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"
76 extern char *version_string;
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))
95 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
97 #define HTTP_STATUS_OK 200
98 #define HTTP_STATUS_CREATED 201
99 #define HTTP_STATUS_ACCEPTED 202
100 #define HTTP_STATUS_NO_CONTENT 204
101 #define HTTP_STATUS_PARTIAL_CONTENTS 206
103 /* Redirection 3xx. */
104 #define HTTP_STATUS_MULTIPLE_CHOICES 300
105 #define HTTP_STATUS_MOVED_PERMANENTLY 301
106 #define HTTP_STATUS_MOVED_TEMPORARILY 302
107 #define HTTP_STATUS_NOT_MODIFIED 304
109 /* Client error 4xx. */
110 #define HTTP_STATUS_BAD_REQUEST 400
111 #define HTTP_STATUS_UNAUTHORIZED 401
112 #define HTTP_STATUS_FORBIDDEN 403
113 #define HTTP_STATUS_NOT_FOUND 404
115 /* Server errors 5xx. */
116 #define HTTP_STATUS_INTERNAL 500
117 #define HTTP_STATUS_NOT_IMPLEMENTED 501
118 #define HTTP_STATUS_BAD_GATEWAY 502
119 #define HTTP_STATUS_UNAVAILABLE 503
122 /* Parse the HTTP status line, which is of format:
124 HTTP-Version SP Status-Code SP Reason-Phrase
126 The function returns the status-code, or -1 if the status line is
127 malformed. The pointer to reason-phrase is returned in RP. */
129 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
131 /* (the variables must not be named `major' and `minor', because
132 that breaks compilation with SunOS4 cc.) */
133 int mjr, mnr, statcode;
136 *reason_phrase_ptr = NULL;
138 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
139 major version, and Y is minor version. */
140 if (strncmp (line, "HTTP/", 5) != 0)
144 /* Calculate major HTTP version. */
146 for (mjr = 0; ISDIGIT (*line); line++)
147 mjr = 10 * mjr + (*line - '0');
148 if (*line != '.' || p == line)
152 /* Calculate minor HTTP version. */
154 for (mnr = 0; ISDIGIT (*line); line++)
155 mnr = 10 * mnr + (*line - '0');
156 if (*line != ' ' || p == line)
158 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
159 minor version can be safely ignored. */
164 /* Calculate status code. */
165 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
167 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
169 /* Set up the reason phrase pointer. */
171 /* RFC2068 requires SPC here, but we allow the string to finish
172 here, in case no reason-phrase is present. */
176 *reason_phrase_ptr = line;
181 *reason_phrase_ptr = line + 1;
186 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
188 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
189 PROMISED_SIZE bytes are sent over the wire -- if the file is
190 longer, read only that much; if the file is shorter, report an error. */
193 post_file (int sock, void *ssl, const char *file_name, long promised_size)
195 static char chunk[8192];
200 /* Only one of SOCK and SSL may be active at the same time. */
201 assert (sock > -1 || ssl != NULL);
202 assert (sock == -1 || ssl == NULL);
204 DEBUGP (("[writing POST file %s ... ", file_name));
206 fp = fopen (file_name, "rb");
209 while (!feof (fp) && written < promised_size)
212 int length = fread (chunk, 1, sizeof (chunk), fp);
215 towrite = WMIN (promised_size - written, length);
218 write_error = ssl_iwrite (ssl, chunk, towrite);
221 write_error = iwrite (sock, chunk, towrite);
231 /* If we've written less than was promised, report a (probably
232 nonsensical) error rather than break the promise. */
233 if (written < promised_size)
239 assert (written == promised_size);
240 DEBUGP (("done]\n"));
244 /* Functions to be used as arguments to header_process(): */
246 struct http_process_range_closure {
252 /* Parse the `Content-Range' header and extract the information it
253 contains. Returns 1 if successful, -1 otherwise. */
255 http_process_range (const char *hdr, void *arg)
257 struct http_process_range_closure *closure
258 = (struct http_process_range_closure *)arg;
261 /* Certain versions of Nutscape proxy server send out
262 `Content-Length' without "bytes" specifier, which is a breach of
263 RFC2068 (as well as the HTTP/1.1 draft which was current at the
264 time). But hell, I must support it... */
265 if (!strncasecmp (hdr, "bytes", 5))
268 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
272 hdr += skip_lws (hdr);
278 for (num = 0; ISDIGIT (*hdr); hdr++)
279 num = 10 * num + (*hdr - '0');
280 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
282 closure->first_byte_pos = num;
284 for (num = 0; ISDIGIT (*hdr); hdr++)
285 num = 10 * num + (*hdr - '0');
286 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
288 closure->last_byte_pos = num;
290 for (num = 0; ISDIGIT (*hdr); hdr++)
291 num = 10 * num + (*hdr - '0');
292 closure->entity_length = num;
296 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
297 Used for `Accept-Ranges'. */
299 http_process_none (const char *hdr, void *arg)
301 int *where = (int *)arg;
303 if (strstr (hdr, "none"))
310 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
312 http_process_type (const char *hdr, void *arg)
314 char **result = (char **)arg;
315 /* Locate P on `;' or the terminating zero, whichever comes first. */
316 const char *p = strchr (hdr, ';');
318 p = hdr + strlen (hdr);
319 while (p > hdr && ISSPACE (*(p - 1)))
321 *result = strdupdelim (hdr, p);
325 /* Check whether the `Connection' header is set to "keep-alive". */
327 http_process_connection (const char *hdr, void *arg)
329 int *flag = (int *)arg;
330 if (!strcasecmp (hdr, "Keep-Alive"))
335 /* Commit the cookie to the cookie jar. */
338 http_process_set_cookie (const char *hdr, void *arg)
340 struct url *u = (struct url *)arg;
342 /* The jar should have been created by now. */
343 assert (wget_cookie_jar != NULL);
345 cookie_jar_process_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
351 /* Persistent connections. Currently, we cache the most recently used
352 connection as persistent, provided that the HTTP server agrees to
353 make it such. The persistence data is stored in the variables
354 below. Ideally, it would be in a structure, and it should be
355 possible to cache an arbitrary fixed number of these connections.
357 I think the code is quite easy to extend in that direction. */
359 /* Whether a persistent connection is active. */
360 static int pc_active_p;
361 /* Host and port of currently active persistent connection. */
362 static struct address_list *pc_last_host_ip;
363 static unsigned short pc_last_port;
365 /* File descriptor of the currently active persistent connection. */
366 static int pc_last_fd;
369 /* Whether a ssl handshake has occoured on this connection */
370 static int pc_active_ssl;
371 /* SSL connection of the currently active persistent connection. */
372 static SSL *pc_last_ssl;
373 #endif /* HAVE_SSL */
375 /* Mark the persistent connection as invalid. This is used by the
376 CLOSE_* macros after they forcefully close a registered persistent
377 connection. This does not close the file descriptor -- it is left
378 to the caller to do that. (Maybe it should, though.) */
381 invalidate_persistent (void)
386 #endif /* HAVE_SSL */
387 if (pc_last_host_ip != NULL)
389 address_list_release (pc_last_host_ip);
390 pc_last_host_ip = NULL;
392 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
395 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
396 persistent. This will enable someone to use the same connection
397 later. In the context of HTTP, this must be called only AFTER the
398 response has been received and the server has promised that the
399 connection will remain alive.
