2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001, 2002
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 In addition, as a special exception, the Free Software Foundation
22 gives permission to link the code of its release of Wget with the
23 OpenSSL project's "OpenSSL" library (or with modified versions of it
24 that use the same license as the "OpenSSL" library), and distribute
25 the linked executables. You must obey the GNU General Public License
26 in all respects for all of the code used other than "OpenSSL". If you
27 modify this file, you may extend this exception to your version of the
28 file, but you are not obligated to do so. If you do not wish to do
29 so, delete this exception statement from your version. */
35 #include <sys/types.h>
46 #if TIME_WITH_SYS_TIME
47 # include <sys/time.h>
51 # include <sys/time.h>
70 # include "gen_sslfunc.h"
78 extern char *version_string;
79 extern LARGE_INT total_downloaded_bytes;
82 static int cookies_loaded_p;
83 struct cookie_jar *wget_cookie_jar;
85 #define TEXTHTML_S "text/html"
86 #define TEXTXHTML_S "application/xhtml+xml"
87 #define HTTP_ACCEPT "*/*"
89 /* Some status code validation macros: */
90 #define H_20X(x) (((x) >= 200) && ((x) < 300))
91 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
92 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
93 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
94 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
96 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
98 #define HTTP_STATUS_OK 200
99 #define HTTP_STATUS_CREATED 201
100 #define HTTP_STATUS_ACCEPTED 202
101 #define HTTP_STATUS_NO_CONTENT 204
102 #define HTTP_STATUS_PARTIAL_CONTENTS 206
104 /* Redirection 3xx. */
105 #define HTTP_STATUS_MULTIPLE_CHOICES 300
106 #define HTTP_STATUS_MOVED_PERMANENTLY 301
107 #define HTTP_STATUS_MOVED_TEMPORARILY 302
108 #define HTTP_STATUS_NOT_MODIFIED 304
109 #define HTTP_STATUS_TEMPORARY_REDIRECT 307
111 /* Client error 4xx. */
112 #define HTTP_STATUS_BAD_REQUEST 400
113 #define HTTP_STATUS_UNAUTHORIZED 401
114 #define HTTP_STATUS_FORBIDDEN 403
115 #define HTTP_STATUS_NOT_FOUND 404
117 /* Server errors 5xx. */
118 #define HTTP_STATUS_INTERNAL 500
119 #define HTTP_STATUS_NOT_IMPLEMENTED 501
120 #define HTTP_STATUS_BAD_GATEWAY 502
121 #define HTTP_STATUS_UNAVAILABLE 503
124 /* Parse the HTTP status line, which is of format:
126 HTTP-Version SP Status-Code SP Reason-Phrase
128 The function returns the status-code, or -1 if the status line is
129 malformed. The pointer to reason-phrase is returned in RP. */
131 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
133 /* (the variables must not be named `major' and `minor', because
134 that breaks compilation with SunOS4 cc.) */
135 int mjr, mnr, statcode;
138 *reason_phrase_ptr = NULL;
140 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
141 major version, and Y is minor version. */
142 if (strncmp (line, "HTTP/", 5) != 0)
146 /* Calculate major HTTP version. */
148 for (mjr = 0; ISDIGIT (*line); line++)
149 mjr = 10 * mjr + (*line - '0');
150 if (*line != '.' || p == line)
154 /* Calculate minor HTTP version. */
156 for (mnr = 0; ISDIGIT (*line); line++)
157 mnr = 10 * mnr + (*line - '0');
158 if (*line != ' ' || p == line)
160 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
161 minor version can be safely ignored. */
166 /* Calculate status code. */
167 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
169 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
171 /* Set up the reason phrase pointer. */
173 /* RFC2068 requires SPC here, but we allow the string to finish
174 here, in case no reason-phrase is present. */
178 *reason_phrase_ptr = line;
183 *reason_phrase_ptr = line + 1;
188 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
190 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
191 PROMISED_SIZE bytes are sent over the wire -- if the file is
