2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
25 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
50 # include <netdb.h> /* for h_errno */
64 # include "gen_sslfunc.h"
68 extern char *version_string;
79 static int cookies_loaded_p;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 hdr += skip_lws (hdr);
211 for (num = 0; ISDIGIT (*hdr); hdr++)
212 num = 10 * num + (*hdr - '0');
213 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
215 closure->first_byte_pos = num;
217 for (num = 0; ISDIGIT (*hdr); hdr++)
218 num = 10 * num + (*hdr - '0');
219 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
221 closure->last_byte_pos = num;
223 for (num = 0; ISDIGIT (*hdr); hdr++)
224 num = 10 * num + (*hdr - '0');
225 closure->entity_length = num;
229 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
230 Used for `Accept-Ranges'. */
232 http_process_none (const char *hdr, void *arg)
234 int *where = (int *)arg;
236 if (strstr (hdr, "none"))
243 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
245 http_process_type (const char *hdr, void *arg)
247 char **result = (char **)arg;
248 /* Locate P on `;' or the terminating zero, whichever comes first. */
249 const char *p = strchr (hdr, ';');
251 p = hdr + strlen (hdr);
252 while (p > hdr && ISSPACE (*(p - 1)))
254 *result = strdupdelim (hdr, p);
258 /* Check whether the `Connection' header is set to "keep-alive". */
260 http_process_connection (const char *hdr, void *arg)
262 int *flag = (int *)arg;
263 if (!strcasecmp (hdr, "Keep-Alive"))
268 /* Persistent connections. Currently, we cache the most recently used
269 connection as persistent, provided that the HTTP server agrees to
270 make it such. The persistence data is stored in the variables
271 below. Ideally, it would be in a structure, and it should be
272 possible to cache an arbitrary fixed number of these connections.
274 I think the code is quite easy to extend in that direction. */
276 /* Whether a persistent connection is active. */
277 static int pc_active_p;
278 /* Host and port of currently active persistent connection. */
279 static unsigned char pc_last_host[4];
280 static unsigned short pc_last_port;
282 /* File descriptor of the currently active persistent connection. */
283 static int pc_last_fd;
286 /* Whether a ssl handshake has occoured on this connection */
287 static int pc_active_ssl;
288 /* SSL connection of the currently active persistent connection. */
289 static SSL *pc_last_ssl;
290 #endif /* HAVE_SSL */
292 /* Mark the persistent connection as invalid. This is used by the
293 CLOSE_* macros after they forcefully close a registered persistent
294 connection. This does not close the file descriptor -- it is left
295 to the caller to do that. (Maybe it should, though.) */
298 invalidate_persistent (void)
303 #endif /* HAVE_SSL */
304 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
307 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
308 persistent. This will enable someone to use the same connection
309 later. In the context of HTTP, this must be called only AFTER the
310 response has been received and the server has promised that the
311 connection will remain alive.
313 If a previous connection was persistent, it is closed. */
316 register_persistent (const char *host, unsigned short port, int fd
326 if (pc_last_fd == fd)
328 /* The connection FD is already registered. Nothing to
334 /* The old persistent connection is still active; let's
335 close it first. This situation arises whenever a
336 persistent connection exists, but we then connect to a
337 different host, and try to register a persistent
338 connection to that one. */
340 /* The ssl disconnect has to take place before the closing
343 shutdown_ssl(pc_last_ssl);
346 invalidate_persistent ();
350 /* This store_hostaddress may not fail, because it has the results
352 success = store_hostaddress (pc_last_host, host);
359 pc_active_ssl = ssl ? 1 : 0;
361 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
364 /* Return non-zero if a persistent connection is available for
365 connecting to HOST:PORT. */
368 persistent_available_p (const char *host, unsigned short port
374 unsigned char this_host[4];
375 /* First, check whether a persistent connection is active at all. */
378 /* Second, check if the active connection pertains to the correct
379 (HOST, PORT) ordered pair. */
380 if (port != pc_last_port)
383 /* Second, a): check if current connection is (not) ssl, too. This
384 test is unlikely to fail because HTTP and HTTPS typicaly use
385 different ports. Yet it is possible, or so I [Christian
386 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
388 if (ssl != pc_active_ssl)
390 #endif /* HAVE_SSL */
391 if (!store_hostaddress (this_host, host))
393 if (memcmp (pc_last_host, this_host, 4))
395 /* Third: check whether the connection is still open. This is
396 important because most server implement a liberal (short) timeout
397 on persistent connections. Wget can of course always reconnect
398 if the connection doesn't work out, but it's nicer to know in
399 advance. This test is a logical followup of the first test, but
400 is "expensive" and therefore placed at the end of the list. */
401 if (!test_socket_open (pc_last_fd))
403 /* Oops, the socket is no longer open. Now that we know that,
404 let's invalidate the persistent connection before returning
407 invalidate_persistent ();
414 # define SHUTDOWN_SSL(ssl) do { \
416 shutdown_ssl (ssl); \
419 # define SHUTDOWN_SSL(ssl)
422 /* The idea behind these two CLOSE macros is to distinguish between
423 two cases: one when the job we've been doing is finished, and we
424 want to close the connection and leave, and two when something is
425 seriously wrong and we're closing the connection as part of
428 In case of keep_alive, CLOSE_FINISH should leave the connection
429 open, while CLOSE_INVALIDATE should still close it.
