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"
70 extern char *version_string;
81 static int cookies_loaded_p;
83 #define TEXTHTML_S "text/html"
84 #define HTTP_ACCEPT "*/*"
86 /* Some status code validation macros: */
87 #define H_20X(x) (((x) >= 200) && ((x) < 300))
88 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
89 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
90 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
92 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
94 #define HTTP_STATUS_OK 200
95 #define HTTP_STATUS_CREATED 201
96 #define HTTP_STATUS_ACCEPTED 202
97 #define HTTP_STATUS_NO_CONTENT 204
98 #define HTTP_STATUS_PARTIAL_CONTENTS 206
100 /* Redirection 3xx. */
101 #define HTTP_STATUS_MULTIPLE_CHOICES 300
102 #define HTTP_STATUS_MOVED_PERMANENTLY 301
103 #define HTTP_STATUS_MOVED_TEMPORARILY 302
104 #define HTTP_STATUS_NOT_MODIFIED 304
106 /* Client error 4xx. */
107 #define HTTP_STATUS_BAD_REQUEST 400
108 #define HTTP_STATUS_UNAUTHORIZED 401
109 #define HTTP_STATUS_FORBIDDEN 403
110 #define HTTP_STATUS_NOT_FOUND 404
112 /* Server errors 5xx. */
113 #define HTTP_STATUS_INTERNAL 500
114 #define HTTP_STATUS_NOT_IMPLEMENTED 501
115 #define HTTP_STATUS_BAD_GATEWAY 502
116 #define HTTP_STATUS_UNAVAILABLE 503
119 /* Parse the HTTP status line, which is of format:
121 HTTP-Version SP Status-Code SP Reason-Phrase
123 The function returns the status-code, or -1 if the status line is
124 malformed. The pointer to reason-phrase is returned in RP. */
126 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
128 /* (the variables must not be named `major' and `minor', because
129 that breaks compilation with SunOS4 cc.) */
130 int mjr, mnr, statcode;
133 *reason_phrase_ptr = NULL;
135 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
136 major version, and Y is minor version. */
137 if (strncmp (line, "HTTP/", 5) != 0)
141 /* Calculate major HTTP version. */
143 for (mjr = 0; ISDIGIT (*line); line++)
144 mjr = 10 * mjr + (*line - '0');
145 if (*line != '.' || p == line)
149 /* Calculate minor HTTP version. */
151 for (mnr = 0; ISDIGIT (*line); line++)
152 mnr = 10 * mnr + (*line - '0');
153 if (*line != ' ' || p == line)
155 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
156 minor version can be safely ignored. */
161 /* Calculate status code. */
162 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
164 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
166 /* Set up the reason phrase pointer. */
168 /* RFC2068 requires SPC here, but we allow the string to finish
169 here, in case no reason-phrase is present. */
173 *reason_phrase_ptr = line;
178 *reason_phrase_ptr = line + 1;
183 /* Functions to be used as arguments to header_process(): */
185 struct http_process_range_closure {
191 /* Parse the `Content-Range' header and extract the information it
192 contains. Returns 1 if successful, -1 otherwise. */
194 http_process_range (const char *hdr, void *arg)
196 struct http_process_range_closure *closure
197 = (struct http_process_range_closure *)arg;
200 /* Certain versions of Nutscape proxy server send out
201 `Content-Length' without "bytes" specifier, which is a breach of
202 RFC2068 (as well as the HTTP/1.1 draft which was current at the
203 time). But hell, I must support it... */
204 if (!strncasecmp (hdr, "bytes", 5))
207 hdr += skip_lws (hdr);
213 for (num = 0; ISDIGIT (*hdr); hdr++)
214 num = 10 * num + (*hdr - '0');
215 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
217 closure->first_byte_pos = num;
219 for (num = 0; ISDIGIT (*hdr); hdr++)
220 num = 10 * num + (*hdr - '0');
221 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
223 closure->last_byte_pos = num;
225 for (num = 0; ISDIGIT (*hdr); hdr++)
226 num = 10 * num + (*hdr - '0');
227 closure->entity_length = num;
231 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
232 Used for `Accept-Ranges'. */
234 http_process_none (const char *hdr, void *arg)
236 int *where = (int *)arg;
238 if (strstr (hdr, "none"))
245 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
247 http_process_type (const char *hdr, void *arg)
249 char **result = (char **)arg;
250 /* Locate P on `;' or the terminating zero, whichever comes first. */
251 const char *p = strchr (hdr, ';');
253 p = hdr + strlen (hdr);
254 while (p > hdr && ISSPACE (*(p - 1)))
256 *result = strdupdelim (hdr, p);
260 /* Check whether the `Connection' header is set to "keep-alive". */
262 http_process_connection (const char *hdr, void *arg)
264 int *flag = (int *)arg;
265 if (!strcasecmp (hdr, "Keep-Alive"))
270 /* Persistent connections. Currently, we cache the most recently used
271 connection as persistent, provided that the HTTP server agrees to
272 make it such. The persistence data is stored in the variables
273 below. Ideally, it would be in a structure, and it should be
274 possible to cache an arbitrary fixed number of these connections.
276 I think the code is quite easy to extend in that direction. */
278 /* Whether a persistent connection is active. */
279 static int pc_active_p;
280 /* Host and port of currently active persistent connection. */
281 static unsigned char pc_last_host[4];
282 static unsigned short pc_last_port;
284 /* File descriptor of the currently active persistent connection. */
285 static int pc_last_fd;
288 /* Whether a ssl handshake has occoured on this connection */
289 static int pc_active_ssl;
290 /* SSL connection of the currently active persistent connection. */
291 static SSL *pc_last_ssl;
292 #endif /* HAVE_SSL */
294 /* Mark the persistent connection as invalid. This is used by the
295 CLOSE_* macros after they forcefully close a registered persistent
296 connection. This does not close the file descriptor -- it is left
297 to the caller to do that. (Maybe it should, though.) */
300 invalidate_persistent (void)
305 #endif /* HAVE_SSL */
306 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
309 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
310 persistent. This will enable someone to use the same connection
311 later. In the context of HTTP, this must be called only AFTER the
312 response has been received and the server has promised that the
313 connection will remain alive.
