2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001, 2002
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.
21 In addition, as a special exception, the Free Software Foundation
22 gives permission to link the code of its release of Wget with the
23 OpenSSL project's "OpenSSL" library (or with modified versions of it
24 that use the same license as the "OpenSSL" library), and distribute
25 the linked executables. You must obey the GNU General Public License
26 in all respects for all of the code used other than "OpenSSL". If you
27 modify this file, you may extend this exception to your version of the
28 file, but you are not obligated to do so. If you do not wish to do
29 so, delete this exception statement from your version. */
35 #include <sys/types.h>
46 #if TIME_WITH_SYS_TIME
47 # include <sys/time.h>
51 # include <sys/time.h>
70 # include "gen_sslfunc.h"
78 extern char *version_string;
79 extern LARGE_INT total_downloaded_bytes;
82 static int cookies_loaded_p;
83 struct cookie_jar *wget_cookie_jar;
85 #define TEXTHTML_S "text/html"
86 #define TEXTXHTML_S "application/xhtml+xml"
87 #define HTTP_ACCEPT "*/*"
89 /* Some status code validation macros: */
90 #define H_20X(x) (((x) >= 200) && ((x) < 300))
91 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
92 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
93 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
94 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
96 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
98 #define HTTP_STATUS_OK 200
99 #define HTTP_STATUS_CREATED 201
100 #define HTTP_STATUS_ACCEPTED 202
101 #define HTTP_STATUS_NO_CONTENT 204
102 #define HTTP_STATUS_PARTIAL_CONTENTS 206
104 /* Redirection 3xx. */
105 #define HTTP_STATUS_MULTIPLE_CHOICES 300
106 #define HTTP_STATUS_MOVED_PERMANENTLY 301
107 #define HTTP_STATUS_MOVED_TEMPORARILY 302
108 #define HTTP_STATUS_NOT_MODIFIED 304
109 #define HTTP_STATUS_TEMPORARY_REDIRECT 307
111 /* Client error 4xx. */
112 #define HTTP_STATUS_BAD_REQUEST 400
113 #define HTTP_STATUS_UNAUTHORIZED 401
114 #define HTTP_STATUS_FORBIDDEN 403
115 #define HTTP_STATUS_NOT_FOUND 404
117 /* Server errors 5xx. */
118 #define HTTP_STATUS_INTERNAL 500
119 #define HTTP_STATUS_NOT_IMPLEMENTED 501
120 #define HTTP_STATUS_BAD_GATEWAY 502
121 #define HTTP_STATUS_UNAVAILABLE 503
124 /* Parse the HTTP status line, which is of format:
126 HTTP-Version SP Status-Code SP Reason-Phrase
128 The function returns the status-code, or -1 if the status line is
129 malformed. The pointer to reason-phrase is returned in RP. */
131 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
133 /* (the variables must not be named `major' and `minor', because
134 that breaks compilation with SunOS4 cc.) */
135 int mjr, mnr, statcode;
138 *reason_phrase_ptr = NULL;
140 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
141 major version, and Y is minor version. */
142 if (strncmp (line, "HTTP/", 5) != 0)
146 /* Calculate major HTTP version. */
148 for (mjr = 0; ISDIGIT (*line); line++)
149 mjr = 10 * mjr + (*line - '0');
150 if (*line != '.' || p == line)
154 /* Calculate minor HTTP version. */
156 for (mnr = 0; ISDIGIT (*line); line++)
157 mnr = 10 * mnr + (*line - '0');
158 if (*line != ' ' || p == line)
160 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
161 minor version can be safely ignored. */
166 /* Calculate status code. */
167 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
169 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
171 /* Set up the reason phrase pointer. */
173 /* RFC2068 requires SPC here, but we allow the string to finish
174 here, in case no reason-phrase is present. */
178 *reason_phrase_ptr = line;
183 *reason_phrase_ptr = line + 1;
188 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
190 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
191 PROMISED_SIZE bytes are sent over the wire -- if the file is
192 longer, read only that much; if the file is shorter, report an error. */
195 post_file (int sock, void *ssl, const char *file_name, long promised_size)
197 static char chunk[8192];
202 /* Only one of SOCK and SSL may be active at the same time. */
203 assert (sock > -1 || ssl != NULL);
204 assert (sock == -1 || ssl == NULL);
206 DEBUGP (("[writing POST file %s ... ", file_name));
208 fp = fopen (file_name, "rb");
211 while (!feof (fp) && written < promised_size)
214 int length = fread (chunk, 1, sizeof (chunk), fp);
217 towrite = WMIN (promised_size - written, length);
220 write_error = ssl_iwrite (ssl, chunk, towrite);
223 write_error = iwrite (sock, chunk, towrite);
233 /* If we've written less than was promised, report a (probably
234 nonsensical) error rather than break the promise. */
235 if (written < promised_size)
241 assert (written == promised_size);
242 DEBUGP (("done]\n"));
246 /* Functions to be used as arguments to header_process(): */
248 struct http_process_range_closure {
254 /* Parse the `Content-Range' header and extract the information it
255 contains. Returns 1 if successful, -1 otherwise. */
257 http_process_range (const char *hdr, void *arg)
259 struct http_process_range_closure *closure
260 = (struct http_process_range_closure *)arg;
263 /* Certain versions of Nutscape proxy server send out
264 `Content-Length' without "bytes" specifier, which is a breach of
265 RFC2068 (as well as the HTTP/1.1 draft which was current at the
266 time). But hell, I must support it... */
267 if (!strncasecmp (hdr, "bytes", 5))
270 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
274 hdr += skip_lws (hdr);
280 for (num = 0; ISDIGIT (*hdr); hdr++)
281 num = 10 * num + (*hdr - '0');
282 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
284 closure->first_byte_pos = num;
286 for (num = 0; ISDIGIT (*hdr); hdr++)
287 num = 10 * num + (*hdr - '0');
288 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
290 closure->last_byte_pos = num;
292 for (num = 0; ISDIGIT (*hdr); hdr++)
293 num = 10 * num + (*hdr - '0');
294 closure->entity_length = num;
298 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
299 Used for `Accept-Ranges'. */
301 http_process_none (const char *hdr, void *arg)
303 int *where = (int *)arg;
305 if (strstr (hdr, "none"))
312 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
314 http_process_type (const char *hdr, void *arg)
316 char **result = (char **)arg;
317 /* Locate P on `;' or the terminating zero, whichever comes first. */
318 const char *p = strchr (hdr, ';');
320 p = hdr + strlen (hdr);
321 while (p > hdr && ISSPACE (*(p - 1)))
323 *result = strdupdelim (hdr, p);
327 /* Check whether the `Connection' header is set to "keep-alive". */
329 http_process_connection (const char *hdr, void *arg)
331 int *flag = (int *)arg;
332 if (!strcasecmp (hdr, "Keep-Alive"))
337 /* Commit the cookie to the cookie jar. */
340 http_process_set_cookie (const char *hdr, void *arg)
342 struct url *u = (struct url *)arg;
344 /* The jar should have been created by now. */
345 assert (wget_cookie_jar != NULL);
347 cookie_jar_process_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
353 /* Persistent connections. Currently, we cache the most recently used
354 connection as persistent, provided that the HTTP server agrees to
355 make it such. The persistence data is stored in the variables
356 below. Ideally, it would be in a structure, and it should be
357 possible to cache an arbitrary fixed number of these connections.