431 Note that the semantics of the flag `keep_alive' is "this
432 connection *will* be reused (the server has promised not to close
433 the connection once we're done)", while the semantics of
434 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
435 active, registered connection". */
437 #define CLOSE_FINISH(fd) do { \
440 SHUTDOWN_SSL (ssl); \
442 if (pc_active_p && (fd) == pc_last_fd) \
443 invalidate_persistent (); \
447 #define CLOSE_INVALIDATE(fd) do { \
448 SHUTDOWN_SSL (ssl); \
450 if (pc_active_p && (fd) == pc_last_fd) \
451 invalidate_persistent (); \
456 long len; /* received length */
457 long contlen; /* expected length */
458 long restval; /* the restart value */
459 int res; /* the result of last read */
460 char *newloc; /* new location (redirection) */
461 char *remote_time; /* remote time-stamp string */
462 char *error; /* textual HTTP error */
463 int statcode; /* status code */
464 long dltime; /* time of the download */
465 int no_truncate; /* whether truncating the file is
469 /* Free the elements of hstat X. */
470 #define FREEHSTAT(x) do \
472 FREE_MAYBE ((x).newloc); \
473 FREE_MAYBE ((x).remote_time); \
474 FREE_MAYBE ((x).error); \
475 (x).newloc = (x).remote_time = (x).error = NULL; \
478 static char *create_authorization_line PARAMS ((const char *, const char *,
479 const char *, const char *,
481 static char *basic_authentication_encode PARAMS ((const char *, const char *,
483 static int known_authentication_scheme_p PARAMS ((const char *));
485 time_t http_atotm PARAMS ((char *));
487 #define BEGINS_WITH(line, string_constant) \
488 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
489 && (ISSPACE (line[sizeof (string_constant) - 1]) \
490 || !line[sizeof (string_constant) - 1]))
492 /* Retrieve a document through HTTP protocol. It recognizes status
493 code, and correctly handles redirections. It closes the network
494 socket. If it receives an error from the functions below it, it
495 will print it if there is enough information to do so (almost
496 always), returning the error to the caller (i.e. http_loop).
498 Various HTTP parameters are stored to hs. Although it parses the
499 response code correctly, it is not used in a sane way. The caller
502 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
503 and u->proxy->url will be given to the proxy server (bad naming,
506 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
508 char *request, *type, *command, *path;
510 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
511 char *authenticate_h;
515 char *request_keep_alive;
516 int sock, hcount, num_written, all_length, remport, statcode;
517 long contlen, contrange;
521 int auth_tried_already;
524 static SSL_CTX *ssl_ctx = NULL;
526 #endif /* HAVE_SSL */
527 struct wget_timer *timer;
528 char *cookies = NULL;
530 /* Whether this connection will be kept alive after the HTTP request
534 /* Flags that detect the two ways of specifying HTTP keep-alive
536 int http_keep_alive_1, http_keep_alive_2;
538 /* Whether keep-alive should be inhibited. */
539 int inhibit_keep_alive;
542 /* initialize ssl_ctx on first run */
545 err=init_ssl (&ssl_ctx);
550 case SSLERRCTXCREATE:
552 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
556 /* try without certfile */
557 logprintf (LOG_NOTQUIET,
558 _("Failed to load certificates from %s\n"),
561 logprintf (LOG_NOTQUIET,
562 _("Trying without the specified certificate\n"));
565 logprintf (LOG_NOTQUIET,
566 _("Failed to get certificate key from %s\n"),
569 logprintf (LOG_NOTQUIET,
570 _("Trying without the specified certificate\n"));
577 #endif /* HAVE_SSL */
579 if (!(*dt & HEAD_ONLY))
580 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
581 know the local filename so we can save to it. */
582 assert (u->local != NULL);
585 auth_tried_already = 0;
587 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
590 /* We need to come back here when the initial attempt to retrieve
591 without authorization header fails. (Expected to happen at least
592 for the Digest authorization scheme.) */
595 http_keep_alive_1 = http_keep_alive_2 = 0;
597 /* Initialize certain elements of struct http_stat. */
602 hs->remote_time = NULL;
605 /* Which structure to use to retrieve the original URL data. */
611 /* First: establish the connection. */
612 if (inhibit_keep_alive
615 !persistent_available_p (u->host, u->port)
617 !persistent_available_p (u->host, u->port, u->scheme == SCHEME_HTTPS)
618 #endif /* HAVE_SSL */
621 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
622 err = make_connection (&sock, u->host, u->port);
626 logputs (LOG_VERBOSE, "\n");
627 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
631 logputs (LOG_VERBOSE, "\n");
632 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
636 logputs (LOG_VERBOSE, "\n");
637 logprintf (LOG_NOTQUIET,
638 _("Connection to %s:%hu refused.\n"), u->host, u->port);
642 logputs (LOG_VERBOSE, "\n");
643 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
648 /* Everything is fine! */
649 logputs (LOG_VERBOSE, _("connected!\n"));
656 if (u->scheme == SCHEME_HTTPS)
657 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
659 logputs (LOG_VERBOSE, "\n");
660 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
665 #endif /* HAVE_SSL */
669 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
670 /* #### pc_last_fd should be accessed through an accessor
675 #endif /* HAVE_SSL */
676 DEBUGP (("Reusing fd %d.\n", sock));
680 path = u->proxy->url;
684 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
688 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
689 sprintf (referer, "Referer: %s\r\n", ou->referer);
691 if (*dt & SEND_NOCACHE)
692 pragma_h = "Pragma: no-cache\r\n";
697 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
698 /* Gag me! Some servers (e.g. WebSitePro) have been known to
699 respond to the following `Range' format by generating a
700 multipart/x-byte-ranges MIME document! This MIME type was
701 present in an old draft of the byteranges specification.
702 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
703 only if multiple non-overlapping ranges are requested --
704 which Wget never does. */
705 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
710 STRDUP_ALLOCA (useragent, opt.useragent);
713 useragent = (char *)alloca (10 + strlen (version_string));
714 sprintf (useragent, "Wget/%s", version_string);
716 /* Construct the authentication, if userid is present. */
719 search_netrc (ou->host, (const char **)&user, (const char **)&passwd, 0);
720 user = user ? user : opt.http_user;
721 passwd = passwd ? passwd : opt.http_passwd;
728 /* We have the username and the password, but haven't tried
729 any authorization yet. Let's see if the "Basic" method
730 works. If not, we'll come back here and construct a
731 proper authorization method with the right challenges.
733 If we didn't employ this kind of logic, every URL that
734 requires authorization would have to be processed twice,
735 which is very suboptimal and generates a bunch of false
736 "unauthorized" errors in the server log.