315 If a previous connection was persistent, it is closed. */
318 register_persistent (const char *host, unsigned short port, int fd
328 if (pc_last_fd == fd)
330 /* The connection FD is already registered. Nothing to
336 /* The old persistent connection is still active; let's
337 close it first. This situation arises whenever a
338 persistent connection exists, but we then connect to a
339 different host, and try to register a persistent
340 connection to that one. */
342 /* The ssl disconnect has to take place before the closing
345 shutdown_ssl(pc_last_ssl);
348 invalidate_persistent ();
352 /* This store_hostaddress may not fail, because it has the results
354 success = store_hostaddress (pc_last_host, host);
361 pc_active_ssl = ssl ? 1 : 0;
363 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
366 /* Return non-zero if a persistent connection is available for
367 connecting to HOST:PORT. */
370 persistent_available_p (const char *host, unsigned short port
376 unsigned char this_host[4];
377 /* First, check whether a persistent connection is active at all. */
380 /* Second, check if the active connection pertains to the correct
381 (HOST, PORT) ordered pair. */
382 if (port != pc_last_port)
385 /* Second, a): check if current connection is (not) ssl, too. This
386 test is unlikely to fail because HTTP and HTTPS typicaly use
387 different ports. Yet it is possible, or so I [Christian
388 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
390 if (ssl != pc_active_ssl)
392 #endif /* HAVE_SSL */
393 if (!store_hostaddress (this_host, host))
395 if (memcmp (pc_last_host, this_host, 4))
397 /* Third: check whether the connection is still open. This is
398 important because most server implement a liberal (short) timeout
399 on persistent connections. Wget can of course always reconnect
400 if the connection doesn't work out, but it's nicer to know in
401 advance. This test is a logical followup of the first test, but
402 is "expensive" and therefore placed at the end of the list. */
403 if (!test_socket_open (pc_last_fd))
405 /* Oops, the socket is no longer open. Now that we know that,
406 let's invalidate the persistent connection before returning
409 invalidate_persistent ();
416 # define SHUTDOWN_SSL(ssl) do { \
418 shutdown_ssl (ssl); \
421 # define SHUTDOWN_SSL(ssl)
424 /* The idea behind these two CLOSE macros is to distinguish between
425 two cases: one when the job we've been doing is finished, and we
426 want to close the connection and leave, and two when something is
427 seriously wrong and we're closing the connection as part of
430 In case of keep_alive, CLOSE_FINISH should leave the connection
431 open, while CLOSE_INVALIDATE should still close it.
433 Note that the semantics of the flag `keep_alive' is "this
434 connection *will* be reused (the server has promised not to close
435 the connection once we're done)", while the semantics of
436 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
437 active, registered connection". */
439 #define CLOSE_FINISH(fd) do { \
442 SHUTDOWN_SSL (ssl); \
444 if (pc_active_p && (fd) == pc_last_fd) \
445 invalidate_persistent (); \
449 #define CLOSE_INVALIDATE(fd) do { \
450 SHUTDOWN_SSL (ssl); \
452 if (pc_active_p && (fd) == pc_last_fd) \
453 invalidate_persistent (); \
458 long len; /* received length */
459 long contlen; /* expected length */
460 long restval; /* the restart value */
461 int res; /* the result of last read */
462 char *newloc; /* new location (redirection) */
463 char *remote_time; /* remote time-stamp string */
464 char *error; /* textual HTTP error */
465 int statcode; /* status code */
466 long dltime; /* time of the download */
467 int no_truncate; /* whether truncating the file is
471 /* Free the elements of hstat X. */
472 #define FREEHSTAT(x) do \
474 FREE_MAYBE ((x).newloc); \
475 FREE_MAYBE ((x).remote_time); \
476 FREE_MAYBE ((x).error); \
477 (x).newloc = (x).remote_time = (x).error = NULL; \
480 static char *create_authorization_line PARAMS ((const char *, const char *,
481 const char *, const char *,
483 static char *basic_authentication_encode PARAMS ((const char *, const char *,
485 static int known_authentication_scheme_p PARAMS ((const char *));
487 time_t http_atotm PARAMS ((char *));
489 #define BEGINS_WITH(line, string_constant) \
490 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
491 && (ISSPACE (line[sizeof (string_constant) - 1]) \
492 || !line[sizeof (string_constant) - 1]))
494 /* Retrieve a document through HTTP protocol. It recognizes status
495 code, and correctly handles redirections. It closes the network
496 socket. If it receives an error from the functions below it, it
497 will print it if there is enough information to do so (almost
498 always), returning the error to the caller (i.e. http_loop).
500 Various HTTP parameters are stored to hs. Although it parses the
501 response code correctly, it is not used in a sane way. The caller
504 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
505 and u->proxy->url will be given to the proxy server (bad naming,
508 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
510 char *request, *type, *command, *path;
512 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
513 char *authenticate_h;
517 char *request_keep_alive;
518 int sock, hcount, num_written, all_length, remport, statcode;
519 long contlen, contrange;
523 int auth_tried_already;
526 static SSL_CTX *ssl_ctx = NULL;
528 #endif /* HAVE_SSL */
529 struct wget_timer *timer;
530 char *cookies = NULL;
532 /* Whether this connection will be kept alive after the HTTP request
536 /* Flags that detect the two ways of specifying HTTP keep-alive
538 int http_keep_alive_1, http_keep_alive_2;
540 /* Whether keep-alive should be inhibited. */
541 int inhibit_keep_alive;
544 /* initialize ssl_ctx on first run */
547 err=init_ssl (&ssl_ctx);
552 case SSLERRCTXCREATE:
554 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
558 /* try without certfile */
559 logprintf (LOG_NOTQUIET,
560 _("Failed to load certificates from %s\n"),
563 logprintf (LOG_NOTQUIET,
564 _("Trying without the specified certificate\n"));
567 logprintf (LOG_NOTQUIET,
568 _("Failed to get certificate key from %s\n"),
571 logprintf (LOG_NOTQUIET,
572 _("Trying without the specified certificate\n"));
579 #endif /* HAVE_SSL */
581 if (!(*dt & HEAD_ONLY))
582 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
583 know the local filename so we can save to it. */
584 assert (u->local != NULL);
587 auth_tried_already = 0;
589 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
592 /* We need to come back here when the initial attempt to retrieve
593 without authorization header fails. (Expected to happen at least
594 for the Digest authorization scheme.) */
597 http_keep_alive_1 = http_keep_alive_2 = 0;
599 /* Initialize certain elements of struct http_stat. */
604 hs->remote_time = NULL;
607 /* Which structure to use to retrieve the original URL data. */
613 /* First: establish the connection. */
614 if (inhibit_keep_alive
617 !persistent_available_p (u->host, u->port)
619 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
620 #endif /* HAVE_SSL */
623 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
624 err = make_connection (&sock, u->host, u->port);
628 logputs (LOG_VERBOSE, "\n");
629 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
633 logputs (LOG_VERBOSE, "\n");
634 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
638 logputs (LOG_VERBOSE, "\n");
639 logprintf (LOG_NOTQUIET,
640 _("Connection to %s:%hu refused.\n"), u->host, u->port);
644 logputs (LOG_VERBOSE, "\n");
645 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
650 /* Everything is fine! */
651 logputs (LOG_VERBOSE, _("connected!\n"));
658 if (u->proto == URLHTTPS)
659 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
661 logputs (LOG_VERBOSE, "\n");
662 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
666 #endif /* HAVE_SSL */
670 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
671 /* #### pc_last_fd should be accessed through an accessor
676 #endif /* HAVE_SSL */
677 DEBUGP (("Reusing fd %d.\n", sock));
681 path = u->proxy->url;
685 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
689 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
690 sprintf (referer, "Referer: %s\r\n", ou->referer);
692 if (*dt & SEND_NOCACHE)
693 pragma_h = "Pragma: no-cache\r\n";
698 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
699 /* Gag me! Some servers (e.g. WebSitePro) have been known to
700 respond to the following `Range' format by generating a
701 multipart/x-byte-ranges MIME document! This MIME type was
702 present in an old draft of the byteranges specification.
