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->proto==URLHTTPS ? 1 : 0))
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->proto == URLHTTPS)
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->proto == URLHTTPS
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,
807 ou->proto == URLHTTPS);
809 /* Allocate the memory for the request. */
810 request = (char *)alloca (strlen (command) + strlen (path)
813 + (port_maybe ? strlen (port_maybe) : 0)
814 + strlen (HTTP_ACCEPT)
815 + (request_keep_alive
816 ? strlen (request_keep_alive) : 0)
817 + (referer ? strlen (referer) : 0)
818 + (cookies ? strlen (cookies) : 0)
819 + (wwwauth ? strlen (wwwauth) : 0)
820 + (proxyauth ? strlen (proxyauth) : 0)
821 + (range ? strlen (range) : 0)
823 + (opt.user_header ? strlen (opt.user_header) : 0)
825 /* Construct the request. */
831 %s%s%s%s%s%s%s%s\r\n",
832 command, path, useragent, remhost,
833 port_maybe ? port_maybe : "",
835 request_keep_alive ? request_keep_alive : "",
836 referer ? referer : "",
837 cookies ? cookies : "",
838 wwwauth ? wwwauth : "",
839 proxyauth ? proxyauth : "",
842 opt.user_header ? opt.user_header : "");
843 DEBUGP (("---request begin---\n%s---request end---\n", request));
844 /* Free the temporary memory. */
845 FREE_MAYBE (wwwauth);
846 FREE_MAYBE (proxyauth);
847 FREE_MAYBE (cookies);
849 /* Send the request to server. */
851 if (u->proto == URLHTTPS)
852 num_written = ssl_iwrite (ssl, request, strlen (request));
854 #endif /* HAVE_SSL */
855 num_written = iwrite (sock, request, strlen (request));
859 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
861 CLOSE_INVALIDATE (sock);
864 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
865 u->proxy ? "Proxy" : "HTTP");
866 contlen = contrange = -1;
871 /* Before reading anything, initialize the rbuf. */
872 rbuf_initialize (&rbuf, sock);
874 if (u->proto == URLHTTPS)
878 #endif /* HAVE_SSL */
881 /* Header-fetching loop. */
889 /* Get the header. */
890 status = header_get (&rbuf, &hdr,
891 /* Disallow continuations for status line. */
892 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
894 /* Check for errors. */
895 if (status == HG_EOF && *hdr)
897 /* This used to be an unconditional error, but that was
898 somewhat controversial, because of a large number of
899 broken CGI's that happily "forget" to send the second EOL
900 before closing the connection of a HEAD request.
902 So, the deal is to check whether the header is empty
903 (*hdr is zero if it is); if yes, it means that the
904 previous header was fully retrieved, and that -- most
905 probably -- the request is complete. "...be liberal in
906 what you accept." Oh boy. */
907 logputs (LOG_VERBOSE, "\n");
908 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
911 FREE_MAYBE (all_headers);
912 CLOSE_INVALIDATE (sock);
915 else if (status == HG_ERROR)
917 logputs (LOG_VERBOSE, "\n");
918 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
922 FREE_MAYBE (all_headers);
923 CLOSE_INVALIDATE (sock);
927 /* If the headers are to be saved to a file later, save them to
929 if (opt.save_headers)
931 int lh = strlen (hdr);
932 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
933 memcpy (all_headers + all_length, hdr, lh);
935 all_headers[all_length++] = '\n';
936 all_headers[all_length] = '\0';
939 /* Check for status line. */
943 /* Parse the first line of server response. */
944 statcode = parse_http_status_line (hdr, &error);
945 hs->statcode = statcode;
946 /* Store the descriptive response. */
947 if (statcode == -1) /* malformed response */
949 /* A common reason for "malformed response" error is the
950 case when no data was actually received. Handle this
953 hs->error = xstrdup (_("No data received"));
955 hs->error = xstrdup (_("Malformed status line"));
960 hs->error = xstrdup (_("(no description)"));
962 hs->error = xstrdup (error);
970 if (opt.server_response)
971 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
973 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
979 /* Exit on empty header. */
986 /* Print the header if requested. */
987 if (opt.server_response && hcount != 1)
988 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
990 /* Try getting content-length. */
991 if (contlen == -1 && !opt.ignore_length)
992 if (header_process (hdr, "Content-Length", header_extract_number,
995 /* Try getting content-type. */
997 if (header_process (hdr, "Content-Type", http_process_type, &type))
999 /* Try getting location. */
1001 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1003 /* Try getting last-modified. */