738 #### But this logic also has a serious problem when used
739 with stronger authentications: we *first* transmit the
740 username and the password in clear text, and *then*
741 attempt a stronger authentication scheme. That cannot be
742 right! We are only fortunate that almost everyone still
743 uses the `Basic' scheme anyway.
745 There should be an option to prevent this from happening,
746 for those who use strong authentication schemes and value
748 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
752 wwwauth = create_authorization_line (authenticate_h, user, passwd,
760 char *proxy_user, *proxy_passwd;
761 /* For normal username and password, URL components override
762 command-line/wgetrc parameters. With proxy authentication,
763 it's the reverse, because proxy URLs are normally the
764 "permanent" ones, so command-line args should take
766 if (opt.proxy_user && opt.proxy_passwd)
768 proxy_user = opt.proxy_user;
769 proxy_passwd = opt.proxy_passwd;
773 proxy_user = u->user;
774 proxy_passwd = u->passwd;
776 /* #### This is junky. Can't the proxy request, say, `Digest'
778 if (proxy_user && proxy_passwd)
779 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
780 "Proxy-Authorization");
785 /* String of the form :PORT. Used only for non-standard ports. */
789 && remport != (u->scheme == SCHEME_HTTPS
790 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
792 && remport != DEFAULT_HTTP_PORT
796 port_maybe = (char *)alloca (numdigit (remport) + 2);
797 sprintf (port_maybe, ":%d", remport);
800 if (!inhibit_keep_alive)
801 request_keep_alive = "Connection: Keep-Alive\r\n";
803 request_keep_alive = NULL;
806 cookies = build_cookies_request (ou->host, ou->port, ou->path,
808 ou->scheme == SCHEME_HTTPS
814 /* Allocate the memory for the request. */
815 request = (char *)alloca (strlen (command) + strlen (path)
818 + (port_maybe ? strlen (port_maybe) : 0)
819 + strlen (HTTP_ACCEPT)
820 + (request_keep_alive
821 ? strlen (request_keep_alive) : 0)
822 + (referer ? strlen (referer) : 0)
823 + (cookies ? strlen (cookies) : 0)
824 + (wwwauth ? strlen (wwwauth) : 0)
825 + (proxyauth ? strlen (proxyauth) : 0)
826 + (range ? strlen (range) : 0)
828 + (opt.user_header ? strlen (opt.user_header) : 0)
830 /* Construct the request. */
836 %s%s%s%s%s%s%s%s\r\n",
837 command, path, useragent, remhost,
838 port_maybe ? port_maybe : "",
840 request_keep_alive ? request_keep_alive : "",
841 referer ? referer : "",
842 cookies ? cookies : "",
843 wwwauth ? wwwauth : "",
844 proxyauth ? proxyauth : "",
847 opt.user_header ? opt.user_header : "");
848 DEBUGP (("---request begin---\n%s---request end---\n", request));
849 /* Free the temporary memory. */
850 FREE_MAYBE (wwwauth);
851 FREE_MAYBE (proxyauth);
852 FREE_MAYBE (cookies);
854 /* Send the request to server. */
856 if (u->scheme == SCHEME_HTTPS)
857 num_written = ssl_iwrite (ssl, request, strlen (request));
859 #endif /* HAVE_SSL */
860 num_written = iwrite (sock, request, strlen (request));
864 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
866 CLOSE_INVALIDATE (sock);
869 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
870 u->proxy ? "Proxy" : "HTTP");
871 contlen = contrange = -1;
876 /* Before reading anything, initialize the rbuf. */
877 rbuf_initialize (&rbuf, sock);
879 if (u->scheme == SCHEME_HTTPS)
883 #endif /* HAVE_SSL */
886 /* Header-fetching loop. */
894 /* Get the header. */
895 status = header_get (&rbuf, &hdr,
896 /* Disallow continuations for status line. */
897 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
899 /* Check for errors. */
900 if (status == HG_EOF && *hdr)
902 /* This used to be an unconditional error, but that was
903 somewhat controversial, because of a large number of
904 broken CGI's that happily "forget" to send the second EOL
905 before closing the connection of a HEAD request.
907 So, the deal is to check whether the header is empty
908 (*hdr is zero if it is); if yes, it means that the
909 previous header was fully retrieved, and that -- most
910 probably -- the request is complete. "...be liberal in
911 what you accept." Oh boy. */
912 logputs (LOG_VERBOSE, "\n");
913 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
916 FREE_MAYBE (all_headers);
917 CLOSE_INVALIDATE (sock);
920 else if (status == HG_ERROR)
922 logputs (LOG_VERBOSE, "\n");
923 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
927 FREE_MAYBE (all_headers);
928 CLOSE_INVALIDATE (sock);
932 /* If the headers are to be saved to a file later, save them to
934 if (opt.save_headers)
936 int lh = strlen (hdr);
937 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
938 memcpy (all_headers + all_length, hdr, lh);
940 all_headers[all_length++] = '\n';
941 all_headers[all_length] = '\0';
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);
975 if (opt.server_response)
976 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
978 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
984 /* Exit on empty header. */
991 /* Print the header if requested. */
992 if (opt.server_response && hcount != 1)
993 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
995 /* Try getting content-length. */
996 if (contlen == -1 && !opt.ignore_length)
997 if (header_process (hdr, "Content-Length", header_extract_number,
1000 /* Try getting content-type. */
1002 if (header_process (hdr, "Content-Type", http_process_type, &type))
1004 /* Try getting location. */
1006 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1008 /* Try getting last-modified. */
1009 if (!hs->remote_time)
1010 if (header_process (hdr, "Last-Modified", header_strdup,
1013 /* Try getting cookies. */
1015 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1017 /* Try getting www-authentication. */
1018 if (!authenticate_h)
1019 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1022 /* Check for accept-ranges header. If it contains the word
1023 `none', disable the ranges. */
1024 if (*dt & ACCEPTRANGES)
1027 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1030 *dt &= ~ACCEPTRANGES;
1034 /* Try getting content-range. */
1035 if (contrange == -1)
1037 struct http_process_range_closure closure;
1038 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1040 contrange = closure.first_byte_pos;
1044 /* Check for keep-alive related responses. */
1045 if (!inhibit_keep_alive)
1047 /* Check for the `Keep-Alive' header. */
1048 if (!http_keep_alive_1)
1050 if (header_process (hdr, "Keep-Alive", header_exists,
1051 &http_keep_alive_1))