703 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
704 only if multiple non-overlapping ranges are requested --
705 which Wget never does. */
706 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
711 STRDUP_ALLOCA (useragent, opt.useragent);
714 useragent = (char *)alloca (10 + strlen (version_string));
715 sprintf (useragent, "Wget/%s", version_string);
717 /* Construct the authentication, if userid is present. */
720 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
721 user = user ? user : opt.http_user;
722 passwd = passwd ? passwd : opt.http_passwd;
729 /* We have the username and the password, but haven't tried
730 any authorization yet. Let's see if the "Basic" method
731 works. If not, we'll come back here and construct a
732 proper authorization method with the right challenges.
734 If we didn't employ this kind of logic, every URL that
735 requires authorization would have to be processed twice,
736 which is very suboptimal and generates a bunch of false
737 "unauthorized" errors in the server log.
739 #### But this logic also has a serious problem when used
740 with stronger authentications: we *first* transmit the
741 username and the password in clear text, and *then*
742 attempt a stronger authentication scheme. That cannot be
743 right! We are only fortunate that almost everyone still
744 uses the `Basic' scheme anyway.
746 There should be an option to prevent this from happening,
747 for those who use strong authentication schemes and value
749 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
753 wwwauth = create_authorization_line (authenticate_h, user, passwd,
761 char *proxy_user, *proxy_passwd;
762 /* For normal username and password, URL components override
763 command-line/wgetrc parameters. With proxy authentication,
764 it's the reverse, because proxy URLs are normally the
765 "permanent" ones, so command-line args should take
767 if (opt.proxy_user && opt.proxy_passwd)
769 proxy_user = opt.proxy_user;
770 proxy_passwd = opt.proxy_passwd;
774 proxy_user = u->user;
775 proxy_passwd = u->passwd;
777 /* #### This is junky. Can't the proxy request, say, `Digest'
779 if (proxy_user && proxy_passwd)
780 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
781 "Proxy-Authorization");
786 /* String of the form :PORT. Used only for non-standard ports. */
790 && remport != (u->proto == URLHTTPS
791 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
793 && remport != DEFAULT_HTTP_PORT
797 port_maybe = (char *)alloca (numdigit (remport) + 2);
798 sprintf (port_maybe, ":%d", remport);
801 if (!inhibit_keep_alive)
802 request_keep_alive = "Connection: Keep-Alive\r\n";
804 request_keep_alive = NULL;
807 cookies = build_cookies_request (ou->host, ou->port, ou->path,
808 ou->proto == URLHTTPS);
810 /* Allocate the memory for the request. */
811 request = (char *)alloca (strlen (command) + strlen (path)
814 + (port_maybe ? strlen (port_maybe) : 0)
815 + strlen (HTTP_ACCEPT)
816 + (request_keep_alive
817 ? strlen (request_keep_alive) : 0)
818 + (referer ? strlen (referer) : 0)
819 + (cookies ? strlen (cookies) : 0)
820 + (wwwauth ? strlen (wwwauth) : 0)
821 + (proxyauth ? strlen (proxyauth) : 0)
822 + (range ? strlen (range) : 0)
824 + (opt.user_header ? strlen (opt.user_header) : 0)
826 /* Construct the request. */
832 %s%s%s%s%s%s%s%s\r\n",
833 command, path, useragent, remhost,
834 port_maybe ? port_maybe : "",
836 request_keep_alive ? request_keep_alive : "",
837 referer ? referer : "",
838 cookies ? cookies : "",
839 wwwauth ? wwwauth : "",
840 proxyauth ? proxyauth : "",
843 opt.user_header ? opt.user_header : "");
844 DEBUGP (("---request begin---\n%s---request end---\n", request));
845 /* Free the temporary memory. */
846 FREE_MAYBE (wwwauth);
847 FREE_MAYBE (proxyauth);
848 FREE_MAYBE (cookies);
850 /* Send the request to server. */
852 if (u->proto == URLHTTPS)
853 num_written = ssl_iwrite (ssl, request, strlen (request));
855 #endif /* HAVE_SSL */
856 num_written = iwrite (sock, request, strlen (request));
860 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
862 CLOSE_INVALIDATE (sock);
865 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
866 u->proxy ? "Proxy" : "HTTP");
867 contlen = contrange = -1;
872 /* Before reading anything, initialize the rbuf. */
873 rbuf_initialize (&rbuf, sock);
875 if (u->proto == URLHTTPS)
879 #endif /* HAVE_SSL */
882 /* Header-fetching loop. */
890 /* Get the header. */
891 status = header_get (&rbuf, &hdr,
892 /* Disallow continuations for status line. */
893 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
895 /* Check for errors. */
896 if (status == HG_EOF && *hdr)
898 /* This used to be an unconditional error, but that was
899 somewhat controversial, because of a large number of
900 broken CGI's that happily "forget" to send the second EOL
901 before closing the connection of a HEAD request.
