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>
68 # include "gen_sslfunc.h"
75 extern char *version_string;
81 static int cookies_loaded_p;
82 struct cookie_jar *wget_cookie_jar;
84 #define TEXTHTML_S "text/html"
85 #define HTTP_ACCEPT "*/*"
87 /* Some status code validation macros: */
88 #define H_20X(x) (((x) >= 200) && ((x) < 300))
89 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
90 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
91 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
93 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
95 #define HTTP_STATUS_OK 200
96 #define HTTP_STATUS_CREATED 201
97 #define HTTP_STATUS_ACCEPTED 202
98 #define HTTP_STATUS_NO_CONTENT 204
99 #define HTTP_STATUS_PARTIAL_CONTENTS 206
101 /* Redirection 3xx. */
102 #define HTTP_STATUS_MULTIPLE_CHOICES 300
103 #define HTTP_STATUS_MOVED_PERMANENTLY 301
104 #define HTTP_STATUS_MOVED_TEMPORARILY 302
105 #define HTTP_STATUS_NOT_MODIFIED 304
107 /* Client error 4xx. */
108 #define HTTP_STATUS_BAD_REQUEST 400
109 #define HTTP_STATUS_UNAUTHORIZED 401
110 #define HTTP_STATUS_FORBIDDEN 403
111 #define HTTP_STATUS_NOT_FOUND 404
113 /* Server errors 5xx. */
114 #define HTTP_STATUS_INTERNAL 500
115 #define HTTP_STATUS_NOT_IMPLEMENTED 501
116 #define HTTP_STATUS_BAD_GATEWAY 502
117 #define HTTP_STATUS_UNAVAILABLE 503
120 /* Parse the HTTP status line, which is of format:
122 HTTP-Version SP Status-Code SP Reason-Phrase
124 The function returns the status-code, or -1 if the status line is
125 malformed. The pointer to reason-phrase is returned in RP. */
127 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
129 /* (the variables must not be named `major' and `minor', because
130 that breaks compilation with SunOS4 cc.) */
131 int mjr, mnr, statcode;
134 *reason_phrase_ptr = NULL;
136 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
137 major version, and Y is minor version. */
138 if (strncmp (line, "HTTP/", 5) != 0)
142 /* Calculate major HTTP version. */
144 for (mjr = 0; ISDIGIT (*line); line++)
145 mjr = 10 * mjr + (*line - '0');
146 if (*line != '.' || p == line)
150 /* Calculate minor HTTP version. */
152 for (mnr = 0; ISDIGIT (*line); line++)
153 mnr = 10 * mnr + (*line - '0');
154 if (*line != ' ' || p == line)
156 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
157 minor version can be safely ignored. */
162 /* Calculate status code. */
163 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
165 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
167 /* Set up the reason phrase pointer. */
169 /* RFC2068 requires SPC here, but we allow the string to finish
170 here, in case no reason-phrase is present. */
174 *reason_phrase_ptr = line;
179 *reason_phrase_ptr = line + 1;
184 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
186 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
187 PROMISED_SIZE bytes are sent over the wire -- if the file is
188 longer, read only that much; if the file is shorter, pad it with
192 post_file (int sock, void *ssl, const char *file_name, long promised_size)
194 static char chunk[8192];
199 /* Only one of SOCK and SSL may be active at the same time. */
200 assert (sock > -1 || ssl != NULL);
201 assert (sock == -1 || ssl == NULL);
203 DEBUGP (("[writing POST file %s ... ", file_name));
205 fp = fopen (file_name, "rb");
208 while (written < promised_size)
211 int length = fread (chunk, 1, sizeof (chunk), fp);
214 towrite = WMIN (promised_size - written, length);
217 write_error = ssl_iwrite (ssl, chunk, towrite);
220 write_error = iwrite (sock, chunk, towrite);
231 if (written < promised_size)
233 /* This highly unlikely case can happen only if the file has
234 shrunk under us. To uphold the promise that exactly
235 promised_size bytes would be delivered, pad the remaining
236 data with zeros. #### Should we abort instead? */
237 DEBUGP (("padding %ld bytes ... ", promised_size - written));
238 memset (chunk, '\0', sizeof (chunk));
239 while (written < promised_size)
241 int towrite = WMIN (promised_size - written, sizeof (chunk));
244 write_error = ssl_iwrite (ssl, chunk, towrite);
247 write_error = iwrite (sock, chunk, towrite);
253 assert (written == promised_size);
254 DEBUGP (("done]\n"));
258 /* Functions to be used as arguments to header_process(): */
260 struct http_process_range_closure {
266 /* Parse the `Content-Range' header and extract the information it
267 contains. Returns 1 if successful, -1 otherwise. */
269 http_process_range (const char *hdr, void *arg)
271 struct http_process_range_closure *closure
272 = (struct http_process_range_closure *)arg;
275 /* Certain versions of Nutscape proxy server send out
276 `Content-Length' without "bytes" specifier, which is a breach of
277 RFC2068 (as well as the HTTP/1.1 draft which was current at the
278 time). But hell, I must support it... */
279 if (!strncasecmp (hdr, "bytes", 5))
282 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
286 hdr += skip_lws (hdr);
292 for (num = 0; ISDIGIT (*hdr); hdr++)
293 num = 10 * num + (*hdr - '0');
294 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
296 closure->first_byte_pos = num;
298 for (num = 0; ISDIGIT (*hdr); hdr++)
299 num = 10 * num + (*hdr - '0');
300 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
302 closure->last_byte_pos = num;
304 for (num = 0; ISDIGIT (*hdr); hdr++)
305 num = 10 * num + (*hdr - '0');
306 closure->entity_length = num;
310 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
311 Used for `Accept-Ranges'. */
313 http_process_none (const char *hdr, void *arg)
315 int *where = (int *)arg;
317 if (strstr (hdr, "none"))
324 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
326 http_process_type (const char *hdr, void *arg)
328 char **result = (char **)arg;
329 /* Locate P on `;' or the terminating zero, whichever comes first. */
330 const char *p = strchr (hdr, ';');
332 p = hdr + strlen (hdr);
333 while (p > hdr && ISSPACE (*(p - 1)))
335 *result = strdupdelim (hdr, p);
339 /* Check whether the `Connection' header is set to "keep-alive". */
341 http_process_connection (const char *hdr, void *arg)
343 int *flag = (int *)arg;
344 if (!strcasecmp (hdr, "Keep-Alive"))
349 /* Commit the cookie to the cookie jar. */
352 http_process_set_cookie (const char *hdr, void *arg)
354 struct url *u = (struct url *)arg;
356 /* The jar should have been created by now. */
357 assert (wget_cookie_jar != NULL);
359 cookie_jar_process_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
365 /* Persistent connections. Currently, we cache the most recently used
366 connection as persistent, provided that the HTTP server agrees to
367 make it such. The persistence data is stored in the variables
368 below. Ideally, it would be in a structure, and it should be
369 possible to cache an arbitrary fixed number of these connections.