401 If a previous connection was persistent, it is closed. */
405 register_persistent (const char *host, unsigned short port, int fd, SSL *ssl)
409 register_persistent (const char *host, unsigned short port, int fd)
414 if (pc_last_fd == fd)
416 /* The connection FD is already registered. Nothing to
422 /* The old persistent connection is still active; let's
423 close it first. This situation arises whenever a
424 persistent connection exists, but we then connect to a
425 different host, and try to register a persistent
426 connection to that one. */
428 /* The ssl disconnect has to take place before the closing
431 shutdown_ssl(pc_last_ssl);
434 invalidate_persistent ();
438 assert (pc_last_host_ip == NULL);
440 /* This lookup_host cannot fail, because it has the results in the
442 pc_last_host_ip = lookup_host (host, 1);
443 assert (pc_last_host_ip != NULL);
450 pc_active_ssl = ssl ? 1 : 0;
452 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
456 # define SHUTDOWN_SSL(ssl) do { \
458 shutdown_ssl (ssl); \
461 # define SHUTDOWN_SSL(ssl)
464 /* Return non-zero if a persistent connection is available for
465 connecting to HOST:PORT. */
469 persistent_available_p (const char *host, unsigned short port, int ssl)
473 persistent_available_p (const char *host, unsigned short port)
477 struct address_list *this_host_ip;
479 /* First, check whether a persistent connection is active at all. */
482 /* Second, check if the active connection pertains to the correct
483 (HOST, PORT) ordered pair. */
484 if (port != pc_last_port)
488 /* Second, a): check if current connection is (not) ssl, too. This
489 test is unlikely to fail because HTTP and HTTPS typicaly use
490 different ports. Yet it is possible, or so I [Christian
491 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
493 if (ssl != pc_active_ssl)
495 #endif /* HAVE_SSL */
497 this_host_ip = lookup_host (host, 1);
501 /* To equate the two host names for the purposes of persistent
502 connections, they need to share all the IP addresses in the
504 success = address_list_match_all (pc_last_host_ip, this_host_ip);
505 address_list_release (this_host_ip);
509 /* Third: check whether the connection is still open. This is
510 important because most server implement a liberal (short) timeout
511 on persistent connections. Wget can of course always reconnect
512 if the connection doesn't work out, but it's nicer to know in
513 advance. This test is a logical followup of the first test, but
514 is "expensive" and therefore placed at the end of the list. */
515 if (!test_socket_open (pc_last_fd))
517 /* Oops, the socket is no longer open. Now that we know that,
518 let's invalidate the persistent connection before returning
522 SHUTDOWN_SSL (pc_last_ssl);
525 invalidate_persistent ();
531 /* The idea behind these two CLOSE macros is to distinguish between
532 two cases: one when the job we've been doing is finished, and we
533 want to close the connection and leave, and two when something is
534 seriously wrong and we're closing the connection as part of
537 In case of keep_alive, CLOSE_FINISH should leave the connection
538 open, while CLOSE_INVALIDATE should still close it.
540 Note that the semantics of the flag `keep_alive' is "this
541 connection *will* be reused (the server has promised not to close
542 the connection once we're done)", while the semantics of
543 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
544 active, registered connection". */
546 #define CLOSE_FINISH(fd) do { \
549 SHUTDOWN_SSL (ssl); \
551 if (pc_active_p && (fd) == pc_last_fd) \
552 invalidate_persistent (); \
556 #define CLOSE_INVALIDATE(fd) do { \
557 SHUTDOWN_SSL (ssl); \
559 if (pc_active_p && (fd) == pc_last_fd) \
560 invalidate_persistent (); \
565 long len; /* received length */
566 long contlen; /* expected length */
567 long restval; /* the restart value */
568 int res; /* the result of last read */
569 char *newloc; /* new location (redirection) */
570 char *remote_time; /* remote time-stamp string */
571 char *error; /* textual HTTP error */
572 int statcode; /* status code */
573 double dltime; /* time of the download in msecs */
574 int no_truncate; /* whether truncating the file is
576 const char *referer; /* value of the referer header. */
577 char **local_file; /* local file. */
581 free_hstat (struct http_stat *hs)
583 FREE_MAYBE (hs->newloc);
584 FREE_MAYBE (hs->remote_time);
585 FREE_MAYBE (hs->error);
587 /* Guard against being called twice. */
589 hs->remote_time = NULL;
593 static char *create_authorization_line PARAMS ((const char *, const char *,
594 const char *, const char *,
596 static char *basic_authentication_encode PARAMS ((const char *, const char *,
598 static int known_authentication_scheme_p PARAMS ((const char *));
600 time_t http_atotm PARAMS ((const char *));
602 #define BEGINS_WITH(line, string_constant) \
603 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
604 && (ISSPACE (line[sizeof (string_constant) - 1]) \
605 || !line[sizeof (string_constant) - 1]))
607 /* Retrieve a document through HTTP protocol. It recognizes status
608 code, and correctly handles redirections. It closes the network
609 socket. If it receives an error from the functions below it, it
610 will print it if there is enough information to do so (almost
611 always), returning the error to the caller (i.e. http_loop).