1054 /* Check for `Connection: Keep-Alive'. */
1055 if (!http_keep_alive_2)
1057 if (header_process (hdr, "Connection", http_process_connection,
1058 &http_keep_alive_2))
1066 logputs (LOG_VERBOSE, "\n");
1069 && (http_keep_alive_1 || http_keep_alive_2))
1071 assert (inhibit_keep_alive == 0);
1075 /* The server has promised that it will not close the connection
1076 when we're done. This means that we can register it. */
1078 register_persistent (u->host, u->port, sock);
1080 register_persistent (u->host, u->port, sock, ssl);
1081 #endif /* HAVE_SSL */
1083 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1086 /* Authorization is required. */
1090 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1091 might be more bytes in the body. */
1092 if (auth_tried_already)
1094 /* If we have tried it already, then there is not point
1097 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1098 xfree (authenticate_h);
1101 else if (!known_authentication_scheme_p (authenticate_h))
1103 xfree (authenticate_h);
1104 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1107 else if (BEGINS_WITH (authenticate_h, "Basic"))
1109 /* The authentication scheme is basic, the one we try by
1110 default, and it failed. There's no sense in trying
1116 auth_tried_already = 1;
1120 /* We do not need this anymore. */
1123 xfree (authenticate_h);
1124 authenticate_h = NULL;
1127 /* 20x responses are counted among successful by default. */
1128 if (H_20X (statcode))
1131 /* Return if redirected. */
1132 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1134 /* RFC2068 says that in case of the 300 (multiple choices)
1135 response, the server can output a preferred URL through
1136 `Location' header; otherwise, the request should be treated
1137 like GET. So, if the location is set, it will be a
1138 redirection; otherwise, just proceed normally. */
1139 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1143 logprintf (LOG_VERBOSE,
1144 _("Location: %s%s\n"),
1145 hs->newloc ? hs->newloc : _("unspecified"),
1146 hs->newloc ? _(" [following]") : "");
1147 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1148 might be more bytes in the body. */
1150 FREE_MAYBE (all_headers);
1155 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1158 /* We don't assume text/html by default. */
1161 if (opt.html_extension && (*dt & TEXTHTML))
1162 /* -E / --html-extension / html_extension = on was specified, and this is a
1163 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1164 already the file's suffix, tack on ".html". */
1166 char* last_period_in_local_filename = strrchr(u->local, '.');
1168 if (last_period_in_local_filename == NULL ||
1169 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1170 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1172 size_t local_filename_len = strlen(u->local);
1174 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1175 strcpy(u->local + local_filename_len, ".html");
1177 *dt |= ADDED_HTML_EXTENSION;
1181 if (contrange == -1)
1183 /* We did not get a content-range header. This means that the
1184 server did not honor our `Range' request. Normally, this
1185 means we should reset hs->restval and continue normally. */
1187 /* However, if `-c' is used, we need to be a bit more careful:
1189 1. If `-c' is specified and the file already existed when
1190 Wget was started, it would be a bad idea for us to start
1191 downloading it from scratch, effectively truncating it. I
1192 believe this cannot happen unless `-c' was specified.
1194 2. If `-c' is used on a file that is already fully
1195 downloaded, we're requesting bytes after the end of file,
1196 which can result in server not honoring `Range'. If this is
1197 the case, `Content-Length' will be equal to the length of the
1199 if (opt.always_rest)
1201 /* Check for condition #2. */
1202 if (hs->restval > 0 /* restart was requested. */
1203 && contlen != -1 /* we got content-length. */
1204 && hs->restval >= contlen /* file fully downloaded
1208 logputs (LOG_VERBOSE, _("\
1209 \n The file is already fully retrieved; nothing to do.\n\n"));
1210 /* In case the caller inspects. */
1214 FREE_MAYBE (all_headers);
1215 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1216 might be more bytes in the body. */
1217 return RETRUNNEEDED;
1220 /* Check for condition #1. */
1221 if (hs->no_truncate)
1223 logprintf (LOG_NOTQUIET,
1226 Continued download failed on this file, which conflicts with `-c'.\n\
1227 Refusing to truncate existing file `%s'.\n\n"), u->local);
1229 FREE_MAYBE (all_headers);
1230 CLOSE_INVALIDATE (sock);
1231 return CONTNOTSUPPORTED;
1239 else if (contrange != hs->restval ||
1240 (H_PARTIAL (statcode) && contrange == -1))
1242 /* This means the whole request was somehow misunderstood by the
1243 server. Bail out. */
1245 FREE_MAYBE (all_headers);
1246 CLOSE_INVALIDATE (sock);
1253 contlen += contrange;
1255 contrange = -1; /* If conent-length was not sent,
1256 content-range will be ignored. */
1258 hs->contlen = contlen;
1262 if ((*dt & RETROKF) && !opt.server_response)
1264 /* No need to print this output if the body won't be
1265 downloaded at all, or if the original server response is
1267 logputs (LOG_VERBOSE, _("Length: "));
1270 logputs (LOG_VERBOSE, legible (contlen));
1271 if (contrange != -1)
1272 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1273 legible (contlen - contrange));
1276 logputs (LOG_VERBOSE,
1277 opt.ignore_length ? _("ignored") : _("unspecified"));
1279 logprintf (LOG_VERBOSE, " [%s]\n", type);
1281 logputs (LOG_VERBOSE, "\n");
1285 type = NULL; /* We don't need it any more. */
1287 /* Return if we have no intention of further downloading. */
1288 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1290 /* In case the caller cares to look... */
1294 FREE_MAYBE (all_headers);
1295 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1296 might be more bytes in the body. */
1297 return RETRFINISHED;
1300 /* Open the local file. */
1303 mkalldirs (u->local);
1305 rotate_backups (u->local);
1306 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1309 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1310 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1311 might be more bytes in the body. */
1312 FREE_MAYBE (all_headers);
1318 extern int global_download_count;
1320 /* To ensure that repeated "from scratch" downloads work for -O
1321 files, we rewind the file pointer, unless restval is
1322 non-zero. (This works only when -O is used on regular files,
1323 but it's still a valuable feature.)
