. WebSitePro) have been known to
694 respond to the following `Range' format by generating a
695 multipart/x-byte-ranges MIME document! This MIME type was
696 present in an old draft of the byteranges specification.
697 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
698 only if multiple non-overlapping ranges are requested --
699 which Wget never does. */
700 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
705 STRDUP_ALLOCA (useragent, opt.useragent);
708 useragent = (char *)alloca (10 + strlen (version_string));
709 sprintf (useragent, "Wget/%s", version_string);
711 /* Construct the authentication, if userid is present. */
714 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
715 user = user ? user : opt.http_user;
716 passwd = passwd ? passwd : opt.http_passwd;
723 /* We have the username and the password, but haven't tried
724 any authorization yet. Let's see if the "Basic" method
725 works. If not, we'll come back here and construct a
726 proper authorization method with the right challenges.
728 If we didn't employ this kind of logic, every URL that
729 requires authorization would have to be processed twice,
730 which is very suboptimal and generates a bunch of false
731 "unauthorized" errors in the server log.
733 #### But this logic also has a serious problem when used
734 with stronger authentications: we *first* transmit the
735 username and the password in clear text, and *then*
736 attempt a stronger authentication scheme. That cannot be
737 right! We are only fortunate that almost everyone still
738 uses the `Basic' scheme anyway.
740 There should be an option to prevent this from happening,
741 for those who use strong authentication schemes and value
743 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
747 wwwauth = create_authorization_line (authenticate_h, user, passwd,
755 char *proxy_user, *proxy_passwd;
756 /* For normal username and password, URL components override
757 command-line/wgetrc parameters. With proxy authentication,
758 it's the reverse, because proxy URLs are normally the
759 "permanent" ones, so command-line args should take
761 if (opt.proxy_user && opt.proxy_passwd)
763 proxy_user = opt.proxy_user;
764 proxy_passwd = opt.proxy_passwd;
768 proxy_user = u->user;
769 proxy_passwd = u->passwd;
771 /* #### This is junky. Can't the proxy request, say, `Digest'
773 if (proxy_user && proxy_passwd)
774 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
775 "Proxy-Authorization");
780 /* String of the form :PORT. Used only for non-standard ports. */
783 if (remport != (u->proto == URLHTTPS ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT) )
785 if (remport != DEFAULT_HTTP_PORT)
788 port_maybe = (char *)alloca (numdigit (remport) + 2);
789 sprintf (port_maybe, ":%d", remport);
792 if (!inhibit_keep_alive)
793 request_keep_alive = "Connection: Keep-Alive\r\n";
795 request_keep_alive = NULL;
797 /* Allocate the memory for the request. */
798 request = (char *)alloca (strlen (command) + strlen (path)
801 + (port_maybe ? strlen (port_maybe) : 0)
802 + strlen (HTTP_ACCEPT)
803 + (request_keep_alive
804 ? strlen (request_keep_alive) : 0)
805 + (referer ? strlen (referer) : 0)
806 + (wwwauth ? strlen (wwwauth) : 0)
807 + (proxyauth ? strlen (proxyauth) : 0)
808 + (range ? strlen (range) : 0)
810 + (opt.user_header ? strlen (opt.user_header) : 0)
812 /* Construct the request. */
819 command, path, useragent, remhost,
820 port_maybe ? port_maybe : "",
822 request_keep_alive ? request_keep_alive : "",
823 referer ? referer : "",
824 wwwauth ? wwwauth : "",
825 proxyauth ? proxyauth : "",
828 opt.user_header ? opt.user_header : "");
829 DEBUGP (("---request begin---\n%s---request end---\n", request));
830 /* Free the temporary memory. */
831 FREE_MAYBE (wwwauth);
832 FREE_MAYBE (proxyauth);
834 /* Send the request to server. */
836 if (u->proto == URLHTTPS)
837 num_written = ssl_iwrite (ssl, request, strlen (request));
839 #endif /* HAVE_SSL */
840 num_written = iwrite (sock, request, strlen (request));
844 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
846 CLOSE_INVALIDATE (sock);
849 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
850 u->proxy ? "Proxy" : "HTTP");
851 contlen = contrange = -1;
856 /* Before reading anything, initialize the rbuf. */
857 rbuf_initialize (&rbuf, sock);
859 if (u->proto == URLHTTPS)
863 #endif /* HAVE_SSL */
866 /* Header-fetching loop. */
874 /* Get the header. */
875 status = header_get (&rbuf, &hdr,
876 /* Disallow continuations for status line. */
877 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
879 /* Check for errors. */
880 if (status == HG_EOF && *hdr)
882 /* This used to be an unconditional error, but that was
883 somewhat controversial, because of a large number of
884 broken CGI's that happily "forget" to send the second EOL
885 before closing the connection of a HEAD request.