613 Various HTTP parameters are stored to hs. Although it parses the
614 response code correctly, it is not used in a sane way. The caller
617 If PROXY is non-NULL, the connection will be made to the proxy
618 server, and u->url will be requested. */
620 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
622 char *request, *type, *command, *full_path;
624 char *pragma_h, *referer, *useragent, *range, *wwwauth;
625 char *authenticate_h;
629 char *request_keep_alive;
630 int sock, hcount, all_length, statcode;
632 long contlen, contrange;
635 int auth_tried_already;
638 static SSL_CTX *ssl_ctx = NULL;
641 char *cookies = NULL;
643 /* Whether this connection will be kept alive after the HTTP request
647 /* Flags that detect the two ways of specifying HTTP keep-alive
649 int http_keep_alive_1, http_keep_alive_2;
651 /* Whether keep-alive should be inhibited. */
652 int inhibit_keep_alive;
654 /* Whether we need to print the host header with braces around host,
655 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
656 "Host: symbolic-name:1234". */
657 int squares_around_host = 0;
659 /* Headers sent when using POST. */
660 char *post_content_type, *post_content_length;
661 long post_data_size = 0;
664 /* initialize ssl_ctx on first run */
667 uerr_t err = init_ssl (&ssl_ctx);
672 case SSLERRCTXCREATE:
674 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
678 /* try without certfile */
679 logprintf (LOG_NOTQUIET,
680 _("Failed to load certificates from %s\n"),
683 logprintf (LOG_NOTQUIET,
684 _("Trying without the specified certificate\n"));
687 logprintf (LOG_NOTQUIET,
688 _("Failed to get certificate key from %s\n"),
691 logprintf (LOG_NOTQUIET,
692 _("Trying without the specified certificate\n"));
699 #endif /* HAVE_SSL */
701 if (!(*dt & HEAD_ONLY))
702 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
703 know the local filename so we can save to it. */
704 assert (*hs->local_file != NULL);
707 auth_tried_already = 0;
709 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
712 /* We need to come back here when the initial attempt to retrieve
713 without authorization header fails. (Expected to happen at least
714 for the Digest authorization scheme.) */
717 http_keep_alive_1 = http_keep_alive_2 = 0;
719 post_content_type = NULL;
720 post_content_length = NULL;
722 /* Initialize certain elements of struct http_stat. */
727 hs->remote_time = NULL;
730 /* If we're using a proxy, we will be connecting to the proxy
732 conn = proxy ? proxy : u;
734 /* First: establish the connection. */
735 if (inhibit_keep_alive
738 !persistent_available_p (conn->host, conn->port)
740 !persistent_available_p (conn->host, conn->port,
741 u->scheme == SCHEME_HTTPS)
742 #endif /* HAVE_SSL */
745 struct address_list *al = lookup_host (conn->host, 0);
748 set_connection_host_name (conn->host);
749 sock = connect_to_many (al, conn->port, 0);
750 set_connection_host_name (NULL);
751 address_list_release (al);
754 return CONNECT_ERROR (errno);
757 if (conn->scheme == SCHEME_HTTPS)
758 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
760 logputs (LOG_VERBOSE, "\n");
761 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
766 #endif /* HAVE_SSL */
770 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
771 conn->host, conn->port);
772 /* #### pc_last_fd should be accessed through an accessor
777 #endif /* HAVE_SSL */
778 DEBUGP (("Reusing fd %d.\n", sock));
783 else if (opt.post_file_name || opt.post_data)
791 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
792 sprintf (referer, "Referer: %s\r\n", hs->referer);
795 if (*dt & SEND_NOCACHE)
796 pragma_h = "Pragma: no-cache\r\n";
802 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
803 /* Gag me! Some servers (e.g. WebSitePro) have been known to
804 respond to the following `Range' format by generating a
805 multipart/x-byte-ranges MIME document! This MIME type was
806 present in an old draft of the byteranges specification.
807 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
808 only if multiple non-overlapping ranges are requested --
809 which Wget never does. */
810 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
815 STRDUP_ALLOCA (useragent, opt.useragent);
818 useragent = (char *)alloca (10 + strlen (version_string));
819 sprintf (useragent, "Wget/%s", version_string);
821 /* Construct the authentication, if userid is present. */
824 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
825 user = user ? user : opt.http_user;
826 passwd = passwd ? passwd : opt.http_passwd;
833 /* We have the username and the password, but haven't tried
834 any authorization yet. Let's see if the "Basic" method
835 works. If not, we'll come back here and construct a
836 proper authorization method with the right challenges.
838 If we didn't employ this kind of logic, every URL that
839 requires authorization would have to be processed twice,
840 which is very suboptimal and generates a bunch of false
841 "unauthorized" errors in the server log.
843 #### But this logic also has a serious problem when used
844 with stronger authentications: we *first* transmit the
845 username and the password in clear text, and *then*
846 attempt a stronger authentication scheme. That cannot be
847 right! We are only fortunate that almost everyone still
848 uses the `Basic' scheme anyway.
850 There should be an option to prevent this from happening,
851 for those who use strong authentication schemes and value
853 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
857 /* Use the full path, i.e. one that includes the leading
858 slash and the query string, but is independent of proxy
860 char *pth = url_full_path (u);
861 wwwauth = create_authorization_line (authenticate_h, user, passwd,
870 char *proxy_user, *proxy_passwd;
871 /* For normal username and password, URL components override
872 command-line/wgetrc parameters. With proxy authentication,
873 it's the reverse, because proxy URLs are normally the
874 "permanent" ones, so command-line args should take
876 if (opt.proxy_user && opt.proxy_passwd)
878 proxy_user = opt.proxy_user;
879 proxy_passwd = opt.proxy_passwd;
883 proxy_user = proxy->user;
884 proxy_passwd = proxy->passwd;
886 /* #### This does not appear right. Can't the proxy request,
887 say, `Digest' authentication? */
888 if (proxy_user && proxy_passwd)
889 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
890 "Proxy-Authorization");
893 /* String of the form :PORT. Used only for non-standard ports. */
895 if (u->port != scheme_default_port (u->scheme))
897 port_maybe = (char *)alloca (numdigit (u->port) + 2);
898 sprintf (port_maybe, ":%d", u->port);
901 if (!inhibit_keep_alive)
902 request_keep_alive = "Connection: Keep-Alive\r\n";
904 request_keep_alive = NULL;
907 cookies = cookie_jar_generate_cookie_header (wget_cookie_jar, u->host,
910 u->scheme == SCHEME_HTTPS
916 if (opt.post_data || opt.post_file_name)
918 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
920 post_data_size = strlen (opt.post_data);
923 post_data_size = file_size (opt.post_file_name);
924 if (post_data_size == -1)
926 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
931 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
932 sprintf (post_content_length,
933 "Content-Length: %ld\r\n", post_data_size);
937 full_path = xstrdup (u->url);
939 /* Use the full path, i.e. one that includes the leading slash and
940 the query string. E.g. if u->path is "foo/bar" and u->query is
941 "param=value", full_path will be "/foo/bar?param=value". */
942 full_path = url_full_path (u);
944 if (strchr (u->host, ':'))
945 squares_around_host = 1;
947 /* Allocate the memory for the request. */
948 request = (char *)alloca (strlen (command)
952 + (port_maybe ? strlen (port_maybe) : 0)
953 + strlen (HTTP_ACCEPT)
954 + (request_keep_alive
955 ? strlen (request_keep_alive) : 0)
956 + (referer ? strlen (referer) : 0)
957 + (cookies ? strlen (cookies) : 0)
958 + (wwwauth ? strlen (wwwauth) : 0)
959 + (proxyauth ? strlen (proxyauth) : 0)
960 + (range ? strlen (range) : 0)
963 ? strlen (post_content_type) : 0)
964 + (post_content_length
965 ? strlen (post_content_length) : 0)
966 + (opt.user_header ? strlen (opt.user_header) : 0)
968 /* Construct the request. */
974 %s%s%s%s%s%s%s%s%s%s\r\n",
977 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
978 port_maybe ? port_maybe : "",
980 request_keep_alive ? request_keep_alive : "",
981 referer ? referer : "",
982 cookies ? cookies : "",
983 wwwauth ? wwwauth : "",
984 proxyauth ? proxyauth : "",
987 post_content_type ? post_content_type : "",
988 post_content_length ? post_content_length : "",
989 opt.user_header ? opt.user_header : "");
990 DEBUGP (("---request begin---\n%s", request));
992 /* Free the temporary memory. */
993 FREE_MAYBE (wwwauth);
994 FREE_MAYBE (proxyauth);
995 FREE_MAYBE (cookies);
998 /* Send the request to server. */
1000 if (conn->scheme == SCHEME_HTTPS)
1001 write_error = ssl_iwrite (ssl, request, strlen (request));
1004 write_error = iwrite (sock, request, strlen (request));
1006 if (write_error >= 0)
1010 DEBUGP (("[POST data: %s]\n", opt.post_data));
1012 if (conn->scheme == SCHEME_HTTPS)
1013 write_error = ssl_iwrite (ssl, opt.post_data, post_data_size);
1016 write_error = iwrite (sock, opt.post_data, post_data_size);
1018 else if (opt.post_file_name)
1021 if (conn->scheme == SCHEME_HTTPS)
1022 write_error = post_file (-1, ssl, opt.post_file_name,
1026 write_error = post_file (sock, NULL, opt.post_file_name,
1030 DEBUGP (("---request end---\n"));
1032 if (write_error < 0)
1034 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1036 CLOSE_INVALIDATE (sock);
1039 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1040 proxy ? "Proxy" : "HTTP");
1041 contlen = contrange = -1;
1046 /* Before reading anything, initialize the rbuf. */
1047 rbuf_initialize (&rbuf, sock);
1049 if (conn->scheme == SCHEME_HTTPS)
1053 #endif /* HAVE_SSL */
1056 /* Header-fetching loop. */
1064 /* Get the header. */
1065 status = header_get (&rbuf, &hdr,
1066 /* Disallow continuations for status line. */
1067 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
1069 /* Check for errors. */
1070 if (status == HG_EOF && *hdr)
1072 /* This used to be an unconditional error, but that was
1073 somewhat controversial, because of a large number of
1074 broken CGI's that happily "forget" to send the second EOL
1075 before closing the connection of a HEAD request.