1004 if (!hs->remote_time)
1005 if (header_process (hdr, "Last-Modified", header_strdup,
1008 /* Try getting cookies. */
1010 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1012 /* Try getting www-authentication. */
1013 if (!authenticate_h)
1014 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1017 /* Check for accept-ranges header. If it contains the word
1018 `none', disable the ranges. */
1019 if (*dt & ACCEPTRANGES)
1022 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1025 *dt &= ~ACCEPTRANGES;
1029 /* Try getting content-range. */
1030 if (contrange == -1)
1032 struct http_process_range_closure closure;
1033 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1035 contrange = closure.first_byte_pos;
1039 /* Check for keep-alive related responses. */
1040 if (!inhibit_keep_alive)
1042 /* Check for the `Keep-Alive' header. */
1043 if (!http_keep_alive_1)
1045 if (header_process (hdr, "Keep-Alive", header_exists,
1046 &http_keep_alive_1))
1049 /* Check for `Connection: Keep-Alive'. */
1050 if (!http_keep_alive_2)
1052 if (header_process (hdr, "Connection", http_process_connection,
1053 &http_keep_alive_2))
1061 logputs (LOG_VERBOSE, "\n");
1064 && (http_keep_alive_1 || http_keep_alive_2))
1066 assert (inhibit_keep_alive == 0);
1070 /* The server has promised that it will not close the connection
1071 when we're done. This means that we can register it. */
1073 register_persistent (u->host, u->port, sock);
1075 register_persistent (u->host, u->port, sock, ssl);
1076 #endif /* HAVE_SSL */
1078 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1081 /* Authorization is required. */
1085 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1086 might be more bytes in the body. */
1087 if (auth_tried_already)
1089 /* If we have tried it already, then there is not point
1092 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1093 xfree (authenticate_h);
1096 else if (!known_authentication_scheme_p (authenticate_h))
1098 xfree (authenticate_h);
1099 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1102 else if (BEGINS_WITH (authenticate_h, "Basic"))
1104 /* The authentication scheme is basic, the one we try by
1105 default, and it failed. There's no sense in trying
1111 auth_tried_already = 1;
1115 /* We do not need this anymore. */
1118 xfree (authenticate_h);
1119 authenticate_h = NULL;
1122 /* 20x responses are counted among successful by default. */
1123 if (H_20X (statcode))
1126 /* Return if redirected. */
1127 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1129 /* RFC2068 says that in case of the 300 (multiple choices)
1130 response, the server can output a preferred URL through
1131 `Location' header; otherwise, the request should be treated
1132 like GET. So, if the location is set, it will be a
1133 redirection; otherwise, just proceed normally. */
1134 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1138 logprintf (LOG_VERBOSE,
1139 _("Location: %s%s\n"),
1140 hs->newloc ? hs->newloc : _("unspecified"),
1141 hs->newloc ? _(" [following]") : "");
1142 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1143 might be more bytes in the body. */
1145 FREE_MAYBE (all_headers);
1150 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1153 /* We don't assume text/html by default. */
1156 if (opt.html_extension && (*dt & TEXTHTML))
1157 /* -E / --html-extension / html_extension = on was specified, and this is a
1158 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1159 already the file's suffix, tack on ".html". */
1161 char* last_period_in_local_filename = strrchr(u->local, '.');
1163 if (last_period_in_local_filename == NULL ||
1164 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1165 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1167 size_t local_filename_len = strlen(u->local);
1169 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1170 strcpy(u->local + local_filename_len, ".html");
1172 *dt |= ADDED_HTML_EXTENSION;
1176 if (contrange == -1)
1178 /* We did not get a content-range header. This means that the
1179 server did not honor our `Range' request. Normally, this
1180 means we should reset hs->restval and continue normally. */
1182 /* However, if `-c' is used, we need to be a bit more careful:
1184 1. If `-c' is specified and the file already existed when
1185 Wget was started, it would be a bad idea for us to start
1186 downloading it from scratch, effectively truncating it. I
1187 believe this cannot happen unless `-c' was specified.