1325 However, this loses when more than one URL is specified on
1326 the command line the second rewinds eradicates the contents
1327 of the first download. Thus we disable the above trick for
1328 all the downloads except the very first one.
1330 #### A possible solution to this would be to remember the
1331 file position in the output document and to seek to that
1332 position, instead of rewinding. */
1333 if (!hs->restval && global_download_count == 0)
1335 /* This will silently fail for streams that don't correspond
1336 to regular files, but that's OK. */
1338 /* ftruncate is needed because opt.dfp is opened in append
1339 mode if opt.always_rest is set. */
1340 ftruncate (fileno (fp), 0);
1345 /* #### This confuses the code that checks for file size. There
1346 should be some overhead information. */
1347 if (opt.save_headers)
1348 fwrite (all_headers, 1, all_length, fp);
1349 timer = wtimer_new ();
1350 /* Get the contents of the document. */
1351 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1352 (contlen != -1 ? contlen : 0),
1354 hs->dltime = wtimer_elapsed (timer);
1355 wtimer_delete (timer);
1357 /* Close or flush the file. We have to be careful to check for
1358 error here. Checking the result of fwrite() is not enough --
1359 errors could go unnoticed! */
1362 flush_res = fclose (fp);
1364 flush_res = fflush (fp);
1365 if (flush_res == EOF)
1368 FREE_MAYBE (all_headers);
1369 CLOSE_FINISH (sock);
1372 return RETRFINISHED;
1375 /* The genuine HTTP loop! This is the part where the retrieval is
1376 retried, and retried, and retried, and... */
1378 http_loop (struct urlinfo *u, char **newloc, int *dt)
1381 int use_ts, got_head = 0; /* time-stamping info */
1382 char *filename_plus_orig_suffix;
1383 char *local_filename = NULL;
1384 char *tms, *suf, *locf, *tmrate;
1386 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1387 long local_size = 0; /* the size of the local file */
1388 size_t filename_len;
1389 struct http_stat hstat; /* HTTP status */
1392 /* This used to be done in main(), but it's a better idea to do it
1393 here so that we don't go through the hoops if we're just using
1395 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1397 load_cookies (opt.cookies_input);
1398 cookies_loaded_p = 1;
1403 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1404 has_wildcards_p because it would also warn on `?', and we know that
1405 shows up in CGI paths a *lot*. */
1406 if (strchr (u->url, '*'))
1407 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1409 /* Determine the local filename. */
1411 u->local = url_filename (u->proxy ? u->proxy : u);
1413 if (!opt.output_document)
1416 locf = opt.output_document;
1418 filename_len = strlen (u->local);
1419 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1421 if (opt.noclobber && file_exists_p (u->local))
1423 /* If opt.noclobber is turned on and file already exists, do not
1424 retrieve the file */
1425 logprintf (LOG_VERBOSE, _("\
1426 File `%s' already there, will not retrieve.\n"), u->local);
1427 /* If the file is there, we suppose it's retrieved OK. */
1430 /* #### Bogusness alert. */
1431 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1432 text/html, a harmless lie. */
1433 if (((suf = suffix (u->local)) != NULL)
1434 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1437 /* Another harmless lie: */
1442 if (opt.timestamping)
1444 boolean local_dot_orig_file_exists = FALSE;
1446 if (opt.backup_converted)
1447 /* If -K is specified, we'll act on the assumption that it was specified
1448 last time these files were downloaded as well, and instead of just
1449 comparing local file X against server file X, we'll compare local
1450 file X.orig (if extant, else X) against server file X. If -K
1451 _wasn't_ specified last time, or the server contains files called
1452 *.orig, -N will be back to not operating correctly with -k. */
1454 /* Would a single s[n]printf() call be faster? --dan
1456 Definitely not. sprintf() is horribly slow. It's a
1457 different question whether the difference between the two
1458 affects a program. Usually I'd say "no", but at one
1459 point I profiled Wget, and found that a measurable and
1460 non-negligible amount of time was lost calling sprintf()
1461 in url.c. Replacing sprintf with inline calls to
1462 strcpy() and long_to_string() made a difference.
1464 memcpy (filename_plus_orig_suffix, u->local, filename_len);
1465 memcpy (filename_plus_orig_suffix + filename_len,
1466 ".orig", sizeof (".orig"));
1468 /* Try to stat() the .orig file. */
1469 if (stat (filename_plus_orig_suffix, &st) == 0)
1471 local_dot_orig_file_exists = TRUE;
1472 local_filename = filename_plus_orig_suffix;
1476 if (!local_dot_orig_file_exists)
1477 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1478 if (stat (u->local, &st) == 0)
1479 local_filename = u->local;
1481 if (local_filename != NULL)
1482 /* There was a local file, so we'll check later to see if the version
1483 the server has is the same version we already have, allowing us to
1488 local_size = st.st_size;
1492 /* Reset the counter. */
1494 *dt = 0 | ACCEPTRANGES;
1498 /* Increment the pass counter. */
1500 sleep_between_retrievals (count);
1501 /* Get the current time string. */
1502 tms = time_str (NULL);
1503 /* Print fetch message, if opt.verbose. */
1506 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1510 sprintf (tmp, _("(try:%2d)"), count);
1511 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1512 tms, hurl, tmp, locf);
1514 ws_changetitle (hurl, 1);
1519 /* Default document type is empty. However, if spider mode is
1520 on or time-stamping is employed, HEAD_ONLY commands is
1521 encoded within *dt. */
1522 if (opt.spider || (use_ts && !got_head))
1526 /* Assume no restarting. */
1528 /* Decide whether or not to restart. */
1529 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1530 /* #### this calls access() and then stat(); could be optimized. */
1531 && file_exists_p (locf))
1532 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1533 hstat.restval = st.st_size;
1535 /* In `-c' is used and the file is existing and non-empty,
1536 refuse to truncate it if the server doesn't support continued
1538 hstat.no_truncate = 0;
1539 if (opt.always_rest && hstat.restval)
1540 hstat.no_truncate = 1;
1542 /* Decide whether to send the no-cache directive. We send it in
1544 a) we're using a proxy, and we're past our first retrieval.