887 So, the deal is to check whether the header is empty
888 (*hdr is zero if it is); if yes, it means that the
889 previous header was fully retrieved, and that -- most
890 probably -- the request is complete. "...be liberal in
891 what you accept." Oh boy. */
892 logputs (LOG_VERBOSE, "\n");
893 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
896 FREE_MAYBE (hs->newloc);
897 FREE_MAYBE (all_headers);
898 CLOSE_INVALIDATE (sock);
901 else if (status == HG_ERROR)
903 logputs (LOG_VERBOSE, "\n");
904 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
908 FREE_MAYBE (hs->newloc);
909 FREE_MAYBE (all_headers);
910 CLOSE_INVALIDATE (sock);
914 /* If the headers are to be saved to a file later, save them to
916 if (opt.save_headers)
918 int lh = strlen (hdr);
919 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
920 memcpy (all_headers + all_length, hdr, lh);
922 all_headers[all_length++] = '\n';
923 all_headers[all_length] = '\0';
926 /* Print the header if requested. */
927 if (opt.server_response && hcount != 1)
928 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
930 /* Check for status line. */
934 /* Parse the first line of server response. */
935 statcode = parse_http_status_line (hdr, &error);
936 hs->statcode = statcode;
937 /* Store the descriptive response. */
938 if (statcode == -1) /* malformed response */
940 /* A common reason for "malformed response" error is the
941 case when no data was actually received. Handle this
944 hs->error = xstrdup (_("No data received"));
946 hs->error = xstrdup (_("Malformed status line"));
951 hs->error = xstrdup (_("(no description)"));
953 hs->error = xstrdup (error);
960 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
965 /* Exit on empty header. */
972 /* Try getting content-length. */
973 if (contlen == -1 && !opt.ignore_length)
974 if (header_process (hdr, "Content-Length", header_extract_number,
977 /* Try getting content-type. */
979 if (header_process (hdr, "Content-Type", http_process_type, &type))
981 /* Try getting location. */
983 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
985 /* Try getting last-modified. */
986 if (!hs->remote_time)
987 if (header_process (hdr, "Last-Modified", header_strdup,
990 /* Try getting www-authentication. */
992 if (header_process (hdr, "WWW-Authenticate", header_strdup,
995 /* Check for accept-ranges header. If it contains the word
996 `none', disable the ranges. */
997 if (*dt & ACCEPTRANGES)
1000 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1003 *dt &= ~ACCEPTRANGES;
1007 /* Try getting content-range. */
1008 if (contrange == -1)
1010 struct http_process_range_closure closure;
1011 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1013 contrange = closure.first_byte_pos;
1017 /* Check for keep-alive related responses. */
1018 if (!inhibit_keep_alive)
1020 /* Check for the `Keep-Alive' header. */
1021 if (!http_keep_alive_1)
1023 if (header_process (hdr, "Keep-Alive", header_exists,
1024 &http_keep_alive_1))
1027 /* Check for `Connection: Keep-Alive'. */
1028 if (!http_keep_alive_2)
1030 if (header_process (hdr, "Connection", http_process_connection,
1031 &http_keep_alive_2))
1039 logputs (LOG_VERBOSE, "\n");
1042 && (http_keep_alive_1 || http_keep_alive_2))
1044 assert (inhibit_keep_alive == 0);
1048 /* The server has promised that it will not close the connection
1049 when we're done. This means that we can register it. */
1051 register_persistent (u->host, u->port, sock);
1053 register_persistent (u->host, u->port, sock, ssl);
1054 #endif /* HAVE_SSL */
1056 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1059 /* Authorization is required. */
1063 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1064 might be more bytes in the body. */
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_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1171 might be more bytes in the body. */
1173 FREE_MAYBE (all_headers);
1179 if ((*dt & RETROKF) && !opt.server_response)
1181 /* No need to print this output if the body won't be
1182 downloaded at all, or if the original server response is
1184 logputs (LOG_VERBOSE, _("Length: "));
1187 logputs (LOG_VERBOSE, legible (contlen));
1188 if (contrange != -1)
1189 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1190 legible (contlen - contrange));
1193 logputs (LOG_VERBOSE,
1194 opt.ignore_length ? _("ignored") : _("unspecified"));
