2 Copyright (C) 1995, 1996, 1997, 1998, 2000 Free Software Foundation, Inc.
4 This file is part of Wget.
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
24 #include <sys/types.h>
35 #if TIME_WITH_SYS_TIME
36 # include <sys/time.h>
40 # include <sys/time.h>
49 # include <netdb.h> /* for h_errno */
66 # include "gen_sslfunc.h"
70 extern char *version_string;
82 #define TEXTHTML_S "text/html"
83 #define HTTP_ACCEPT "*/*"
85 /* Some status code validation macros: */
86 #define H_20X(x) (((x) >= 200) && ((x) < 300))
87 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
88 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
89 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
91 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
93 #define HTTP_STATUS_OK 200
94 #define HTTP_STATUS_CREATED 201
95 #define HTTP_STATUS_ACCEPTED 202
96 #define HTTP_STATUS_NO_CONTENT 204
97 #define HTTP_STATUS_PARTIAL_CONTENTS 206
99 /* Redirection 3xx. */
100 #define HTTP_STATUS_MULTIPLE_CHOICES 300
101 #define HTTP_STATUS_MOVED_PERMANENTLY 301
102 #define HTTP_STATUS_MOVED_TEMPORARILY 302
103 #define HTTP_STATUS_NOT_MODIFIED 304
105 /* Client error 4xx. */
106 #define HTTP_STATUS_BAD_REQUEST 400
107 #define HTTP_STATUS_UNAUTHORIZED 401
108 #define HTTP_STATUS_FORBIDDEN 403
109 #define HTTP_STATUS_NOT_FOUND 404
111 /* Server errors 5xx. */
112 #define HTTP_STATUS_INTERNAL 500
113 #define HTTP_STATUS_NOT_IMPLEMENTED 501
114 #define HTTP_STATUS_BAD_GATEWAY 502
115 #define HTTP_STATUS_UNAVAILABLE 503
118 /* Parse the HTTP status line, which is of format:
120 HTTP-Version SP Status-Code SP Reason-Phrase
122 The function returns the status-code, or -1 if the status line is
123 malformed. The pointer to reason-phrase is returned in RP. */
125 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
127 /* (the variables must not be named `major' and `minor', because
128 that breaks compilation with SunOS4 cc.) */
129 int mjr, mnr, statcode;
132 *reason_phrase_ptr = NULL;
134 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
135 major version, and Y is minor version. */
136 if (strncmp (line, "HTTP/", 5) != 0)
140 /* Calculate major HTTP version. */
142 for (mjr = 0; ISDIGIT (*line); line++)
143 mjr = 10 * mjr + (*line - '0');
144 if (*line != '.' || p == line)
148 /* Calculate minor HTTP version. */
150 for (mnr = 0; ISDIGIT (*line); line++)
151 mnr = 10 * mnr + (*line - '0');
152 if (*line != ' ' || p == line)
154 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
155 minor version can be safely ignored. */
160 /* Calculate status code. */
161 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
163 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
165 /* Set up the reason phrase pointer. */
167 /* RFC2068 requires SPC here, but we allow the string to finish
168 here, in case no reason-phrase is present. */
172 *reason_phrase_ptr = line;
177 *reason_phrase_ptr = line + 1;
182 /* Functions to be used as arguments to header_process(): */
184 struct http_process_range_closure {
190 /* Parse the `Content-Range' header and extract the information it
191 contains. Returns 1 if successful, -1 otherwise. */
193 http_process_range (const char *hdr, void *arg)
195 struct http_process_range_closure *closure
196 = (struct http_process_range_closure *)arg;
199 /* Certain versions of Nutscape proxy server send out
200 `Content-Length' without "bytes" specifier, which is a breach of
201 RFC2068 (as well as the HTTP/1.1 draft which was current at the
202 time). But hell, I must support it... */
203 if (!strncasecmp (hdr, "bytes", 5))
206 hdr += skip_lws (hdr);
212 for (num = 0; ISDIGIT (*hdr); hdr++)
213 num = 10 * num + (*hdr - '0');
214 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
216 closure->first_byte_pos = num;
218 for (num = 0; ISDIGIT (*hdr); hdr++)
219 num = 10 * num + (*hdr - '0');
220 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
222 closure->last_byte_pos = num;
224 for (num = 0; ISDIGIT (*hdr); hdr++)
225 num = 10 * num + (*hdr - '0');
226 closure->entity_length = num;
230 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
231 Used for `Accept-Ranges'. */
233 http_process_none (const char *hdr, void *arg)
235 int *where = (int *)arg;
237 if (strstr (hdr, "none"))
244 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
246 http_process_type (const char *hdr, void *arg)
248 char **result = (char **)arg;
249 /* Locate P on `;' or the terminating zero, whichever comes first. */
250 const char *p = strchr (hdr, ';');
252 p = hdr + strlen (hdr);
253 while (p > hdr && ISSPACE (*(p - 1)))
255 *result = strdupdelim (hdr, p);
259 /* Check whether the `Connection' header is set to "keep-alive". */
261 http_process_connection (const char *hdr, void *arg)
263 int *flag = (int *)arg;
264 if (!strcasecmp (hdr, "Keep-Alive"))
269 /* Persistent connections. Currently, we cache the most recently used
270 connection as persistent, provided that the HTTP server agrees to
271 make it such. The persistence data is stored in the variables
272 below. Ideally, it would be in a structure, and it should be
273 possible to cache an arbitrary fixed number of these connections.
275 I think the code is quite easy to extend in that direction. */
277 /* Whether a persistent connection is active. */
278 static int pc_active_p;
279 /* Host and port of currently active persistent connection. */
280 static unsigned char pc_last_host[4];
281 static unsigned short pc_last_port;
283 /* File descriptor of the currently active persistent connection. */
284 static int pc_last_fd;
287 /* Whether a ssl handshake has occoured on this connection */
288 static int pc_active_ssl;
289 /* SSL connection of the currently active persistent connection. */
290 static SSL *pc_last_ssl;
291 #endif /* HAVE_SSL */
293 /* Mark the persistent connection as invalid. This is used by the
294 CLOSE_* macros after they forcefully close a registered persistent
295 connection. This does not close the file descriptor -- it is left
296 to the caller to do that. (Maybe it should, though.) */
299 invalidate_persistent (void)
304 #endif /* HAVE_SSL */
305 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
308 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
309 persistent. This will enable someone to use the same connection
310 later. In the context of HTTP, this must be called only AFTER the
311 response has been received and the server has promised that the
312 connection will remain alive.