371 I think the code is quite easy to extend in that direction. */
373 /* Whether a persistent connection is active. */
374 static int pc_active_p;
375 /* Host and port of currently active persistent connection. */
376 static struct address_list *pc_last_host_ip;
377 static unsigned short pc_last_port;
379 /* File descriptor of the currently active persistent connection. */
380 static int pc_last_fd;
383 /* Whether a ssl handshake has occoured on this connection */
384 static int pc_active_ssl;
385 /* SSL connection of the currently active persistent connection. */
386 static SSL *pc_last_ssl;
387 #endif /* HAVE_SSL */
389 /* Mark the persistent connection as invalid. This is used by the
390 CLOSE_* macros after they forcefully close a registered persistent
391 connection. This does not close the file descriptor -- it is left
392 to the caller to do that. (Maybe it should, though.) */
395 invalidate_persistent (void)
400 #endif /* HAVE_SSL */
401 if (pc_last_host_ip != NULL)
403 address_list_release (pc_last_host_ip);
404 pc_last_host_ip = NULL;
406 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
409 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
410 persistent. This will enable someone to use the same connection
411 later. In the context of HTTP, this must be called only AFTER the
412 response has been received and the server has promised that the
413 connection will remain alive.
415 If a previous connection was persistent, it is closed. */
419 register_persistent (const char *host, unsigned short port, int fd, SSL *ssl)
423 register_persistent (const char *host, unsigned short port, int fd)
428 if (pc_last_fd == fd)
430 /* The connection FD is already registered. Nothing to
436 /* The old persistent connection is still active; let's
437 close it first. This situation arises whenever a
438 persistent connection exists, but we then connect to a
439 different host, and try to register a persistent
440 connection to that one. */
442 /* The ssl disconnect has to take place before the closing
445 shutdown_ssl(pc_last_ssl);
448 invalidate_persistent ();
452 assert (pc_last_host_ip == NULL);
454 /* This lookup_host cannot fail, because it has the results in the
456 pc_last_host_ip = lookup_host (host, 1);
457 assert (pc_last_host_ip != NULL);
464 pc_active_ssl = ssl ? 1 : 0;
466 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
470 # define SHUTDOWN_SSL(ssl) do { \
472 shutdown_ssl (ssl); \
475 # define SHUTDOWN_SSL(ssl)
478 /* Return non-zero if a persistent connection is available for
479 connecting to HOST:PORT. */
483 persistent_available_p (const char *host, unsigned short port, int ssl)
487 persistent_available_p (const char *host, unsigned short port)
491 struct address_list *this_host_ip;
493 /* First, check whether a persistent connection is active at all. */
496 /* Second, check if the active connection pertains to the correct
497 (HOST, PORT) ordered pair. */
498 if (port != pc_last_port)
502 /* Second, a): check if current connection is (not) ssl, too. This
503 test is unlikely to fail because HTTP and HTTPS typicaly use
504 different ports. Yet it is possible, or so I [Christian
505 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
507 if (ssl != pc_active_ssl)
509 #endif /* HAVE_SSL */
511 this_host_ip = lookup_host (host, 1);
515 /* To equate the two host names for the purposes of persistent
516 connections, they need to share all the IP addresses in the
518 success = address_list_match_all (pc_last_host_ip, this_host_ip);
519 address_list_release (this_host_ip);
523 /* Third: check whether the connection is still open. This is
524 important because most server implement a liberal (short) timeout
525 on persistent connections. Wget can of course always reconnect
526 if the connection doesn't work out, but it's nicer to know in
527 advance. This test is a logical followup of the first test, but
528 is "expensive" and therefore placed at the end of the list. */
529 if (!test_socket_open (pc_last_fd))
531 /* Oops, the socket is no longer open. Now that we know that,
532 let's invalidate the persistent connection before returning
536 SHUTDOWN_SSL (pc_last_ssl);
539 invalidate_persistent ();
545 /* The idea behind these two CLOSE macros is to distinguish between
546 two cases: one when the job we've been doing is finished, and we
547 want to close the connection and leave, and two when something is
548 seriously wrong and we're closing the connection as part of
551 In case of keep_alive, CLOSE_FINISH should leave the connection
552 open, while CLOSE_INVALIDATE should still close it.
554 Note that the semantics of the flag `keep_alive' is "this
555 connection *will* be reused (the server has promised not to close
556 the connection once we're done)", while the semantics of
557 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
558 active, registered connection". */
560 #define CLOSE_FINISH(fd) do { \
563 SHUTDOWN_SSL (ssl); \
565 if (pc_active_p && (fd) == pc_last_fd) \
566 invalidate_persistent (); \
570 #define CLOSE_INVALIDATE(fd) do { \
571 SHUTDOWN_SSL (ssl); \
573 if (pc_active_p && (fd) == pc_last_fd) \
574 invalidate_persistent (); \
579 long len; /* received length */
580 long contlen; /* expected length */
581 long restval; /* the restart value */
582 int res; /* the result of last read */
583 char *newloc; /* new location (redirection) */
584 char *remote_time; /* remote time-stamp string */
585 char *error; /* textual HTTP error */
586 int statcode; /* status code */
587 double dltime; /* time of the download in msecs */
588 int no_truncate; /* whether truncating the file is
590 const char *referer; /* value of the referer header. */
591 char **local_file; /* local file. */
595 free_hstat (struct http_stat *hs)
597 FREE_MAYBE (hs->newloc);
598 FREE_MAYBE (hs->remote_time);
599 FREE_MAYBE (hs->error);
601 /* Guard against being called twice. */
603 hs->remote_time = NULL;
607 static char *create_authorization_line PARAMS ((const char *, const char *,
608 const char *, const char *,
610 static char *basic_authentication_encode PARAMS ((const char *, const char *,
612 static int known_authentication_scheme_p PARAMS ((const char *));
614 time_t http_atotm PARAMS ((const char *));
616 #define BEGINS_WITH(line, string_constant) \
617 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
618 && (ISSPACE (line[sizeof (string_constant) - 1]) \
619 || !line[sizeof (string_constant) - 1]))
621 /* Retrieve a document through HTTP protocol. It recognizes status
622 code, and correctly handles redirections. It closes the network
623 socket. If it receives an error from the functions below it, it
624 will print it if there is enough information to do so (almost
625 always), returning the error to the caller (i.e. http_loop).