1077 So, the deal is to check whether the header is empty
1078 (*hdr is zero if it is); if yes, it means that the
1079 previous header was fully retrieved, and that -- most
1080 probably -- the request is complete. "...be liberal in
1081 what you accept." Oh boy. */
1082 logputs (LOG_VERBOSE, "\n");
1083 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1086 FREE_MAYBE (all_headers);
1087 CLOSE_INVALIDATE (sock);
1090 else if (status == HG_ERROR)
1092 logputs (LOG_VERBOSE, "\n");
1093 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1097 FREE_MAYBE (all_headers);
1098 CLOSE_INVALIDATE (sock);
1102 /* If the headers are to be saved to a file later, save them to
1104 if (opt.save_headers)
1106 int lh = strlen (hdr);
1107 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1108 memcpy (all_headers + all_length, hdr, lh);
1110 all_headers[all_length++] = '\n';
1111 all_headers[all_length] = '\0';
1114 /* Check for status line. */
1118 /* Parse the first line of server response. */
1119 statcode = parse_http_status_line (hdr, &error);
1120 hs->statcode = statcode;
1121 /* Store the descriptive response. */
1122 if (statcode == -1) /* malformed response */
1124 /* A common reason for "malformed response" error is the
1125 case when no data was actually received. Handle this
1128 hs->error = xstrdup (_("No data received"));
1130 hs->error = xstrdup (_("Malformed status line"));
1135 hs->error = xstrdup (_("(no description)"));
1137 hs->error = xstrdup (error);
1139 if ((statcode != -1)
1145 if (opt.server_response)
1146 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1148 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1154 /* Exit on empty header. */
1161 /* Print the header if requested. */
1162 if (opt.server_response && hcount != 1)
1163 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1165 /* Try getting content-length. */
1166 if (contlen == -1 && !opt.ignore_length)
1167 if (header_process (hdr, "Content-Length", header_extract_number,
1170 /* Try getting content-type. */
1172 if (header_process (hdr, "Content-Type", http_process_type, &type))
1174 /* Try getting location. */
1176 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1178 /* Try getting last-modified. */
1179 if (!hs->remote_time)
1180 if (header_process (hdr, "Last-Modified", header_strdup,
1183 /* Try getting cookies. */
1185 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1187 /* Try getting www-authentication. */
1188 if (!authenticate_h)
1189 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1192 /* Check for accept-ranges header. If it contains the word
1193 `none', disable the ranges. */
1194 if (*dt & ACCEPTRANGES)
1197 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1200 *dt &= ~ACCEPTRANGES;
1204 /* Try getting content-range. */
1205 if (contrange == -1)
1207 struct http_process_range_closure closure;
1208 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1210 contrange = closure.first_byte_pos;
1214 /* Check for keep-alive related responses. */
1215 if (!inhibit_keep_alive)
1217 /* Check for the `Keep-Alive' header. */
1218 if (!http_keep_alive_1)
1220 if (header_process (hdr, "Keep-Alive", header_exists,
1221 &http_keep_alive_1))
1224 /* Check for `Connection: Keep-Alive'. */
1225 if (!http_keep_alive_2)
1227 if (header_process (hdr, "Connection", http_process_connection,
1228 &http_keep_alive_2))
1236 logputs (LOG_VERBOSE, "\n");
1239 && (http_keep_alive_1 || http_keep_alive_2))
1241 assert (inhibit_keep_alive == 0);
1245 /* The server has promised that it will not close the connection
1246 when we're done. This means that we can register it. */
1248 register_persistent (conn->host, conn->port, sock);
1250 register_persistent (conn->host, conn->port, sock, ssl);
1251 #endif /* HAVE_SSL */
1253 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1256 /* Authorization is required. */
1260 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1261 might be more bytes in the body. */
1262 if (auth_tried_already)
1264 /* If we have tried it already, then there is not point
1267 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1268 xfree (authenticate_h);
1271 else if (!known_authentication_scheme_p (authenticate_h))
1273 xfree (authenticate_h);
1274 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1277 else if (BEGINS_WITH (authenticate_h, "Basic"))
1279 /* The authentication scheme is basic, the one we try by
1280 default, and it failed. There's no sense in trying
1286 auth_tried_already = 1;
1290 /* We do not need this anymore. */
1293 xfree (authenticate_h);
1294 authenticate_h = NULL;
1297 /* 20x responses are counted among successful by default. */
1298 if (H_20X (statcode))
1301 /* Return if redirected. */
1302 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1304 /* RFC2068 says that in case of the 300 (multiple choices)
1305 response, the server can output a preferred URL through
1306 `Location' header; otherwise, the request should be treated
1307 like GET. So, if the location is set, it will be a
1308 redirection; otherwise, just proceed normally. */
1309 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1313 logprintf (LOG_VERBOSE,
1314 _("Location: %s%s\n"),
1315 hs->newloc ? hs->newloc : _("unspecified"),
1316 hs->newloc ? _(" [following]") : "");
1317 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1318 might be more bytes in the body. */
1320 FREE_MAYBE (all_headers);
1325 /* If content-type is not given, assume text/html. This is because
1326 of the multitude of broken CGI's that "forget" to generate the
1329 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1330 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1335 if (opt.html_extension && (*dt & TEXTHTML))
1336 /* -E / --html-extension / html_extension = on was specified, and this is a
1337 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1338 already the file's suffix, tack on ".html". */
1340 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1342 if (last_period_in_local_filename == NULL ||
1343 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1344 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1346 size_t local_filename_len = strlen(*hs->local_file);
1348 *hs->local_file = xrealloc(*hs->local_file,
1349 local_filename_len + sizeof(".html"));
1350 strcpy(*hs->local_file + local_filename_len, ".html");
1352 *dt |= ADDED_HTML_EXTENSION;
1356 if (contrange == -1)
1358 /* We did not get a content-range header. This means that the
1359 server did not honor our `Range' request. Normally, this
1360 means we should reset hs->restval and continue normally. */
1362 /* However, if `-c' is used, we need to be a bit more careful:
1364 1. If `-c' is specified and the file already existed when
1365 Wget was started, it would be a bad idea for us to start
1366 downloading it from scratch, effectively truncating it. I
1367 believe this cannot happen unless `-c' was specified.