1196 logprintf (LOG_VERBOSE, " [%s]\n", type);
1198 logputs (LOG_VERBOSE, "\n");
1202 type = NULL; /* We don't need it any more. */
1204 /* Return if we have no intention of further downloading. */
1205 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1207 /* In case someone cares to look... */
1211 FREE_MAYBE (all_headers);
1212 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1213 might be more bytes in the body. */
1214 return RETRFINISHED;
1217 /* Open the local file. */
1220 mkalldirs (u->local);
1222 rotate_backups (u->local);
1223 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1226 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1227 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1228 might be more bytes in the body. */
1229 FREE_MAYBE (all_headers);
1238 /* This will silently fail for streams that don't correspond
1239 to regular files, but that's OK. */
1245 /* #### This confuses the code that checks for file size. There
1246 should be some overhead information. */
1247 if (opt.save_headers)
1248 fwrite (all_headers, 1, all_length, fp);
1250 /* Get the contents of the document. */
1251 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1252 (contlen != -1 ? contlen : 0),
1254 hs->dltime = elapsed_time ();
1256 /* Close or flush the file. We have to be careful to check for
1257 error here. Checking the result of fwrite() is not enough --
1258 errors could go unnoticed! */
1261 flush_res = fclose (fp);
1263 flush_res = fflush (fp);
1264 if (flush_res == EOF)
1267 FREE_MAYBE (all_headers);
1268 CLOSE_FINISH (sock);
1271 return RETRFINISHED;
1274 /* The genuine HTTP loop! This is the part where the retrieval is
1275 retried, and retried, and retried, and... */
1277 http_loop (struct urlinfo *u, char **newloc, int *dt)
1280 int use_ts, got_head = 0; /* time-stamping info */
1281 char *filename_plus_orig_suffix;
1282 char *local_filename = NULL;
1283 char *tms, *suf, *locf, *tmrate;
1285 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1286 long local_size = 0; /* the size of the local file */
1287 size_t filename_len;
1288 struct http_stat hstat; /* HTTP status */
1293 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1294 has_wildcards_p because it would also warn on `?', and we know that
1295 shows up in CGI paths a *lot*. */
1296 if (strchr (u->url, '*'))
1297 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1299 /* Determine the local filename. */
1301 u->local = url_filename (u->proxy ? u->proxy : u);
1303 if (!opt.output_document)
1306 locf = opt.output_document;
1308 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1309 xmalloc() here before EACH return. This is one reason it's better to set
1310 flags that influence flow control and then return once at the end. */
1311 filename_len = strlen(u->local);
1312 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1314 if (opt.noclobber && file_exists_p (u->local))
1316 /* If opt.noclobber is turned on and file already exists, do not
1317 retrieve the file */
1318 logprintf (LOG_VERBOSE, _("\
1319 File `%s' already there, will not retrieve.\n"), u->local);
1320 /* If the file is there, we suppose it's retrieved OK. */
1323 /* #### Bogusness alert. */
1324 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1325 text/html, a harmless lie. */
1326 if (((suf = suffix (u->local)) != NULL)
1327 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1330 xfree (filename_plus_orig_suffix); /* must precede every return! */
1331 /* Another harmless lie: */
1336 if (opt.timestamping)
1338 boolean local_dot_orig_file_exists = FALSE;
1340 if (opt.backup_converted)
1341 /* If -K is specified, we'll act on the assumption that it was specified
1342 last time these files were downloaded as well, and instead of just
1343 comparing local file X against server file X, we'll compare local
1344 file X.orig (if extant, else X) against server file X. If -K
1345 _wasn't_ specified last time, or the server contains files called
1346 *.orig, -N will be back to not operating correctly with -k. */
1348 /* Would a single s[n]printf() call be faster? --dan
1350 It wouldn't. sprintf() is horribly slow. At one point I
1351 profiled Wget, and found that a measurable and
1352 non-negligible amount of time was lost calling sprintf()
1353 in url.c. Replacing sprintf with inline calls to
1354 strcpy() and long_to_string() made a difference.