359 I think the code is quite easy to extend in that direction. */
361 /* Whether a persistent connection is active. */
362 static int pc_active_p;
363 /* Host and port of currently active persistent connection. */
364 static struct address_list *pc_last_host_ip;
365 static unsigned short pc_last_port;
367 /* File descriptor of the currently active persistent connection. */
368 static int pc_last_fd;
371 /* Whether a ssl handshake has occoured on this connection */
372 static int pc_active_ssl;
373 /* SSL connection of the currently active persistent connection. */
374 static SSL *pc_last_ssl;
375 #endif /* HAVE_SSL */
377 /* Mark the persistent connection as invalid. This is used by the
378 CLOSE_* macros after they forcefully close a registered persistent
379 connection. This does not close the file descriptor -- it is left
380 to the caller to do that. (Maybe it should, though.) */
383 invalidate_persistent (void)
388 #endif /* HAVE_SSL */
389 if (pc_last_host_ip != NULL)
391 address_list_release (pc_last_host_ip);
392 pc_last_host_ip = NULL;
394 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
397 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
398 persistent. This will enable someone to use the same connection
399 later. In the context of HTTP, this must be called only AFTER the
400 response has been received and the server has promised that the
401 connection will remain alive.
403 If a previous connection was persistent, it is closed. */
407 register_persistent (const char *host, unsigned short port, int fd, SSL *ssl)
411 register_persistent (const char *host, unsigned short port, int fd)
416 if (pc_last_fd == fd)
418 /* The connection FD is already registered. Nothing to
424 /* The old persistent connection is still active; let's
425 close it first. This situation arises whenever a
426 persistent connection exists, but we then connect to a
427 different host, and try to register a persistent
428 connection to that one. */
430 /* The ssl disconnect has to take place before the closing
433 shutdown_ssl(pc_last_ssl);
436 invalidate_persistent ();
440 assert (pc_last_host_ip == NULL);
442 /* This lookup_host cannot fail, because it has the results in the
444 pc_last_host_ip = lookup_host (host, LH_SILENT);
445 assert (pc_last_host_ip != NULL);
452 pc_active_ssl = ssl ? 1 : 0;
454 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
458 # define SHUTDOWN_SSL(ssl) do { \
460 shutdown_ssl (ssl); \
463 # define SHUTDOWN_SSL(ssl)
466 /* Return non-zero if a persistent connection is available for
467 connecting to HOST:PORT. */
471 persistent_available_p (const char *host, unsigned short port, int ssl)
475 persistent_available_p (const char *host, unsigned short port)
479 struct address_list *this_host_ip;
481 /* First, check whether a persistent connection is active at all. */
484 /* Second, check if the active connection pertains to the correct
485 (HOST, PORT) ordered pair. */
486 if (port != pc_last_port)
490 /* Second, a): check if current connection is (not) ssl, too. This
491 test is unlikely to fail because HTTP and HTTPS typicaly use
492 different ports. Yet it is possible, or so I [Christian
493 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
495 if (ssl != pc_active_ssl)
497 #endif /* HAVE_SSL */
499 this_host_ip = lookup_host (host, LH_SILENT);
503 /* To equate the two host names for the purposes of persistent
504 connections, they need to share all the IP addresses in the
506 success = address_list_match_all (pc_last_host_ip, this_host_ip);
507 address_list_release (this_host_ip);
511 /* Third: check whether the connection is still open. This is
512 important because most server implement a liberal (short) timeout
513 on persistent connections. Wget can of course always reconnect
514 if the connection doesn't work out, but it's nicer to know in
515 advance. This test is a logical followup of the first test, but
516 is "expensive" and therefore placed at the end of the list. */
517 if (!test_socket_open (pc_last_fd))
519 /* Oops, the socket is no longer open. Now that we know that,
520 let's invalidate the persistent connection before returning
524 SHUTDOWN_SSL (pc_last_ssl);
527 invalidate_persistent ();
533 /* The idea behind these two CLOSE macros is to distinguish between
534 two cases: one when the job we've been doing is finished, and we
535 want to close the connection and leave, and two when something is
536 seriously wrong and we're closing the connection as part of
539 In case of keep_alive, CLOSE_FINISH should leave the connection
540 open, while CLOSE_INVALIDATE should still close it.
542 Note that the semantics of the flag `keep_alive' is "this
543 connection *will* be reused (the server has promised not to close
544 the connection once we're done)", while the semantics of
545 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
546 active, registered connection". */
548 #define CLOSE_FINISH(fd) do { \
551 SHUTDOWN_SSL (ssl); \
553 if (pc_active_p && (fd) == pc_last_fd) \
554 invalidate_persistent (); \
558 #define CLOSE_INVALIDATE(fd) do { \
559 SHUTDOWN_SSL (ssl); \
561 if (pc_active_p && (fd) == pc_last_fd) \
562 invalidate_persistent (); \
567 long len; /* received length */
568 long contlen; /* expected length */
569 long restval; /* the restart value */
570 int res; /* the result of last read */
571 char *newloc; /* new location (redirection) */
572 char *remote_time; /* remote time-stamp string */
573 char *error; /* textual HTTP error */
574 int statcode; /* status code */
575 double dltime; /* time of the download in msecs */
576 int no_truncate; /* whether truncating the file is
578 const char *referer; /* value of the referer header. */
579 char **local_file; /* local file. */
583 free_hstat (struct http_stat *hs)
585 FREE_MAYBE (hs->newloc);
586 FREE_MAYBE (hs->remote_time);
587 FREE_MAYBE (hs->error);
589 /* Guard against being called twice. */
591 hs->remote_time = NULL;
595 static char *create_authorization_line PARAMS ((const char *, const char *,
596 const char *, const char *,
598 static char *basic_authentication_encode PARAMS ((const char *, const char *,
600 static int known_authentication_scheme_p PARAMS ((const char *));
602 time_t http_atotm PARAMS ((const char *));
604 #define BEGINS_WITH(line, string_constant) \
605 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
606 && (ISSPACE (line[sizeof (string_constant) - 1]) \
607 || !line[sizeof (string_constant) - 1]))
609 /* Retrieve a document through HTTP protocol. It recognizes status
610 code, and correctly handles redirections. It closes the network
611 socket. If it receives an error from the functions below it, it
612 will print it if there is enough information to do so (almost
613 always), returning the error to the caller (i.e. http_loop).