314 If a previous connection was persistent, it is closed. */
317 register_persistent (const char *host, unsigned short port, int fd
327 if (pc_last_fd == fd)
329 /* The connection FD is already registered. Nothing to
335 /* The old persistent connection is still active; let's
336 close it first. This situation arises whenever a
337 persistent connection exists, but we then connect to a
338 different host, and try to register a persistent
339 connection to that one. */
341 /* The ssl disconnect has to take place before the closing
344 shutdown_ssl(pc_last_ssl);
347 invalidate_persistent ();
351 /* This store_hostaddress may not fail, because it has the results
353 success = store_hostaddress (pc_last_host, host);
360 pc_active_ssl = ssl ? 1 : 0;
362 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
365 /* Return non-zero if a persistent connection is available for
366 connecting to HOST:PORT. */
369 persistent_available_p (const char *host, unsigned short port
375 unsigned char this_host[4];
376 /* First, check whether a persistent connection is active at all. */
379 /* Second, check if the active connection pertains to the correct
380 (HOST, PORT) ordered pair. */
381 if (port != pc_last_port)
384 /* Second, a): check if current connection is (not) ssl, too. This
385 test is unlikely to fail because HTTP and HTTPS typicaly use
386 different ports. Yet it is possible, or so I [Christian
387 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
389 if (ssl != pc_active_ssl)
391 #endif /* HAVE_SSL */
392 if (!store_hostaddress (this_host, host))
394 if (memcmp (pc_last_host, this_host, 4))
396 /* Third: check whether the connection is still open. This is
397 important because most server implement a liberal (short) timeout
398 on persistent connections. Wget can of course always reconnect
399 if the connection doesn't work out, but it's nicer to know in
400 advance. This test is a logical followup of the first test, but
401 is "expensive" and therefore placed at the end of the list. */
402 if (!test_socket_open (pc_last_fd))
404 /* Oops, the socket is no longer open. Now that we know that,
405 let's invalidate the persistent connection before returning
408 invalidate_persistent ();
415 # define SHUTDOWN_SSL(ssl) do { \
417 shutdown_ssl (ssl); \
420 # define SHUTDOWN_SSL(ssl)
423 /* The idea behind these two CLOSE macros is to distinguish between
424 two cases: one when the job we've been doing is finished, and we
425 want to close the connection and leave, and two when something is
426 seriously wrong and we're closing the connection as part of
429 In case of keep_alive, CLOSE_FINISH should leave the connection
430 open, while CLOSE_INVALIDATE should still close it.
432 Note that the semantics of the flag `keep_alive' is "this
433 connection *will* be reused (the server has promised not to close
434 the connection once we're done)", while the semantics of
435 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
436 active, registered connection". */
438 #define CLOSE_FINISH(fd) do { \
441 SHUTDOWN_SSL (ssl); \
443 if (pc_active_p && (fd) == pc_last_fd) \
444 invalidate_persistent (); \
448 #define CLOSE_INVALIDATE(fd) do { \
449 SHUTDOWN_SSL (ssl); \
451 if (pc_active_p && (fd) == pc_last_fd) \
452 invalidate_persistent (); \
457 long len; /* received length */
458 long contlen; /* expected length */
459 long restval; /* the restart value */
460 int res; /* the result of last read */
461 char *newloc; /* new location (redirection) */
462 char *remote_time; /* remote time-stamp string */
463 char *error; /* textual HTTP error */
464 int statcode; /* status code */
465 long dltime; /* time of the download */
466 int no_truncate; /* whether truncating the file is
470 /* Free the elements of hstat X. */
471 #define FREEHSTAT(x) do \
473 FREE_MAYBE ((x).newloc); \
474 FREE_MAYBE ((x).remote_time); \
475 FREE_MAYBE ((x).error); \
476 (x).newloc = (x).remote_time = (x).error = NULL; \
479 static char *create_authorization_line PARAMS ((const char *, const char *,
480 const char *, const char *,
482 static char *basic_authentication_encode PARAMS ((const char *, const char *,
484 static int known_authentication_scheme_p PARAMS ((const char *));
486 time_t http_atotm PARAMS ((char *));
488 #define BEGINS_WITH(line, string_constant) \
489 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
490 && (ISSPACE (line[sizeof (string_constant) - 1]) \
491 || !line[sizeof (string_constant) - 1]))
493 /* Retrieve a document through HTTP protocol. It recognizes status
494 code, and correctly handles redirections. It closes the network
495 socket. If it receives an error from the functions below it, it
496 will print it if there is enough information to do so (almost
497 always), returning the error to the caller (i.e. http_loop).
499 Various HTTP parameters are stored to hs. Although it parses the
500 response code correctly, it is not used in a sane way. The caller
503 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
504 and u->proxy->url will be given to the proxy server (bad naming,
507 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
509 char *request, *type, *command, *path;
511 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
512 char *authenticate_h;
516 char *request_keep_alive;
517 int sock, hcount, num_written, all_length, remport, statcode;
518 long contlen, contrange;
522 int auth_tried_already;
525 static SSL_CTX *ssl_ctx = NULL;
527 #endif /* HAVE_SSL */
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;
598 cookies = build_cookies_request (u->host, u->port, u->path,
599 u->proto == URLHTTPS);
601 /* Initialize certain elements of struct http_stat. */
606 hs->remote_time = NULL;
609 /* Which structure to use to retrieve the original URL data. */
615 /* First: establish the connection. */
616 if (inhibit_keep_alive
619 !persistent_available_p (u->host, u->port)
621 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
622 #endif /* HAVE_SSL */
625 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
626 err = make_connection (&sock, u->host, u->port);
630 logputs (LOG_VERBOSE, "\n");
631 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
635 logputs (LOG_VERBOSE, "\n");
636 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
640 logputs (LOG_VERBOSE, "\n");
641 logprintf (LOG_NOTQUIET,
642 _("Connection to %s:%hu refused.\n"), u->host, u->port);
646 logputs (LOG_VERBOSE, "\n");
647 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
652 /* Everything is fine! */
653 logputs (LOG_VERBOSE, _("connected!\n"));
660 if (u->proto == URLHTTPS)
661 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
663 logputs (LOG_VERBOSE, "\n");
664 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
668 #endif /* HAVE_SSL */
672 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
673 /* #### pc_last_fd should be accessed through an accessor
678 #endif /* HAVE_SSL */
679 DEBUGP (("Reusing fd %d.\n", sock));
683 path = u->proxy->url;
687 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
691 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
692 sprintf (referer, "Referer: %s\r\n", ou->referer);
694 if (*dt & SEND_NOCACHE)
695 pragma_h = "Pragma: no-cache\r\n";
700 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
701 /* Gag me! Some servers (e.g. WebSitePro) have been known to
702 respond to the following `Range' format by generating a
703 multipart/x-byte-ranges MIME document! This MIME type was
704 present in an old draft of the byteranges specification.