627 Various HTTP parameters are stored to hs. Although it parses the
628 response code correctly, it is not used in a sane way. The caller
631 If PROXY is non-NULL, the connection will be made to the proxy
632 server, and u->url will be requested. */
634 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
636 char *request, *type, *command, *full_path;
638 char *pragma_h, *referer, *useragent, *range, *wwwauth;
639 char *authenticate_h;
643 char *request_keep_alive;
644 int sock, hcount, all_length, statcode;
646 long contlen, contrange;
649 int auth_tried_already;
652 static SSL_CTX *ssl_ctx = NULL;
655 char *cookies = NULL;
657 /* Whether this connection will be kept alive after the HTTP request
661 /* Flags that detect the two ways of specifying HTTP keep-alive
663 int http_keep_alive_1, http_keep_alive_2;
665 /* Whether keep-alive should be inhibited. */
666 int inhibit_keep_alive;
668 /* Whether we need to print the host header with braces around host,
669 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
670 "Host: symbolic-name:1234". */
671 int squares_around_host = 0;
673 /* Headers sent when using POST. */
674 char *post_content_type, *post_content_length;
675 long post_data_size = 0;
678 /* initialize ssl_ctx on first run */
681 uerr_t err = init_ssl (&ssl_ctx);
686 case SSLERRCTXCREATE:
688 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
692 /* try without certfile */
693 logprintf (LOG_NOTQUIET,
694 _("Failed to load certificates from %s\n"),
697 logprintf (LOG_NOTQUIET,
698 _("Trying without the specified certificate\n"));
701 logprintf (LOG_NOTQUIET,
702 _("Failed to get certificate key from %s\n"),
705 logprintf (LOG_NOTQUIET,
706 _("Trying without the specified certificate\n"));
713 #endif /* HAVE_SSL */
715 if (!(*dt & HEAD_ONLY))
716 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
717 know the local filename so we can save to it. */
718 assert (*hs->local_file != NULL);
721 auth_tried_already = 0;
723 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
726 /* We need to come back here when the initial attempt to retrieve
727 without authorization header fails. (Expected to happen at least
728 for the Digest authorization scheme.) */
731 http_keep_alive_1 = http_keep_alive_2 = 0;
733 post_content_type = NULL;
734 post_content_length = NULL;
736 /* Initialize certain elements of struct http_stat. */
741 hs->remote_time = NULL;
744 /* If we're using a proxy, we will be connecting to the proxy
746 conn = proxy ? proxy : u;
748 /* First: establish the connection. */
749 if (inhibit_keep_alive
752 !persistent_available_p (conn->host, conn->port)
754 !persistent_available_p (conn->host, conn->port,
755 u->scheme == SCHEME_HTTPS)
756 #endif /* HAVE_SSL */
759 struct address_list *al = lookup_host (conn->host, 0);
762 set_connection_host_name (conn->host);
763 sock = connect_to_many (al, conn->port, 0);
764 set_connection_host_name (NULL);
765 address_list_release (al);
768 return CONNECT_ERROR (errno);
771 if (conn->scheme == SCHEME_HTTPS)
772 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
774 logputs (LOG_VERBOSE, "\n");
775 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
780 #endif /* HAVE_SSL */
784 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
785 conn->host, conn->port);
786 /* #### pc_last_fd should be accessed through an accessor
791 #endif /* HAVE_SSL */
792 DEBUGP (("Reusing fd %d.\n", sock));
797 else if (opt.post_file_name || opt.post_data)
805 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
806 sprintf (referer, "Referer: %s\r\n", hs->referer);
809 if (*dt & SEND_NOCACHE)
810 pragma_h = "Pragma: no-cache\r\n";
816 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
817 /* Gag me! Some servers (e.g. WebSitePro) have been known to
818 respond to the following `Range' format by generating a
819 multipart/x-byte-ranges MIME document! This MIME type was
820 present in an old draft of the byteranges specification.
821 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
822 only if multiple non-overlapping ranges are requested --
823 which Wget never does. */
824 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
829 STRDUP_ALLOCA (useragent, opt.useragent);
832 useragent = (char *)alloca (10 + strlen (version_string));
833 sprintf (useragent, "Wget/%s", version_string);
835 /* Construct the authentication, if userid is present. */
838 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
839 user = user ? user : opt.http_user;
840 passwd = passwd ? passwd : opt.http_passwd;
847 /* We have the username and the password, but haven't tried
848 any authorization yet. Let's see if the "Basic" method
849 works. If not, we'll come back here and construct a
850 proper authorization method with the right challenges.
852 If we didn't employ this kind of logic, every URL that
853 requires authorization would have to be processed twice,
854 which is very suboptimal and generates a bunch of false
855 "unauthorized" errors in the server log.
857 #### But this logic also has a serious problem when used
858 with stronger authentications: we *first* transmit the
859 username and the password in clear text, and *then*
860 attempt a stronger authentication scheme. That cannot be
861 right! We are only fortunate that almost everyone still
862 uses the `Basic' scheme anyway.
864 There should be an option to prevent this from happening,
865 for those who use strong authentication schemes and value
867 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
871 /* Use the full path, i.e. one that includes the leading
872 slash and the query string, but is independent of proxy
874 char *pth = url_full_path (u);
875 wwwauth = create_authorization_line (authenticate_h, user, passwd,
884 char *proxy_user, *proxy_passwd;
885 /* For normal username and password, URL components override
886 command-line/wgetrc parameters. With proxy authentication,
887 it's the reverse, because proxy URLs are normally the
888 "permanent" ones, so command-line args should take
890 if (opt.proxy_user && opt.proxy_passwd)
892 proxy_user = opt.proxy_user;
893 proxy_passwd = opt.proxy_passwd;
897 proxy_user = proxy->user;
898 proxy_passwd = proxy->passwd;
900 /* #### This does not appear right. Can't the proxy request,
901 say, `Digest' authentication? */
902 if (proxy_user && proxy_passwd)
903 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
904 "Proxy-Authorization");
907 /* String of the form :PORT. Used only for non-standard ports. */
909 if (u->port != scheme_default_port (u->scheme))
911 port_maybe = (char *)alloca (numdigit (u->port) + 2);
912 sprintf (port_maybe, ":%d", u->port);
915 if (!inhibit_keep_alive)
916 request_keep_alive = "Connection: Keep-Alive\r\n";
918 request_keep_alive = NULL;
921 cookies = cookie_jar_generate_cookie_header (wget_cookie_jar, u->host,
924 u->scheme == SCHEME_HTTPS
930 if (opt.post_data || opt.post_file_name)
932 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
934 post_data_size = strlen (opt.post_data);
937 post_data_size = file_size (opt.post_file_name);
938 if (post_data_size == -1)
940 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
945 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
946 sprintf (post_content_length,
947 "Content-Length: %ld\r\n", post_data_size);
951 full_path = xstrdup (u->url);
953 /* Use the full path, i.e. one that includes the leading slash and
954 the query string. E.g. if u->path is "foo/bar" and u->query is
955 "param=value", full_path will be "/foo/bar?param=value". */
956 full_path = url_full_path (u);
958 if (strchr (u->host, ':'))
959 squares_around_host = 1;
961 /* Allocate the memory for the request. */
962 request = (char *)alloca (strlen (command)
966 + (port_maybe ? strlen (port_maybe) : 0)
967 + strlen (HTTP_ACCEPT)
968 + (request_keep_alive
969 ? strlen (request_keep_alive) : 0)
970 + (referer ? strlen (referer) : 0)
971 + (cookies ? strlen (cookies) : 0)
972 + (wwwauth ? strlen (wwwauth) : 0)
973 + (proxyauth ? strlen (proxyauth) : 0)
974 + (range ? strlen (range) : 0)
977 ? strlen (post_content_type) : 0)
978 + (post_content_length
979 ? strlen (post_content_length) : 0)
980 + (opt.user_header ? strlen (opt.user_header) : 0)
982 /* Construct the request. */
988 %s%s%s%s%s%s%s%s%s%s\r\n",
991 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
992 port_maybe ? port_maybe : "",
994 request_keep_alive ? request_keep_alive : "",
995 referer ? referer : "",
996 cookies ? cookies : "",
997 wwwauth ? wwwauth : "",
998 proxyauth ? proxyauth : "",
1001 post_content_type ? post_content_type : "",
1002 post_content_length ? post_content_length : "",
1003 opt.user_header ? opt.user_header : "");
1004 DEBUGP (("---request begin---\n%s", request));
1006 /* Free the temporary memory. */
1007 FREE_MAYBE (wwwauth);
1008 FREE_MAYBE (proxyauth);
1009 FREE_MAYBE (cookies);
1012 /* Send the request to server. */
1014 if (conn->scheme == SCHEME_HTTPS)
1015 write_error = ssl_iwrite (ssl, request, strlen (request));
1018 write_error = iwrite (sock, request, strlen (request));
1020 if (write_error >= 0)
1024 DEBUGP (("[POST data: %s]\n", opt.post_data));
1026 if (conn->scheme == SCHEME_HTTPS)
1027 write_error = ssl_iwrite (ssl, opt.post_data, post_data_size);
1030 write_error = iwrite (sock, opt.post_data, post_data_size);
1032 else if (opt.post_file_name)
1035 if (conn->scheme == SCHEME_HTTPS)
1036 write_error = post_file (-1, ssl, opt.post_file_name,
1040 write_error = post_file (sock, NULL, opt.post_file_name,
1044 DEBUGP (("---request end---\n"));
1046 if (write_error < 0)
1048 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1050 CLOSE_INVALIDATE (sock);
1053 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1054 proxy ? "Proxy" : "HTTP");
1055 contlen = contrange = -1;
1060 /* Before reading anything, initialize the rbuf. */
1061 rbuf_initialize (&rbuf, sock);
1063 if (conn->scheme == SCHEME_HTTPS)
1067 #endif /* HAVE_SSL */
1070 /* Header-fetching loop. */
1078 /* Get the header. */
1079 status = header_get (&rbuf, &hdr,
1080 /* Disallow continuations for status line. */
1081 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
1083 /* Check for errors. */
1084 if (status == HG_EOF && *hdr)
1086 /* This used to be an unconditional error, but that was
1087 somewhat controversial, because of a large number of
1088 broken CGI's that happily "forget" to send the second EOL
1089 before closing the connection of a HEAD request.