615 Various HTTP parameters are stored to hs. Although it parses the
616 response code correctly, it is not used in a sane way. The caller
619 If PROXY is non-NULL, the connection will be made to the proxy
620 server, and u->url will be requested. */
622 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
624 char *request, *type, *command, *full_path;
626 char *pragma_h, *referer, *useragent, *range, *wwwauth;
627 char *authenticate_h;
631 char *request_keep_alive;
632 int sock, hcount, all_length, statcode;
634 long contlen, contrange;
637 int auth_tried_already;
640 static SSL_CTX *ssl_ctx = NULL;
643 char *cookies = NULL;
645 /* Whether this connection will be kept alive after the HTTP request
649 /* Flags that detect the two ways of specifying HTTP keep-alive
651 int http_keep_alive_1, http_keep_alive_2;
653 /* Whether keep-alive should be inhibited. */
654 int inhibit_keep_alive;
656 /* Whether we need to print the host header with braces around host,
657 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
658 "Host: symbolic-name:1234". */
659 int squares_around_host = 0;
661 /* Headers sent when using POST. */
662 char *post_content_type, *post_content_length;
663 long post_data_size = 0;
666 /* initialize ssl_ctx on first run */
669 uerr_t err = init_ssl (&ssl_ctx);
674 case SSLERRCTXCREATE:
676 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
680 /* try without certfile */
681 logprintf (LOG_NOTQUIET,
682 _("Failed to load certificates from %s\n"),
685 logprintf (LOG_NOTQUIET,
686 _("Trying without the specified certificate\n"));
689 logprintf (LOG_NOTQUIET,
690 _("Failed to get certificate key from %s\n"),
693 logprintf (LOG_NOTQUIET,
694 _("Trying without the specified certificate\n"));
701 #endif /* HAVE_SSL */
703 if (!(*dt & HEAD_ONLY))
704 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
705 know the local filename so we can save to it. */
706 assert (*hs->local_file != NULL);
709 auth_tried_already = 0;
711 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
714 /* We need to come back here when the initial attempt to retrieve
715 without authorization header fails. (Expected to happen at least
716 for the Digest authorization scheme.) */
719 http_keep_alive_1 = http_keep_alive_2 = 0;
721 post_content_type = NULL;
722 post_content_length = NULL;
724 /* Initialize certain elements of struct http_stat. */
729 hs->remote_time = NULL;
732 /* If we're using a proxy, we will be connecting to the proxy
734 conn = proxy ? proxy : u;
736 /* First: establish the connection. */
737 if (inhibit_keep_alive
740 !persistent_available_p (conn->host, conn->port)
742 !persistent_available_p (conn->host, conn->port,
743 u->scheme == SCHEME_HTTPS)
744 #endif /* HAVE_SSL */
747 sock = connect_to_host (conn->host, conn->port);
751 return CONNECT_ERROR (errno);
754 if (conn->scheme == SCHEME_HTTPS)
755 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
757 logputs (LOG_VERBOSE, "\n");
758 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
763 #endif /* HAVE_SSL */
767 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
768 conn->host, conn->port);
769 /* #### pc_last_fd should be accessed through an accessor
774 #endif /* HAVE_SSL */
775 DEBUGP (("Reusing fd %d.\n", sock));
780 else if (opt.post_file_name || opt.post_data)
788 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
789 sprintf (referer, "Referer: %s\r\n", hs->referer);
792 if (*dt & SEND_NOCACHE)
793 pragma_h = "Pragma: no-cache\r\n";
799 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
800 /* Gag me! Some servers (e.g. WebSitePro) have been known to
801 respond to the following `Range' format by generating a
802 multipart/x-byte-ranges MIME document! This MIME type was
803 present in an old draft of the byteranges specification.
804 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
805 only if multiple non-overlapping ranges are requested --
806 which Wget never does. */
807 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
812 STRDUP_ALLOCA (useragent, opt.useragent);
815 useragent = (char *)alloca (10 + strlen (version_string));
816 sprintf (useragent, "Wget/%s", version_string);
818 /* Construct the authentication, if userid is present. */
821 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
822 user = user ? user : opt.http_user;
823 passwd = passwd ? passwd : opt.http_passwd;
830 /* We have the username and the password, but haven't tried
831 any authorization yet. Let's see if the "Basic" method
832 works. If not, we'll come back here and construct a
833 proper authorization method with the right challenges.
835 If we didn't employ this kind of logic, every URL that
836 requires authorization would have to be processed twice,
837 which is very suboptimal and generates a bunch of false
838 "unauthorized" errors in the server log.
840 #### But this logic also has a serious problem when used
841 with stronger authentications: we *first* transmit the
842 username and the password in clear text, and *then*
843 attempt a stronger authentication scheme. That cannot be
844 right! We are only fortunate that almost everyone still
845 uses the `Basic' scheme anyway.
847 There should be an option to prevent this from happening,
848 for those who use strong authentication schemes and value
850 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
854 /* Use the full path, i.e. one that includes the leading
855 slash and the query string, but is independent of proxy
857 char *pth = url_full_path (u);
858 wwwauth = create_authorization_line (authenticate_h, user, passwd,
867 char *proxy_user, *proxy_passwd;
868 /* For normal username and password, URL components override
869 command-line/wgetrc parameters. With proxy authentication,
870 it's the reverse, because proxy URLs are normally the
871 "permanent" ones, so command-line args should take
873 if (opt.proxy_user && opt.proxy_passwd)
875 proxy_user = opt.proxy_user;
876 proxy_passwd = opt.proxy_passwd;
880 proxy_user = proxy->user;
881 proxy_passwd = proxy->passwd;
883 /* #### This does not appear right. Can't the proxy request,
884 say, `Digest' authentication? */
885 if (proxy_user && proxy_passwd)
886 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
887 "Proxy-Authorization");
890 /* String of the form :PORT. Used only for non-standard ports. */
892 if (u->port != scheme_default_port (u->scheme))
894 port_maybe = (char *)alloca (numdigit (u->port) + 2);
895 sprintf (port_maybe, ":%d", u->port);
898 if (!inhibit_keep_alive)
899 request_keep_alive = "Connection: Keep-Alive\r\n";
901 request_keep_alive = NULL;
904 cookies = cookie_jar_generate_cookie_header (wget_cookie_jar, u->host,
907 u->scheme == SCHEME_HTTPS
913 if (opt.post_data || opt.post_file_name)
915 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
917 post_data_size = strlen (opt.post_data);
920 post_data_size = file_size (opt.post_file_name);
921 if (post_data_size == -1)
923 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
928 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
929 sprintf (post_content_length,
930 "Content-Length: %ld\r\n", post_data_size);
934 full_path = xstrdup (u->url);
936 /* Use the full path, i.e. one that includes the leading slash and
937 the query string. E.g. if u->path is "foo/bar" and u->query is
938 "param=value", full_path will be "/foo/bar?param=value". */
939 full_path = url_full_path (u);
941 if (strchr (u->host, ':'))
942 squares_around_host = 1;
944 /* Allocate the memory for the request. */
945 request = (char *)alloca (strlen (command)
949 + (port_maybe ? strlen (port_maybe) : 0)
950 + strlen (HTTP_ACCEPT)
951 + (request_keep_alive
952 ? strlen (request_keep_alive) : 0)
953 + (referer ? strlen (referer) : 0)
954 + (cookies ? strlen (cookies) : 0)
955 + (wwwauth ? strlen (wwwauth) : 0)
956 + (proxyauth ? strlen (proxyauth) : 0)
957 + (range ? strlen (range) : 0)
960 ? strlen (post_content_type) : 0)
961 + (post_content_length
962 ? strlen (post_content_length) : 0)
963 + (opt.user_header ? strlen (opt.user_header) : 0)
965 /* Construct the request. */
971 %s%s%s%s%s%s%s%s%s%s\r\n",
974 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
975 port_maybe ? port_maybe : "",
977 request_keep_alive ? request_keep_alive : "",
978 referer ? referer : "",
979 cookies ? cookies : "",
980 wwwauth ? wwwauth : "",
981 proxyauth ? proxyauth : "",
984 post_content_type ? post_content_type : "",
985 post_content_length ? post_content_length : "",
986 opt.user_header ? opt.user_header : "");
987 DEBUGP (("---request begin---\n%s", request));
989 /* Free the temporary memory. */
990 FREE_MAYBE (wwwauth);
991 FREE_MAYBE (proxyauth);
992 FREE_MAYBE (cookies);
995 /* Send the request to server. */
997 if (conn->scheme == SCHEME_HTTPS)
998 write_error = ssl_iwrite (ssl, request, strlen (request));
1001 write_error = iwrite (sock, request, strlen (request));
1003 if (write_error >= 0)
1007 DEBUGP (("[POST data: %s]\n", opt.post_data));
1009 if (conn->scheme == SCHEME_HTTPS)
1010 write_error = ssl_iwrite (ssl, opt.post_data, post_data_size);
1013 write_error = iwrite (sock, opt.post_data, post_data_size);
1015 else if (opt.post_file_name && post_data_size != 0)
1018 if (conn->scheme == SCHEME_HTTPS)
1019 write_error = post_file (-1, ssl, opt.post_file_name,
1023 write_error = post_file (sock, NULL, opt.post_file_name,
1027 DEBUGP (("---request end---\n"));
1029 if (write_error < 0)
1031 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1033 CLOSE_INVALIDATE (sock);
1036 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1037 proxy ? "Proxy" : "HTTP");
1038 contlen = contrange = -1;
1043 /* Before reading anything, initialize the rbuf. */
1044 rbuf_initialize (&rbuf, sock);
1046 if (conn->scheme == SCHEME_HTTPS)
1050 #endif /* HAVE_SSL */
1053 /* Header-fetching loop. */
1061 /* Get the header. */
1062 status = header_get (&rbuf, &hdr,
1063 /* Disallow continuations for status line. */
1064 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
1066 /* Check for errors. */
1067 if (status == HG_EOF && *hdr)
1069 /* This used to be an unconditional error, but that was
1070 somewhat controversial, because of a large number of
1071 broken CGI's that happily "forget" to send the second EOL
1072 before closing the connection of a HEAD request.