1189 2. If `-c' is used on a file that is already fully
1190 downloaded, we're requesting bytes after the end of file,
1191 which can result in server not honoring `Range'. If this is
1192 the case, `Content-Length' will be equal to the length of the
1194 if (opt.always_rest)
1196 /* Check for condition #2. */
1197 if (hs->restval > 0 /* restart was requested. */
1198 && contlen != -1 /* we got content-length. */
1199 && hs->restval >= contlen /* file fully downloaded
1203 logputs (LOG_VERBOSE, _("\
1204 \n The file is already fully retrieved; nothing to do.\n\n"));
1205 /* In case the caller inspects. */
1209 FREE_MAYBE (all_headers);
1210 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1211 might be more bytes in the body. */
1212 return RETRUNNEEDED;
1215 /* Check for condition #1. */
1216 if (hs->no_truncate)
1218 logprintf (LOG_NOTQUIET,
1221 Continued download failed on this file, which conflicts with `-c'.\n\
1222 Refusing to truncate existing file `%s'.\n\n"), u->local);
1224 FREE_MAYBE (all_headers);
1225 CLOSE_INVALIDATE (sock);
1226 return CONTNOTSUPPORTED;
1234 else if (contrange != hs->restval ||
1235 (H_PARTIAL (statcode) && contrange == -1))
1237 /* This means the whole request was somehow misunderstood by the
1238 server. Bail out. */
1240 FREE_MAYBE (all_headers);
1241 CLOSE_INVALIDATE (sock);
1248 contlen += contrange;
1250 contrange = -1; /* If conent-length was not sent,
1251 content-range will be ignored. */
1253 hs->contlen = contlen;
1257 if ((*dt & RETROKF) && !opt.server_response)
1259 /* No need to print this output if the body won't be
1260 downloaded at all, or if the original server response is
1262 logputs (LOG_VERBOSE, _("Length: "));
1265 logputs (LOG_VERBOSE, legible (contlen));
1266 if (contrange != -1)
1267 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1268 legible (contlen - contrange));
1271 logputs (LOG_VERBOSE,
1272 opt.ignore_length ? _("ignored") : _("unspecified"));
1274 logprintf (LOG_VERBOSE, " [%s]\n", type);
1276 logputs (LOG_VERBOSE, "\n");
1280 type = NULL; /* We don't need it any more. */
1282 /* Return if we have no intention of further downloading. */
1283 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1285 /* In case the caller cares to look... */
1289 FREE_MAYBE (all_headers);
1290 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1291 might be more bytes in the body. */
1292 return RETRFINISHED;
1295 /* Open the local file. */
1298 mkalldirs (u->local);
1300 rotate_backups (u->local);
1301 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1304 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1305 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1306 might be more bytes in the body. */
1307 FREE_MAYBE (all_headers);
1313 extern int global_download_count;
1315 /* To ensure that repeated "from scratch" downloads work for -O
1316 files, we rewind the file pointer, unless restval is
1317 non-zero. (This works only when -O is used on regular files,
1318 but it's still a valuable feature.)
1320 However, this loses when more than one URL is specified on
1321 the command line the second rewinds eradicates the contents
1322 of the first download. Thus we disable the above trick for
1323 all the downloads except the very first one.
1325 #### A possible solution to this would be to remember the
1326 file position in the output document and to seek to that
1327 position, instead of rewinding. */
1328 if (!hs->restval && global_download_count == 0)
1330 /* This will silently fail for streams that don't correspond
1331 to regular files, but that's OK. */
1333 /* ftruncate is needed because opt.dfp is opened in append
1334 mode if opt.always_rest is set. */
1335 ftruncate (fileno (fp), 0);
1340 /* #### This confuses the code that checks for file size. There
1341 should be some overhead information. */
1342 if (opt.save_headers)
1343 fwrite (all_headers, 1, all_length, fp);
1344 timer = wtimer_new ();
1345 /* Get the contents of the document. */
1346 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1347 (contlen != -1 ? contlen : 0),
1349 hs->dltime = wtimer_elapsed (timer);
1350 wtimer_delete (timer);
1352 /* Close or flush the file. We have to be careful to check for
1353 error here. Checking the result of fwrite() is not enough --
1354 errors could go unnoticed! */
1357 flush_res = fclose (fp);
1359 flush_res = fflush (fp);
1360 if (flush_res == EOF)
1363 FREE_MAYBE (all_headers);
1364 CLOSE_FINISH (sock);
1367 return RETRFINISHED;
1370 /* The genuine HTTP loop! This is the part where the retrieval is
1371 retried, and retried, and retried, and... */
1373 http_loop (struct urlinfo *u, char **newloc, int *dt)
1376 int use_ts, got_head = 0; /* time-stamping info */
1377 char *filename_plus_orig_suffix;