903 So, the deal is to check whether the header is empty
904 (*hdr is zero if it is); if yes, it means that the
905 previous header was fully retrieved, and that -- most
906 probably -- the request is complete. "...be liberal in
907 what you accept." Oh boy. */
908 logputs (LOG_VERBOSE, "\n");
909 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
912 FREE_MAYBE (all_headers);
913 CLOSE_INVALIDATE (sock);
916 else if (status == HG_ERROR)
918 logputs (LOG_VERBOSE, "\n");
919 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
923 FREE_MAYBE (all_headers);
924 CLOSE_INVALIDATE (sock);
928 /* If the headers are to be saved to a file later, save them to
930 if (opt.save_headers)
932 int lh = strlen (hdr);
933 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
934 memcpy (all_headers + all_length, hdr, lh);
936 all_headers[all_length++] = '\n';
937 all_headers[all_length] = '\0';
940 /* Print the header if requested. */
941 if (opt.server_response && hcount != 1)
942 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
944 /* Check for status line. */
948 /* Parse the first line of server response. */
949 statcode = parse_http_status_line (hdr, &error);
950 hs->statcode = statcode;
951 /* Store the descriptive response. */
952 if (statcode == -1) /* malformed response */
954 /* A common reason for "malformed response" error is the
955 case when no data was actually received. Handle this
958 hs->error = xstrdup (_("No data received"));
960 hs->error = xstrdup (_("Malformed status line"));
965 hs->error = xstrdup (_("(no description)"));
967 hs->error = xstrdup (error);
974 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
979 /* Exit on empty header. */
986 /* Try getting content-length. */
987 if (contlen == -1 && !opt.ignore_length)
988 if (header_process (hdr, "Content-Length", header_extract_number,
991 /* Try getting content-type. */
993 if (header_process (hdr, "Content-Type", http_process_type, &type))
995 /* Try getting location. */
997 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
999 /* Try getting last-modified. */
1000 if (!hs->remote_time)
1001 if (header_process (hdr, "Last-Modified", header_strdup,
1004 /* Try getting cookies. */
1006 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1008 /* Try getting www-authentication. */
1009 if (!authenticate_h)
1010 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1013 /* Check for accept-ranges header. If it contains the word
1014 `none', disable the ranges. */
1015 if (*dt & ACCEPTRANGES)
1018 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1021 *dt &= ~ACCEPTRANGES;
1025 /* Try getting content-range. */
1026 if (contrange == -1)
1028 struct http_process_range_closure closure;
1029 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1031 contrange = closure.first_byte_pos;
1035 /* Check for keep-alive related responses. */
1036 if (!inhibit_keep_alive)
1038 /* Check for the `Keep-Alive' header. */
1039 if (!http_keep_alive_1)
1041 if (header_process (hdr, "Keep-Alive", header_exists,
1042 &http_keep_alive_1))
1045 /* Check for `Connection: Keep-Alive'. */
1046 if (!http_keep_alive_2)
1048 if (header_process (hdr, "Connection", http_process_connection,
1049 &http_keep_alive_2))
1057 logputs (LOG_VERBOSE, "\n");
1060 && (http_keep_alive_1 || http_keep_alive_2))
1062 assert (inhibit_keep_alive == 0);
1066 /* The server has promised that it will not close the connection
1067 when we're done. This means that we can register it. */
1069 register_persistent (u->host, u->port, sock);
1071 register_persistent (u->host, u->port, sock, ssl);
1072 #endif /* HAVE_SSL */
1074 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1077 /* Authorization is required. */
1081 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1082 might be more bytes in the body. */
1083 if (auth_tried_already)
1085 /* If we have tried it already, then there is not point
1088 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1089 xfree (authenticate_h);
1092 else if (!known_authentication_scheme_p (authenticate_h))
1094 xfree (authenticate_h);
1095 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1098 else if (BEGINS_WITH (authenticate_h, "Basic"))
1100 /* The authentication scheme is basic, the one we try by
1101 default, and it failed. There's no sense in trying
1107 auth_tried_already = 1;
1111 /* We do not need this anymore. */
1114 xfree (authenticate_h);
1115 authenticate_h = NULL;
1118 /* 20x responses are counted among successful by default. */
1119 if (H_20X (statcode))
1122 /* Return if redirected. */
1123 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1125 /* RFC2068 says that in case of the 300 (multiple choices)
1126 response, the server can output a preferred URL through
1127 `Location' header; otherwise, the request should be treated
1128 like GET. So, if the location is set, it will be a
1129 redirection; otherwise, just proceed normally. */
1130 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1134 logprintf (LOG_VERBOSE,
1135 _("Location: %s%s\n"),
1136 hs->newloc ? hs->newloc : _("unspecified"),
1137 hs->newloc ? _(" [following]") : "");
1138 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1139 might be more bytes in the body. */
1141 FREE_MAYBE (all_headers);
1146 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1149 /* We don't assume text/html by default. */
1152 if (opt.html_extension && (*dt & TEXTHTML))
1153 /* -E / --html-extension / html_extension = on was specified, and this is a
1154 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1155 already the file's suffix, tack on ".html". */
1157 char* last_period_in_local_filename = strrchr(u->local, '.');
1159 if (last_period_in_local_filename == NULL ||
1160 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1161 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1163 size_t local_filename_len = strlen(u->local);
1165 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1166 strcpy(u->local + local_filename_len, ".html");
1168 *dt |= ADDED_HTML_EXTENSION;
1172 if (contrange == -1)
1174 /* We did not get a content-range header. This means that the
1175 server did not honor our `Range' request. Normally, this
1176 means we should reset hs->restval and continue normally. */
1178 /* However, if `-c' is used, we need to be a bit more careful:
1180 1. If `-c' is specified and the file already existed when
1181 Wget was started, it would be a bad idea for us to start
1182 downloading it from scratch, effectively truncating it. I
1183 believe this cannot happen unless `-c' was specified.