705 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
706 only if multiple non-overlapping ranges are requested --
707 which Wget never does. */
708 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
713 STRDUP_ALLOCA (useragent, opt.useragent);
716 useragent = (char *)alloca (10 + strlen (version_string));
717 sprintf (useragent, "Wget/%s", version_string);
719 /* Construct the authentication, if userid is present. */
722 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
723 user = user ? user : opt.http_user;
724 passwd = passwd ? passwd : opt.http_passwd;
731 /* We have the username and the password, but haven't tried
732 any authorization yet. Let's see if the "Basic" method
733 works. If not, we'll come back here and construct a
734 proper authorization method with the right challenges.
736 If we didn't employ this kind of logic, every URL that
737 requires authorization would have to be processed twice,
738 which is very suboptimal and generates a bunch of false
739 "unauthorized" errors in the server log.
741 #### But this logic also has a serious problem when used
742 with stronger authentications: we *first* transmit the
743 username and the password in clear text, and *then*
744 attempt a stronger authentication scheme. That cannot be
745 right! We are only fortunate that almost everyone still
746 uses the `Basic' scheme anyway.
748 There should be an option to prevent this from happening,
749 for those who use strong authentication schemes and value
751 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
755 wwwauth = create_authorization_line (authenticate_h, user, passwd,
763 char *proxy_user, *proxy_passwd;
764 /* For normal username and password, URL components override
765 command-line/wgetrc parameters. With proxy authentication,
766 it's the reverse, because proxy URLs are normally the
767 "permanent" ones, so command-line args should take
769 if (opt.proxy_user && opt.proxy_passwd)
771 proxy_user = opt.proxy_user;
772 proxy_passwd = opt.proxy_passwd;
776 proxy_user = u->user;
777 proxy_passwd = u->passwd;
779 /* #### This is junky. Can't the proxy request, say, `Digest'
781 if (proxy_user && proxy_passwd)
782 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
783 "Proxy-Authorization");
788 /* String of the form :PORT. Used only for non-standard ports. */
792 && remport != (u->proto == URLHTTPS
793 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
795 && remport != DEFAULT_HTTP_PORT
799 port_maybe = (char *)alloca (numdigit (remport) + 2);
800 sprintf (port_maybe, ":%d", remport);
803 if (!inhibit_keep_alive)
804 request_keep_alive = "Connection: Keep-Alive\r\n";
806 request_keep_alive = NULL;
808 /* Allocate the memory for the request. */
809 request = (char *)alloca (strlen (command) + strlen (path)
812 + (port_maybe ? strlen (port_maybe) : 0)
813 + strlen (HTTP_ACCEPT)
814 + (request_keep_alive
815 ? strlen (request_keep_alive) : 0)
816 + (referer ? strlen (referer) : 0)
817 + (cookies ? strlen (cookies) : 0)
818 + (wwwauth ? strlen (wwwauth) : 0)
819 + (proxyauth ? strlen (proxyauth) : 0)
820 + (range ? strlen (range) : 0)
822 + (opt.user_header ? strlen (opt.user_header) : 0)
824 /* Construct the request. */
830 %s%s%s%s%s%s%s%s\r\n",
831 command, path, useragent, remhost,
832 port_maybe ? port_maybe : "",
834 request_keep_alive ? request_keep_alive : "",
835 referer ? referer : "",
836 cookies ? cookies : "",
837 wwwauth ? wwwauth : "",
838 proxyauth ? proxyauth : "",
841 opt.user_header ? opt.user_header : "");
842 DEBUGP (("---request begin---\n%s---request end---\n", request));
843 /* Free the temporary memory. */
844 FREE_MAYBE (wwwauth);
845 FREE_MAYBE (proxyauth);
846 FREE_MAYBE (cookies);
848 /* Send the request to server. */
850 if (u->proto == URLHTTPS)
851 num_written = ssl_iwrite (ssl, request, strlen (request));
853 #endif /* HAVE_SSL */
854 num_written = iwrite (sock, request, strlen (request));
858 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
860 CLOSE_INVALIDATE (sock);
863 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
864 u->proxy ? "Proxy" : "HTTP");
865 contlen = contrange = -1;
870 /* Before reading anything, initialize the rbuf. */
871 rbuf_initialize (&rbuf, sock);
873 if (u->proto == URLHTTPS)
877 #endif /* HAVE_SSL */
880 /* Header-fetching loop. */
888 /* Get the header. */
889 status = header_get (&rbuf, &hdr,
890 /* Disallow continuations for status line. */
891 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
893 /* Check for errors. */
894 if (status == HG_EOF && *hdr)
896 /* This used to be an unconditional error, but that was
897 somewhat controversial, because of a large number of
898 broken CGI's that happily "forget" to send the second EOL
899 before closing the connection of a HEAD request.