1074 So, the deal is to check whether the header is empty
1075 (*hdr is zero if it is); if yes, it means that the
1076 previous header was fully retrieved, and that -- most
1077 probably -- the request is complete. "...be liberal in
1078 what you accept." Oh boy. */
1079 logputs (LOG_VERBOSE, "\n");
1080 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1083 FREE_MAYBE (all_headers);
1084 CLOSE_INVALIDATE (sock);
1087 else if (status == HG_ERROR)
1089 logputs (LOG_VERBOSE, "\n");
1090 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1094 FREE_MAYBE (all_headers);
1095 CLOSE_INVALIDATE (sock);
1099 /* If the headers are to be saved to a file later, save them to
1101 if (opt.save_headers)
1103 int lh = strlen (hdr);
1104 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1105 memcpy (all_headers + all_length, hdr, lh);
1107 all_headers[all_length++] = '\n';
1108 all_headers[all_length] = '\0';
1111 /* Check for status line. */
1115 /* Parse the first line of server response. */
1116 statcode = parse_http_status_line (hdr, &error);
1117 hs->statcode = statcode;
1118 /* Store the descriptive response. */
1119 if (statcode == -1) /* malformed response */
1121 /* A common reason for "malformed response" error is the
1122 case when no data was actually received. Handle this
1125 hs->error = xstrdup (_("No data received"));
1127 hs->error = xstrdup (_("Malformed status line"));
1132 hs->error = xstrdup (_("(no description)"));
1134 hs->error = xstrdup (error);
1136 if ((statcode != -1)
1142 if (opt.server_response)
1143 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1145 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1151 /* Exit on empty header. */
1158 /* Print the header if requested. */
1159 if (opt.server_response && hcount != 1)
1160 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1162 /* Try getting content-length. */
1163 if (contlen == -1 && !opt.ignore_length)
1164 if (header_process (hdr, "Content-Length", header_extract_number,
1167 /* Try getting content-type. */
1169 if (header_process (hdr, "Content-Type", http_process_type, &type))
1171 /* Try getting location. */
1173 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1175 /* Try getting last-modified. */
1176 if (!hs->remote_time)
1177 if (header_process (hdr, "Last-Modified", header_strdup,
1180 /* Try getting cookies. */
1182 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1184 /* Try getting www-authentication. */
1185 if (!authenticate_h)
1186 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1189 /* Check for accept-ranges header. If it contains the word
1190 `none', disable the ranges. */
1191 if (*dt & ACCEPTRANGES)
1194 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1197 *dt &= ~ACCEPTRANGES;
1201 /* Try getting content-range. */
1202 if (contrange == -1)
1204 struct http_process_range_closure closure;
1205 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1207 contrange = closure.first_byte_pos;
1211 /* Check for keep-alive related responses. */
1212 if (!inhibit_keep_alive)
1214 /* Check for the `Keep-Alive' header. */
1215 if (!http_keep_alive_1)
1217 if (header_process (hdr, "Keep-Alive", header_exists,
1218 &http_keep_alive_1))
1221 /* Check for `Connection: Keep-Alive'. */
1222 if (!http_keep_alive_2)
1224 if (header_process (hdr, "Connection", http_process_connection,
1225 &http_keep_alive_2))
1233 logputs (LOG_VERBOSE, "\n");
1236 && (http_keep_alive_1 || http_keep_alive_2))
1238 assert (inhibit_keep_alive == 0);
1242 /* The server has promised that it will not close the connection
1243 when we're done. This means that we can register it. */
1245 register_persistent (conn->host, conn->port, sock);
1247 register_persistent (conn->host, conn->port, sock, ssl);
1248 #endif /* HAVE_SSL */
1250 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1253 /* Authorization is required. */
1257 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1258 might be more bytes in the body. */
1259 if (auth_tried_already)
1261 /* If we have tried it already, then there is not point
1264 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1265 xfree (authenticate_h);
1268 else if (!known_authentication_scheme_p (authenticate_h))
1270 xfree (authenticate_h);
1271 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1274 else if (BEGINS_WITH (authenticate_h, "Basic"))
1276 /* The authentication scheme is basic, the one we try by
1277 default, and it failed. There's no sense in trying
1283 auth_tried_already = 1;
1287 /* We do not need this anymore. */
1290 xfree (authenticate_h);
1291 authenticate_h = NULL;
1294 /* 20x responses are counted among successful by default. */
1295 if (H_20X (statcode))
1298 /* Return if redirected. */
1299 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1301 /* RFC2068 says that in case of the 300 (multiple choices)
1302 response, the server can output a preferred URL through
1303 `Location' header; otherwise, the request should be treated
1304 like GET. So, if the location is set, it will be a
1305 redirection; otherwise, just proceed normally. */
1306 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1310 logprintf (LOG_VERBOSE,
1311 _("Location: %s%s\n"),
1312 hs->newloc ? hs->newloc : _("unspecified"),
1313 hs->newloc ? _(" [following]") : "");
1314 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1315 might be more bytes in the body. */
1317 FREE_MAYBE (all_headers);
1322 /* If content-type is not given, assume text/html. This is because
1323 of the multitude of broken CGI's that "forget" to generate the
1326 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1327 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1332 if (opt.html_extension && (*dt & TEXTHTML))
1333 /* -E / --html-extension / html_extension = on was specified, and this is a
1334 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1335 already the file's suffix, tack on ".html". */
1337 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1339 if (last_period_in_local_filename == NULL
1340 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1341 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1343 size_t local_filename_len = strlen(*hs->local_file);
1345 *hs->local_file = xrealloc(*hs->local_file,
1346 local_filename_len + sizeof(".html"));
1347 strcpy(*hs->local_file + local_filename_len, ".html");
1349 *dt |= ADDED_HTML_EXTENSION;
1353 if (contrange == -1)
1355 /* We did not get a content-range header. This means that the
1356 server did not honor our `Range' request. Normally, this
1357 means we should reset hs->restval and continue normally. */
1359 /* However, if `-c' is used, we need to be a bit more careful:
1361 1. If `-c' is specified and the file already existed when
1362 Wget was started, it would be a bad idea for us to start
1363 downloading it from scratch, effectively truncating it. I
1364 believe this cannot happen unless `-c' was specified.