1545 Some proxies are notorious for caching incomplete data, so
1546 we require a fresh get.
1547 b) caching is explicitly inhibited. */
1548 if ((u->proxy && count > 1) /* a */
1549 || !opt.allow_cache /* b */
1551 *dt |= SEND_NOCACHE;
1553 *dt &= ~SEND_NOCACHE;
1555 /* Try fetching the document, or at least its head. :-) */
1556 err = gethttp (u, &hstat, dt);
1558 /* It's unfortunate that wget determines the local filename before finding
1559 out the Content-Type of the file. Barring a major restructuring of the
1560 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1561 u->local to tack on ".html". */
1562 if (!opt.output_document)
1565 locf = opt.output_document;
1568 tms = time_str (NULL);
1569 /* Get the new location (with or without the redirection). */
1571 *newloc = xstrdup (hstat.newloc);
1574 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1575 case CONERROR: case READERR: case WRITEFAILED:
1577 /* Non-fatal errors continue executing the loop, which will
1578 bring them to "while" statement at the end, to judge
1579 whether the number of tries was exceeded. */
1581 printwhat (count, opt.ntry);
1584 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1585 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1586 /* Fatal errors just return from the function. */
1590 case FWRITEERR: case FOPENERR:
1591 /* Another fatal error. */
1592 logputs (LOG_VERBOSE, "\n");
1593 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1594 u->local, strerror (errno));
1599 /* Another fatal error. */
1600 logputs (LOG_VERBOSE, "\n");
1601 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1606 /* Return the new location to the caller. */
1609 logprintf (LOG_NOTQUIET,
1610 _("ERROR: Redirection (%d) without location.\n"),
1618 /* The file was already fully retrieved. */
1623 /* Deal with you later. */
1626 /* All possibilities should have been exhausted. */
1629 if (!(*dt & RETROKF))
1633 /* #### Ugly ugly ugly! */
1634 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1635 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1638 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1639 tms, hstat.statcode, hstat.error);
1640 logputs (LOG_VERBOSE, "\n");
1645 /* Did we get the time-stamp? */
1648 if (opt.timestamping && !hstat.remote_time)
1650 logputs (LOG_NOTQUIET, _("\
1651 Last-modified header missing -- time-stamps turned off.\n"));
1653 else if (hstat.remote_time)
1655 /* Convert the date-string into struct tm. */
1656 tmr = http_atotm (hstat.remote_time);
1657 if (tmr == (time_t) (-1))
1658 logputs (LOG_VERBOSE, _("\
1659 Last-modified header invalid -- time-stamp ignored.\n"));
1663 /* The time-stamping section. */
1668 use_ts = 0; /* no more time-stamping */
1669 count = 0; /* the retrieve count for HEAD is
1671 if (hstat.remote_time && tmr != (time_t) (-1))
1673 /* Now time-stamping can be used validly. Time-stamping
1674 means that if the sizes of the local and remote file
1675 match, and local file is newer than the remote file,
1676 it will not be retrieved. Otherwise, the normal
1677 download procedure is resumed. */
1679 (hstat.contlen == -1 || local_size == hstat.contlen))
1681 logprintf (LOG_VERBOSE, _("\
1682 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1687 else if (tml >= tmr)
1688 logprintf (LOG_VERBOSE, _("\
1689 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1691 logputs (LOG_VERBOSE,
1692 _("Remote file is newer, retrieving.\n"));
1697 if ((tmr != (time_t) (-1))
1699 && ((hstat.len == hstat.contlen) ||
1700 ((hstat.res == 0) &&
1701 ((hstat.contlen == -1) ||
1702 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1704 /* #### This code repeats in http.c and ftp.c. Move it to a
1706 const char *fl = NULL;
1707 if (opt.output_document)
1709 if (opt.od_known_regular)
1710 fl = opt.output_document;
1717 /* End of time-stamping section. */
1721 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1725 /* It is now safe to free the remainder of hstat, since the
1726 strings within it will no longer be used. */
1729 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1731 if (hstat.len == hstat.contlen)
1735 logprintf (LOG_VERBOSE,
1736 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1737 tms, tmrate, locf, hstat.len, hstat.contlen);
1738 logprintf (LOG_NONVERBOSE,
1739 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1740 tms, u->url, hstat.len, hstat.contlen, locf, count);
1743 downloaded_increase (hstat.len);
1745 /* Remember that we downloaded the file for later ".orig" code. */
1746 if (*dt & ADDED_HTML_EXTENSION)
1747 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1749 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1753 else if (hstat.res == 0) /* No read error */
1755 if (hstat.contlen == -1) /* We don't know how much we were supposed
1756 to get, so assume we succeeded. */
1760 logprintf (LOG_VERBOSE,
1761 _("%s (%s) - `%s' saved [%ld]\n\n"),
1762 tms, tmrate, locf, hstat.len);
1763 logprintf (LOG_NONVERBOSE,
1764 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1765 tms, u->url, hstat.len, locf, count);
1768 downloaded_increase (hstat.len);
1770 /* Remember that we downloaded the file for later ".orig" code. */
1771 if (*dt & ADDED_HTML_EXTENSION)
1772 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1774 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1778 else if (hstat.len < hstat.contlen) /* meaning we lost the
1779 connection too soon */
1781 logprintf (LOG_VERBOSE,
1782 _("%s (%s) - Connection closed at byte %ld. "),
1783 tms, tmrate, hstat.len);
1784 printwhat (count, opt.ntry);
1787 else if (!opt.kill_longer) /* meaning we got more than expected */
1789 logprintf (LOG_VERBOSE,
1790 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1791 tms, tmrate, locf, hstat.len, hstat.contlen);
1792 logprintf (LOG_NONVERBOSE,
1793 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1794 tms, u->url, hstat.len, hstat.contlen, locf, count);
1796 downloaded_increase (hstat.len);
1798 /* Remember that we downloaded the file for later ".orig" code. */
1799 if (*dt & ADDED_HTML_EXTENSION)
1800 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1802 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1806 else /* the same, but not accepted */
1808 logprintf (LOG_VERBOSE,
1809 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1810 tms, tmrate, hstat.len, hstat.contlen);
1811 printwhat (count, opt.ntry);
1815 else /* now hstat.res can only be -1 */
1817 if (hstat.contlen == -1)
1819 logprintf (LOG_VERBOSE,
1820 _("%s (%s) - Read error at byte %ld (%s)."),
1821 tms, tmrate, hstat.len, strerror (errno));
1822 printwhat (count, opt.ntry);
1825 else /* hstat.res == -1 and contlen is given */
1827 logprintf (LOG_VERBOSE,
1828 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1829 tms, tmrate, hstat.len, hstat.contlen,
1831 printwhat (count, opt.ntry);
1838 while (!opt.ntry || (count < opt.ntry));
1842 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1843 than local timezone.