1369 2. If `-c' is used on a file that is already fully
1370 downloaded, we're requesting bytes after the end of file,
1371 which can result in server not honoring `Range'. If this is
1372 the case, `Content-Length' will be equal to the length of the
1374 if (opt.always_rest)
1376 /* Check for condition #2. */
1377 if (hs->restval > 0 /* restart was requested. */
1378 && contlen != -1 /* we got content-length. */
1379 && hs->restval >= contlen /* file fully downloaded
1383 logputs (LOG_VERBOSE, _("\
1384 \n The file is already fully retrieved; nothing to do.\n\n"));
1385 /* In case the caller inspects. */
1388 /* Mark as successfully retrieved. */
1391 FREE_MAYBE (all_headers);
1392 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1393 might be more bytes in the body. */
1394 return RETRUNNEEDED;
1397 /* Check for condition #1. */
1398 if (hs->no_truncate)
1400 logprintf (LOG_NOTQUIET,
1403 Continued download failed on this file, which conflicts with `-c'.\n\
1404 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1406 FREE_MAYBE (all_headers);
1407 CLOSE_INVALIDATE (sock);
1408 return CONTNOTSUPPORTED;
1416 else if (contrange != hs->restval ||
1417 (H_PARTIAL (statcode) && contrange == -1))
1419 /* This means the whole request was somehow misunderstood by the
1420 server. Bail out. */
1422 FREE_MAYBE (all_headers);
1423 CLOSE_INVALIDATE (sock);
1430 contlen += contrange;
1432 contrange = -1; /* If conent-length was not sent,
1433 content-range will be ignored. */
1435 hs->contlen = contlen;
1439 if ((*dt & RETROKF) && !opt.server_response)
1441 /* No need to print this output if the body won't be
1442 downloaded at all, or if the original server response is
1444 logputs (LOG_VERBOSE, _("Length: "));
1447 logputs (LOG_VERBOSE, legible (contlen));
1448 if (contrange != -1)
1449 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1450 legible (contlen - contrange));
1453 logputs (LOG_VERBOSE,
1454 opt.ignore_length ? _("ignored") : _("unspecified"));
1456 logprintf (LOG_VERBOSE, " [%s]\n", type);
1458 logputs (LOG_VERBOSE, "\n");
1462 type = NULL; /* We don't need it any more. */
1464 /* Return if we have no intention of further downloading. */
1465 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1467 /* In case the caller cares to look... */
1471 FREE_MAYBE (all_headers);
1472 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1473 might be more bytes in the body. */
1474 return RETRFINISHED;
1477 /* Open the local file. */
1480 mkalldirs (*hs->local_file);
1482 rotate_backups (*hs->local_file);
1483 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1486 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1487 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1488 might be more bytes in the body. */
1489 FREE_MAYBE (all_headers);
1495 extern int global_download_count;
1497 /* To ensure that repeated "from scratch" downloads work for -O
1498 files, we rewind the file pointer, unless restval is
1499 non-zero. (This works only when -O is used on regular files,
1500 but it's still a valuable feature.)
1502 However, this loses when more than one URL is specified on
1503 the command line the second rewinds eradicates the contents
1504 of the first download. Thus we disable the above trick for
1505 all the downloads except the very first one.
1507 #### A possible solution to this would be to remember the
1508 file position in the output document and to seek to that
1509 position, instead of rewinding.
1511 We don't truncate stdout, since that breaks
1512 "wget -O - [...] >> foo".
1514 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1516 /* This will silently fail for streams that don't correspond
1517 to regular files, but that's OK. */
1519 /* ftruncate is needed because opt.dfp is opened in append
1520 mode if opt.always_rest is set. */
1521 ftruncate (fileno (fp), 0);
1526 /* #### This confuses the code that checks for file size. There
1527 should be some overhead information. */
1528 if (opt.save_headers)
1529 fwrite (all_headers, 1, all_length, fp);
1531 /* Get the contents of the document. */
1532 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1533 (contlen != -1 ? contlen : 0),
1534 &rbuf, keep_alive, &hs->dltime);
1537 CLOSE_FINISH (sock);
1539 CLOSE_INVALIDATE (sock);
1542 /* Close or flush the file. We have to be careful to check for
1543 error here. Checking the result of fwrite() is not enough --
1544 errors could go unnoticed! */
1547 flush_res = fclose (fp);
1549 flush_res = fflush (fp);
1550 if (flush_res == EOF)
1553 FREE_MAYBE (all_headers);
1556 return RETRFINISHED;
1559 /* The genuine HTTP loop! This is the part where the retrieval is
1560 retried, and retried, and retried, and... */
1562 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1563 int *dt, struct url *proxy)
1566 int use_ts, got_head = 0; /* time-stamping info */
1567 char *filename_plus_orig_suffix;
1568 char *local_filename = NULL;
1569 char *tms, *locf, *tmrate;
1571 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1572 long local_size = 0; /* the size of the local file */
1573 size_t filename_len;
1574 struct http_stat hstat; /* HTTP status */
1578 /* This used to be done in main(), but it's a better idea to do it
1579 here so that we don't go through the hoops if we're just using
1583 if (!wget_cookie_jar)
1584 wget_cookie_jar = cookie_jar_new ();
1585 if (opt.cookies_input && !cookies_loaded_p)
1587 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1588 cookies_loaded_p = 1;
1594 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1595 has_wildcards_p because it would also warn on `?', and we know that
1596 shows up in CGI paths a *lot*. */
1597 if (strchr (u->url, '*'))
1598 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1600 /* Determine the local filename. */
1601 if (local_file && *local_file)
1602 hstat.local_file = local_file;
1603 else if (local_file)
1605 *local_file = url_file_name (u);
1606 hstat.local_file = local_file;
1610 dummy = url_file_name (u);
1611 hstat.local_file = &dummy;
1614 if (!opt.output_document)
1615 locf = *hstat.local_file;
1617 locf = opt.output_document;
1619 hstat.referer = referer;
1621 filename_len = strlen (*hstat.local_file);
1622 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1624 if (opt.noclobber && file_exists_p (*hstat.local_file))
1626 /* If opt.noclobber is turned on and file already exists, do not
1627 retrieve the file */
1628 logprintf (LOG_VERBOSE, _("\
1629 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1630 /* If the file is there, we suppose it's retrieved OK. */
1633 /* #### Bogusness alert. */
1634 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1635 if (has_html_suffix_p (*hstat.local_file))
1643 if (opt.timestamping)
1645 boolean local_dot_orig_file_exists = FALSE;
1647 if (opt.backup_converted)
1648 /* If -K is specified, we'll act on the assumption that it was specified
1649 last time these files were downloaded as well, and instead of just
1650 comparing local file X against server file X, we'll compare local
1651 file X.orig (if extant, else X) against server file X. If -K
1652 _wasn't_ specified last time, or the server contains files called
1653 *.orig, -N will be back to not operating correctly with -k. */
1655 /* Would a single s[n]printf() call be faster? --dan
1657 Definitely not. sprintf() is horribly slow. It's a
1658 different question whether the difference between the two
1659 affects a program. Usually I'd say "no", but at one
1660 point I profiled Wget, and found that a measurable and
1661 non-negligible amount of time was lost calling sprintf()
1662 in url.c. Replacing sprintf with inline calls to
1663 strcpy() and long_to_string() made a difference.