1356 strcpy(filename_plus_orig_suffix, u->local);
1357 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1359 /* Try to stat() the .orig file. */
1360 if (stat(filename_plus_orig_suffix, &st) == 0)
1362 local_dot_orig_file_exists = TRUE;
1363 local_filename = filename_plus_orig_suffix;
1367 if (!local_dot_orig_file_exists)
1368 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1369 if (stat (u->local, &st) == 0)
1370 local_filename = u->local;
1372 if (local_filename != NULL)
1373 /* There was a local file, so we'll check later to see if the version
1374 the server has is the same version we already have, allowing us to
1379 local_size = st.st_size;
1383 /* Reset the counter. */
1385 *dt = 0 | ACCEPTRANGES;
1389 /* Increment the pass counter. */
1391 sleep_between_retrievals (count);
1392 /* Get the current time string. */
1393 tms = time_str (NULL);
1394 /* Print fetch message, if opt.verbose. */
1397 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1401 sprintf (tmp, _("(try:%2d)"), count);
1402 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1403 tms, hurl, tmp, locf);
1405 ws_changetitle (hurl, 1);
1410 /* Default document type is empty. However, if spider mode is
1411 on or time-stamping is employed, HEAD_ONLY commands is
1412 encoded within *dt. */
1413 if (opt.spider || (use_ts && !got_head))
1417 /* Assume no restarting. */
1419 /* Decide whether or not to restart. */
1420 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1421 && file_exists_p (u->local))
1422 if (stat (u->local, &st) == 0)
1423 hstat.restval = st.st_size;
1424 /* Decide whether to send the no-cache directive. */
1425 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1426 *dt |= SEND_NOCACHE;
1428 *dt &= ~SEND_NOCACHE;
1430 /* Try fetching the document, or at least its head. :-) */
1431 err = gethttp (u, &hstat, dt);
1433 /* It's unfortunate that wget determines the local filename before finding
1434 out the Content-Type of the file. Barring a major restructuring of the
1435 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1436 u->local to tack on ".html". */
1437 if (!opt.output_document)
1440 locf = opt.output_document;
1443 tms = time_str (NULL);
1444 /* Get the new location (with or without the redirection). */
1446 *newloc = xstrdup (hstat.newloc);
1449 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1450 case CONERROR: case READERR: case WRITEFAILED:
1452 /* Non-fatal errors continue executing the loop, which will
1453 bring them to "while" statement at the end, to judge
1454 whether the number of tries was exceeded. */
1456 printwhat (count, opt.ntry);
1459 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1460 case SSLERRCTXCREATE:
1461 /* Fatal errors just return from the function. */
1463 xfree (filename_plus_orig_suffix); /* must precede every return! */
1466 case FWRITEERR: case FOPENERR:
1467 /* Another fatal error. */
1468 logputs (LOG_VERBOSE, "\n");
1469 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1470 u->local, strerror (errno));
1475 /* Another fatal error. */
1476 logputs (LOG_VERBOSE, "\n");
1477 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1479 xfree (filename_plus_orig_suffix); /* must precede every return! */
1483 /* Return the new location to the caller. */
1486 logprintf (LOG_NOTQUIET,
1487 _("ERROR: Redirection (%d) without location.\n"),
1489 xfree (filename_plus_orig_suffix); /* must precede every return! */
1493 xfree (filename_plus_orig_suffix); /* must precede every return! */
1497 /* Deal with you later. */
1500 /* All possibilities should have been exhausted. */
1503 if (!(*dt & RETROKF))
1507 /* #### Ugly ugly ugly! */
1508 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1509 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1512 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1513 tms, hstat.statcode, hstat.error);
1514 logputs (LOG_VERBOSE, "\n");
1516 xfree (filename_plus_orig_suffix); /* must precede every return! */
1520 /* Did we get the time-stamp? */
1523 if (opt.timestamping && !hstat.remote_time)