1185 2. If `-c' is used on a file that is already fully
1186 downloaded, we're requesting bytes after the end of file,
1187 which can result in server not honoring `Range'. If this is
1188 the case, `Content-Length' will be equal to the length of the
1190 if (opt.always_rest)
1192 /* Check for condition #2. */
1193 if (hs->restval > 0 /* restart was requested. */
1194 && contlen != -1 /* we got content-length. */
1195 && hs->restval >= contlen /* file fully downloaded
1199 logputs (LOG_VERBOSE, _("\
1200 \n The file is already fully retrieved; nothing to do.\n\n"));
1201 /* In case the caller inspects. */
1205 FREE_MAYBE (all_headers);
1206 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1207 might be more bytes in the body. */
1208 return RETRUNNEEDED;
1211 /* Check for condition #1. */
1212 if (hs->no_truncate)
1214 logprintf (LOG_NOTQUIET,
1217 The server does not support continued downloads, which conflicts with `-c'.\n\
1218 Refusing to truncate existing file `%s'.\n\n"), u->local);
1220 FREE_MAYBE (all_headers);
1221 CLOSE_INVALIDATE (sock);
1222 return CONTNOTSUPPORTED;
1230 else if (contrange != hs->restval ||
1231 (H_PARTIAL (statcode) && contrange == -1))
1233 /* This means the whole request was somehow misunderstood by the
1234 server. Bail out. */
1236 FREE_MAYBE (all_headers);
1237 CLOSE_INVALIDATE (sock);
1244 contlen += contrange;
1246 contrange = -1; /* If conent-length was not sent,
1247 content-range will be ignored. */
1249 hs->contlen = contlen;
1253 if ((*dt & RETROKF) && !opt.server_response)
1255 /* No need to print this output if the body won't be
1256 downloaded at all, or if the original server response is
1258 logputs (LOG_VERBOSE, _("Length: "));
1261 logputs (LOG_VERBOSE, legible (contlen));
1262 if (contrange != -1)
1263 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1264 legible (contlen - contrange));
1267 logputs (LOG_VERBOSE,
1268 opt.ignore_length ? _("ignored") : _("unspecified"));
1270 logprintf (LOG_VERBOSE, " [%s]\n", type);
1272 logputs (LOG_VERBOSE, "\n");
1276 type = NULL; /* We don't need it any more. */
1278 /* Return if we have no intention of further downloading. */
1279 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1281 /* In case the caller cares to look... */
1285 FREE_MAYBE (all_headers);
1286 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1287 might be more bytes in the body. */
1288 return RETRFINISHED;
1291 /* Open the local file. */
1294 mkalldirs (u->local);
1296 rotate_backups (u->local);
1297 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1300 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1301 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1302 might be more bytes in the body. */
1303 FREE_MAYBE (all_headers);
1309 extern int global_download_count;
1311 /* To ensure that repeated "from scratch" downloads work for -O
1312 files, we rewind the file pointer, unless restval is
1313 non-zero. (This works only when -O is used on regular files,
1314 but it's still a valuable feature.)
1316 However, this loses when more than one URL is specified on
1317 the command line the second rewinds eradicates the contents
1318 of the first download. Thus we disable the above trick for
1319 all the downloads except the very first one.
1321 #### A possible solution to this would be to remember the
1322 file position in the output document and to seek to that
1323 position, instead of rewinding. */
1324 if (!hs->restval && global_download_count == 0)
1326 /* This will silently fail for streams that don't correspond
1327 to regular files, but that's OK. */
1329 /* ftruncate is needed because opt.dfp is opened in append
1330 mode if opt.always_rest is set. */
1331 ftruncate (fileno (fp), 0);
1336 /* #### This confuses the code that checks for file size. There
1337 should be some overhead information. */
1338 if (opt.save_headers)
1339 fwrite (all_headers, 1, all_length, fp);
1340 timer = wtimer_new ();
1341 /* Get the contents of the document. */
1342 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1343 (contlen != -1 ? contlen : 0),
1345 hs->dltime = wtimer_elapsed (timer);
1346 wtimer_delete (timer);
1348 /* Close or flush the file. We have to be careful to check for
1349 error here. Checking the result of fwrite() is not enough --
1350 errors could go unnoticed! */
1353 flush_res = fclose (fp);
1355 flush_res = fflush (fp);
1356 if (flush_res == EOF)
1359 FREE_MAYBE (all_headers);
1360 CLOSE_FINISH (sock);
1363 return RETRFINISHED;
1366 /* The genuine HTTP loop! This is the part where the retrieval is
1367 retried, and retried, and retried, and... */
1369 http_loop (struct urlinfo *u, char **newloc, int *dt)
1372 int use_ts, got_head = 0; /* time-stamping info */
1373 char *filename_plus_orig_suffix;
1374 char *local_filename = NULL;
1375 char *tms, *suf, *locf, *tmrate;
1377 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1378 long local_size = 0; /* the size of the local file */
1379 size_t filename_len;
1380 struct http_stat hstat; /* HTTP status */
1383 /* This used to be done in main(), but it's a better idea to do it
1384 here so that we don't go through the hoops if we're just using
1386 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1388 load_cookies (opt.cookies_input);
1389 cookies_loaded_p = 1;
1394 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1395 has_wildcards_p because it would also warn on `?', and we know that
1396 shows up in CGI paths a *lot*. */
1397 if (strchr (u->url, '*'))
1398 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1400 /* Determine the local filename. */
1402 u->local = url_filename (u->proxy ? u->proxy : u);
1404 if (!opt.output_document)
1407 locf = opt.output_document;
1409 filename_len = strlen (u->local);
1410 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1412 if (opt.noclobber && file_exists_p (u->local))
1414 /* If opt.noclobber is turned on and file already exists, do not
1415 retrieve the file */
1416 logprintf (LOG_VERBOSE, _("\
1417 File `%s' already there, will not retrieve.\n"), u->local);
1418 /* If the file is there, we suppose it's retrieved OK. */
1421 /* #### Bogusness alert. */
1422 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1423 text/html, a harmless lie. */
1424 if (((suf = suffix (u->local)) != NULL)
1425 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1428 /* Another harmless lie: */
1433 if (opt.timestamping)
1435 boolean local_dot_orig_file_exists = FALSE;
1437 if (opt.backup_converted)
1438 /* If -K is specified, we'll act on the assumption that it was specified
1439 last time these files were downloaded as well, and instead of just
1440 comparing local file X against server file X, we'll compare local
1441 file X.orig (if extant, else X) against server file X. If -K
1442 _wasn't_ specified last time, or the server contains files called
1443 *.orig, -N will be back to not operating correctly with -k. */
1445 /* Would a single s[n]printf() call be faster? --dan
1447 Definitely not. sprintf() is horribly slow. It's a
1448 different question whether the difference between the two
1449 affects a program. Usually I'd say "no", but at one
1450 point I profiled Wget, and found that a measurable and
1451 non-negligible amount of time was lost calling sprintf()
1452 in url.c. Replacing sprintf with inline calls to
1453 strcpy() and long_to_string() made a difference.