1366 2. If `-c' is used on a file that is already fully
1367 downloaded, we're requesting bytes after the end of file,
1368 which can result in server not honoring `Range'. If this is
1369 the case, `Content-Length' will be equal to the length of the
1371 if (opt.always_rest)
1373 /* Check for condition #2. */
1374 if (hs->restval > 0 /* restart was requested. */
1375 && contlen != -1 /* we got content-length. */
1376 && hs->restval >= contlen /* file fully downloaded
1380 logputs (LOG_VERBOSE, _("\
1381 \n The file is already fully retrieved; nothing to do.\n\n"));
1382 /* In case the caller inspects. */
1385 /* Mark as successfully retrieved. */
1388 FREE_MAYBE (all_headers);
1389 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1390 might be more bytes in the body. */
1391 return RETRUNNEEDED;
1394 /* Check for condition #1. */
1395 if (hs->no_truncate)
1397 logprintf (LOG_NOTQUIET,
1400 Continued download failed on this file, which conflicts with `-c'.\n\
1401 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1403 FREE_MAYBE (all_headers);
1404 CLOSE_INVALIDATE (sock);
1405 return CONTNOTSUPPORTED;
1413 else if (contrange != hs->restval ||
1414 (H_PARTIAL (statcode) && contrange == -1))
1416 /* This means the whole request was somehow misunderstood by the
1417 server. Bail out. */
1419 FREE_MAYBE (all_headers);
1420 CLOSE_INVALIDATE (sock);
1427 contlen += contrange;
1429 contrange = -1; /* If conent-length was not sent,
1430 content-range will be ignored. */
1432 hs->contlen = contlen;
1436 if ((*dt & RETROKF) && !opt.server_response)
1438 /* No need to print this output if the body won't be
1439 downloaded at all, or if the original server response is
1441 logputs (LOG_VERBOSE, _("Length: "));
1444 logputs (LOG_VERBOSE, legible (contlen));
1445 if (contrange != -1)
1446 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1447 legible (contlen - contrange));
1450 logputs (LOG_VERBOSE,
1451 opt.ignore_length ? _("ignored") : _("unspecified"));
1453 logprintf (LOG_VERBOSE, " [%s]\n", type);
1455 logputs (LOG_VERBOSE, "\n");
1459 type = NULL; /* We don't need it any more. */
1461 /* Return if we have no intention of further downloading. */
1462 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1464 /* In case the caller cares to look... */
1468 FREE_MAYBE (all_headers);
1469 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1470 might be more bytes in the body. */
1471 return RETRFINISHED;
1474 /* Open the local file. */
1477 mkalldirs (*hs->local_file);
1479 rotate_backups (*hs->local_file);
1480 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1483 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1484 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1485 might be more bytes in the body. */
1486 FREE_MAYBE (all_headers);
1492 extern int global_download_count;
1494 /* To ensure that repeated "from scratch" downloads work for -O
1495 files, we rewind the file pointer, unless restval is
1496 non-zero. (This works only when -O is used on regular files,
1497 but it's still a valuable feature.)
1499 However, this loses when more than one URL is specified on
1500 the command line the second rewinds eradicates the contents
1501 of the first download. Thus we disable the above trick for
1502 all the downloads except the very first one.
1504 #### A possible solution to this would be to remember the
1505 file position in the output document and to seek to that
1506 position, instead of rewinding.
1508 We don't truncate stdout, since that breaks
1509 "wget -O - [...] >> foo".
1511 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1513 /* This will silently fail for streams that don't correspond
1514 to regular files, but that's OK. */
1516 /* ftruncate is needed because opt.dfp is opened in append
1517 mode if opt.always_rest is set. */
1518 ftruncate (fileno (fp), 0);
1523 /* #### This confuses the code that checks for file size. There
1524 should be some overhead information. */
1525 if (opt.save_headers)
1526 fwrite (all_headers, 1, all_length, fp);
1528 /* Get the contents of the document. */
1529 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1530 (contlen != -1 ? contlen : 0),
1531 &rbuf, keep_alive, &hs->dltime);
1534 CLOSE_FINISH (sock);
1536 CLOSE_INVALIDATE (sock);
1539 /* Close or flush the file. We have to be careful to check for
1540 error here. Checking the result of fwrite() is not enough --
1541 errors could go unnoticed! */
1544 flush_res = fclose (fp);
1546 flush_res = fflush (fp);
1547 if (flush_res == EOF)
1550 FREE_MAYBE (all_headers);
1553 return RETRFINISHED;
1556 /* The genuine HTTP loop! This is the part where the retrieval is
1557 retried, and retried, and retried, and... */
1559 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1560 int *dt, struct url *proxy)
1563 int use_ts, got_head = 0; /* time-stamping info */
1564 char *filename_plus_orig_suffix;
1565 char *local_filename = NULL;
1566 char *tms, *locf, *tmrate;
1568 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1569 long local_size = 0; /* the size of the local file */
1570 size_t filename_len;
1571 struct http_stat hstat; /* HTTP status */
1575 /* This used to be done in main(), but it's a better idea to do it
1576 here so that we don't go through the hoops if we're just using
1580 if (!wget_cookie_jar)
1581 wget_cookie_jar = cookie_jar_new ();
1582 if (opt.cookies_input && !cookies_loaded_p)
1584 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1585 cookies_loaded_p = 1;
1591 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1592 has_wildcards_p because it would also warn on `?', and we know that
1593 shows up in CGI paths a *lot*. */
1594 if (strchr (u->url, '*'))
1595 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1597 /* Determine the local filename. */
1598 if (local_file && *local_file)
1599 hstat.local_file = local_file;
1600 else if (local_file)
1602 *local_file = url_file_name (u);
1603 hstat.local_file = local_file;
1607 dummy = url_file_name (u);
1608 hstat.local_file = &dummy;
1611 if (!opt.output_document)
1612 locf = *hstat.local_file;
1614 locf = opt.output_document;
1616 hstat.referer = referer;
1618 filename_len = strlen (*hstat.local_file);
1619 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1621 if (opt.noclobber && file_exists_p (*hstat.local_file))
1623 /* If opt.noclobber is turned on and file already exists, do not
1624 retrieve the file */
1625 logprintf (LOG_VERBOSE, _("\
1626 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1627 /* If the file is there, we suppose it's retrieved OK. */
1630 /* #### Bogusness alert. */
1631 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1632 if (has_html_suffix_p (*hstat.local_file))
1640 if (opt.timestamping)
1642 boolean local_dot_orig_file_exists = FALSE;
1644 if (opt.backup_converted)
1645 /* If -K is specified, we'll act on the assumption that it was specified
1646 last time these files were downloaded as well, and instead of just
1647 comparing local file X against server file X, we'll compare local
1648 file X.orig (if extant, else X) against server file X. If -K
1649 _wasn't_ specified last time, or the server contains files called
1650 *.orig, -N will be back to not operating correctly with -k. */
1652 /* Would a single s[n]printf() call be faster? --dan
1654 Definitely not. sprintf() is horribly slow. It's a
1655 different question whether the difference between the two
1656 affects a program. Usually I'd say "no", but at one
1657 point I profiled Wget, and found that a measurable and
1658 non-negligible amount of time was lost calling sprintf()
1659 in url.c. Replacing sprintf with inline calls to
1660 strcpy() and long_to_string() made a difference.