901 So, the deal is to check whether the header is empty
902 (*hdr is zero if it is); if yes, it means that the
903 previous header was fully retrieved, and that -- most
904 probably -- the request is complete. "...be liberal in
905 what you accept." Oh boy. */
906 logputs (LOG_VERBOSE, "\n");
907 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
910 FREE_MAYBE (hs->newloc);
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 (hs->newloc);
923 FREE_MAYBE (all_headers);
924 CLOSE_INVALIDATE (sock);
928 /* If the headers are to be saved to a file later, save them to
930 if (opt.save_headers)
932 int lh = strlen (hdr);
933 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
934 memcpy (all_headers + all_length, hdr, lh);
936 all_headers[all_length++] = '\n';
937 all_headers[all_length] = '\0';
940 /* Print the header if requested. */
941 if (opt.server_response && hcount != 1)
942 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
944 /* Check for status line. */
948 /* Parse the first line of server response. */
949 statcode = parse_http_status_line (hdr, &error);
950 hs->statcode = statcode;
951 /* Store the descriptive response. */
952 if (statcode == -1) /* malformed response */
954 /* A common reason for "malformed response" error is the
955 case when no data was actually received. Handle this
958 hs->error = xstrdup (_("No data received"));
960 hs->error = xstrdup (_("Malformed status line"));
965 hs->error = xstrdup (_("(no description)"));
967 hs->error = xstrdup (error);
974 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
979 /* Exit on empty header. */
986 /* Try getting content-length. */
987 if (contlen == -1 && !opt.ignore_length)
988 if (header_process (hdr, "Content-Length", header_extract_number,
991 /* Try getting content-type. */
993 if (header_process (hdr, "Content-Type", http_process_type, &type))
995 /* Try getting location. */
997 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
999 /* Try getting last-modified. */
1000 if (!hs->remote_time)
1001 if (header_process (hdr, "Last-Modified", header_strdup,
1004 /* Try getting cookies. */
1006 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1008 /* Try getting www-authentication. */
1009 if (!authenticate_h)
1010 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1013 /* Check for accept-ranges header. If it contains the word
1014 `none', disable the ranges. */
1015 if (*dt & ACCEPTRANGES)
1018 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1021 *dt &= ~ACCEPTRANGES;
1025 /* Try getting content-range. */
1026 if (contrange == -1)
1028 struct http_process_range_closure closure;
1029 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1031 contrange = closure.first_byte_pos;
1035 /* Check for keep-alive related responses. */
1036 if (!inhibit_keep_alive)
1038 /* Check for the `Keep-Alive' header. */
1039 if (!http_keep_alive_1)
1041 if (header_process (hdr, "Keep-Alive", header_exists,
1042 &http_keep_alive_1))
1045 /* Check for `Connection: Keep-Alive'. */
1046 if (!http_keep_alive_2)
1048 if (header_process (hdr, "Connection", http_process_connection,
1049 &http_keep_alive_2))
1057 logputs (LOG_VERBOSE, "\n");
1060 && (http_keep_alive_1 || http_keep_alive_2))
1062 assert (inhibit_keep_alive == 0);
1066 /* The server has promised that it will not close the connection
1067 when we're done. This means that we can register it. */
1069 register_persistent (u->host, u->port, sock);
1071 register_persistent (u->host, u->port, sock, ssl);
1072 #endif /* HAVE_SSL */
1074 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1077 /* Authorization is required. */
1081 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1082 might be more bytes in the body. */
1083 if (auth_tried_already)
1085 /* If we have tried it already, then there is not point
1088 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1089 xfree (authenticate_h);
1092 else if (!known_authentication_scheme_p (authenticate_h))
1094 xfree (authenticate_h);
1095 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1098 else if (BEGINS_WITH (authenticate_h, "Basic"))
1100 /* The authentication scheme is basic, the one we try by
1101 default, and it failed. There's no sense in trying
1107 auth_tried_already = 1;
1111 /* We do not need this anymore. */
1114 xfree (authenticate_h);
1115 authenticate_h = NULL;
1118 /* 20x responses are counted among successful by default. */
1119 if (H_20X (statcode))
1122 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1125 /* We don't assume text/html by default. */
1128 if (opt.html_extension && (*dt & TEXTHTML))
1129 /* -E / --html-extension / html_extension = on was specified, and this is a
1130 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1131 already the file's suffix, tack on ".html". */
1133 char* last_period_in_local_filename = strrchr(u->local, '.');
1135 if (last_period_in_local_filename == NULL ||
1136 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1137 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1139 size_t local_filename_len = strlen(u->local);
1141 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1142 strcpy(u->local + local_filename_len, ".html");
1144 *dt |= ADDED_HTML_EXTENSION;
1148 if (contrange == -1)
1150 /* We did not get a content-range header. This means that the
1151 server did not honor our `Range' request. Normally, this
1152 means we should reset hs->restval and continue normally. */
1154 /* However, if `-c' is used, we need to be a bit more careful:
1156 1. If `-c' is specified and the file already existed when
1157 Wget was started, it would be a bad idea for us to start
1158 downloading it from scratch, effectively truncating it. I
1159 believe this cannot happen unless `-c' was specified.
1161 2. If `-c' is used on a file that is already fully
1162 downloaded, we're requesting bytes after the end of file,
1163 which can result in server not honoring `Range'. If this is
1164 the case, `Content-Length' will be equal to the length of the
1166 if (opt.always_rest)
1168 /* Check for condition #2. */
1169 if (hs->restval == contlen)
1171 logputs (LOG_VERBOSE, _("\
1172 \n The file is already fully retrieved; nothing to do.\n\n"));
1173 /* In case the caller inspects. */
1177 FREE_MAYBE (hs->newloc);
1178 FREE_MAYBE (all_headers);
1179 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1180 might be more bytes in the body. */
1181 return RETRFINISHED;
1184 /* Check for condition #1. */
1185 if (hs->no_truncate)
1187 logprintf (LOG_NOTQUIET,
1190 The server does not support continued download;\n\
1191 refusing to truncate `%s'.\n\n"), u->local);
1192 return CONTNOTSUPPORTED;
1201 else if (contrange != hs->restval ||
1202 (H_PARTIAL (statcode) && contrange == -1))
1204 /* This means the whole request was somehow misunderstood by the
1205 server. Bail out. */
1207 FREE_MAYBE (hs->newloc);
1208 FREE_MAYBE (all_headers);
1209 CLOSE_INVALIDATE (sock);
1216 contlen += contrange;
1218 contrange = -1; /* If conent-length was not sent,
1219 content-range will be ignored. */
1221 hs->contlen = contlen;
1223 /* Return if redirected. */
1224 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1226 /* RFC2068 says that in case of the 300 (multiple choices)
1227 response, the server can output a preferred URL through
1228 `Location' header; otherwise, the request should be treated
1229 like GET. So, if the location is set, it will be a
1230 redirection; otherwise, just proceed normally. */
1231 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1235 logprintf (LOG_VERBOSE,
1236 _("Location: %s%s\n"),
1237 hs->newloc ? hs->newloc : _("unspecified"),
1238 hs->newloc ? _(" [following]") : "");
1239 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1240 might be more bytes in the body. */
1242 FREE_MAYBE (all_headers);
1248 if ((*dt & RETROKF) && !opt.server_response)
1250 /* No need to print this output if the body won't be
1251 downloaded at all, or if the original server response is
1253 logputs (LOG_VERBOSE, _("Length: "));
1256 logputs (LOG_VERBOSE, legible (contlen));
1257 if (contrange != -1)
1258 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1259 legible (contlen - contrange));
1262 logputs (LOG_VERBOSE,
1263 opt.ignore_length ? _("ignored") : _("unspecified"));
1265 logprintf (LOG_VERBOSE, " [%s]\n", type);
1267 logputs (LOG_VERBOSE, "\n");
1271 type = NULL; /* We don't need it any more. */
1273 /* Return if we have no intention of further downloading. */
1274 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1276 /* In case the caller cares to look... */
1280 FREE_MAYBE (all_headers);
1281 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1282 might be more bytes in the body. */
1283 return RETRFINISHED;
1286 /* Open the local file. */
1289 mkalldirs (u->local);
1291 rotate_backups (u->local);
1292 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1295 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1296 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1297 might be more bytes in the body. */
1298 FREE_MAYBE (all_headers);
1304 extern int global_download_count;
1306 /* To ensure that repeated "from scratch" downloads work for -O
1307 files, we rewind the file pointer, unless restval is
1308 non-zero. (This works only when -O is used on regular files,
1309 but it's still a valuable feature.)