1091 So, the deal is to check whether the header is empty
1092 (*hdr is zero if it is); if yes, it means that the
1093 previous header was fully retrieved, and that -- most
1094 probably -- the request is complete. "...be liberal in
1095 what you accept." Oh boy. */
1096 logputs (LOG_VERBOSE, "\n");
1097 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1100 FREE_MAYBE (all_headers);
1101 CLOSE_INVALIDATE (sock);
1104 else if (status == HG_ERROR)
1106 logputs (LOG_VERBOSE, "\n");
1107 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1111 FREE_MAYBE (all_headers);
1112 CLOSE_INVALIDATE (sock);
1116 /* If the headers are to be saved to a file later, save them to
1118 if (opt.save_headers)
1120 int lh = strlen (hdr);
1121 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1122 memcpy (all_headers + all_length, hdr, lh);
1124 all_headers[all_length++] = '\n';
1125 all_headers[all_length] = '\0';
1128 /* Check for status line. */
1132 /* Parse the first line of server response. */
1133 statcode = parse_http_status_line (hdr, &error);
1134 hs->statcode = statcode;
1135 /* Store the descriptive response. */
1136 if (statcode == -1) /* malformed response */
1138 /* A common reason for "malformed response" error is the
1139 case when no data was actually received. Handle this
1142 hs->error = xstrdup (_("No data received"));
1144 hs->error = xstrdup (_("Malformed status line"));
1149 hs->error = xstrdup (_("(no description)"));
1151 hs->error = xstrdup (error);
1153 if ((statcode != -1)
1159 if (opt.server_response)
1160 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1162 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1168 /* Exit on empty header. */
1175 /* Print the header if requested. */
1176 if (opt.server_response && hcount != 1)
1177 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1179 /* Try getting content-length. */
1180 if (contlen == -1 && !opt.ignore_length)
1181 if (header_process (hdr, "Content-Length", header_extract_number,
1184 /* Try getting content-type. */
1186 if (header_process (hdr, "Content-Type", http_process_type, &type))
1188 /* Try getting location. */
1190 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1192 /* Try getting last-modified. */
1193 if (!hs->remote_time)
1194 if (header_process (hdr, "Last-Modified", header_strdup,
1197 /* Try getting cookies. */
1199 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1201 /* Try getting www-authentication. */
1202 if (!authenticate_h)
1203 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1206 /* Check for accept-ranges header. If it contains the word
1207 `none', disable the ranges. */
1208 if (*dt & ACCEPTRANGES)
1211 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1214 *dt &= ~ACCEPTRANGES;
1218 /* Try getting content-range. */
1219 if (contrange == -1)
1221 struct http_process_range_closure closure;
1222 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1224 contrange = closure.first_byte_pos;
1228 /* Check for keep-alive related responses. */
1229 if (!inhibit_keep_alive)
1231 /* Check for the `Keep-Alive' header. */
1232 if (!http_keep_alive_1)
1234 if (header_process (hdr, "Keep-Alive", header_exists,
1235 &http_keep_alive_1))
1238 /* Check for `Connection: Keep-Alive'. */
1239 if (!http_keep_alive_2)
1241 if (header_process (hdr, "Connection", http_process_connection,
1242 &http_keep_alive_2))
1250 logputs (LOG_VERBOSE, "\n");
1253 && (http_keep_alive_1 || http_keep_alive_2))
1255 assert (inhibit_keep_alive == 0);
1259 /* The server has promised that it will not close the connection
1260 when we're done. This means that we can register it. */
1262 register_persistent (conn->host, conn->port, sock);
1264 register_persistent (conn->host, conn->port, sock, ssl);
1265 #endif /* HAVE_SSL */
1267 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1270 /* Authorization is required. */
1274 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1275 might be more bytes in the body. */
1276 if (auth_tried_already)
1278 /* If we have tried it already, then there is not point
1281 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1282 xfree (authenticate_h);
1285 else if (!known_authentication_scheme_p (authenticate_h))
1287 xfree (authenticate_h);
1288 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1291 else if (BEGINS_WITH (authenticate_h, "Basic"))
1293 /* The authentication scheme is basic, the one we try by
1294 default, and it failed. There's no sense in trying
1300 auth_tried_already = 1;
1304 /* We do not need this anymore. */
1307 xfree (authenticate_h);
1308 authenticate_h = NULL;
1311 /* 20x responses are counted among successful by default. */
1312 if (H_20X (statcode))
1315 /* Return if redirected. */
1316 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1318 /* RFC2068 says that in case of the 300 (multiple choices)
1319 response, the server can output a preferred URL through
1320 `Location' header; otherwise, the request should be treated
1321 like GET. So, if the location is set, it will be a
1322 redirection; otherwise, just proceed normally. */
1323 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1327 logprintf (LOG_VERBOSE,
1328 _("Location: %s%s\n"),
1329 hs->newloc ? hs->newloc : _("unspecified"),
1330 hs->newloc ? _(" [following]") : "");
1331 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1332 might be more bytes in the body. */
1334 FREE_MAYBE (all_headers);
1339 /* If content-type is not given, assume text/html. This is because
1340 of the multitude of broken CGI's that "forget" to generate the
1342 if (!type || 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1347 if (opt.html_extension && (*dt & TEXTHTML))
1348 /* -E / --html-extension / html_extension = on was specified, and this is a
1349 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1350 already the file's suffix, tack on ".html". */
1352 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1354 if (last_period_in_local_filename == NULL ||
1355 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1356 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1358 size_t local_filename_len = strlen(*hs->local_file);
1360 *hs->local_file = xrealloc(*hs->local_file,
1361 local_filename_len + sizeof(".html"));
1362 strcpy(*hs->local_file + local_filename_len, ".html");
1364 *dt |= ADDED_HTML_EXTENSION;
1368 if (contrange == -1)
1370 /* We did not get a content-range header. This means that the
1371 server did not honor our `Range' request. Normally, this
1372 means we should reset hs->restval and continue normally. */
1374 /* However, if `-c' is used, we need to be a bit more careful:
1376 1. If `-c' is specified and the file already existed when
1377 Wget was started, it would be a bad idea for us to start
1378 downloading it from scratch, effectively truncating it. I
1379 believe this cannot happen unless `-c' was specified.