1662 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1663 memcpy (filename_plus_orig_suffix + filename_len,
1664 ".orig", sizeof (".orig"));
1666 /* Try to stat() the .orig file. */
1667 if (stat (filename_plus_orig_suffix, &st) == 0)
1669 local_dot_orig_file_exists = TRUE;
1670 local_filename = filename_plus_orig_suffix;
1674 if (!local_dot_orig_file_exists)
1675 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1676 if (stat (*hstat.local_file, &st) == 0)
1677 local_filename = *hstat.local_file;
1679 if (local_filename != NULL)
1680 /* There was a local file, so we'll check later to see if the version
1681 the server has is the same version we already have, allowing us to
1687 /* Modification time granularity is 2 seconds for Windows, so
1688 increase local time by 1 second for later comparison. */
1691 local_size = st.st_size;
1695 /* Reset the counter. */
1697 *dt = 0 | ACCEPTRANGES;
1701 /* Increment the pass counter. */
1703 sleep_between_retrievals (count);
1704 /* Get the current time string. */
1705 tms = time_str (NULL);
1706 /* Print fetch message, if opt.verbose. */
1709 char *hurl = url_string (u, 1);
1713 sprintf (tmp, _("(try:%2d)"), count);
1714 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1715 tms, hurl, tmp, locf);
1717 ws_changetitle (hurl, 1);
1722 /* Default document type is empty. However, if spider mode is
1723 on or time-stamping is employed, HEAD_ONLY commands is
1724 encoded within *dt. */
1725 if (opt.spider || (use_ts && !got_head))
1729 /* Assume no restarting. */
1731 /* Decide whether or not to restart. */
1732 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1733 /* #### this calls access() and then stat(); could be optimized. */
1734 && file_exists_p (locf))
1735 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1736 hstat.restval = st.st_size;
1738 /* In `-c' is used and the file is existing and non-empty,
1739 refuse to truncate it if the server doesn't support continued
1741 hstat.no_truncate = 0;
1742 if (opt.always_rest && hstat.restval)
1743 hstat.no_truncate = 1;
1745 /* Decide whether to send the no-cache directive. We send it in
1747 a) we're using a proxy, and we're past our first retrieval.
1748 Some proxies are notorious for caching incomplete data, so
1749 we require a fresh get.
1750 b) caching is explicitly inhibited. */
1751 if ((proxy && count > 1) /* a */
1752 || !opt.allow_cache /* b */
1754 *dt |= SEND_NOCACHE;
1756 *dt &= ~SEND_NOCACHE;
1758 /* Try fetching the document, or at least its head. */
1759 err = gethttp (u, &hstat, dt, proxy);
1761 /* It's unfortunate that wget determines the local filename before finding
1762 out the Content-Type of the file. Barring a major restructuring of the
1763 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1764 *hstat.local_file to tack on ".html". */
1765 if (!opt.output_document)
1766 locf = *hstat.local_file;
1768 locf = opt.output_document;
1771 tms = time_str (NULL);
1772 /* Get the new location (with or without the redirection). */
1774 *newloc = xstrdup (hstat.newloc);
1777 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1778 case CONERROR: case READERR: case WRITEFAILED:
1780 /* Non-fatal errors continue executing the loop, which will
1781 bring them to "while" statement at the end, to judge
1782 whether the number of tries was exceeded. */
1783 free_hstat (&hstat);
1784 printwhat (count, opt.ntry);
1787 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1788 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1789 /* Fatal errors just return from the function. */
1790 free_hstat (&hstat);
1794 case FWRITEERR: case FOPENERR:
1795 /* Another fatal error. */
1796 logputs (LOG_VERBOSE, "\n");
1797 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1798 *hstat.local_file, strerror (errno));
1799 free_hstat (&hstat);
1804 /* Another fatal error. */
1805 logputs (LOG_VERBOSE, "\n");
1806 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1807 free_hstat (&hstat);
1812 /* Return the new location to the caller. */
1815 logprintf (LOG_NOTQUIET,
1816 _("ERROR: Redirection (%d) without location.\n"),
1818 free_hstat (&hstat);
1822 free_hstat (&hstat);
1827 /* The file was already fully retrieved. */
1828 free_hstat (&hstat);
1833 /* Deal with you later. */
1836 /* All possibilities should have been exhausted. */
1839 if (!(*dt & RETROKF))
1843 /* #### Ugly ugly ugly! */
1844 char *hurl = url_string (u, 1);
1845 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1848 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1849 tms, hstat.statcode, hstat.error);
1850 logputs (LOG_VERBOSE, "\n");
1851 free_hstat (&hstat);
1856 /* Did we get the time-stamp? */
1859 if (opt.timestamping && !hstat.remote_time)
1861 logputs (LOG_NOTQUIET, _("\
1862 Last-modified header missing -- time-stamps turned off.\n"));
1864 else if (hstat.remote_time)
1866 /* Convert the date-string into struct tm. */
1867 tmr = http_atotm (hstat.remote_time);
1868 if (tmr == (time_t) (-1))
1869 logputs (LOG_VERBOSE, _("\
1870 Last-modified header invalid -- time-stamp ignored.\n"));
1874 /* The time-stamping section. */
1879 use_ts = 0; /* no more time-stamping */
1880 count = 0; /* the retrieve count for HEAD is
1882 if (hstat.remote_time && tmr != (time_t) (-1))
1884 /* Now time-stamping can be used validly. Time-stamping
1885 means that if the sizes of the local and remote file
1886 match, and local file is newer than the remote file,
1887 it will not be retrieved. Otherwise, the normal
1888 download procedure is resumed. */
1890 (hstat.contlen == -1 || local_size == hstat.contlen))
1892 logprintf (LOG_VERBOSE, _("\
1893 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1895 free_hstat (&hstat);
1899 else if (tml >= tmr)
1900 logprintf (LOG_VERBOSE, _("\
1901 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1903 logputs (LOG_VERBOSE,
1904 _("Remote file is newer, retrieving.\n"));
1906 free_hstat (&hstat);
1909 if ((tmr != (time_t) (-1))
1911 && ((hstat.len == hstat.contlen) ||
1912 ((hstat.res == 0) &&
1913 ((hstat.contlen == -1) ||
1914 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1916 /* #### This code repeats in http.c and ftp.c. Move it to a
1918 const char *fl = NULL;
1919 if (opt.output_document)
1921 if (opt.od_known_regular)
1922 fl = opt.output_document;
1925 fl = *hstat.local_file;
1929 /* End of time-stamping section. */
1933 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1938 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1940 if (hstat.len == hstat.contlen)
1944 logprintf (LOG_VERBOSE,
1945 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1946 tms, tmrate, locf, hstat.len, hstat.contlen);
1947 logprintf (LOG_NONVERBOSE,
1948 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1949 tms, u->url, hstat.len, hstat.contlen, locf, count);
1952 total_downloaded_bytes += hstat.len;
1954 /* Remember that we downloaded the file for later ".orig" code. */
1955 if (*dt & ADDED_HTML_EXTENSION)
1956 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1958 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1960 free_hstat (&hstat);