1378 char *local_filename = NULL;
1379 char *tms, *suf, *locf, *tmrate;
1381 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1382 long local_size = 0; /* the size of the local file */
1383 size_t filename_len;
1384 struct http_stat hstat; /* HTTP status */
1387 /* This used to be done in main(), but it's a better idea to do it
1388 here so that we don't go through the hoops if we're just using
1390 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1392 load_cookies (opt.cookies_input);
1393 cookies_loaded_p = 1;
1398 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1399 has_wildcards_p because it would also warn on `?', and we know that
1400 shows up in CGI paths a *lot*. */
1401 if (strchr (u->url, '*'))
1402 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1404 /* Determine the local filename. */
1406 u->local = url_filename (u->proxy ? u->proxy : u);
1408 if (!opt.output_document)
1411 locf = opt.output_document;
1413 filename_len = strlen (u->local);
1414 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1416 if (opt.noclobber && file_exists_p (u->local))
1418 /* If opt.noclobber is turned on and file already exists, do not
1419 retrieve the file */
1420 logprintf (LOG_VERBOSE, _("\
1421 File `%s' already there, will not retrieve.\n"), u->local);
1422 /* If the file is there, we suppose it's retrieved OK. */
1425 /* #### Bogusness alert. */
1426 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1427 text/html, a harmless lie. */
1428 if (((suf = suffix (u->local)) != NULL)
1429 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1432 /* Another harmless lie: */
1437 if (opt.timestamping)
1439 boolean local_dot_orig_file_exists = FALSE;
1441 if (opt.backup_converted)
1442 /* If -K is specified, we'll act on the assumption that it was specified
1443 last time these files were downloaded as well, and instead of just
1444 comparing local file X against server file X, we'll compare local
1445 file X.orig (if extant, else X) against server file X. If -K
1446 _wasn't_ specified last time, or the server contains files called
1447 *.orig, -N will be back to not operating correctly with -k. */
1449 /* Would a single s[n]printf() call be faster? --dan
1451 Definitely not. sprintf() is horribly slow. It's a
1452 different question whether the difference between the two
1453 affects a program. Usually I'd say "no", but at one
1454 point I profiled Wget, and found that a measurable and
1455 non-negligible amount of time was lost calling sprintf()
1456 in url.c. Replacing sprintf with inline calls to
1457 strcpy() and long_to_string() made a difference.
1459 memcpy (filename_plus_orig_suffix, u->local, filename_len);
1460 memcpy (filename_plus_orig_suffix + filename_len,
1461 ".orig", sizeof (".orig"));
1463 /* Try to stat() the .orig file. */
1464 if (stat (filename_plus_orig_suffix, &st) == 0)
1466 local_dot_orig_file_exists = TRUE;
1467 local_filename = filename_plus_orig_suffix;
1471 if (!local_dot_orig_file_exists)
1472 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1473 if (stat (u->local, &st) == 0)
1474 local_filename = u->local;
1476 if (local_filename != NULL)
1477 /* There was a local file, so we'll check later to see if the version
1478 the server has is the same version we already have, allowing us to
1483 local_size = st.st_size;
1487 /* Reset the counter. */
1489 *dt = 0 | ACCEPTRANGES;
1493 /* Increment the pass counter. */
1495 sleep_between_retrievals (count);
1496 /* Get the current time string. */
1497 tms = time_str (NULL);
1498 /* Print fetch message, if opt.verbose. */
1501 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1505 sprintf (tmp, _("(try:%2d)"), count);
1506 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1507 tms, hurl, tmp, locf);
1509 ws_changetitle (hurl, 1);
1514 /* Default document type is empty. However, if spider mode is
1515 on or time-stamping is employed, HEAD_ONLY commands is
1516 encoded within *dt. */
1517 if (opt.spider || (use_ts && !got_head))
1521 /* Assume no restarting. */
1523 /* Decide whether or not to restart. */
1524 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1525 /* #### this calls access() and then stat(); could be optimized. */
1526 && file_exists_p (locf))
1527 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1528 hstat.restval = st.st_size;
1530 /* In `-c' is used and the file is existing and non-empty,
1531 refuse to truncate it if the server doesn't support continued
1533 hstat.no_truncate = 0;
1534 if (opt.always_rest && hstat.restval)
1535 hstat.no_truncate = 1;
1537 /* Decide whether to send the no-cache directive. We send it in
1539 a) we're using a proxy, and we're past our first retrieval.
1540 Some proxies are notorious for caching incomplete data, so
1541 we require a fresh get.