1525 logputs (LOG_NOTQUIET, _("\
1526 Last-modified header missing -- time-stamps turned off.\n"));
1528 else if (hstat.remote_time)
1530 /* Convert the date-string into struct tm. */
1531 tmr = http_atotm (hstat.remote_time);
1532 if (tmr == (time_t) (-1))
1533 logputs (LOG_VERBOSE, _("\
1534 Last-modified header invalid -- time-stamp ignored.\n"));
1538 /* The time-stamping section. */
1543 use_ts = 0; /* no more time-stamping */
1544 count = 0; /* the retrieve count for HEAD is
1546 if (hstat.remote_time && tmr != (time_t) (-1))
1548 /* Now time-stamping can be used validly. Time-stamping
1549 means that if the sizes of the local and remote file
1550 match, and local file is newer than the remote file,
1551 it will not be retrieved. Otherwise, the normal
1552 download procedure is resumed. */
1554 (hstat.contlen == -1 || local_size == hstat.contlen))
1556 logprintf (LOG_VERBOSE, _("\
1557 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1560 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1563 else if (tml >= tmr)
1564 logprintf (LOG_VERBOSE, _("\
1565 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1567 logputs (LOG_VERBOSE,
1568 _("Remote file is newer, retrieving.\n"));
1573 if ((tmr != (time_t) (-1))
1575 && ((hstat.len == hstat.contlen) ||
1576 ((hstat.res == 0) &&
1577 ((hstat.contlen == -1) ||
1578 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1580 /* #### This code repeats in http.c and ftp.c. Move it to a
1582 const char *fl = NULL;
1583 if (opt.output_document)
1585 if (opt.od_known_regular)
1586 fl = opt.output_document;
1593 /* End of time-stamping section. */
1597 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1598 xfree (filename_plus_orig_suffix); /* must precede every return! */
1602 /* It is now safe to free the remainder of hstat, since the
1603 strings within it will no longer be used. */
1606 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1608 if (hstat.len == hstat.contlen)
1612 logprintf (LOG_VERBOSE,
1613 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1614 tms, tmrate, locf, hstat.len, hstat.contlen);
1615 logprintf (LOG_NONVERBOSE,
1616 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1617 tms, u->url, hstat.len, hstat.contlen, locf, count);
1620 downloaded_increase (hstat.len);
1622 /* Remember that we downloaded the file for later ".orig" code. */
1623 if (*dt & ADDED_HTML_EXTENSION)
1624 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1626 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1628 xfree(filename_plus_orig_suffix); /* must precede every return! */
1631 else if (hstat.res == 0) /* No read error */
1633 if (hstat.contlen == -1) /* We don't know how much we were supposed
1634 to get, so assume we succeeded. */
1638 logprintf (LOG_VERBOSE,
1639 _("%s (%s) - `%s' saved [%ld]\n\n"),
1640 tms, tmrate, locf, hstat.len);
1641 logprintf (LOG_NONVERBOSE,
1642 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1643 tms, u->url, hstat.len, locf, count);
1646 downloaded_increase (hstat.len);
1648 /* Remember that we downloaded the file for later ".orig" code. */
1649 if (*dt & ADDED_HTML_EXTENSION)
1650 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1652 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1654 xfree (filename_plus_orig_suffix); /* must precede every return! */
1657 else if (hstat.len < hstat.contlen) /* meaning we lost the
1658 connection too soon */
1660 logprintf (LOG_VERBOSE,
1661 _("%s (%s) - Connection closed at byte %ld. "),
1662 tms, tmrate, hstat.len);
1663 printwhat (count, opt.ntry);
1666 else if (!opt.kill_longer) /* meaning we got more than expected */
1668 logprintf (LOG_VERBOSE,
1669 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1670 tms, tmrate, locf, hstat.len, hstat.contlen);
1671 logprintf (LOG_NONVERBOSE,
1672 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1673 tms, u->url, hstat.len, hstat.contlen, locf, count);
1675 downloaded_increase (hstat.len);
1677 /* Remember that we downloaded the file for later ".orig" code. */