1845 mktime is similar but assumes struct tm, also known as the
1846 "broken-down" form of time, is in local time zone. mktime_from_utc
1847 uses mktime to make the conversion understanding that an offset
1848 will be introduced by the local time assumption.
1850 mktime_from_utc then measures the introduced offset by applying
1851 gmtime to the initial result and applying mktime to the resulting
1852 "broken-down" form. The difference between the two mktime results
1853 is the measured offset which is then subtracted from the initial
1854 mktime result to yield a calendar time which is the value returned.
1856 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1857 consistent offset (the non DST offset) since tm and tm+o might be
1858 on opposite sides of a DST change.
1860 Some implementations of mktime return -1 for the nonexistent
1861 localtime hour at the beginning of DST. In this event, use
1862 mktime(tm - 1hr) + 3600.
1866 gmtime(t+o) --> tm+o
1867 mktime(tm+o) --> t+2o
1868 t+o - (t+2o - t+o) = t
1870 Note that glibc contains a function of the same purpose named
1871 `timegm' (reverse of gmtime). But obviously, it is not universally
1872 available, and unfortunately it is not straightforwardly
1873 extractable for use here. Perhaps configure should detect timegm
1874 and use it where available.
1876 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1877 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1878 Further improved by Roger with assistance from Edward J. Sabol
1879 based on input by Jamie Zawinski. */
1882 mktime_from_utc (struct tm *t)
1893 return -1; /* can't deal with output from strptime */
1904 return -1; /* can't deal with output from gmtime */
1907 return (tl - (tb - tl));
1910 /* Check whether the result of strptime() indicates success.
1911 strptime() returns the pointer to how far it got to in the string.
1912 The processing has been successful if the string is at `GMT' or
1913 `+X', or at the end of the string.
1915 In extended regexp parlance, the function returns 1 if P matches
1916 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1917 can return) is considered a failure and 0 is returned. */
1919 check_end (const char *p)
1923 while (ISSPACE (*p))
1926 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1927 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1933 /* Convert the textual specification of time in TIME_STRING to the
1934 number of seconds since the Epoch.
1936 TIME_STRING can be in any of the three formats RFC2068 allows the
1937 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1938 Timezones are ignored, and should be GMT.
1940 Return the computed time_t representation, or -1 if the conversion
1943 This function uses strptime with various string formats for parsing
1944 TIME_STRING. This results in a parser that is not as lenient in
1945 interpreting TIME_STRING as I would like it to be. Being based on
1946 strptime, it always allows shortened months, one-digit days, etc.,
1947 but due to the multitude of formats in which time can be
1948 represented, an ideal HTTP time parser would be even more
1949 forgiving. It should completely ignore things like week days and
1950 concentrate only on the various forms of representing years,
1951 months, days, hours, minutes, and seconds. For example, it would
1952 be nice if it accepted ISO 8601 out of the box.
1954 I've investigated free and PD code for this purpose, but none was
1955 usable. getdate was big and unwieldy, and had potential copyright
1956 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1957 distributed with phttpd, is excellent, but we cannot use it because
1958 it is not assigned to the FSF. So I stuck it with strptime. */
1961 http_atotm (char *time_string)
1963 /* NOTE: Solaris strptime man page claims that %n and %t match white
1964 space, but that's not universally available. Instead, we simply
1965 use ` ' to mean "skip all WS", which works under all strptime
1966 implementations I've tested. */
1968 static const char *time_formats[] = {
1969 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1970 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1971 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1972 (google.com uses this for their cookies.) */
1973 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1979 /* According to Roger Beeman, we need to initialize tm_isdst, since
1980 strptime won't do it. */
1983 /* Note that under foreign locales Solaris strptime() fails to
1984 recognize English dates, which renders this function useless. We
1985 solve this by being careful not to affect LC_TIME when
1986 initializing locale.
1988 Another solution would be to temporarily set locale to C, invoke
1989 strptime(), and restore it back. This is slow and dirty,
1990 however, and locale support other than LC_MESSAGES can mess other
1991 things, so I rather chose to stick with just setting LC_MESSAGES.
1993 GNU strptime does not have this problem because it recognizes
1994 both international and local dates. */
1996 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1997 if (check_end (strptime (time_string, time_formats[i], &t)))
1998 return mktime_from_utc (&t);
2000 /* All formats have failed. */
2004 /* Authorization support: We support two authorization schemes:
2006 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2008 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2009 consisting of answering to the server's challenge with the proper
2012 /* How many bytes it will take to store LEN bytes in base64. */
2013 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2015 /* Encode the string S of length LENGTH to base64 format and place it
2016 to STORE. STORE will be 0-terminated, and must point to a writable
2017 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2019 base64_encode (const char *s, char *store, int length)
2021 /* Conversion table. */
2022 static char tbl[64] = {
2023 'A','B','C','D','E','F','G','H',
2024 'I','J','K','L','M','N','O','P',
2025 'Q','R','S','T','U','V','W','X',
2026 'Y','Z','a','b','c','d','e','f',
2027 'g','h','i','j','k','l','m','n',
2028 'o','p','q','r','s','t','u','v',
2029 'w','x','y','z','0','1','2','3',
2030 '4','5','6','7','8','9','+','/'
2033 unsigned char *p = (unsigned char *)store;
2035 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2036 for (i = 0; i < length; i += 3)
2038 *p++ = tbl[s[0] >> 2];
2039 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2040 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2041 *p++ = tbl[s[2] & 0x3f];
2044 /* Pad the result if necessary... */
2045 if (i == length + 1)
2047 else if (i == length + 2)
2048 *(p - 1) = *(p - 2) = '=';
2049 /* ...and zero-terminate it. */
2053 /* Create the authentication header contents for the `Basic' scheme.