1964 else if (hstat.res == 0) /* No read error */
1966 if (hstat.contlen == -1) /* We don't know how much we were supposed
1967 to get, so assume we succeeded. */
1971 logprintf (LOG_VERBOSE,
1972 _("%s (%s) - `%s' saved [%ld]\n\n"),
1973 tms, tmrate, locf, hstat.len);
1974 logprintf (LOG_NONVERBOSE,
1975 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1976 tms, u->url, hstat.len, locf, count);
1979 total_downloaded_bytes += hstat.len;
1981 /* Remember that we downloaded the file for later ".orig" code. */
1982 if (*dt & ADDED_HTML_EXTENSION)
1983 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1985 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1987 free_hstat (&hstat);
1991 else if (hstat.len < hstat.contlen) /* meaning we lost the
1992 connection too soon */
1994 logprintf (LOG_VERBOSE,
1995 _("%s (%s) - Connection closed at byte %ld. "),
1996 tms, tmrate, hstat.len);
1997 printwhat (count, opt.ntry);
1998 free_hstat (&hstat);
2001 else if (!opt.kill_longer) /* meaning we got more than expected */
2003 logprintf (LOG_VERBOSE,
2004 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2005 tms, tmrate, locf, hstat.len, hstat.contlen);
2006 logprintf (LOG_NONVERBOSE,
2007 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2008 tms, u->url, hstat.len, hstat.contlen, locf, count);
2010 total_downloaded_bytes += hstat.len;
2012 /* Remember that we downloaded the file for later ".orig" code. */
2013 if (*dt & ADDED_HTML_EXTENSION)
2014 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2016 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2018 free_hstat (&hstat);
2022 else /* the same, but not accepted */
2024 logprintf (LOG_VERBOSE,
2025 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2026 tms, tmrate, hstat.len, hstat.contlen);
2027 printwhat (count, opt.ntry);
2028 free_hstat (&hstat);
2032 else /* now hstat.res can only be -1 */
2034 if (hstat.contlen == -1)
2036 logprintf (LOG_VERBOSE,
2037 _("%s (%s) - Read error at byte %ld (%s)."),
2038 tms, tmrate, hstat.len, strerror (errno));
2039 printwhat (count, opt.ntry);
2040 free_hstat (&hstat);
2043 else /* hstat.res == -1 and contlen is given */
2045 logprintf (LOG_VERBOSE,
2046 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2047 tms, tmrate, hstat.len, hstat.contlen,
2049 printwhat (count, opt.ntry);
2050 free_hstat (&hstat);
2057 while (!opt.ntry || (count < opt.ntry));
2061 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2062 than local timezone.
2064 mktime is similar but assumes struct tm, also known as the
2065 "broken-down" form of time, is in local time zone. mktime_from_utc
2066 uses mktime to make the conversion understanding that an offset
2067 will be introduced by the local time assumption.
2069 mktime_from_utc then measures the introduced offset by applying
2070 gmtime to the initial result and applying mktime to the resulting
2071 "broken-down" form. The difference between the two mktime results
2072 is the measured offset which is then subtracted from the initial
2073 mktime result to yield a calendar time which is the value returned.
2075 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2076 consistent offset (the non DST offset) since tm and tm+o might be
2077 on opposite sides of a DST change.
2079 Some implementations of mktime return -1 for the nonexistent
2080 localtime hour at the beginning of DST. In this event, use
2081 mktime(tm - 1hr) + 3600.
2085 gmtime(t+o) --> tm+o
2086 mktime(tm+o) --> t+2o
2087 t+o - (t+2o - t+o) = t
2089 Note that glibc contains a function of the same purpose named
2090 `timegm' (reverse of gmtime). But obviously, it is not universally
2091 available, and unfortunately it is not straightforwardly
2092 extractable for use here. Perhaps configure should detect timegm
2093 and use it where available.
2095 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2096 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2097 Further improved by Roger with assistance from Edward J. Sabol
2098 based on input by Jamie Zawinski. */
2101 mktime_from_utc (struct tm *t)
2112 return -1; /* can't deal with output from strptime */
2123 return -1; /* can't deal with output from gmtime */
2126 return (tl - (tb - tl));
2129 /* Check whether the result of strptime() indicates success.
2130 strptime() returns the pointer to how far it got to in the string.
2131 The processing has been successful if the string is at `GMT' or
2132 `+X', or at the end of the string.
2134 In extended regexp parlance, the function returns 1 if P matches
2135 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2136 can return) is considered a failure and 0 is returned. */
2138 check_end (const char *p)
2142 while (ISSPACE (*p))
2145 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2146 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2152 /* Convert the textual specification of time in TIME_STRING to the
2153 number of seconds since the Epoch.
2155 TIME_STRING can be in any of the three formats RFC2068 allows the
2156 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2157 Timezones are ignored, and should be GMT.
2159 Return the computed time_t representation, or -1 if the conversion
2162 This function uses strptime with various string formats for parsing
2163 TIME_STRING. This results in a parser that is not as lenient in
2164 interpreting TIME_STRING as I would like it to be. Being based on
2165 strptime, it always allows shortened months, one-digit days, etc.,
2166 but due to the multitude of formats in which time can be
2167 represented, an ideal HTTP time parser would be even more
2168 forgiving. It should completely ignore things like week days and
2169 concentrate only on the various forms of representing years,
2170 months, days, hours, minutes, and seconds. For example, it would
2171 be nice if it accepted ISO 8601 out of the box.
2173 I've investigated free and PD code for this purpose, but none was
2174 usable. getdate was big and unwieldy, and had potential copyright
2175 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2176 distributed with phttpd, is excellent, but we cannot use it because
2177 it is not assigned to the FSF. So I stuck it with strptime. */
2180 http_atotm (const char *time_string)
2182 /* NOTE: Solaris strptime man page claims that %n and %t match white
2183 space, but that's not universally available. Instead, we simply
2184 use ` ' to mean "skip all WS", which works under all strptime
2185 implementations I've tested. */
2187 static const char *time_formats[] = {
2188 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2189 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2190 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2191 (google.com uses this for their cookies.) */
2192 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2198 /* According to Roger Beeman, we need to initialize tm_isdst, since
2199 strptime won't do it. */
2202 /* Note that under foreign locales Solaris strptime() fails to
2203 recognize English dates, which renders this function useless. We
2204 solve this by being careful not to affect LC_TIME when
2205 initializing locale.