1678 if (*dt & ADDED_HTML_EXTENSION)
1679 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1681 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1683 xfree (filename_plus_orig_suffix); /* must precede every return! */
1686 else /* the same, but not accepted */
1688 logprintf (LOG_VERBOSE,
1689 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1690 tms, tmrate, hstat.len, hstat.contlen);
1691 printwhat (count, opt.ntry);
1695 else /* now hstat.res can only be -1 */
1697 if (hstat.contlen == -1)
1699 logprintf (LOG_VERBOSE,
1700 _("%s (%s) - Read error at byte %ld (%s)."),
1701 tms, tmrate, hstat.len, strerror (errno));
1702 printwhat (count, opt.ntry);
1705 else /* hstat.res == -1 and contlen is given */
1707 logprintf (LOG_VERBOSE,
1708 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1709 tms, tmrate, hstat.len, hstat.contlen,
1711 printwhat (count, opt.ntry);
1718 while (!opt.ntry || (count < opt.ntry));
1719 xfree (filename_plus_orig_suffix); /* must precede every return! */
1723 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1724 than local timezone (mktime assumes the latter).
1726 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1727 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1729 mktime_from_utc (struct tm *t)
1736 tb = mktime (gmtime (&tl));
1737 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1740 /* Check whether the result of strptime() indicates success.
1741 strptime() returns the pointer to how far it got to in the string.
1742 The processing has been successful if the string is at `GMT' or
1743 `+X', or at the end of the string.
1745 In extended regexp parlance, the function returns 1 if P matches
1746 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1747 strptime()) is considered a failure and 0 is returned. */
1749 check_end (const char *p)
1753 while (ISSPACE (*p))
1756 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1757 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1763 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1764 the three formats RFC2068 allows the HTTP servers to emit --
1765 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1768 We use strptime() to recognize various dates, which makes it a
1769 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1770 allows shortened dates and months, one-digit days, etc.). It also
1771 allows more than one space anywhere where the specs require one SP.
1772 The routine should probably be even more forgiving (as recommended
1773 by RFC2068), but I do not have the time to write one.
1775 Return the computed time_t representation, or -1 if all the
1778 Needless to say, what we *really* need here is something like
1779 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1780 and does not use strptime(). atotm() is to be found in the sources
1781 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1783 http_atotm (char *time_string)
1787 /* Roger Beeman says: "This function dynamically allocates struct tm
1788 t, but does no initialization. The only field that actually
1789 needs initialization is tm_isdst, since the others will be set by
1790 strptime. Since strptime does not set tm_isdst, it will return
1791 the data structure with whatever data was in tm_isdst to begin
1792 with. For those of us in timezones where DST can occur, there
1793 can be a one hour shift depending on the previous contents of the
1794 data area where the data structure is allocated." */
1797 /* Note that under foreign locales Solaris strptime() fails to
1798 recognize English dates, which renders this function useless. I
1799 assume that other non-GNU strptime's are plagued by the same
1800 disease. We solve this by setting only LC_MESSAGES in
1801 i18n_initialize(), instead of LC_ALL.
1803 Another solution could be to temporarily set locale to C, invoke
1804 strptime(), and restore it back. This is slow and dirty,
1805 however, and locale support other than LC_MESSAGES can mess other
1806 things, so I rather chose to stick with just setting LC_MESSAGES.