1311 However, this loses when more than one URL is specified on
1312 the command line the second rewinds eradicates the contents
1313 of the first download. Thus we disable the above trick for
1314 all the downloads except the very first one.
1316 #### A possible solution to this would be to remember the
1317 file position in the output document and to seek to that
1318 position, instead of rewinding. */
1319 if (!hs->restval && global_download_count == 0)
1321 /* This will silently fail for streams that don't correspond
1322 to regular files, but that's OK. */
1324 /* ftruncate is needed because opt.dfp is opened in append
1325 mode if opt.always_rest is set. */
1326 ftruncate (fileno (fp), 0);
1331 /* #### This confuses the code that checks for file size. There
1332 should be some overhead information. */
1333 if (opt.save_headers)
1334 fwrite (all_headers, 1, all_length, fp);
1336 /* Get the contents of the document. */
1337 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1338 (contlen != -1 ? contlen : 0),
1340 hs->dltime = elapsed_time ();
1342 /* Close or flush the file. We have to be careful to check for
1343 error here. Checking the result of fwrite() is not enough --
1344 errors could go unnoticed! */
1347 flush_res = fclose (fp);
1349 flush_res = fflush (fp);
1350 if (flush_res == EOF)
1353 FREE_MAYBE (all_headers);
1354 CLOSE_FINISH (sock);
1357 return RETRFINISHED;
1360 /* The genuine HTTP loop! This is the part where the retrieval is
1361 retried, and retried, and retried, and... */
1363 http_loop (struct urlinfo *u, char **newloc, int *dt)
1366 int use_ts, got_head = 0; /* time-stamping info */
1367 char *filename_plus_orig_suffix;
1368 char *local_filename = NULL;
1369 char *tms, *suf, *locf, *tmrate;
1371 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1372 long local_size = 0; /* the size of the local file */
1373 size_t filename_len;
1374 struct http_stat hstat; /* HTTP status */
1379 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1380 has_wildcards_p because it would also warn on `?', and we know that
1381 shows up in CGI paths a *lot*. */
1382 if (strchr (u->url, '*'))
1383 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1385 /* Determine the local filename. */
1387 u->local = url_filename (u->proxy ? u->proxy : u);
1389 if (!opt.output_document)
1392 locf = opt.output_document;
1394 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1395 xmalloc() here before EACH return. This is one reason it's better to set
1396 flags that influence flow control and then return once at the end. */
1397 filename_len = strlen(u->local);
1398 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1400 if (opt.noclobber && file_exists_p (u->local))
1402 /* If opt.noclobber is turned on and file already exists, do not
1403 retrieve the file */
1404 logprintf (LOG_VERBOSE, _("\
1405 File `%s' already there, will not retrieve.\n"), u->local);
1406 /* If the file is there, we suppose it's retrieved OK. */
1409 /* #### Bogusness alert. */
1410 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1411 text/html, a harmless lie. */
1412 if (((suf = suffix (u->local)) != NULL)
1413 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1416 xfree (filename_plus_orig_suffix); /* must precede every return! */
1417 /* Another harmless lie: */
1422 if (opt.timestamping)
1424 boolean local_dot_orig_file_exists = FALSE;
1426 if (opt.backup_converted)
1427 /* If -K is specified, we'll act on the assumption that it was specified
1428 last time these files were downloaded as well, and instead of just
1429 comparing local file X against server file X, we'll compare local
1430 file X.orig (if extant, else X) against server file X. If -K
1431 _wasn't_ specified last time, or the server contains files called
1432 *.orig, -N will be back to not operating correctly with -k. */
1434 /* Would a single s[n]printf() call be faster? --dan
1436 Definitely not. sprintf() is horribly slow. It's a
1437 different question whether the difference between the two
1438 affects a program. Usually I'd say "no", but at one
1439 point I profiled Wget, and found that a measurable and
1440 non-negligible amount of time was lost calling sprintf()
1441 in url.c. Replacing sprintf with inline calls to
1442 strcpy() and long_to_string() made a difference.