2054 This is done by encoding the string `USER:PASS' in base64 and
2055 prepending `HEADER: Basic ' to it. */
2057 basic_authentication_encode (const char *user, const char *passwd,
2060 char *t1, *t2, *res;
2061 int len1 = strlen (user) + 1 + strlen (passwd);
2062 int len2 = BASE64_LENGTH (len1);
2064 t1 = (char *)alloca (len1 + 1);
2065 sprintf (t1, "%s:%s", user, passwd);
2066 t2 = (char *)alloca (1 + len2);
2067 base64_encode (t1, t2, len1);
2068 res = (char *)xmalloc (len2 + 11 + strlen (header));
2069 sprintf (res, "%s: Basic %s\r\n", header, t2);
2075 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2076 of a field in such a header. If the field is the one specified by
2077 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2078 digest authorization code), extract its value in the (char*)
2079 variable pointed by RET. Returns negative on a malformed header,
2080 or number of bytes that have been parsed by this call. */
2082 extract_header_attr (const char *au, const char *attr_name, char **ret)
2084 const char *cp, *ep;
2088 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2090 cp += strlen (attr_name);
2093 cp += skip_lws (cp);
2098 cp += skip_lws (cp);
2103 for (ep = cp; *ep && *ep != '\"'; ep++)
2108 *ret = strdupdelim (cp, ep);
2115 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2116 an array of 16 bytes containing the hash keys, and BUF should be a
2117 buffer of 33 writable characters (32 for hex digits plus one for
2118 zero termination). */
2120 dump_hash (unsigned char *buf, const unsigned char *hash)
2124 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2126 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2127 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2132 /* Take the line apart to find the challenge, and compose a digest
2133 authorization header. See RFC2069 section 2.1.2. */
2135 digest_authentication_encode (const char *au, const char *user,
2136 const char *passwd, const char *method,
2139 static char *realm, *opaque, *nonce;
2144 { "realm", &realm },
2145 { "opaque", &opaque },
2150 realm = opaque = nonce = NULL;
2152 au += 6; /* skip over `Digest' */
2157 au += skip_lws (au);
2158 for (i = 0; i < ARRAY_SIZE (options); i++)
2160 int skip = extract_header_attr (au, options[i].name,
2161 options[i].variable);
2165 FREE_MAYBE (opaque);
2175 if (i == ARRAY_SIZE (options))
2177 while (*au && *au != '=')
2181 au += skip_lws (au);
2185 while (*au && *au != '\"')
2192 while (*au && *au != ',')
2197 if (!realm || !nonce || !user || !passwd || !path || !method)
2200 FREE_MAYBE (opaque);
2205 /* Calculate the digest value. */
2207 MD5_CONTEXT_TYPE ctx;
2208 unsigned char hash[MD5_HASHLEN];
2209 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2210 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2212 /* A1BUF = H(user ":" realm ":" password) */
2214 MD5_UPDATE (user, strlen (user), &ctx);
2215 MD5_UPDATE (":", 1, &ctx);
2216 MD5_UPDATE (realm, strlen (realm), &ctx);
2217 MD5_UPDATE (":", 1, &ctx);
2218 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2219 MD5_FINISH (&ctx, hash);
2220 dump_hash (a1buf, hash);
2222 /* A2BUF = H(method ":" path) */
2224 MD5_UPDATE (method, strlen (method), &ctx);
2225 MD5_UPDATE (":", 1, &ctx);
2226 MD5_UPDATE (path, strlen (path), &ctx);
2227 MD5_FINISH (&ctx, hash);
2228 dump_hash (a2buf, hash);
2230 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2232 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2233 MD5_UPDATE (":", 1, &ctx);
2234 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2235 MD5_UPDATE (":", 1, &ctx);
2236 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2237 MD5_FINISH (&ctx, hash);
2238 dump_hash (response_digest, hash);
2240 res = (char*) xmalloc (strlen (user)
2245 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2246 + (opaque ? strlen (opaque) : 0)
2248 sprintf (res, "Authorization: Digest \
2249 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2250 user, realm, nonce, path, response_digest);
2253 char *p = res + strlen (res);
2254 strcat (p, ", opaque=\"");
2258 strcat (res, "\r\n");
2262 #endif /* USE_DIGEST */
2265 #define BEGINS_WITH(line, string_constant) \
2266 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2267 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2268 || !line[sizeof (string_constant) - 1]))
2271 known_authentication_scheme_p (const char *au)
2273 return BEGINS_WITH (au, "Basic")
2274 || BEGINS_WITH (au, "Digest")
2275 || BEGINS_WITH (au, "NTLM");
2280 /* Create the HTTP authorization request header. When the
2281 `WWW-Authenticate' response header is seen, according to the
2282 authorization scheme specified in that header (`Basic' and `Digest'
2283 are supported by the current implementation), produce an
2284 appropriate HTTP authorization request header. */
2286 create_authorization_line (const char *au, const char *user,
2287 const char *passwd, const char *method,
2290 char *wwwauth = NULL;
2292 if (!strncasecmp (au, "Basic", 5))
2293 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2294 if (!strncasecmp (au, "NTLM", 4))
2295 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2297 else if (!strncasecmp (au, "Digest", 6))
2298 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2299 #endif /* USE_DIGEST */