1808 Also note that none of this is necessary under GNU strptime(),
1809 because it recognizes both international and local dates. */
1811 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1812 it to eat all white space up to (and including) a newline, and
1813 the function fails if there is no newline (!).
1815 Let's hope all strptime() implementations use ` ' to skip *all*
1816 whitespace instead of just one (it works that way on all the
1817 systems I've tested it on). */
1819 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1820 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1821 return mktime_from_utc (&t);
1822 /* RFC850: Thu, 29-Jan-98 22:12:57 */
1823 if (check_end (strptime (time_string, "%a, %d-%b-%y %T", &t)))
1824 return mktime_from_utc (&t);
1825 /* asctime: Thu Jan 29 22:12:57 1998 */
1826 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1827 return mktime_from_utc (&t);
1832 /* Authorization support: We support two authorization schemes:
1834 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1836 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1837 consisting of answering to the server's challenge with the proper
1840 /* How many bytes it will take to store LEN bytes in base64. */
1841 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1843 /* Encode the string S of length LENGTH to base64 format and place it
1844 to STORE. STORE will be 0-terminated, and must point to a writable
1845 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1847 base64_encode (const char *s, char *store, int length)
1849 /* Conversion table. */
1850 static char tbl[64] = {
1851 'A','B','C','D','E','F','G','H',
1852 'I','J','K','L','M','N','O','P',
1853 'Q','R','S','T','U','V','W','X',
1854 'Y','Z','a','b','c','d','e','f',
1855 'g','h','i','j','k','l','m','n',
1856 'o','p','q','r','s','t','u','v',
1857 'w','x','y','z','0','1','2','3',
1858 '4','5','6','7','8','9','+','/'
1861 unsigned char *p = (unsigned char *)store;
1863 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1864 for (i = 0; i < length; i += 3)
1866 *p++ = tbl[s[0] >> 2];
1867 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1868 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1869 *p++ = tbl[s[2] & 0x3f];
1872 /* Pad the result if necessary... */
1873 if (i == length + 1)
1875 else if (i == length + 2)
1876 *(p - 1) = *(p - 2) = '=';
1877 /* ...and zero-terminate it. */
1881 /* Create the authentication header contents for the `Basic' scheme.
1882 This is done by encoding the string `USER:PASS' in base64 and
1883 prepending `HEADER: Basic ' to it. */
1885 basic_authentication_encode (const char *user, const char *passwd,
1888 char *t1, *t2, *res;
1889 int len1 = strlen (user) + 1 + strlen (passwd);
1890 int len2 = BASE64_LENGTH (len1);
1892 t1 = (char *)alloca (len1 + 1);
1893 sprintf (t1, "%s:%s", user, passwd);
1894 t2 = (char *)alloca (1 + len2);
1895 base64_encode (t1, t2, len1);
1896 res = (char *)xmalloc (len2 + 11 + strlen (header));
1897 sprintf (res, "%s: Basic %s\r\n", header, t2);
1903 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
1904 of a field in such a header. If the field is the one specified by
1905 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
1906 digest authorization code), extract its value in the (char*)
1907 variable pointed by RET. Returns negative on a malformed header,
1908 or number of bytes that have been parsed by this call. */
1910 extract_header_attr (const char *au, const char *attr_name, char **ret)
1912 const char *cp, *ep;
1916 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
1918 cp += strlen (attr_name);
1921 cp += skip_lws (cp);
1926 cp += skip_lws (cp);
1931 for (ep = cp; *ep && *ep != '\"'; ep++)
1936 *ret = strdupdelim (cp, ep);
1943 /* Response value needs to be in lowercase, so we cannot use HEXD2ASC
1944 from url.h. See RFC 2069 2.1.2 for the syntax of response-digest. */
1945 #define HEXD2asc(x) (((x) < 10) ? ((x) + '0') : ((x) - 10 + 'a'))
1947 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