2207 Another solution would be to temporarily set locale to C, invoke
2208 strptime(), and restore it back. This is slow and dirty,
2209 however, and locale support other than LC_MESSAGES can mess other
2210 things, so I rather chose to stick with just setting LC_MESSAGES.
2212 GNU strptime does not have this problem because it recognizes
2213 both international and local dates. */
2215 for (i = 0; i < countof (time_formats); i++)
2216 if (check_end (strptime (time_string, time_formats[i], &t)))
2217 return mktime_from_utc (&t);
2219 /* All formats have failed. */
2223 /* Authorization support: We support two authorization schemes:
2225 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2227 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2228 consisting of answering to the server's challenge with the proper
2231 /* How many bytes it will take to store LEN bytes in base64. */
2232 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2234 /* Encode the string S of length LENGTH to base64 format and place it
2235 to STORE. STORE will be 0-terminated, and must point to a writable
2236 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2238 base64_encode (const char *s, char *store, int length)
2240 /* Conversion table. */
2241 static char tbl[64] = {
2242 'A','B','C','D','E','F','G','H',
2243 'I','J','K','L','M','N','O','P',
2244 'Q','R','S','T','U','V','W','X',
2245 'Y','Z','a','b','c','d','e','f',
2246 'g','h','i','j','k','l','m','n',
2247 'o','p','q','r','s','t','u','v',
2248 'w','x','y','z','0','1','2','3',
2249 '4','5','6','7','8','9','+','/'
2252 unsigned char *p = (unsigned char *)store;
2254 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2255 for (i = 0; i < length; i += 3)
2257 *p++ = tbl[s[0] >> 2];
2258 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2259 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2260 *p++ = tbl[s[2] & 0x3f];
2263 /* Pad the result if necessary... */
2264 if (i == length + 1)
2266 else if (i == length + 2)
2267 *(p - 1) = *(p - 2) = '=';
2268 /* ...and zero-terminate it. */
2272 /* Create the authentication header contents for the `Basic' scheme.
2273 This is done by encoding the string `USER:PASS' in base64 and
2274 prepending `HEADER: Basic ' to it. */
2276 basic_authentication_encode (const char *user, const char *passwd,
2279 char *t1, *t2, *res;
2280 int len1 = strlen (user) + 1 + strlen (passwd);
2281 int len2 = BASE64_LENGTH (len1);
2283 t1 = (char *)alloca (len1 + 1);
2284 sprintf (t1, "%s:%s", user, passwd);
2285 t2 = (char *)alloca (1 + len2);
2286 base64_encode (t1, t2, len1);
2287 res = (char *)xmalloc (len2 + 11 + strlen (header));
2288 sprintf (res, "%s: Basic %s\r\n", header, t2);
2294 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2295 of a field in such a header. If the field is the one specified by
2296 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2297 digest authorization code), extract its value in the (char*)
2298 variable pointed by RET. Returns negative on a malformed header,
2299 or number of bytes that have been parsed by this call. */
2301 extract_header_attr (const char *au, const char *attr_name, char **ret)
2303 const char *cp, *ep;
2307 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2309 cp += strlen (attr_name);
2312 cp += skip_lws (cp);
2317 cp += skip_lws (cp);
2322 for (ep = cp; *ep && *ep != '\"'; ep++)
2327 *ret = strdupdelim (cp, ep);
2334 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2335 an array of 16 bytes containing the hash keys, and BUF should be a
2336 buffer of 33 writable characters (32 for hex digits plus one for
2337 zero termination). */
2339 dump_hash (unsigned char *buf, const unsigned char *hash)
2343 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2345 *buf++ = XNUM_TO_digit (*hash >> 4);
2346 *buf++ = XNUM_TO_digit (*hash & 0xf);
2351 /* Take the line apart to find the challenge, and compose a digest
2352 authorization header. See RFC2069 section 2.1.2. */
2354 digest_authentication_encode (const char *au, const char *user,
2355 const char *passwd, const char *method,
2358 static char *realm, *opaque, *nonce;
2363 { "realm", &realm },
2364 { "opaque", &opaque },
2369 realm = opaque = nonce = NULL;
2371 au += 6; /* skip over `Digest' */
2376 au += skip_lws (au);
2377 for (i = 0; i < countof (options); i++)
2379 int skip = extract_header_attr (au, options[i].name,
2380 options[i].variable);
2384 FREE_MAYBE (opaque);
2394 if (i == countof (options))
2396 while (*au && *au != '=')
2400 au += skip_lws (au);
2404 while (*au && *au != '\"')
2411 while (*au && *au != ',')
2416 if (!realm || !nonce || !user || !passwd || !path || !method)
2419 FREE_MAYBE (opaque);
2424 /* Calculate the digest value. */
2426 ALLOCA_MD5_CONTEXT (ctx);
2427 unsigned char hash[MD5_HASHLEN];
2428 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2429 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2431 /* A1BUF = H(user ":" realm ":" password) */
2433 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2434 gen_md5_update ((unsigned char *)":", 1, ctx);
2435 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2436 gen_md5_update ((unsigned char *)":", 1, ctx);
2437 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2438 gen_md5_finish (ctx, hash);
2439 dump_hash (a1buf, hash);
2441 /* A2BUF = H(method ":" path) */
2443 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2444 gen_md5_update ((unsigned char *)":", 1, ctx);
2445 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2446 gen_md5_finish (ctx, hash);
2447 dump_hash (a2buf, hash);
2449 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2451 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2452 gen_md5_update ((unsigned char *)":", 1, ctx);
2453 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2454 gen_md5_update ((unsigned char *)":", 1, ctx);
2455 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2456 gen_md5_finish (ctx, hash);
2457 dump_hash (response_digest, hash);
2459 res = (char*) xmalloc (strlen (user)
2464 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2465 + (opaque ? strlen (opaque) : 0)
2467 sprintf (res, "Authorization: Digest \
2468 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2469 user, realm, nonce, path, response_digest);
2472 char *p = res + strlen (res);
2473 strcat (p, ", opaque=\"");
2477 strcat (res, "\r\n");
2481 #endif /* USE_DIGEST */
2484 #define BEGINS_WITH(line, string_constant) \
2485 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2486 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2487 || !line[sizeof (string_constant) - 1]))
2490 known_authentication_scheme_p (const char *au)
2492 return BEGINS_WITH (au, "Basic")
2493 || BEGINS_WITH (au, "Digest")
2494 || BEGINS_WITH (au, "NTLM");
2499 /* Create the HTTP authorization request header. When the
2500 `WWW-Authenticate' response header is seen, according to the
2501 authorization scheme specified in that header (`Basic' and `Digest'
2502 are supported by the current implementation), produce an
2503 appropriate HTTP authorization request header. */
2505 create_authorization_line (const char *au, const char *user,
2506 const char *passwd, const char *method,
2509 char *wwwauth = NULL;
2511 if (!strncasecmp (au, "Basic", 5))
2512 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2513 if (!strncasecmp (au, "NTLM", 4))
2514 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2516 else if (!strncasecmp (au, "Digest", 6))
2517 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2518 #endif /* USE_DIGEST */
2525 if (pc_last_host_ip)
2526 address_list_release (pc_last_host_ip);