1542 b) caching is explicitly inhibited. */
1543 if ((u->proxy && count > 1) /* a */
1544 || !opt.allow_cache /* b */
1546 *dt |= SEND_NOCACHE;
1548 *dt &= ~SEND_NOCACHE;
1550 /* Try fetching the document, or at least its head. :-) */
1551 err = gethttp (u, &hstat, dt);
1553 /* It's unfortunate that wget determines the local filename before finding
1554 out the Content-Type of the file. Barring a major restructuring of the
1555 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1556 u->local to tack on ".html". */
1557 if (!opt.output_document)
1560 locf = opt.output_document;
1563 tms = time_str (NULL);
1564 /* Get the new location (with or without the redirection). */
1566 *newloc = xstrdup (hstat.newloc);
1569 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1570 case CONERROR: case READERR: case WRITEFAILED:
1572 /* Non-fatal errors continue executing the loop, which will
1573 bring them to "while" statement at the end, to judge
1574 whether the number of tries was exceeded. */
1576 printwhat (count, opt.ntry);
1579 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1580 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1581 /* Fatal errors just return from the function. */
1585 case FWRITEERR: case FOPENERR:
1586 /* Another fatal error. */
1587 logputs (LOG_VERBOSE, "\n");
1588 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1589 u->local, strerror (errno));
1594 /* Another fatal error. */
1595 logputs (LOG_VERBOSE, "\n");
1596 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1601 /* Return the new location to the caller. */
1604 logprintf (LOG_NOTQUIET,
1605 _("ERROR: Redirection (%d) without location.\n"),
1613 /* The file was already fully retrieved. */
1618 /* Deal with you later. */
1621 /* All possibilities should have been exhausted. */
1624 if (!(*dt & RETROKF))
1628 /* #### Ugly ugly ugly! */
1629 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1630 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1633 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1634 tms, hstat.statcode, hstat.error);
1635 logputs (LOG_VERBOSE, "\n");
1640 /* Did we get the time-stamp? */
1643 if (opt.timestamping && !hstat.remote_time)
1645 logputs (LOG_NOTQUIET, _("\
1646 Last-modified header missing -- time-stamps turned off.\n"));
1648 else if (hstat.remote_time)
1650 /* Convert the date-string into struct tm. */
1651 tmr = http_atotm (hstat.remote_time);
1652 if (tmr == (time_t) (-1))
1653 logputs (LOG_VERBOSE, _("\
1654 Last-modified header invalid -- time-stamp ignored.\n"));
1658 /* The time-stamping section. */
1663 use_ts = 0; /* no more time-stamping */
1664 count = 0; /* the retrieve count for HEAD is
1666 if (hstat.remote_time && tmr != (time_t) (-1))
1668 /* Now time-stamping can be used validly. Time-stamping
1669 means that if the sizes of the local and remote file
1670 match, and local file is newer than the remote file,
1671 it will not be retrieved. Otherwise, the normal
1672 download procedure is resumed. */
1674 (hstat.contlen == -1 || local_size == hstat.contlen))
1676 logprintf (LOG_VERBOSE, _("\
1677 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1682 else if (tml >= tmr)
1683 logprintf (LOG_VERBOSE, _("\
1684 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1686 logputs (LOG_VERBOSE,
1687 _("Remote file is newer, retrieving.\n"));
1692 if ((tmr != (time_t) (-1))
1694 && ((hstat.len == hstat.contlen) ||
1695 ((hstat.res == 0) &&
1696 ((hstat.contlen == -1) ||
1697 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1699 /* #### This code repeats in http.c and ftp.c. Move it to a
1701 const char *fl = NULL;
1702 if (opt.output_document)
1704 if (opt.od_known_regular)
1705 fl = opt.output_document;
1712 /* End of time-stamping section. */
1716 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1720 /* It is now safe to free the remainder of hstat, since the
1721 strings within it will no longer be used. */
1724 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1726 if (hstat.len == hstat.contlen)
1730 logprintf (LOG_VERBOSE,
1731 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1732 tms, tmrate, locf, hstat.len, hstat.contlen);
1733 logprintf (LOG_NONVERBOSE,
1734 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1735 tms, u->url, hstat.len, hstat.contlen, locf, count);
1738 downloaded_increase (hstat.len);
1740 /* Remember that we downloaded the file for later ".orig" code. */
1741 if (*dt & ADDED_HTML_EXTENSION)
1742 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1744 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1748 else if (hstat.res == 0) /* No read error */
1750 if (hstat.contlen == -1) /* We don't know how much we were supposed