1665 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1666 memcpy (filename_plus_orig_suffix + filename_len,
1667 ".orig", sizeof (".orig"));
1669 /* Try to stat() the .orig file. */
1670 if (stat (filename_plus_orig_suffix, &st) == 0)
1672 local_dot_orig_file_exists = TRUE;
1673 local_filename = filename_plus_orig_suffix;
1677 if (!local_dot_orig_file_exists)
1678 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1679 if (stat (*hstat.local_file, &st) == 0)
1680 local_filename = *hstat.local_file;
1682 if (local_filename != NULL)
1683 /* There was a local file, so we'll check later to see if the version
1684 the server has is the same version we already have, allowing us to
1690 /* Modification time granularity is 2 seconds for Windows, so
1691 increase local time by 1 second for later comparison. */
1694 local_size = st.st_size;
1698 /* Reset the counter. */
1700 *dt = 0 | ACCEPTRANGES;
1704 /* Increment the pass counter. */
1706 sleep_between_retrievals (count);
1707 /* Get the current time string. */
1708 tms = time_str (NULL);
1709 /* Print fetch message, if opt.verbose. */
1712 char *hurl = url_string (u, 1);
1716 sprintf (tmp, _("(try:%2d)"), count);
1717 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1718 tms, hurl, tmp, locf);
1720 ws_changetitle (hurl, 1);
1725 /* Default document type is empty. However, if spider mode is
1726 on or time-stamping is employed, HEAD_ONLY commands is
1727 encoded within *dt. */
1728 if (opt.spider || (use_ts && !got_head))
1732 /* Assume no restarting. */
1734 /* Decide whether or not to restart. */
1735 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1736 /* #### this calls access() and then stat(); could be optimized. */
1737 && file_exists_p (locf))
1738 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1739 hstat.restval = st.st_size;
1741 /* In `-c' is used and the file is existing and non-empty,
1742 refuse to truncate it if the server doesn't support continued
1744 hstat.no_truncate = 0;
1745 if (opt.always_rest && hstat.restval)
1746 hstat.no_truncate = 1;
1748 /* Decide whether to send the no-cache directive. We send it in
1750 a) we're using a proxy, and we're past our first retrieval.
1751 Some proxies are notorious for caching incomplete data, so
1752 we require a fresh get.
1753 b) caching is explicitly inhibited. */
1754 if ((proxy && count > 1) /* a */
1755 || !opt.allow_cache /* b */
1757 *dt |= SEND_NOCACHE;
1759 *dt &= ~SEND_NOCACHE;
1761 /* Try fetching the document, or at least its head. */
1762 err = gethttp (u, &hstat, dt, proxy);
1764 /* It's unfortunate that wget determines the local filename before finding
1765 out the Content-Type of the file. Barring a major restructuring of the
1766 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1767 *hstat.local_file to tack on ".html". */
1768 if (!opt.output_document)
1769 locf = *hstat.local_file;
1771 locf = opt.output_document;
1774 tms = time_str (NULL);
1775 /* Get the new location (with or without the redirection). */
1777 *newloc = xstrdup (hstat.newloc);
1780 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1781 case CONERROR: case READERR: case WRITEFAILED:
1783 /* Non-fatal errors continue executing the loop, which will
1784 bring them to "while" statement at the end, to judge
1785 whether the number of tries was exceeded. */
1786 free_hstat (&hstat);
1787 printwhat (count, opt.ntry);
1790 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1791 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1792 /* Fatal errors just return from the function. */
1793 free_hstat (&hstat);
1797 case FWRITEERR: case FOPENERR:
1798 /* Another fatal error. */
1799 logputs (LOG_VERBOSE, "\n");
1800 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1801 *hstat.local_file, strerror (errno));
1802 free_hstat (&hstat);
1807 /* Another fatal error. */
1808 logputs (LOG_VERBOSE, "\n");
1809 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1810 free_hstat (&hstat);
1815 /* Return the new location to the caller. */
1818 logprintf (LOG_NOTQUIET,
1819 _("ERROR: Redirection (%d) without location.\n"),
1821 free_hstat (&hstat);
1825 free_hstat (&hstat);
1830 /* The file was already fully retrieved. */
1831 free_hstat (&hstat);
1836 /* Deal with you later. */
1839 /* All possibilities should have been exhausted. */
1842 if (!(*dt & RETROKF))
1846 /* #### Ugly ugly ugly! */
1847 char *hurl = url_string (u, 1);
1848 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1851 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1852 tms, hstat.statcode, hstat.error);
1853 logputs (LOG_VERBOSE, "\n");
1854 free_hstat (&hstat);
1859 /* Did we get the time-stamp? */
1862 if (opt.timestamping && !hstat.remote_time)
1864 logputs (LOG_NOTQUIET, _("\
1865 Last-modified header missing -- time-stamps turned off.\n"));
1867 else if (hstat.remote_time)
1869 /* Convert the date-string into struct tm. */
1870 tmr = http_atotm (hstat.remote_time);
1871 if (tmr == (time_t) (-1))
1872 logputs (LOG_VERBOSE, _("\
1873 Last-modified header invalid -- time-stamp ignored.\n"));
1877 /* The time-stamping section. */
1882 use_ts = 0; /* no more time-stamping */
1883 count = 0; /* the retrieve count for HEAD is
1885 if (hstat.remote_time && tmr != (time_t) (-1))
1887 /* Now time-stamping can be used validly. Time-stamping
1888 means that if the sizes of the local and remote file
1889 match, and local file is newer than the remote file,
1890 it will not be retrieved. Otherwise, the normal
1891 download procedure is resumed. */
1893 (hstat.contlen == -1 || local_size == hstat.contlen))
1895 logprintf (LOG_VERBOSE, _("\
1896 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1898 free_hstat (&hstat);
1902 else if (tml >= tmr)
1903 logprintf (LOG_VERBOSE, _("\
1904 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1906 logputs (LOG_VERBOSE,
1907 _("Remote file is newer, retrieving.\n"));
1909 free_hstat (&hstat);
1912 if ((tmr != (time_t) (-1))
1914 && ((hstat.len == hstat.contlen) ||
1915 ((hstat.res == 0) &&
1916 ((hstat.contlen == -1) ||
1917 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1919 /* #### This code repeats in http.c and ftp.c. Move it to a
1921 const char *fl = NULL;
1922 if (opt.output_document)
1924 if (opt.od_known_regular)
1925 fl = opt.output_document;
1928 fl = *hstat.local_file;
1932 /* End of time-stamping section. */
1936 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1941 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1943 if (hstat.len == hstat.contlen)
1947 logprintf (LOG_VERBOSE,
1948 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1949 tms, tmrate, locf, hstat.len, hstat.contlen);
1950 logprintf (LOG_NONVERBOSE,
1951 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1952 tms, u->url, hstat.len, hstat.contlen, locf, count);
1955 downloaded_increase (hstat.len);
1957 /* Remember that we downloaded the file for later ".orig" code. */
1958 if (*dt & ADDED_HTML_EXTENSION)
1959 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1961 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1963 free_hstat (&hstat);