1381 2. If `-c' is used on a file that is already fully
1382 downloaded, we're requesting bytes after the end of file,
1383 which can result in server not honoring `Range'. If this is
1384 the case, `Content-Length' will be equal to the length of the
1386 if (opt.always_rest)
1388 /* Check for condition #2. */
1389 if (hs->restval > 0 /* restart was requested. */
1390 && contlen != -1 /* we got content-length. */
1391 && hs->restval >= contlen /* file fully downloaded
1395 logputs (LOG_VERBOSE, _("\
1396 \n The file is already fully retrieved; nothing to do.\n\n"));
1397 /* In case the caller inspects. */
1400 /* Mark as successfully retrieved. */
1403 FREE_MAYBE (all_headers);
1404 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1405 might be more bytes in the body. */
1406 return RETRUNNEEDED;
1409 /* Check for condition #1. */
1410 if (hs->no_truncate)
1412 logprintf (LOG_NOTQUIET,
1415 Continued download failed on this file, which conflicts with `-c'.\n\
1416 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1418 FREE_MAYBE (all_headers);
1419 CLOSE_INVALIDATE (sock);
1420 return CONTNOTSUPPORTED;
1428 else if (contrange != hs->restval ||
1429 (H_PARTIAL (statcode) && contrange == -1))
1431 /* This means the whole request was somehow misunderstood by the
1432 server. Bail out. */
1434 FREE_MAYBE (all_headers);
1435 CLOSE_INVALIDATE (sock);
1442 contlen += contrange;
1444 contrange = -1; /* If conent-length was not sent,
1445 content-range will be ignored. */
1447 hs->contlen = contlen;
1451 if ((*dt & RETROKF) && !opt.server_response)
1453 /* No need to print this output if the body won't be
1454 downloaded at all, or if the original server response is
1456 logputs (LOG_VERBOSE, _("Length: "));
1459 logputs (LOG_VERBOSE, legible (contlen));
1460 if (contrange != -1)
1461 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1462 legible (contlen - contrange));
1465 logputs (LOG_VERBOSE,
1466 opt.ignore_length ? _("ignored") : _("unspecified"));
1468 logprintf (LOG_VERBOSE, " [%s]\n", type);
1470 logputs (LOG_VERBOSE, "\n");
1474 type = NULL; /* We don't need it any more. */
1476 /* Return if we have no intention of further downloading. */
1477 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1479 /* In case the caller cares to look... */
1483 FREE_MAYBE (all_headers);
1484 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1485 might be more bytes in the body. */
1486 return RETRFINISHED;
1489 /* Open the local file. */
1492 mkalldirs (*hs->local_file);
1494 rotate_backups (*hs->local_file);
1495 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1498 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1499 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1500 might be more bytes in the body. */
1501 FREE_MAYBE (all_headers);
1507 extern int global_download_count;
1509 /* To ensure that repeated "from scratch" downloads work for -O
1510 files, we rewind the file pointer, unless restval is
1511 non-zero. (This works only when -O is used on regular files,
1512 but it's still a valuable feature.)
1514 However, this loses when more than one URL is specified on
1515 the command line the second rewinds eradicates the contents
1516 of the first download. Thus we disable the above trick for
1517 all the downloads except the very first one.
1519 #### A possible solution to this would be to remember the
1520 file position in the output document and to seek to that
1521 position, instead of rewinding.
1523 We don't truncate stdout, since that breaks
1524 "wget -O - [...] >> foo".
1526 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1528 /* This will silently fail for streams that don't correspond
1529 to regular files, but that's OK. */
1531 /* ftruncate is needed because opt.dfp is opened in append
1532 mode if opt.always_rest is set. */
1533 ftruncate (fileno (fp), 0);
1538 /* #### This confuses the code that checks for file size. There
1539 should be some overhead information. */
1540 if (opt.save_headers)
1541 fwrite (all_headers, 1, all_length, fp);
1543 /* Get the contents of the document. */
1544 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1545 (contlen != -1 ? contlen : 0),
1546 &rbuf, keep_alive, &hs->dltime);
1549 CLOSE_FINISH (sock);
1551 CLOSE_INVALIDATE (sock);
1554 /* Close or flush the file. We have to be careful to check for
1555 error here. Checking the result of fwrite() is not enough --
1556 errors could go unnoticed! */
1559 flush_res = fclose (fp);
1561 flush_res = fflush (fp);
1562 if (flush_res == EOF)
1565 FREE_MAYBE (all_headers);
1568 return RETRFINISHED;
1571 /* The genuine HTTP loop! This is the part where the retrieval is
1572 retried, and retried, and retried, and... */
1574 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1575 int *dt, struct url *proxy)
1578 int use_ts, got_head = 0; /* time-stamping info */
1579 char *filename_plus_orig_suffix;
1580 char *local_filename = NULL;
1581 char *tms, *locf, *tmrate;
1583 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1584 long local_size = 0; /* the size of the local file */
1585 size_t filename_len;
1586 struct http_stat hstat; /* HTTP status */
1590 /* This used to be done in main(), but it's a better idea to do it
1591 here so that we don't go through the hoops if we're just using
1595 if (!wget_cookie_jar)
1596 wget_cookie_jar = cookie_jar_new ();
1597 if (opt.cookies_input && !cookies_loaded_p)
1599 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1600 cookies_loaded_p = 1;
1606 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1607 has_wildcards_p because it would also warn on `?', and we know that
1608 shows up in CGI paths a *lot*. */
1609 if (strchr (u->url, '*'))
1610 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1612 /* Determine the local filename. */
1613 if (local_file && *local_file)
1614 hstat.local_file = local_file;
1615 else if (local_file)
1617 *local_file = url_file_name (u);
1618 hstat.local_file = local_file;
1622 dummy = url_file_name (u);
1623 hstat.local_file = &dummy;
1626 if (!opt.output_document)
1627 locf = *hstat.local_file;
1629 locf = opt.output_document;
1631 hstat.referer = referer;
1633 filename_len = strlen (*hstat.local_file);
1634 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1636 if (opt.noclobber && file_exists_p (*hstat.local_file))
1638 /* If opt.noclobber is turned on and file already exists, do not
1639 retrieve the file */
1640 logprintf (LOG_VERBOSE, _("\
1641 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1642 /* If the file is there, we suppose it's retrieved OK. */
1645 /* #### Bogusness alert. */
1646 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1647 if (has_html_suffix_p (*hstat.local_file))
1655 if (opt.timestamping)
1657 boolean local_dot_orig_file_exists = FALSE;
1659 if (opt.backup_converted)
1660 /* If -K is specified, we'll act on the assumption that it was specified
1661 last time these files were downloaded as well, and instead of just
1662 comparing local file X against server file X, we'll compare local
1663 file X.orig (if extant, else X) against server file X. If -K
1664 _wasn't_ specified last time, or the server contains files called
1665 *.orig, -N will be back to not operating correctly with -k. */
1667 /* Would a single s[n]printf() call be faster? --dan
1669 Definitely not. sprintf() is horribly slow. It's a
1670 different question whether the difference between the two
1671 affects a program. Usually I'd say "no", but at one
1672 point I profiled Wget, and found that a measurable and
1673 non-negligible amount of time was lost calling sprintf()
1674 in url.c. Replacing sprintf with inline calls to
1675 strcpy() and long_to_string() made a difference.