1751 to get, so assume we succeeded. */
1755 logprintf (LOG_VERBOSE,
1756 _("%s (%s) - `%s' saved [%ld]\n\n"),
1757 tms, tmrate, locf, hstat.len);
1758 logprintf (LOG_NONVERBOSE,
1759 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1760 tms, u->url, hstat.len, locf, count);
1763 downloaded_increase (hstat.len);
1765 /* Remember that we downloaded the file for later ".orig" code. */
1766 if (*dt & ADDED_HTML_EXTENSION)
1767 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1769 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1773 else if (hstat.len < hstat.contlen) /* meaning we lost the
1774 connection too soon */
1776 logprintf (LOG_VERBOSE,
1777 _("%s (%s) - Connection closed at byte %ld. "),
1778 tms, tmrate, hstat.len);
1779 printwhat (count, opt.ntry);
1782 else if (!opt.kill_longer) /* meaning we got more than expected */
1784 logprintf (LOG_VERBOSE,
1785 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1786 tms, tmrate, locf, hstat.len, hstat.contlen);
1787 logprintf (LOG_NONVERBOSE,
1788 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1789 tms, u->url, hstat.len, hstat.contlen, locf, count);
1791 downloaded_increase (hstat.len);
1793 /* Remember that we downloaded the file for later ".orig" code. */
1794 if (*dt & ADDED_HTML_EXTENSION)
1795 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1797 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1801 else /* the same, but not accepted */
1803 logprintf (LOG_VERBOSE,
1804 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1805 tms, tmrate, hstat.len, hstat.contlen);
1806 printwhat (count, opt.ntry);
1810 else /* now hstat.res can only be -1 */
1812 if (hstat.contlen == -1)
1814 logprintf (LOG_VERBOSE,
1815 _("%s (%s) - Read error at byte %ld (%s)."),
1816 tms, tmrate, hstat.len, strerror (errno));
1817 printwhat (count, opt.ntry);
1820 else /* hstat.res == -1 and contlen is given */
1822 logprintf (LOG_VERBOSE,
1823 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1824 tms, tmrate, hstat.len, hstat.contlen,
1826 printwhat (count, opt.ntry);
1833 while (!opt.ntry || (count < opt.ntry));
1837 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1838 than local timezone.
1840 mktime is similar but assumes struct tm, also known as the
1841 "broken-down" form of time, is in local time zone. mktime_from_utc
1842 uses mktime to make the conversion understanding that an offset
1843 will be introduced by the local time assumption.
1845 mktime_from_utc then measures the introduced offset by applying
1846 gmtime to the initial result and applying mktime to the resulting
1847 "broken-down" form. The difference between the two mktime results
1848 is the measured offset which is then subtracted from the initial
1849 mktime result to yield a calendar time which is the value returned.
1851 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1852 consistent offset (the non DST offset) since tm and tm+o might be
1853 on opposite sides of a DST change.
1855 Some implementations of mktime return -1 for the nonexistent
1856 localtime hour at the beginning of DST. In this event, use
1857 mktime(tm - 1hr) + 3600.
1861 gmtime(t+o) --> tm+o
1862 mktime(tm+o) --> t+2o
1863 t+o - (t+2o - t+o) = t
1865 Note that glibc contains a function of the same purpose named
1866 `timegm' (reverse of gmtime). But obviously, it is not universally
1867 available, and unfortunately it is not straightforwardly
1868 extractable for use here. Perhaps configure should detect timegm
1869 and use it where available.
1871 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1872 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1873 Further improved by Roger with assistance from Edward J. Sabol
1874 based on input by Jamie Zawinski. */
1877 mktime_from_utc (struct tm *t)
1888 return -1; /* can't deal with output from strptime */
1899 return -1; /* can't deal with output from gmtime */
1902 return (tl - (tb - tl));
1905 /* Check whether the result of strptime() indicates success.
1906 strptime() returns the pointer to how far it got to in the string.
1907 The processing has been successful if the string is at `GMT' or
1908 `+X', or at the end of the string.
1910 In extended regexp parlance, the function returns 1 if P matches
1911 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1912 can return) is considered a failure and 0 is returned. */
1914 check_end (const char *p)
1918 while (ISSPACE (*p))
1921 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1922 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1928 /* Convert the textual specification of time in TIME_STRING to the
1929 number of seconds since the Epoch.