1967 else if (hstat.res == 0) /* No read error */
1969 if (hstat.contlen == -1) /* We don't know how much we were supposed
1970 to get, so assume we succeeded. */
1974 logprintf (LOG_VERBOSE,
1975 _("%s (%s) - `%s' saved [%ld]\n\n"),
1976 tms, tmrate, locf, hstat.len);
1977 logprintf (LOG_NONVERBOSE,
1978 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1979 tms, u->url, hstat.len, locf, count);
1982 downloaded_increase (hstat.len);
1984 /* Remember that we downloaded the file for later ".orig" code. */
1985 if (*dt & ADDED_HTML_EXTENSION)
1986 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1988 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1990 free_hstat (&hstat);
1994 else if (hstat.len < hstat.contlen) /* meaning we lost the
1995 connection too soon */
1997 logprintf (LOG_VERBOSE,
1998 _("%s (%s) - Connection closed at byte %ld. "),
1999 tms, tmrate, hstat.len);
2000 printwhat (count, opt.ntry);
2001 free_hstat (&hstat);
2004 else if (!opt.kill_longer) /* meaning we got more than expected */
2006 logprintf (LOG_VERBOSE,
2007 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2008 tms, tmrate, locf, hstat.len, hstat.contlen);
2009 logprintf (LOG_NONVERBOSE,
2010 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2011 tms, u->url, hstat.len, hstat.contlen, locf, count);
2013 downloaded_increase (hstat.len);
2015 /* Remember that we downloaded the file for later ".orig" code. */
2016 if (*dt & ADDED_HTML_EXTENSION)
2017 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2019 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2021 free_hstat (&hstat);
2025 else /* the same, but not accepted */
2027 logprintf (LOG_VERBOSE,
2028 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2029 tms, tmrate, hstat.len, hstat.contlen);
2030 printwhat (count, opt.ntry);
2031 free_hstat (&hstat);
2035 else /* now hstat.res can only be -1 */
2037 if (hstat.contlen == -1)
2039 logprintf (LOG_VERBOSE,
2040 _("%s (%s) - Read error at byte %ld (%s)."),
2041 tms, tmrate, hstat.len, strerror (errno));
2042 printwhat (count, opt.ntry);
2043 free_hstat (&hstat);
2046 else /* hstat.res == -1 and contlen is given */
2048 logprintf (LOG_VERBOSE,
2049 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2050 tms, tmrate, hstat.len, hstat.contlen,
2052 printwhat (count, opt.ntry);
2053 free_hstat (&hstat);
2060 while (!opt.ntry || (count < opt.ntry));
2064 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2065 than local timezone.
2067 mktime is similar but assumes struct tm, also known as the
2068 "broken-down" form of time, is in local time zone. mktime_from_utc
2069 uses mktime to make the conversion understanding that an offset
2070 will be introduced by the local time assumption.
2072 mktime_from_utc then measures the introduced offset by applying
2073 gmtime to the initial result and applying mktime to the resulting
2074 "broken-down" form. The difference between the two mktime results
2075 is the measured offset which is then subtracted from the initial
2076 mktime result to yield a calendar time which is the value returned.
2078 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2079 consistent offset (the non DST offset) since tm and tm+o might be
2080 on opposite sides of a DST change.
2082 Some implementations of mktime return -1 for the nonexistent
2083 localtime hour at the beginning of DST. In this event, use
2084 mktime(tm - 1hr) + 3600.
2088 gmtime(t+o) --> tm+o
2089 mktime(tm+o) --> t+2o
2090 t+o - (t+2o - t+o) = t
2092 Note that glibc contains a function of the same purpose named
2093 `timegm' (reverse of gmtime). But obviously, it is not universally
2094 available, and unfortunately it is not straightforwardly
2095 extractable for use here. Perhaps configure should detect timegm
2096 and use it where available.
2098 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2099 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2100 Further improved by Roger with assistance from Edward J. Sabol
2101 based on input by Jamie Zawinski. */
2104 mktime_from_utc (struct tm *t)
2115 return -1; /* can't deal with output from strptime */
2126 return -1; /* can't deal with output from gmtime */
2129 return (tl - (tb - tl));
2132 /* Check whether the result of strptime() indicates success.
2133 strptime() returns the pointer to how far it got to in the string.
2134 The processing has been successful if the string is at `GMT' or
2135 `+X', or at the end of the string.
2137 In extended regexp parlance, the function returns 1 if P matches
2138 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2139 can return) is considered a failure and 0 is returned. */
2141 check_end (const char *p)
2145 while (ISSPACE (*p))
2148 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2149 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2155 /* Convert the textual specification of time in TIME_STRING to the
2156 number of seconds since the Epoch.
2158 TIME_STRING can be in any of the three formats RFC2068 allows the
2159 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2160 Timezones are ignored, and should be GMT.
2162 Return the computed time_t representation, or -1 if the conversion
2165 This function uses strptime with various string formats for parsing
2166 TIME_STRING. This results in a parser that is not as lenient in
2167 interpreting TIME_STRING as I would like it to be. Being based on
2168 strptime, it always allows shortened months, one-digit days, etc.,
2169 but due to the multitude of formats in which time can be
2170 represented, an ideal HTTP time parser would be even more
2171 forgiving. It should completely ignore things like week days and
2172 concentrate only on the various forms of representing years,
2173 months, days, hours, minutes, and seconds. For example, it would
2174 be nice if it accepted ISO 8601 out of the box.
2176 I've investigated free and PD code for this purpose, but none was
2177 usable. getdate was big and unwieldy, and had potential copyright
2178 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2179 distributed with phttpd, is excellent, but we cannot use it because
2180 it is not assigned to the FSF. So I stuck it with strptime. */
2183 http_atotm (const char *time_string)
2185 /* NOTE: Solaris strptime man page claims that %n and %t match white
2186 space, but that's not universally available. Instead, we simply
2187 use ` ' to mean "skip all WS", which works under all strptime
2188 implementations I've tested. */
2190 static const char *time_formats[] = {
2191 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2192 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2193 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2194 (google.com uses this for their cookies.) */
2195 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2201 /* According to Roger Beeman, we need to initialize tm_isdst, since
2202 strptime won't do it. */
2205 /* Note that under foreign locales Solaris strptime() fails to
2206 recognize English dates, which renders this function useless. We
2207 solve this by being careful not to affect LC_TIME when
2208 initializing locale.