1677 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1678 memcpy (filename_plus_orig_suffix + filename_len,
1679 ".orig", sizeof (".orig"));
1681 /* Try to stat() the .orig file. */
1682 if (stat (filename_plus_orig_suffix, &st) == 0)
1684 local_dot_orig_file_exists = TRUE;
1685 local_filename = filename_plus_orig_suffix;
1689 if (!local_dot_orig_file_exists)
1690 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1691 if (stat (*hstat.local_file, &st) == 0)
1692 local_filename = *hstat.local_file;
1694 if (local_filename != NULL)
1695 /* There was a local file, so we'll check later to see if the version
1696 the server has is the same version we already have, allowing us to
1702 /* Modification time granularity is 2 seconds for Windows, so
1703 increase local time by 1 second for later comparison. */
1706 local_size = st.st_size;
1710 /* Reset the counter. */
1712 *dt = 0 | ACCEPTRANGES;
1716 /* Increment the pass counter. */
1718 sleep_between_retrievals (count);
1719 /* Get the current time string. */
1720 tms = time_str (NULL);
1721 /* Print fetch message, if opt.verbose. */
1724 char *hurl = url_string (u, 1);
1728 sprintf (tmp, _("(try:%2d)"), count);
1729 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1730 tms, hurl, tmp, locf);
1732 ws_changetitle (hurl, 1);
1737 /* Default document type is empty. However, if spider mode is
1738 on or time-stamping is employed, HEAD_ONLY commands is
1739 encoded within *dt. */
1740 if (opt.spider || (use_ts && !got_head))
1744 /* Assume no restarting. */
1746 /* Decide whether or not to restart. */
1747 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1748 /* #### this calls access() and then stat(); could be optimized. */
1749 && file_exists_p (locf))
1750 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1751 hstat.restval = st.st_size;
1753 /* In `-c' is used and the file is existing and non-empty,
1754 refuse to truncate it if the server doesn't support continued
1756 hstat.no_truncate = 0;
1757 if (opt.always_rest && hstat.restval)
1758 hstat.no_truncate = 1;
1760 /* Decide whether to send the no-cache directive. We send it in
1762 a) we're using a proxy, and we're past our first retrieval.
1763 Some proxies are notorious for caching incomplete data, so
1764 we require a fresh get.
1765 b) caching is explicitly inhibited. */
1766 if ((proxy && count > 1) /* a */
1767 || !opt.allow_cache /* b */
1769 *dt |= SEND_NOCACHE;
1771 *dt &= ~SEND_NOCACHE;
1773 /* Try fetching the document, or at least its head. */
1774 err = gethttp (u, &hstat, dt, proxy);
1776 /* It's unfortunate that wget determines the local filename before finding
1777 out the Content-Type of the file. Barring a major restructuring of the
1778 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1779 *hstat.local_file to tack on ".html". */
1780 if (!opt.output_document)
1781 locf = *hstat.local_file;
1783 locf = opt.output_document;
1786 tms = time_str (NULL);
1787 /* Get the new location (with or without the redirection). */
1789 *newloc = xstrdup (hstat.newloc);
1792 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1793 case CONERROR: case READERR: case WRITEFAILED:
1795 /* Non-fatal errors continue executing the loop, which will
1796 bring them to "while" statement at the end, to judge
1797 whether the number of tries was exceeded. */
1798 free_hstat (&hstat);
1799 printwhat (count, opt.ntry);
1802 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1803 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1804 /* Fatal errors just return from the function. */
1805 free_hstat (&hstat);
1809 case FWRITEERR: case FOPENERR:
1810 /* Another fatal error. */
1811 logputs (LOG_VERBOSE, "\n");
1812 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1813 *hstat.local_file, strerror (errno));
1814 free_hstat (&hstat);
1819 /* Another fatal error. */
1820 logputs (LOG_VERBOSE, "\n");
1821 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1822 free_hstat (&hstat);
1827 /* Return the new location to the caller. */
1830 logprintf (LOG_NOTQUIET,
1831 _("ERROR: Redirection (%d) without location.\n"),
1833 free_hstat (&hstat);
1837 free_hstat (&hstat);
1842 /* The file was already fully retrieved. */
1843 free_hstat (&hstat);
1848 /* Deal with you later. */
1851 /* All possibilities should have been exhausted. */
1854 if (!(*dt & RETROKF))
1858 /* #### Ugly ugly ugly! */
1859 char *hurl = url_string (u, 1);
1860 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1863 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1864 tms, hstat.statcode, hstat.error);
1865 logputs (LOG_VERBOSE, "\n");
1866 free_hstat (&hstat);
1871 /* Did we get the time-stamp? */
1874 if (opt.timestamping && !hstat.remote_time)
1876 logputs (LOG_NOTQUIET, _("\
1877 Last-modified header missing -- time-stamps turned off.\n"));
1879 else if (hstat.remote_time)
1881 /* Convert the date-string into struct tm. */
1882 tmr = http_atotm (hstat.remote_time);
1883 if (tmr == (time_t) (-1))
1884 logputs (LOG_VERBOSE, _("\
1885 Last-modified header invalid -- time-stamp ignored.\n"));
1889 /* The time-stamping section. */
1894 use_ts = 0; /* no more time-stamping */
1895 count = 0; /* the retrieve count for HEAD is
1897 if (hstat.remote_time && tmr != (time_t) (-1))
1899 /* Now time-stamping can be used validly. Time-stamping
1900 means that if the sizes of the local and remote file
1901 match, and local file is newer than the remote file,
1902 it will not be retrieved. Otherwise, the normal
1903 download procedure is resumed. */
1905 (hstat.contlen == -1 || local_size == hstat.contlen))
1907 logprintf (LOG_VERBOSE, _("\
1908 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1910 free_hstat (&hstat);
1914 else if (tml >= tmr)
1915 logprintf (LOG_VERBOSE, _("\
1916 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1918 logputs (LOG_VERBOSE,
1919 _("Remote file is newer, retrieving.\n"));
1921 free_hstat (&hstat);
1924 if ((tmr != (time_t) (-1))
1926 && ((hstat.len == hstat.contlen) ||
1927 ((hstat.res == 0) &&
1928 ((hstat.contlen == -1) ||
1929 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1931 /* #### This code repeats in http.c and ftp.c. Move it to a
1933 const char *fl = NULL;
1934 if (opt.output_document)
1936 if (opt.od_known_regular)
1937 fl = opt.output_document;
1940 fl = *hstat.local_file;
1944 /* End of time-stamping section. */
1948 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1953 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1955 if (hstat.len == hstat.contlen)
1959 logprintf (LOG_VERBOSE,
1960 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1961 tms, tmrate, locf, hstat.len, hstat.contlen);
1962 logprintf (LOG_NONVERBOSE,
1963 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1964 tms, u->url, hstat.len, hstat.contlen, locf, count);
1967 downloaded_increase (hstat.len);
1969 /* Remember that we downloaded the file for later ".orig" code. */
1970 if (*dt & ADDED_HTML_EXTENSION)
1971 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1973 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1975 free_hstat (&hstat);