1444 strcpy(filename_plus_orig_suffix, u->local);
1445 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1447 /* Try to stat() the .orig file. */
1448 if (stat(filename_plus_orig_suffix, &st) == 0)
1450 local_dot_orig_file_exists = TRUE;
1451 local_filename = filename_plus_orig_suffix;
1455 if (!local_dot_orig_file_exists)
1456 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1457 if (stat (u->local, &st) == 0)
1458 local_filename = u->local;
1460 if (local_filename != NULL)
1461 /* There was a local file, so we'll check later to see if the version
1462 the server has is the same version we already have, allowing us to
1467 local_size = st.st_size;
1471 /* Reset the counter. */
1473 *dt = 0 | ACCEPTRANGES;
1477 /* Increment the pass counter. */
1479 sleep_between_retrievals (count);
1480 /* Get the current time string. */
1481 tms = time_str (NULL);
1482 /* Print fetch message, if opt.verbose. */
1485 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1489 sprintf (tmp, _("(try:%2d)"), count);
1490 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1491 tms, hurl, tmp, locf);
1493 ws_changetitle (hurl, 1);
1498 /* Default document type is empty. However, if spider mode is
1499 on or time-stamping is employed, HEAD_ONLY commands is
1500 encoded within *dt. */
1501 if (opt.spider || (use_ts && !got_head))
1505 /* Assume no restarting. */
1507 /* Decide whether or not to restart. */
1508 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1509 && file_exists_p (locf))
1510 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1511 hstat.restval = st.st_size;
1512 /* Decide whether to send the no-cache directive. */
1513 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1514 *dt |= SEND_NOCACHE;
1516 *dt &= ~SEND_NOCACHE;
1518 /* Try fetching the document, or at least its head. :-) */
1519 err = gethttp (u, &hstat, dt);
1521 /* It's unfortunate that wget determines the local filename before finding
1522 out the Content-Type of the file. Barring a major restructuring of the
1523 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1524 u->local to tack on ".html". */
1525 if (!opt.output_document)
1528 locf = opt.output_document;
1530 /* In `-c' is used, check whether the file we're writing to
1531 exists before we've done anything. If so, we'll refuse to
1532 truncate it if the server doesn't support continued
1534 if (opt.always_rest)
1535 hstat.no_truncate = file_exists_p (locf);
1538 tms = time_str (NULL);
1539 /* Get the new location (with or without the redirection). */
1541 *newloc = xstrdup (hstat.newloc);
1544 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1545 case CONERROR: case READERR: case WRITEFAILED:
1547 /* Non-fatal errors continue executing the loop, which will
1548 bring them to "while" statement at the end, to judge
1549 whether the number of tries was exceeded. */
1551 printwhat (count, opt.ntry);
1554 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1555 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1556 /* Fatal errors just return from the function. */
1558 xfree (filename_plus_orig_suffix); /* must precede every return! */
1561 case FWRITEERR: case FOPENERR:
1562 /* Another fatal error. */
1563 logputs (LOG_VERBOSE, "\n");
1564 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1565 u->local, strerror (errno));
1570 /* Another fatal error. */
1571 logputs (LOG_VERBOSE, "\n");
1572 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1574 xfree (filename_plus_orig_suffix); /* must precede every return! */
1578 /* Return the new location to the caller. */
1581 logprintf (LOG_NOTQUIET,
1582 _("ERROR: Redirection (%d) without location.\n"),
1584 xfree (filename_plus_orig_suffix); /* must precede every return! */
1588 xfree (filename_plus_orig_suffix); /* must precede every return! */
1592 /* Deal with you later. */
1595 /* All possibilities should have been exhausted. */
1598 if (!(*dt & RETROKF))
1602 /* #### Ugly ugly ugly! */
1603 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1604 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1607 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1608 tms, hstat.statcode, hstat.error);
1609 logputs (LOG_VERBOSE, "\n");
1611 xfree (filename_plus_orig_suffix); /* must precede every return! */
1615 /* Did we get the time-stamp? */
1618 if (opt.timestamping && !hstat.remote_time)
1620 logputs (LOG_NOTQUIET, _("\
1621 Last-modified header missing -- time-stamps turned off.\n"));
1623 else if (hstat.remote_time)
1625 /* Convert the date-string into struct tm. */
1626 tmr = http_atotm (hstat.remote_time);
1627 if (tmr == (time_t) (-1))
1628 logputs (LOG_VERBOSE, _("\
1629 Last-modified header invalid -- time-stamp ignored.\n"));
1633 /* The time-stamping section. */
1638 use_ts = 0; /* no more time-stamping */
1639 count = 0; /* the retrieve count for HEAD is
1641 if (hstat.remote_time && tmr != (time_t) (-1))
1643 /* Now time-stamping can be used validly. Time-stamping
1644 means that if the sizes of the local and remote file
1645 match, and local file is newer than the remote file,
1646 it will not be retrieved. Otherwise, the normal
1647 download procedure is resumed. */
1649 (hstat.contlen == -1 || local_size == hstat.contlen))
1651 logprintf (LOG_VERBOSE, _("\
1652 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1655 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1658 else if (tml >= tmr)
1659 logprintf (LOG_VERBOSE, _("\
1660 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1662 logputs (LOG_VERBOSE,
1663 _("Remote file is newer, retrieving.\n"));
1668 if ((tmr != (time_t) (-1))
1670 && ((hstat.len == hstat.contlen) ||
1671 ((hstat.res == 0) &&
1672 ((hstat.contlen == -1) ||
1673 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1675 /* #### This code repeats in http.c and ftp.c. Move it to a
1677 const char *fl = NULL;
1678 if (opt.output_document)
1680 if (opt.od_known_regular)
1681 fl = opt.output_document;
1688 /* End of time-stamping section. */
1692 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1693 xfree (filename_plus_orig_suffix); /* must precede every return! */
1697 /* It is now safe to free the remainder of hstat, since the
1698 strings within it will no longer be used. */
1701 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1703 if (hstat.len == hstat.contlen)
1707 logprintf (LOG_VERBOSE,
1708 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1709 tms, tmrate, locf, hstat.len, hstat.contlen);
1710 logprintf (LOG_NONVERBOSE,
1711 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1712 tms, u->url, hstat.len, hstat.contlen, locf, count);
1715 downloaded_increase (hstat.len);
1717 /* Remember that we downloaded the file for later ".orig" code. */
1718 if (*dt & ADDED_HTML_EXTENSION)
1719 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1721 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1723 xfree(filename_plus_orig_suffix); /* must precede every return! */
1726 else if (hstat.res == 0) /* No read error */
1728 if (hstat.contlen == -1) /* We don't know how much we were supposed
1729 to get, so assume we succeeded. */
1733 logprintf (LOG_VERBOSE,
1734 _("%s (%s) - `%s' saved [%ld]\n\n"),
1735 tms, tmrate, locf, hstat.len);
1736 logprintf (LOG_NONVERBOSE,
1737 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1738 tms, u->url, hstat.len, locf, count);
1741 downloaded_increase (hstat.len);
1743 /* Remember that we downloaded the file for later ".orig" code. */
1744 if (*dt & ADDED_HTML_EXTENSION)