2210 Another solution would be to temporarily set locale to C, invoke
2211 strptime(), and restore it back. This is slow and dirty,
2212 however, and locale support other than LC_MESSAGES can mess other
2213 things, so I rather chose to stick with just setting LC_MESSAGES.
2215 GNU strptime does not have this problem because it recognizes
2216 both international and local dates. */
2218 for (i = 0; i < countof (time_formats); i++)
2219 if (check_end (strptime (time_string, time_formats[i], &t)))
2220 return mktime_from_utc (&t);
2222 /* All formats have failed. */
2226 /* Authorization support: We support two authorization schemes:
2228 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2230 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2231 consisting of answering to the server's challenge with the proper
2234 /* How many bytes it will take to store LEN bytes in base64. */
2235 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2237 /* Encode the string S of length LENGTH to base64 format and place it
2238 to STORE. STORE will be 0-terminated, and must point to a writable
2239 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2241 base64_encode (const char *s, char *store, int length)
2243 /* Conversion table. */
2244 static char tbl[64] = {
2245 'A','B','C','D','E','F','G','H',
2246 'I','J','K','L','M','N','O','P',
2247 'Q','R','S','T','U','V','W','X',
2248 'Y','Z','a','b','c','d','e','f',
2249 'g','h','i','j','k','l','m','n',
2250 'o','p','q','r','s','t','u','v',
2251 'w','x','y','z','0','1','2','3',
2252 '4','5','6','7','8','9','+','/'
2255 unsigned char *p = (unsigned char *)store;
2257 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2258 for (i = 0; i < length; i += 3)
2260 *p++ = tbl[s[0] >> 2];
2261 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2262 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2263 *p++ = tbl[s[2] & 0x3f];
2266 /* Pad the result if necessary... */
2267 if (i == length + 1)
2269 else if (i == length + 2)
2270 *(p - 1) = *(p - 2) = '=';
2271 /* ...and zero-terminate it. */
2275 /* Create the authentication header contents for the `Basic' scheme.
2276 This is done by encoding the string `USER:PASS' in base64 and
2277 prepending `HEADER: Basic ' to it. */
2279 basic_authentication_encode (const char *user, const char *passwd,
2282 char *t1, *t2, *res;
2283 int len1 = strlen (user) + 1 + strlen (passwd);
2284 int len2 = BASE64_LENGTH (len1);
2286 t1 = (char *)alloca (len1 + 1);
2287 sprintf (t1, "%s:%s", user, passwd);
2288 t2 = (char *)alloca (1 + len2);
2289 base64_encode (t1, t2, len1);
2290 res = (char *)xmalloc (len2 + 11 + strlen (header));
2291 sprintf (res, "%s: Basic %s\r\n", header, t2);
2297 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2298 of a field in such a header. If the field is the one specified by
2299 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2300 digest authorization code), extract its value in the (char*)
2301 variable pointed by RET. Returns negative on a malformed header,
2302 or number of bytes that have been parsed by this call. */
2304 extract_header_attr (const char *au, const char *attr_name, char **ret)
2306 const char *cp, *ep;
2310 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2312 cp += strlen (attr_name);
2315 cp += skip_lws (cp);
2320 cp += skip_lws (cp);
2325 for (ep = cp; *ep && *ep != '\"'; ep++)
2330 *ret = strdupdelim (cp, ep);
2337 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2338 an array of 16 bytes containing the hash keys, and BUF should be a
2339 buffer of 33 writable characters (32 for hex digits plus one for
2340 zero termination). */
2342 dump_hash (unsigned char *buf, const unsigned char *hash)
2346 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2348 *buf++ = XNUM_TO_digit (*hash >> 4);
2349 *buf++ = XNUM_TO_digit (*hash & 0xf);
2354 /* Take the line apart to find the challenge, and compose a digest
2355 authorization header. See RFC2069 section 2.1.2. */
2357 digest_authentication_encode (const char *au, const char *user,
2358 const char *passwd, const char *method,
2361 static char *realm, *opaque, *nonce;
2366 { "realm", &realm },
2367 { "opaque", &opaque },
2372 realm = opaque = nonce = NULL;
2374 au += 6; /* skip over `Digest' */
2379 au += skip_lws (au);
2380 for (i = 0; i < countof (options); i++)
2382 int skip = extract_header_attr (au, options[i].name,
2383 options[i].variable);
2387 FREE_MAYBE (opaque);
2397 if (i == countof (options))
2399 while (*au && *au != '=')
2403 au += skip_lws (au);
2407 while (*au && *au != '\"')
2414 while (*au && *au != ',')
2419 if (!realm || !nonce || !user || !passwd || !path || !method)
2422 FREE_MAYBE (opaque);
2427 /* Calculate the digest value. */
2429 ALLOCA_MD5_CONTEXT (ctx);
2430 unsigned char hash[MD5_HASHLEN];
2431 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2432 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2434 /* A1BUF = H(user ":" realm ":" password) */
2436 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2437 gen_md5_update ((unsigned char *)":", 1, ctx);
2438 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2439 gen_md5_update ((unsigned char *)":", 1, ctx);
2440 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2441 gen_md5_finish (ctx, hash);
2442 dump_hash (a1buf, hash);
2444 /* A2BUF = H(method ":" path) */
2446 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2447 gen_md5_update ((unsigned char *)":", 1, ctx);
2448 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2449 gen_md5_finish (ctx, hash);
2450 dump_hash (a2buf, hash);
2452 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2454 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2455 gen_md5_update ((unsigned char *)":", 1, ctx);
2456 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2457 gen_md5_update ((unsigned char *)":", 1, ctx);
2458 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2459 gen_md5_finish (ctx, hash);
2460 dump_hash (response_digest, hash);
2462 res = (char*) xmalloc (strlen (user)
2467 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2468 + (opaque ? strlen (opaque) : 0)
2470 sprintf (res, "Authorization: Digest \
2471 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2472 user, realm, nonce, path, response_digest);
2475 char *p = res + strlen (res);
2476 strcat (p, ", opaque=\"");
2480 strcat (res, "\r\n");
2484 #endif /* USE_DIGEST */
2487 #define BEGINS_WITH(line, string_constant) \
2488 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2489 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2490 || !line[sizeof (string_constant) - 1]))
2493 known_authentication_scheme_p (const char *au)
2495 return BEGINS_WITH (au, "Basic")
2496 || BEGINS_WITH (au, "Digest")
2497 || BEGINS_WITH (au, "NTLM");
2502 /* Create the HTTP authorization request header. When the
2503 `WWW-Authenticate' response header is seen, according to the
2504 authorization scheme specified in that header (`Basic' and `Digest'
2505 are supported by the current implementation), produce an
2506 appropriate HTTP authorization request header. */
2508 create_authorization_line (const char *au, const char *user,
2509 const char *passwd, const char *method,
2512 char *wwwauth = NULL;
2514 if (!strncasecmp (au, "Basic", 5))
2515 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2516 if (!strncasecmp (au, "NTLM", 4))
2517 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2519 else if (!strncasecmp (au, "Digest", 6))
2520 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2521 #endif /* USE_DIGEST */
2528 if (pc_last_host_ip)
2529 address_list_release (pc_last_host_ip);