1931 TIME_STRING can be in any of the three formats RFC2068 allows the
1932 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1933 Timezones are ignored, and should be GMT.
1935 Return the computed time_t representation, or -1 if the conversion
1938 This function uses strptime with various string formats for parsing
1939 TIME_STRING. This results in a parser that is not as lenient in
1940 interpreting TIME_STRING as I would like it to be. Being based on
1941 strptime, it always allows shortened months, one-digit days, etc.,
1942 but due to the multitude of formats in which time can be
1943 represented, an ideal HTTP time parser would be even more
1944 forgiving. It should completely ignore things like week days and
1945 concentrate only on the various forms of representing years,
1946 months, days, hours, minutes, and seconds. For example, it would
1947 be nice if it accepted ISO 8601 out of the box.
1949 I've investigated free and PD code for this purpose, but none was
1950 usable. getdate was big and unwieldy, and had potential copyright
1951 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1952 distributed with phttpd, is excellent, but we cannot use it because
1953 it is not assigned to the FSF. So I stuck it with strptime. */
1956 http_atotm (char *time_string)
1958 /* NOTE: Solaris strptime man page claims that %n and %t match white
1959 space, but that's not universally available. Instead, we simply
1960 use ` ' to mean "skip all WS", which works under all strptime
1961 implementations I've tested. */
1963 static const char *time_formats[] = {
1964 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1965 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1966 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1967 (google.com uses this for their cookies.) */
1968 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1974 /* According to Roger Beeman, we need to initialize tm_isdst, since
1975 strptime won't do it. */
1978 /* Note that under foreign locales Solaris strptime() fails to
1979 recognize English dates, which renders this function useless. We
1980 solve this by being careful not to affect LC_TIME when
1981 initializing locale.
1983 Another solution would be to temporarily set locale to C, invoke
1984 strptime(), and restore it back. This is slow and dirty,
1985 however, and locale support other than LC_MESSAGES can mess other
1986 things, so I rather chose to stick with just setting LC_MESSAGES.
1988 GNU strptime does not have this problem because it recognizes
1989 both international and local dates. */
1991 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1992 if (check_end (strptime (time_string, time_formats[i], &t)))
1993 return mktime_from_utc (&t);
1995 /* All formats have failed. */
1999 /* Authorization support: We support two authorization schemes:
2001 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2003 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2004 consisting of answering to the server's challenge with the proper
2007 /* How many bytes it will take to store LEN bytes in base64. */
2008 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2010 /* Encode the string S of length LENGTH to base64 format and place it
2011 to STORE. STORE will be 0-terminated, and must point to a writable
2012 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2014 base64_encode (const char *s, char *store, int length)
2016 /* Conversion table. */
2017 static char tbl[64] = {
2018 'A','B','C','D','E','F','G','H',
2019 'I','J','K','L','M','N','O','P',
2020 'Q','R','S','T','U','V','W','X',
2021 'Y','Z','a','b','c','d','e','f',
2022 'g','h','i','j','k','l','m','n',
2023 'o','p','q','r','s','t','u','v',
2024 'w','x','y','z','0','1','2','3',
2025 '4','5','6','7','8','9','+','/'
2028 unsigned char *p = (unsigned char *)store;
2030 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2031 for (i = 0; i < length; i += 3)
2033 *p++ = tbl[s[0] >> 2];
2034 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2035 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2036 *p++ = tbl[s[2] & 0x3f];
2039 /* Pad the result if necessary... */
2040 if (i == length + 1)
2042 else if (i == length + 2)
2043 *(p - 1) = *(p - 2) = '=';
2044 /* ...and zero-terminate it. */
2048 /* Create the authentication header contents for the `Basic' scheme.