1745 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1747 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1749 xfree (filename_plus_orig_suffix); /* must precede every return! */
1752 else if (hstat.len < hstat.contlen) /* meaning we lost the
1753 connection too soon */
1755 logprintf (LOG_VERBOSE,
1756 _("%s (%s) - Connection closed at byte %ld. "),
1757 tms, tmrate, hstat.len);
1758 printwhat (count, opt.ntry);
1761 else if (!opt.kill_longer) /* meaning we got more than expected */
1763 logprintf (LOG_VERBOSE,
1764 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1765 tms, tmrate, locf, hstat.len, hstat.contlen);
1766 logprintf (LOG_NONVERBOSE,
1767 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1768 tms, u->url, hstat.len, hstat.contlen, locf, count);
1770 downloaded_increase (hstat.len);
1772 /* Remember that we downloaded the file for later ".orig" code. */
1773 if (*dt & ADDED_HTML_EXTENSION)
1774 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1776 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1778 xfree (filename_plus_orig_suffix); /* must precede every return! */
1781 else /* the same, but not accepted */
1783 logprintf (LOG_VERBOSE,
1784 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1785 tms, tmrate, hstat.len, hstat.contlen);
1786 printwhat (count, opt.ntry);
1790 else /* now hstat.res can only be -1 */
1792 if (hstat.contlen == -1)
1794 logprintf (LOG_VERBOSE,
1795 _("%s (%s) - Read error at byte %ld (%s)."),
1796 tms, tmrate, hstat.len, strerror (errno));
1797 printwhat (count, opt.ntry);
1800 else /* hstat.res == -1 and contlen is given */
1802 logprintf (LOG_VERBOSE,
1803 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1804 tms, tmrate, hstat.len, hstat.contlen,
1806 printwhat (count, opt.ntry);
1813 while (!opt.ntry || (count < opt.ntry));
1814 xfree (filename_plus_orig_suffix); /* must precede every return! */
1818 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1819 than local timezone (mktime assumes the latter).
1821 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1822 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1824 mktime_from_utc (struct tm *t)
1831 tb = mktime (gmtime (&tl));
1832 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1835 /* Check whether the result of strptime() indicates success.
1836 strptime() returns the pointer to how far it got to in the string.
1837 The processing has been successful if the string is at `GMT' or
1838 `+X', or at the end of the string.
1840 In extended regexp parlance, the function returns 1 if P matches
1841 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1842 strptime()) is considered a failure and 0 is returned. */
1844 check_end (const char *p)
1848 while (ISSPACE (*p))
1851 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1852 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1858 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1859 the three formats RFC2068 allows the HTTP servers to emit --
1860 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1863 We use strptime() to recognize various dates, which makes it a
1864 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1865 allows shortened dates and months, one-digit days, etc.). It also
1866 allows more than one space anywhere where the specs require one SP.
1867 The routine should probably be even more forgiving (as recommended
1868 by RFC2068), but I do not have the time to write one.
1870 Return the computed time_t representation, or -1 if all the
1873 Needless to say, what we *really* need here is something like
1874 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1875 and does not use strptime(). atotm() is to be found in the sources
1876 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1878 http_atotm (char *time_string)
1882 /* Roger Beeman says: "This function dynamically allocates struct tm
1883 t, but does no initialization. The only field that actually
1884 needs initialization is tm_isdst, since the others will be set by
1885 strptime. Since strptime does not set tm_isdst, it will return
1886 the data structure with whatever data was in tm_isdst to begin
1887 with. For those of us in timezones where DST can occur, there
1888 can be a one hour shift depending on the previous contents of the
1889 data area where the data structure is allocated." */
1892 /* Note that under foreign locales Solaris strptime() fails to
1893 recognize English dates, which renders this function useless. I
1894 assume that other non-GNU strptime's are plagued by the same
1895 disease. We solve this by setting only LC_MESSAGES in
1896 i18n_initialize(), instead of LC_ALL.
1898 Another solution could be to temporarily set locale to C, invoke
1899 strptime(), and restore it back. This is slow and dirty,
1900 however, and locale support other than LC_MESSAGES can mess other
1901 things, so I rather chose to stick with just setting LC_MESSAGES.
1903 Also note that none of this is necessary under GNU strptime(),
1904 because it recognizes both international and local dates. */
1906 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1907 it to eat all white space up to (and including) a newline, and
1908 the function fails if there is no newline (!).
1910 Let's hope all strptime() implementations use ` ' to skip *all*
1911 whitespace instead of just one (it works that way on all the
1912 systems I've tested it on). */
1914 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1915 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1916 return mktime_from_utc (&t);
1917 /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1918 if (check_end (strptime (time_string, "%A, %d-%b-%y %T", &t)))
1919 return mktime_from_utc (&t);
1920 /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1921 (google.com uses this for their cookies.)*/
1922 if (check_end (strptime (time_string, "%a, %d-%b-%Y %T", &t)))
1923 return mktime_from_utc (&t);
1924 /* asctime: Thu Jan 29 22:12:57 1998 */
1925 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1926 return mktime_from_utc (&t);
1931 /* Authorization support: We support two authorization schemes:
1933 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1935 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1936 consisting of answering to the server's challenge with the proper
1939 /* How many bytes it will take to store LEN bytes in base64. */
1940 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1942 /* Encode the string S of length LENGTH to base64 format and place it
1943 to STORE. STORE will be 0-terminated, and must point to a writable
1944 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1946 base64_encode (const char *s, char *store, int length)
1948 /* Conversion table. */
1949 static char tbl[64] = {
1950 'A','B','C','D','E','F','G','H',
1951 'I','J','K','L','M','N','O','P',
1952 'Q','R','S','T','U','V','W','X',
1953 'Y','Z','a','b','c','d','e','f',
1954 'g','h','i','j','k','l','m','n',
1955 'o','p','q','r','s','t','u','v',
1956 'w','x','y','z','0','1','2','3',
1957 '4','5','6','7','8','9','+','/'
1960 unsigned char *p = (unsigned char *)store;
1962 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1963 for (i = 0; i < length; i += 3)
1965 *p++ = tbl[s[0] >> 2];
1966 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1967 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1968 *p++ = tbl[s[2] & 0x3f];
1971 /* Pad the result if necessary... */
1972 if (i == length + 1)
1974 else if (i == length + 2)
1975 *(p - 1) = *(p - 2) = '=';
1976 /* ...and zero-terminate it. */
1980 /* Create the authentication header contents for the `Basic' scheme.