1455 memcpy (filename_plus_orig_suffix, u->local, filename_len);
1456 memcpy (filename_plus_orig_suffix + filename_len,
1457 ".orig", sizeof (".orig"));
1459 /* Try to stat() the .orig file. */
1460 if (stat (filename_plus_orig_suffix, &st) == 0)
1462 local_dot_orig_file_exists = TRUE;
1463 local_filename = filename_plus_orig_suffix;
1467 if (!local_dot_orig_file_exists)
1468 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1469 if (stat (u->local, &st) == 0)
1470 local_filename = u->local;
1472 if (local_filename != NULL)
1473 /* There was a local file, so we'll check later to see if the version
1474 the server has is the same version we already have, allowing us to
1479 local_size = st.st_size;
1483 /* Reset the counter. */
1485 *dt = 0 | ACCEPTRANGES;
1489 /* Increment the pass counter. */
1491 sleep_between_retrievals (count);
1492 /* Get the current time string. */
1493 tms = time_str (NULL);
1494 /* Print fetch message, if opt.verbose. */
1497 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1501 sprintf (tmp, _("(try:%2d)"), count);
1502 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1503 tms, hurl, tmp, locf);
1505 ws_changetitle (hurl, 1);
1510 /* Default document type is empty. However, if spider mode is
1511 on or time-stamping is employed, HEAD_ONLY commands is
1512 encoded within *dt. */
1513 if (opt.spider || (use_ts && !got_head))
1517 /* Assume no restarting. */
1519 /* Decide whether or not to restart. */
1520 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1521 /* #### this calls access() and then stat(); could be optimized. */
1522 && file_exists_p (locf))
1523 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1524 hstat.restval = st.st_size;
1526 /* In `-c' is used and the file is existing and non-empty,
1527 refuse to truncate it if the server doesn't support continued
1529 hstat.no_truncate = 0;
1530 if (opt.always_rest && hstat.restval)
1531 hstat.no_truncate = 1;
1533 /* Decide whether to send the no-cache directive. We send it in
1535 a) we're using a proxy, and we're past our first retrieval.
1536 Some proxies are notorious for caching incomplete data, so
1537 we require a fresh get.
1538 b) caching is explicitly inhibited. */
1539 if ((u->proxy && count > 1) /* a */
1540 || !opt.allow_cache /* b */
1542 *dt |= SEND_NOCACHE;
1544 *dt &= ~SEND_NOCACHE;
1546 /* Try fetching the document, or at least its head. :-) */
1547 err = gethttp (u, &hstat, dt);
1549 /* It's unfortunate that wget determines the local filename before finding
1550 out the Content-Type of the file. Barring a major restructuring of the
1551 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1552 u->local to tack on ".html". */
1553 if (!opt.output_document)
1556 locf = opt.output_document;
1559 tms = time_str (NULL);
1560 /* Get the new location (with or without the redirection). */
1562 *newloc = xstrdup (hstat.newloc);
1565 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1566 case CONERROR: case READERR: case WRITEFAILED:
1568 /* Non-fatal errors continue executing the loop, which will
1569 bring them to "while" statement at the end, to judge
1570 whether the number of tries was exceeded. */
1572 printwhat (count, opt.ntry);
1575 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1576 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1577 /* Fatal errors just return from the function. */
1581 case FWRITEERR: case FOPENERR:
1582 /* Another fatal error. */
1583 logputs (LOG_VERBOSE, "\n");
1584 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1585 u->local, strerror (errno));
1590 /* Another fatal error. */
1591 logputs (LOG_VERBOSE, "\n");
1592 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1597 /* Return the new location to the caller. */
1600 logprintf (LOG_NOTQUIET,
1601 _("ERROR: Redirection (%d) without location.\n"),
1609 /* The file was already fully retrieved. */
1614 /* Deal with you later. */
1617 /* All possibilities should have been exhausted. */
1620 if (!(*dt & RETROKF))
1624 /* #### Ugly ugly ugly! */
1625 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1626 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1629 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1630 tms, hstat.statcode, hstat.error);
1631 logputs (LOG_VERBOSE, "\n");
1636 /* Did we get the time-stamp? */
1639 if (opt.timestamping && !hstat.remote_time)
1641 logputs (LOG_NOTQUIET, _("\
1642 Last-modified header missing -- time-stamps turned off.\n"));
1644 else if (hstat.remote_time)
1646 /* Convert the date-string into struct tm. */
1647 tmr = http_atotm (hstat.remote_time);
1648 if (tmr == (time_t) (-1))
1649 logputs (LOG_VERBOSE, _("\
1650 Last-modified header invalid -- time-stamp ignored.\n"));
1654 /* The time-stamping section. */
1659 use_ts = 0; /* no more time-stamping */
1660 count = 0; /* the retrieve count for HEAD is
1662 if (hstat.remote_time && tmr != (time_t) (-1))
1664 /* Now time-stamping can be used validly. Time-stamping
1665 means that if the sizes of the local and remote file
1666 match, and local file is newer than the remote file,
1667 it will not be retrieved. Otherwise, the normal
1668 download procedure is resumed. */
1670 (hstat.contlen == -1 || local_size == hstat.contlen))
1672 logprintf (LOG_VERBOSE, _("\
1673 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1678 else if (tml >= tmr)
1679 logprintf (LOG_VERBOSE, _("\
1680 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1682 logputs (LOG_VERBOSE,
1683 _("Remote file is newer, retrieving.\n"));
1688 if ((tmr != (time_t) (-1))
1690 && ((hstat.len == hstat.contlen) ||
1691 ((hstat.res == 0) &&
1692 ((hstat.contlen == -1) ||
1693 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1695 /* #### This code repeats in http.c and ftp.c. Move it to a
1697 const char *fl = NULL;
1698 if (opt.output_document)
1700 if (opt.od_known_regular)
1701 fl = opt.output_document;
1708 /* End of time-stamping section. */
1712 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1716 /* It is now safe to free the remainder of hstat, since the
1717 strings within it will no longer be used. */
1720 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1722 if (hstat.len == hstat.contlen)
1726 logprintf (LOG_VERBOSE,
1727 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1728 tms, tmrate, locf, hstat.len, hstat.contlen);
1729 logprintf (LOG_NONVERBOSE,
1730 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1731 tms, u->url, hstat.len, hstat.contlen, locf, count);
1734 downloaded_increase (hstat.len);
1736 /* Remember that we downloaded the file for later ".orig" code. */
1737 if (*dt & ADDED_HTML_EXTENSION)
1738 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1740 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1744 else if (hstat.res == 0) /* No read error */
1746 if (hstat.contlen == -1) /* We don't know how much we were supposed
1747 to get, so assume we succeeded. */
1751 logprintf (LOG_VERBOSE,
1752 _("%s (%s) - `%s' saved [%ld]\n\n"),
1753 tms, tmrate, locf, hstat.len);
1754 logprintf (LOG_NONVERBOSE,
1755 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1756 tms, u->url, hstat.len, locf, count);
1759 downloaded_increase (hstat.len);
1761 /* Remember that we downloaded the file for later ".orig" code. */
1762 if (*dt & ADDED_HTML_EXTENSION)
1763 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1765 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1769 else if (hstat.len < hstat.contlen) /* meaning we lost the
1770 connection too soon */
1772 logprintf (LOG_VERBOSE,
1773 _("%s (%s) - Connection closed at byte %ld. "),
1774 tms, tmrate, hstat.len);
1775 printwhat (count, opt.ntry);
1778 else if (!opt.kill_longer) /* meaning we got more than expected */
1780 logprintf (LOG_VERBOSE,
1781 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1782 tms, tmrate, locf, hstat.len, hstat.contlen);
1783 logprintf (LOG_NONVERBOSE,
1784 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1785 tms, u->url, hstat.len, hstat.contlen, locf, count);
1787 downloaded_increase (hstat.len);
1789 /* Remember that we downloaded the file for later ".orig" code. */
1790 if (*dt & ADDED_HTML_EXTENSION)
1791 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1793 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1797 else /* the same, but not accepted */
1799 logprintf (LOG_VERBOSE,
1800 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1801 tms, tmrate, hstat.len, hstat.contlen);
1802 printwhat (count, opt.ntry);
1806 else /* now hstat.res can only be -1 */
1808 if (hstat.contlen == -1)
1810 logprintf (LOG_VERBOSE,
1811 _("%s (%s) - Read error at byte %ld (%s)."),
1812 tms, tmrate, hstat.len, strerror (errno));
1813 printwhat (count, opt.ntry);
1816 else /* hstat.res == -1 and contlen is given */
1818 logprintf (LOG_VERBOSE,
1819 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1820 tms, tmrate, hstat.len, hstat.contlen,
1822 printwhat (count, opt.ntry);
1829 while (!opt.ntry || (count < opt.ntry));
1833 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1834 than local timezone.