1979 else if (hstat.res == 0) /* No read error */
1981 if (hstat.contlen == -1) /* We don't know how much we were supposed
1982 to get, so assume we succeeded. */
1986 logprintf (LOG_VERBOSE,
1987 _("%s (%s) - `%s' saved [%ld]\n\n"),
1988 tms, tmrate, locf, hstat.len);
1989 logprintf (LOG_NONVERBOSE,
1990 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1991 tms, u->url, hstat.len, locf, count);
1994 downloaded_increase (hstat.len);
1996 /* Remember that we downloaded the file for later ".orig" code. */
1997 if (*dt & ADDED_HTML_EXTENSION)
1998 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2000 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2002 free_hstat (&hstat);
2006 else if (hstat.len < hstat.contlen) /* meaning we lost the
2007 connection too soon */
2009 logprintf (LOG_VERBOSE,
2010 _("%s (%s) - Connection closed at byte %ld. "),
2011 tms, tmrate, hstat.len);
2012 printwhat (count, opt.ntry);
2013 free_hstat (&hstat);
2016 else if (!opt.kill_longer) /* meaning we got more than expected */
2018 logprintf (LOG_VERBOSE,
2019 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2020 tms, tmrate, locf, hstat.len, hstat.contlen);
2021 logprintf (LOG_NONVERBOSE,
2022 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2023 tms, u->url, hstat.len, hstat.contlen, locf, count);
2025 downloaded_increase (hstat.len);
2027 /* Remember that we downloaded the file for later ".orig" code. */
2028 if (*dt & ADDED_HTML_EXTENSION)
2029 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2031 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2033 free_hstat (&hstat);
2037 else /* the same, but not accepted */
2039 logprintf (LOG_VERBOSE,
2040 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2041 tms, tmrate, hstat.len, hstat.contlen);
2042 printwhat (count, opt.ntry);
2043 free_hstat (&hstat);
2047 else /* now hstat.res can only be -1 */
2049 if (hstat.contlen == -1)
2051 logprintf (LOG_VERBOSE,
2052 _("%s (%s) - Read error at byte %ld (%s)."),
2053 tms, tmrate, hstat.len, strerror (errno));
2054 printwhat (count, opt.ntry);
2055 free_hstat (&hstat);
2058 else /* hstat.res == -1 and contlen is given */
2060 logprintf (LOG_VERBOSE,
2061 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2062 tms, tmrate, hstat.len, hstat.contlen,
2064 printwhat (count, opt.ntry);
2065 free_hstat (&hstat);
2072 while (!opt.ntry || (count < opt.ntry));
2076 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2077 than local timezone.
2079 mktime is similar but assumes struct tm, also known as the
2080 "broken-down" form of time, is in local time zone. mktime_from_utc
2081 uses mktime to make the conversion understanding that an offset
2082 will be introduced by the local time assumption.
2084 mktime_from_utc then measures the introduced offset by applying
2085 gmtime to the initial result and applying mktime to the resulting
2086 "broken-down" form. The difference between the two mktime results
2087 is the measured offset which is then subtracted from the initial
2088 mktime result to yield a calendar time which is the value returned.
2090 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2091 consistent offset (the non DST offset) since tm and tm+o might be
2092 on opposite sides of a DST change.
2094 Some implementations of mktime return -1 for the nonexistent
2095 localtime hour at the beginning of DST. In this event, use
2096 mktime(tm - 1hr) + 3600.
2100 gmtime(t+o) --> tm+o
2101 mktime(tm+o) --> t+2o
2102 t+o - (t+2o - t+o) = t
2104 Note that glibc contains a function of the same purpose named
2105 `timegm' (reverse of gmtime). But obviously, it is not universally
2106 available, and unfortunately it is not straightforwardly
2107 extractable for use here. Perhaps configure should detect timegm
2108 and use it where available.
2110 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2111 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2112 Further improved by Roger with assistance from Edward J. Sabol
2113 based on input by Jamie Zawinski. */
2116 mktime_from_utc (struct tm *t)
2127 return -1; /* can't deal with output from strptime */
2138 return -1; /* can't deal with output from gmtime */
2141 return (tl - (tb - tl));
2144 /* Check whether the result of strptime() indicates success.
2145 strptime() returns the pointer to how far it got to in the string.
2146 The processing has been successful if the string is at `GMT' or
2147 `+X', or at the end of the string.
2149 In extended regexp parlance, the function returns 1 if P matches
2150 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2151 can return) is considered a failure and 0 is returned. */
2153 check_end (const char *p)
2157 while (ISSPACE (*p))
2160 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2161 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2167 /* Convert the textual specification of time in TIME_STRING to the
2168 number of seconds since the Epoch.
2170 TIME_STRING can be in any of the three formats RFC2068 allows the
2171 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2172 Timezones are ignored, and should be GMT.
2174 Return the computed time_t representation, or -1 if the conversion
2177 This function uses strptime with various string formats for parsing
2178 TIME_STRING. This results in a parser that is not as lenient in
2179 interpreting TIME_STRING as I would like it to be. Being based on
2180 strptime, it always allows shortened months, one-digit days, etc.,
2181 but due to the multitude of formats in which time can be
2182 represented, an ideal HTTP time parser would be even more
2183 forgiving. It should completely ignore things like week days and
2184 concentrate only on the various forms of representing years,
2185 months, days, hours, minutes, and seconds. For example, it would
2186 be nice if it accepted ISO 8601 out of the box.
2188 I've investigated free and PD code for this purpose, but none was
2189 usable. getdate was big and unwieldy, and had potential copyright
2190 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2191 distributed with phttpd, is excellent, but we cannot use it because
2192 it is not assigned to the FSF. So I stuck it with strptime. */
2195 http_atotm (const char *time_string)
2197 /* NOTE: Solaris strptime man page claims that %n and %t match white
2198 space, but that's not universally available. Instead, we simply
2199 use ` ' to mean "skip all WS", which works under all strptime
2200 implementations I've tested. */
2202 static const char *time_formats[] = {
2203 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2204 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2205 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2206 (google.com uses this for their cookies.) */
2207 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2213 /* According to Roger Beeman, we need to initialize tm_isdst, since
2214 strptime won't do it. */
2217 /* Note that under foreign locales Solaris strptime() fails to
2218 recognize English dates, which renders this function useless. We
2219 solve this by being careful not to affect LC_TIME when
2220 initializing locale.