1981 This is done by encoding the string `USER:PASS' in base64 and
1982 prepending `HEADER: Basic ' to it. */
1984 basic_authentication_encode (const char *user, const char *passwd,
1987 char *t1, *t2, *res;
1988 int len1 = strlen (user) + 1 + strlen (passwd);
1989 int len2 = BASE64_LENGTH (len1);
1991 t1 = (char *)alloca (len1 + 1);
1992 sprintf (t1, "%s:%s", user, passwd);
1993 t2 = (char *)alloca (1 + len2);
1994 base64_encode (t1, t2, len1);
1995 res = (char *)xmalloc (len2 + 11 + strlen (header));
1996 sprintf (res, "%s: Basic %s\r\n", header, t2);
2002 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2003 of a field in such a header. If the field is the one specified by
2004 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2005 digest authorization code), extract its value in the (char*)
2006 variable pointed by RET. Returns negative on a malformed header,
2007 or number of bytes that have been parsed by this call. */
2009 extract_header_attr (const char *au, const char *attr_name, char **ret)
2011 const char *cp, *ep;
2015 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2017 cp += strlen (attr_name);
2020 cp += skip_lws (cp);
2025 cp += skip_lws (cp);
2030 for (ep = cp; *ep && *ep != '\"'; ep++)
2035 *ret = strdupdelim (cp, ep);
2042 /* Response value needs to be in lowercase, so we cannot use HEXD2ASC
2043 from url.h. See RFC 2069 2.1.2 for the syntax of response-digest. */
2044 #define HEXD2asc(x) (((x) < 10) ? ((x) + '0') : ((x) - 10 + 'a'))
2046 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2047 an array of 16 bytes containing the hash keys, and BUF should be a
2048 buffer of 33 writable characters (32 for hex digits plus one for
2049 zero termination). */
2051 dump_hash (unsigned char *buf, const unsigned char *hash)
2055 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2057 *buf++ = HEXD2asc (*hash >> 4);
2058 *buf++ = HEXD2asc (*hash & 0xf);
2063 /* Take the line apart to find the challenge, and compose a digest
2064 authorization header. See RFC2069 section 2.1.2. */
2066 digest_authentication_encode (const char *au, const char *user,
2067 const char *passwd, const char *method,
2070 static char *realm, *opaque, *nonce;
2075 { "realm", &realm },
2076 { "opaque", &opaque },
2081 realm = opaque = nonce = NULL;
2083 au += 6; /* skip over `Digest' */
2088 au += skip_lws (au);
2089 for (i = 0; i < ARRAY_SIZE (options); i++)
2091 int skip = extract_header_attr (au, options[i].name,
2092 options[i].variable);
2096 FREE_MAYBE (opaque);
2106 if (i == ARRAY_SIZE (options))
2108 while (*au && *au != '=')
2112 au += skip_lws (au);
2116 while (*au && *au != '\"')
2123 while (*au && *au != ',')
2128 if (!realm || !nonce || !user || !passwd || !path || !method)
2131 FREE_MAYBE (opaque);
2136 /* Calculate the digest value. */
2139 unsigned char hash[MD5_HASHLEN];
2140 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2141 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2143 /* A1BUF = H(user ":" realm ":" password) */
2144 md5_init_ctx (&ctx);
2145 md5_process_bytes (user, strlen (user), &ctx);
2146 md5_process_bytes (":", 1, &ctx);
2147 md5_process_bytes (realm, strlen (realm), &ctx);
2148 md5_process_bytes (":", 1, &ctx);
2149 md5_process_bytes (passwd, strlen (passwd), &ctx);
2150 md5_finish_ctx (&ctx, hash);
2151 dump_hash (a1buf, hash);
2153 /* A2BUF = H(method ":" path) */
2154 md5_init_ctx (&ctx);
2155 md5_process_bytes (method, strlen (method), &ctx);
2156 md5_process_bytes (":", 1, &ctx);
2157 md5_process_bytes (path, strlen (path), &ctx);
2158 md5_finish_ctx (&ctx, hash);
2159 dump_hash (a2buf, hash);
2161 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2162 md5_init_ctx (&ctx);
2163 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2164 md5_process_bytes (":", 1, &ctx);
2165 md5_process_bytes (nonce, strlen (nonce), &ctx);
2166 md5_process_bytes (":", 1, &ctx);
2167 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2168 md5_finish_ctx (&ctx, hash);
2169 dump_hash (response_digest, hash);
2171 res = (char*) xmalloc (strlen (user)
2176 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2177 + (opaque ? strlen (opaque) : 0)
2179 sprintf (res, "Authorization: Digest \
2180 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2181 user, realm, nonce, path, response_digest);
2184 char *p = res + strlen (res);
2185 strcat (p, ", opaque=\"");
2189 strcat (res, "\r\n");
2193 #endif /* USE_DIGEST */
2196 #define BEGINS_WITH(line, string_constant) \
2197 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2198 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2199 || !line[sizeof (string_constant) - 1]))
2202 known_authentication_scheme_p (const char *au)
2204 return BEGINS_WITH (au, "Basic")
2205 || BEGINS_WITH (au, "Digest")
2206 || BEGINS_WITH (au, "NTLM");
2211 /* Create the HTTP authorization request header. When the
2212 `WWW-Authenticate' response header is seen, according to the
2213 authorization scheme specified in that header (`Basic' and `Digest'
2214 are supported by the current implementation), produce an
2215 appropriate HTTP authorization request header. */
2217 create_authorization_line (const char *au, const char *user,
2218 const char *passwd, const char *method,
2221 char *wwwauth = NULL;
2223 if (!strncasecmp (au, "Basic", 5))
2224 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2225 if (!strncasecmp (au, "NTLM", 4))
2226 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2228 else if (!strncasecmp (au, "Digest", 6))
2229 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2230 #endif /* USE_DIGEST */