2049 This is done by encoding the string `USER:PASS' in base64 and
2050 prepending `HEADER: Basic ' to it. */
2052 basic_authentication_encode (const char *user, const char *passwd,
2055 char *t1, *t2, *res;
2056 int len1 = strlen (user) + 1 + strlen (passwd);
2057 int len2 = BASE64_LENGTH (len1);
2059 t1 = (char *)alloca (len1 + 1);
2060 sprintf (t1, "%s:%s", user, passwd);
2061 t2 = (char *)alloca (1 + len2);
2062 base64_encode (t1, t2, len1);
2063 res = (char *)xmalloc (len2 + 11 + strlen (header));
2064 sprintf (res, "%s: Basic %s\r\n", header, t2);
2070 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2071 of a field in such a header. If the field is the one specified by
2072 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2073 digest authorization code), extract its value in the (char*)
2074 variable pointed by RET. Returns negative on a malformed header,
2075 or number of bytes that have been parsed by this call. */
2077 extract_header_attr (const char *au, const char *attr_name, char **ret)
2079 const char *cp, *ep;
2083 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2085 cp += strlen (attr_name);
2088 cp += skip_lws (cp);
2093 cp += skip_lws (cp);
2098 for (ep = cp; *ep && *ep != '\"'; ep++)
2103 *ret = strdupdelim (cp, ep);
2110 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2111 an array of 16 bytes containing the hash keys, and BUF should be a
2112 buffer of 33 writable characters (32 for hex digits plus one for
2113 zero termination). */
2115 dump_hash (unsigned char *buf, const unsigned char *hash)
2119 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2121 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2122 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2127 /* Take the line apart to find the challenge, and compose a digest
2128 authorization header. See RFC2069 section 2.1.2. */
2130 digest_authentication_encode (const char *au, const char *user,
2131 const char *passwd, const char *method,
2134 static char *realm, *opaque, *nonce;
2139 { "realm", &realm },
2140 { "opaque", &opaque },
2145 realm = opaque = nonce = NULL;
2147 au += 6; /* skip over `Digest' */
2152 au += skip_lws (au);
2153 for (i = 0; i < ARRAY_SIZE (options); i++)
2155 int skip = extract_header_attr (au, options[i].name,
2156 options[i].variable);
2160 FREE_MAYBE (opaque);
2170 if (i == ARRAY_SIZE (options))
2172 while (*au && *au != '=')
2176 au += skip_lws (au);
2180 while (*au && *au != '\"')
2187 while (*au && *au != ',')
2192 if (!realm || !nonce || !user || !passwd || !path || !method)
2195 FREE_MAYBE (opaque);
2200 /* Calculate the digest value. */
2202 MD5_CONTEXT_TYPE ctx;
2203 unsigned char hash[MD5_HASHLEN];
2204 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2205 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2207 /* A1BUF = H(user ":" realm ":" password) */
2209 MD5_UPDATE (user, strlen (user), &ctx);
2210 MD5_UPDATE (":", 1, &ctx);
2211 MD5_UPDATE (realm, strlen (realm), &ctx);
2212 MD5_UPDATE (":", 1, &ctx);
2213 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2214 MD5_FINISH (&ctx, hash);
2215 dump_hash (a1buf, hash);
2217 /* A2BUF = H(method ":" path) */
2219 MD5_UPDATE (method, strlen (method), &ctx);
2220 MD5_UPDATE (":", 1, &ctx);
2221 MD5_UPDATE (path, strlen (path), &ctx);
2222 MD5_FINISH (&ctx, hash);
2223 dump_hash (a2buf, hash);
2225 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2227 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2228 MD5_UPDATE (":", 1, &ctx);
2229 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2230 MD5_UPDATE (":", 1, &ctx);
2231 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2232 MD5_FINISH (&ctx, hash);
2233 dump_hash (response_digest, hash);
2235 res = (char*) xmalloc (strlen (user)
2240 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2241 + (opaque ? strlen (opaque) : 0)
2243 sprintf (res, "Authorization: Digest \
2244 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2245 user, realm, nonce, path, response_digest);
2248 char *p = res + strlen (res);
2249 strcat (p, ", opaque=\"");
2253 strcat (res, "\r\n");
2257 #endif /* USE_DIGEST */
2260 #define BEGINS_WITH(line, string_constant) \
2261 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2262 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2263 || !line[sizeof (string_constant) - 1]))
2266 known_authentication_scheme_p (const char *au)
2268 return BEGINS_WITH (au, "Basic")
2269 || BEGINS_WITH (au, "Digest")
2270 || BEGINS_WITH (au, "NTLM");
2275 /* Create the HTTP authorization request header. When the
2276 `WWW-Authenticate' response header is seen, according to the
2277 authorization scheme specified in that header (`Basic' and `Digest'
2278 are supported by the current implementation), produce an
2279 appropriate HTTP authorization request header. */
2281 create_authorization_line (const char *au, const char *user,
2282 const char *passwd, const char *method,
2285 char *wwwauth = NULL;
2287 if (!strncasecmp (au, "Basic", 5))
2288 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2289 if (!strncasecmp (au, "NTLM", 4))
2290 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2292 else if (!strncasecmp (au, "Digest", 6))
2293 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2294 #endif /* USE_DIGEST */