2222 Another solution would be to temporarily set locale to C, invoke
2223 strptime(), and restore it back. This is slow and dirty,
2224 however, and locale support other than LC_MESSAGES can mess other
2225 things, so I rather chose to stick with just setting LC_MESSAGES.
2227 GNU strptime does not have this problem because it recognizes
2228 both international and local dates. */
2230 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2231 if (check_end (strptime (time_string, time_formats[i], &t)))
2232 return mktime_from_utc (&t);
2234 /* All formats have failed. */
2238 /* Authorization support: We support two authorization schemes:
2240 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2242 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2243 consisting of answering to the server's challenge with the proper
2246 /* How many bytes it will take to store LEN bytes in base64. */
2247 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2249 /* Encode the string S of length LENGTH to base64 format and place it
2250 to STORE. STORE will be 0-terminated, and must point to a writable
2251 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2253 base64_encode (const char *s, char *store, int length)
2255 /* Conversion table. */
2256 static char tbl[64] = {
2257 'A','B','C','D','E','F','G','H',
2258 'I','J','K','L','M','N','O','P',
2259 'Q','R','S','T','U','V','W','X',
2260 'Y','Z','a','b','c','d','e','f',
2261 'g','h','i','j','k','l','m','n',
2262 'o','p','q','r','s','t','u','v',
2263 'w','x','y','z','0','1','2','3',
2264 '4','5','6','7','8','9','+','/'
2267 unsigned char *p = (unsigned char *)store;
2269 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2270 for (i = 0; i < length; i += 3)
2272 *p++ = tbl[s[0] >> 2];
2273 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2274 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2275 *p++ = tbl[s[2] & 0x3f];
2278 /* Pad the result if necessary... */
2279 if (i == length + 1)
2281 else if (i == length + 2)
2282 *(p - 1) = *(p - 2) = '=';
2283 /* ...and zero-terminate it. */
2287 /* Create the authentication header contents for the `Basic' scheme.
2288 This is done by encoding the string `USER:PASS' in base64 and
2289 prepending `HEADER: Basic ' to it. */
2291 basic_authentication_encode (const char *user, const char *passwd,
2294 char *t1, *t2, *res;
2295 int len1 = strlen (user) + 1 + strlen (passwd);
2296 int len2 = BASE64_LENGTH (len1);
2298 t1 = (char *)alloca (len1 + 1);
2299 sprintf (t1, "%s:%s", user, passwd);
2300 t2 = (char *)alloca (1 + len2);
2301 base64_encode (t1, t2, len1);
2302 res = (char *)xmalloc (len2 + 11 + strlen (header));
2303 sprintf (res, "%s: Basic %s\r\n", header, t2);
2309 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2310 of a field in such a header. If the field is the one specified by
2311 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2312 digest authorization code), extract its value in the (char*)
2313 variable pointed by RET. Returns negative on a malformed header,
2314 or number of bytes that have been parsed by this call. */
2316 extract_header_attr (const char *au, const char *attr_name, char **ret)
2318 const char *cp, *ep;
2322 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2324 cp += strlen (attr_name);
2327 cp += skip_lws (cp);
2332 cp += skip_lws (cp);
2337 for (ep = cp; *ep && *ep != '\"'; ep++)
2342 *ret = strdupdelim (cp, ep);
2349 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2350 an array of 16 bytes containing the hash keys, and BUF should be a
2351 buffer of 33 writable characters (32 for hex digits plus one for
2352 zero termination). */
2354 dump_hash (unsigned char *buf, const unsigned char *hash)
2358 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2360 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2361 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2366 /* Take the line apart to find the challenge, and compose a digest
2367 authorization header. See RFC2069 section 2.1.2. */
2369 digest_authentication_encode (const char *au, const char *user,
2370 const char *passwd, const char *method,
2373 static char *realm, *opaque, *nonce;
2378 { "realm", &realm },
2379 { "opaque", &opaque },
2384 realm = opaque = nonce = NULL;
2386 au += 6; /* skip over `Digest' */
2391 au += skip_lws (au);
2392 for (i = 0; i < ARRAY_SIZE (options); i++)
2394 int skip = extract_header_attr (au, options[i].name,
2395 options[i].variable);
2399 FREE_MAYBE (opaque);
2409 if (i == ARRAY_SIZE (options))
2411 while (*au && *au != '=')
2415 au += skip_lws (au);
2419 while (*au && *au != '\"')
2426 while (*au && *au != ',')
2431 if (!realm || !nonce || !user || !passwd || !path || !method)
2434 FREE_MAYBE (opaque);
2439 /* Calculate the digest value. */
2441 ALLOCA_MD5_CONTEXT (ctx);
2442 unsigned char hash[MD5_HASHLEN];
2443 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2444 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2446 /* A1BUF = H(user ":" realm ":" password) */
2448 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2449 gen_md5_update ((unsigned char *)":", 1, ctx);
2450 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2451 gen_md5_update ((unsigned char *)":", 1, ctx);
2452 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2453 gen_md5_finish (ctx, hash);
2454 dump_hash (a1buf, hash);
2456 /* A2BUF = H(method ":" path) */
2458 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2459 gen_md5_update ((unsigned char *)":", 1, ctx);
2460 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2461 gen_md5_finish (ctx, hash);
2462 dump_hash (a2buf, hash);
2464 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2466 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2467 gen_md5_update ((unsigned char *)":", 1, ctx);
2468 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2469 gen_md5_update ((unsigned char *)":", 1, ctx);
2470 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2471 gen_md5_finish (ctx, hash);
2472 dump_hash (response_digest, hash);
2474 res = (char*) xmalloc (strlen (user)
2479 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2480 + (opaque ? strlen (opaque) : 0)
2482 sprintf (res, "Authorization: Digest \
2483 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2484 user, realm, nonce, path, response_digest);
2487 char *p = res + strlen (res);
2488 strcat (p, ", opaque=\"");
2492 strcat (res, "\r\n");
2496 #endif /* USE_DIGEST */
2499 #define BEGINS_WITH(line, string_constant) \
2500 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2501 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2502 || !line[sizeof (string_constant) - 1]))
2505 known_authentication_scheme_p (const char *au)
2507 return BEGINS_WITH (au, "Basic")
2508 || BEGINS_WITH (au, "Digest")
2509 || BEGINS_WITH (au, "NTLM");
2514 /* Create the HTTP authorization request header. When the
2515 `WWW-Authenticate' response header is seen, according to the
2516 authorization scheme specified in that header (`Basic' and `Digest'
2517 are supported by the current implementation), produce an
2518 appropriate HTTP authorization request header. */
2520 create_authorization_line (const char *au, const char *user,
2521 const char *passwd, const char *method,
2524 char *wwwauth = NULL;
2526 if (!strncasecmp (au, "Basic", 5))
2527 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2528 if (!strncasecmp (au, "NTLM", 4))
2529 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2531 else if (!strncasecmp (au, "Digest", 6))
2532 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2533 #endif /* USE_DIGEST */
2540 if (pc_last_host_ip)
2541 address_list_release (pc_last_host_ip);