1948 an array of 16 bytes containing the hash keys, and BUF should be a
1949 buffer of 33 writable characters (32 for hex digits plus one for
1950 zero termination). */
1952 dump_hash (unsigned char *buf, const unsigned char *hash)
1956 for (i = 0; i < MD5_HASHLEN; i++, hash++)
1958 *buf++ = HEXD2asc (*hash >> 4);
1959 *buf++ = HEXD2asc (*hash & 0xf);
1964 /* Take the line apart to find the challenge, and compose a digest
1965 authorization header. See RFC2069 section 2.1.2. */
1967 digest_authentication_encode (const char *au, const char *user,
1968 const char *passwd, const char *method,
1971 static char *realm, *opaque, *nonce;
1976 { "realm", &realm },
1977 { "opaque", &opaque },
1982 realm = opaque = nonce = NULL;
1984 au += 6; /* skip over `Digest' */
1989 au += skip_lws (au);
1990 for (i = 0; i < ARRAY_SIZE (options); i++)
1992 int skip = extract_header_attr (au, options[i].name,
1993 options[i].variable);
1997 FREE_MAYBE (opaque);
2007 if (i == ARRAY_SIZE (options))
2009 while (*au && *au != '=')
2013 au += skip_lws (au);
2017 while (*au && *au != '\"')
2024 while (*au && *au != ',')
2029 if (!realm || !nonce || !user || !passwd || !path || !method)
2032 FREE_MAYBE (opaque);
2037 /* Calculate the digest value. */
2040 unsigned char hash[MD5_HASHLEN];
2041 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2042 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2044 /* A1BUF = H(user ":" realm ":" password) */
2045 md5_init_ctx (&ctx);
2046 md5_process_bytes (user, strlen (user), &ctx);
2047 md5_process_bytes (":", 1, &ctx);
2048 md5_process_bytes (realm, strlen (realm), &ctx);
2049 md5_process_bytes (":", 1, &ctx);
2050 md5_process_bytes (passwd, strlen (passwd), &ctx);
2051 md5_finish_ctx (&ctx, hash);
2052 dump_hash (a1buf, hash);
2054 /* A2BUF = H(method ":" path) */
2055 md5_init_ctx (&ctx);
2056 md5_process_bytes (method, strlen (method), &ctx);
2057 md5_process_bytes (":", 1, &ctx);
2058 md5_process_bytes (path, strlen (path), &ctx);
2059 md5_finish_ctx (&ctx, hash);
2060 dump_hash (a2buf, hash);
2062 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2063 md5_init_ctx (&ctx);
2064 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2065 md5_process_bytes (":", 1, &ctx);
2066 md5_process_bytes (nonce, strlen (nonce), &ctx);
2067 md5_process_bytes (":", 1, &ctx);
2068 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2069 md5_finish_ctx (&ctx, hash);
2070 dump_hash (response_digest, hash);
2072 res = (char*) xmalloc (strlen (user)
2077 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2078 + (opaque ? strlen (opaque) : 0)
2080 sprintf (res, "Authorization: Digest \
2081 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2082 user, realm, nonce, path, response_digest);
2085 char *p = res + strlen (res);
2086 strcat (p, ", opaque=\"");
2090 strcat (res, "\r\n");
2094 #endif /* USE_DIGEST */
2097 #define BEGINS_WITH(line, string_constant) \
2098 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2099 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2100 || !line[sizeof (string_constant) - 1]))
2103 known_authentication_scheme_p (const char *au)
2105 return BEGINS_WITH (au, "Basic")
2106 || BEGINS_WITH (au, "Digest")
2107 || BEGINS_WITH (au, "NTLM");
2112 /* Create the HTTP authorization request header. When the
2113 `WWW-Authenticate' response header is seen, according to the
2114 authorization scheme specified in that header (`Basic' and `Digest'
2115 are supported by the current implementation), produce an
2116 appropriate HTTP authorization request header. */
2118 create_authorization_line (const char *au, const char *user,
2119 const char *passwd, const char *method,
2122 char *wwwauth = NULL;
2124 if (!strncasecmp (au, "Basic", 5))
2125 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2126 if (!strncasecmp (au, "NTLM", 4))
2127 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2129 else if (!strncasecmp (au, "Digest", 6))
2130 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2131 #endif /* USE_DIGEST */