1836 mktime is similar but assumes struct tm, also known as the
1837 "broken-down" form of time, is in local time zone. mktime_from_utc
1838 uses mktime to make the conversion understanding that an offset
1839 will be introduced by the local time assumption.
1841 mktime_from_utc then measures the introduced offset by applying
1842 gmtime to the initial result and applying mktime to the resulting
1843 "broken-down" form. The difference between the two mktime results
1844 is the measured offset which is then subtracted from the initial
1845 mktime result to yield a calendar time which is the value returned.
1847 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1848 consistent offset (the non DST offset) since tm and tm+o might be
1849 on opposite sides of a DST change.
1851 Some implementations of mktime return -1 for the nonexistent
1852 localtime hour at the beginning of DST. In this event, use
1853 mktime(tm - 1hr) + 3600.
1857 gmtime(t+o) --> tm+o
1858 mktime(tm+o) --> t+2o
1859 t+o - (t+2o - t+o) = t
1861 Note that glibc contains a function of the same purpose named
1862 `timegm' (reverse of gmtime). But obviously, it is not universally
1863 available, and unfortunately it is not straightforwardly
1864 extractable for use here. Perhaps configure should detect timegm
1865 and use it where available.
1867 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1868 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1869 Further improved by Roger with assistance from Edward J. Sabol
1870 based on input by Jamie Zawinski. */
1873 mktime_from_utc (struct tm *t)
1884 return -1; /* can't deal with output from strptime */
1895 return -1; /* can't deal with output from gmtime */
1898 return (tl - (tb - tl));
1901 /* Check whether the result of strptime() indicates success.
1902 strptime() returns the pointer to how far it got to in the string.
1903 The processing has been successful if the string is at `GMT' or
1904 `+X', or at the end of the string.
1906 In extended regexp parlance, the function returns 1 if P matches
1907 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1908 can return) is considered a failure and 0 is returned. */
1910 check_end (const char *p)
1914 while (ISSPACE (*p))
1917 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1918 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1924 /* Convert the textual specification of time in TIME_STRING to the
1925 number of seconds since the Epoch.
1927 TIME_STRING can be in any of the three formats RFC2068 allows the
1928 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1929 Timezones are ignored, and should be GMT.
1931 Return the computed time_t representation, or -1 if the conversion
1934 This function uses strptime with various string formats for parsing
1935 TIME_STRING. This results in a parser that is not as lenient in
1936 interpreting TIME_STRING as I would like it to be. Being based on
1937 strptime, it always allows shortened months, one-digit days, etc.,
1938 but due to the multitude of formats in which time can be
1939 represented, an ideal HTTP time parser would be even more
1940 forgiving. It should completely ignore things like week days and
1941 concentrate only on the various forms of representing years,
1942 months, days, hours, minutes, and seconds. For example, it would
1943 be nice if it accepted ISO 8601 out of the box.
1945 I've investigated free and PD code for this purpose, but none was
1946 usable. getdate was big and unwieldy, and had potential copyright
1947 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1948 distributed with phttpd, is excellent, but we cannot use it because
1949 it is not assigned to the FSF. So I stuck it with strptime. */
1952 http_atotm (char *time_string)
1954 /* NOTE: Solaris strptime man page claims that %n and %t match white
1955 space, but that's not universally available. Instead, we simply
1956 use ` ' to mean "skip all WS", which works under all strptime
1957 implementations I've tested. */
1959 static const char *time_formats[] = {
1960 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1961 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1962 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1963 (google.com uses this for their cookies.) */
1964 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1970 /* According to Roger Beeman, we need to initialize tm_isdst, since
1971 strptime won't do it. */
1974 /* Note that under foreign locales Solaris strptime() fails to
1975 recognize English dates, which renders this function useless. We
1976 solve this by being careful not to affect LC_TIME when
1977 initializing locale.
1979 Another solution would be to temporarily set locale to C, invoke
1980 strptime(), and restore it back. This is slow and dirty,
1981 however, and locale support other than LC_MESSAGES can mess other
1982 things, so I rather chose to stick with just setting LC_MESSAGES.
1984 GNU strptime does not have this problem because it recognizes
1985 both international and local dates. */
1987 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1988 if (check_end (strptime (time_string, time_formats[i], &t)))
1989 return mktime_from_utc (&t);
1991 /* All formats have failed. */
1995 /* Authorization support: We support two authorization schemes:
1997 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1999 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2000 consisting of answering to the server's challenge with the proper
2003 /* How many bytes it will take to store LEN bytes in base64. */
2004 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2006 /* Encode the string S of length LENGTH to base64 format and place it
2007 to STORE. STORE will be 0-terminated, and must point to a writable
2008 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2010 base64_encode (const char *s, char *store, int length)
2012 /* Conversion table. */
2013 static char tbl[64] = {
2014 'A','B','C','D','E','F','G','H',
2015 'I','J','K','L','M','N','O','P',
2016 'Q','R','S','T','U','V','W','X',
2017 'Y','Z','a','b','c','d','e','f',
2018 'g','h','i','j','k','l','m','n',
2019 'o','p','q','r','s','t','u','v',
2020 'w','x','y','z','0','1','2','3',
2021 '4','5','6','7','8','9','+','/'
2024 unsigned char *p = (unsigned char *)store;
2026 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2027 for (i = 0; i < length; i += 3)
2029 *p++ = tbl[s[0] >> 2];
2030 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2031 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2032 *p++ = tbl[s[2] & 0x3f];
2035 /* Pad the result if necessary... */
2036 if (i == length + 1)
2038 else if (i == length + 2)
2039 *(p - 1) = *(p - 2) = '=';
2040 /* ...and zero-terminate it. */
2044 /* Create the authentication header contents for the `Basic' scheme.
2045 This is done by encoding the string `USER:PASS' in base64 and
2046 prepending `HEADER: Basic ' to it. */
2048 basic_authentication_encode (const char *user, const char *passwd,
2051 char *t1, *t2, *res;
2052 int len1 = strlen (user) + 1 + strlen (passwd);
2053 int len2 = BASE64_LENGTH (len1);
2055 t1 = (char *)alloca (len1 + 1);
2056 sprintf (t1, "%s:%s", user, passwd);
2057 t2 = (char *)alloca (1 + len2);
2058 base64_encode (t1, t2, len1);
2059 res = (char *)xmalloc (len2 + 11 + strlen (header));
2060 sprintf (res, "%s: Basic %s\r\n", header, t2);
2066 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2067 of a field in such a header. If the field is the one specified by
2068 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2069 digest authorization code), extract its value in the (char*)
2070 variable pointed by RET. Returns negative on a malformed header,
2071 or number of bytes that have been parsed by this call. */
2073 extract_header_attr (const char *au, const char *attr_name, char **ret)
2075 const char *cp, *ep;
2079 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2081 cp += strlen (attr_name);
2084 cp += skip_lws (cp);
2089 cp += skip_lws (cp);
2094 for (ep = cp; *ep && *ep != '\"'; ep++)
2099 *ret = strdupdelim (cp, ep);
2106 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2107 an array of 16 bytes containing the hash keys, and BUF should be a
2108 buffer of 33 writable characters (32 for hex digits plus one for
2109 zero termination). */
2111 dump_hash (unsigned char *buf, const unsigned char *hash)
2115 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2117 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2118 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2123 /* Take the line apart to find the challenge, and compose a digest
2124 authorization header. See RFC2069 section 2.1.2. */
2126 digest_authentication_encode (const char *au, const char *user,
2127 const char *passwd, const char *method,
2130 static char *realm, *opaque, *nonce;
2135 { "realm", &realm },
2136 { "opaque", &opaque },
2141 realm = opaque = nonce = NULL;
2143 au += 6; /* skip over `Digest' */
2148 au += skip_lws (au);
2149 for (i = 0; i < ARRAY_SIZE (options); i++)
2151 int skip = extract_header_attr (au, options[i].name,
2152 options[i].variable);
2156 FREE_MAYBE (opaque);
2166 if (i == ARRAY_SIZE (options))
2168 while (*au && *au != '=')
2172 au += skip_lws (au);
2176 while (*au && *au != '\"')
2183 while (*au && *au != ',')
2188 if (!realm || !nonce || !user || !passwd || !path || !method)
2191 FREE_MAYBE (opaque);
2196 /* Calculate the digest value. */
2199 unsigned char hash[MD5_HASHLEN];
2200 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2201 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2203 /* A1BUF = H(user ":" realm ":" password) */
2204 md5_init_ctx (&ctx);
2205 md5_process_bytes (user, strlen (user), &ctx);
2206 md5_process_bytes (":", 1, &ctx);
2207 md5_process_bytes (realm, strlen (realm), &ctx);
2208 md5_process_bytes (":", 1, &ctx);
2209 md5_process_bytes (passwd, strlen (passwd), &ctx);
2210 md5_finish_ctx (&ctx, hash);
2211 dump_hash (a1buf, hash);
2213 /* A2BUF = H(method ":" path) */
2214 md5_init_ctx (&ctx);
2215 md5_process_bytes (method, strlen (method), &ctx);
2216 md5_process_bytes (":", 1, &ctx);
2217 md5_process_bytes (path, strlen (path), &ctx);
2218 md5_finish_ctx (&ctx, hash);
2219 dump_hash (a2buf, hash);
2221 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2222 md5_init_ctx (&ctx);
2223 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2224 md5_process_bytes (":", 1, &ctx);
2225 md5_process_bytes (nonce, strlen (nonce), &ctx);
2226 md5_process_bytes (":", 1, &ctx);
2227 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2228 md5_finish_ctx (&ctx, hash);
2229 dump_hash (response_digest, hash);
2231 res = (char*) xmalloc (strlen (user)
2236 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2237 + (opaque ? strlen (opaque) : 0)
2239 sprintf (res, "Authorization: Digest \
2240 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2241 user, realm, nonce, path, response_digest);
2244 char *p = res + strlen (res);
2245 strcat (p, ", opaque=\"");
2249 strcat (res, "\r\n");
2253 #endif /* USE_DIGEST */
2256 #define BEGINS_WITH(line, string_constant) \
2257 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2258 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2259 || !line[sizeof (string_constant) - 1]))
2262 known_authentication_scheme_p (const char *au)
2264 return BEGINS_WITH (au, "Basic")
2265 || BEGINS_WITH (au, "Digest")
2266 || BEGINS_WITH (au, "NTLM");
2271 /* Create the HTTP authorization request header. When the
2272 `WWW-Authenticate' response header is seen, according to the
2273 authorization scheme specified in that header (`Basic' and `Digest'
2274 are supported by the current implementation), produce an
2275 appropriate HTTP authorization request header. */
2277 create_authorization_line (const char *au, const char *user,
2278 const char *passwd, const char *method,
2281 char *wwwauth = NULL;
2283 if (!strncasecmp (au, "Basic", 5))
2284 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2285 if (!strncasecmp (au, "NTLM", 4))
2286 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2288 else if (!strncasecmp (au, "Digest", 6))
2289 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2290 #endif /* USE_DIGEST */