192 longer, read only that much; if the file is shorter, report an error. */
195 post_file (int sock, const char *file_name, long promised_size)
197 static char chunk[8192];
202 DEBUGP (("[writing POST file %s ... ", file_name));
204 fp = fopen (file_name, "rb");
207 while (!feof (fp) && written < promised_size)
210 int length = fread (chunk, 1, sizeof (chunk), fp);
213 towrite = WMIN (promised_size - written, length);
214 write_error = xwrite (sock, chunk, towrite, -1);
224 /* If we've written less than was promised, report a (probably
225 nonsensical) error rather than break the promise. */
226 if (written < promised_size)
232 assert (written == promised_size);
233 DEBUGP (("done]\n"));
237 /* Functions to be used as arguments to header_process(): */
239 struct http_process_range_closure {
245 /* Parse the `Content-Range' header and extract the information it
246 contains. Returns 1 if successful, -1 otherwise. */
248 http_process_range (const char *hdr, void *arg)
250 struct http_process_range_closure *closure
251 = (struct http_process_range_closure *)arg;
254 /* Certain versions of Nutscape proxy server send out
255 `Content-Length' without "bytes" specifier, which is a breach of
256 RFC2068 (as well as the HTTP/1.1 draft which was current at the
257 time). But hell, I must support it... */
258 if (!strncasecmp (hdr, "bytes", 5))
261 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
265 hdr += skip_lws (hdr);
271 for (num = 0; ISDIGIT (*hdr); hdr++)
272 num = 10 * num + (*hdr - '0');
273 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
275 closure->first_byte_pos = num;
277 for (num = 0; ISDIGIT (*hdr); hdr++)
278 num = 10 * num + (*hdr - '0');
279 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
281 closure->last_byte_pos = num;
283 for (num = 0; ISDIGIT (*hdr); hdr++)
284 num = 10 * num + (*hdr - '0');
285 closure->entity_length = num;
289 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
290 Used for `Accept-Ranges'. */
292 http_process_none (const char *hdr, void *arg)
294 int *where = (int *)arg;
296 if (strstr (hdr, "none"))
303 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
305 http_process_type (const char *hdr, void *arg)
307 char **result = (char **)arg;
308 /* Locate P on `;' or the terminating zero, whichever comes first. */
309 const char *p = strchr (hdr, ';');
311 p = hdr + strlen (hdr);
312 while (p > hdr && ISSPACE (*(p - 1)))
314 *result = strdupdelim (hdr, p);
318 /* Check whether the `Connection' header is set to "keep-alive". */
320 http_process_connection (const char *hdr, void *arg)
322 int *flag = (int *)arg;
323 if (!strcasecmp (hdr, "Keep-Alive"))
328 /* Commit the cookie to the cookie jar. */
331 http_process_set_cookie (const char *hdr, void *arg)
333 struct url *u = (struct url *)arg;
335 /* The jar should have been created by now. */
336 assert (wget_cookie_jar != NULL);
338 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path, hdr);
343 /* Persistent connections. Currently, we cache the most recently used
344 connection as persistent, provided that the HTTP server agrees to
345 make it such. The persistence data is stored in the variables
346 below. Ideally, it should be possible to cache an arbitrary fixed
347 number of these connections. */
349 /* Whether a persistent connection is active. */
350 static int pconn_active;
353 /* The socket of the connection. */
356 /* Host and port of the currently active persistent connection. */
360 /* Whether a ssl handshake has occoured on this connection. */
364 /* Mark the persistent connection as invalid and free the resources it
365 uses. This is used by the CLOSE_* macros after they forcefully
366 close a registered persistent connection. */
369 invalidate_persistent (void)
371 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
373 xclose (pconn.socket);
378 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
379 persistent. This will enable someone to use the same connection
380 later. In the context of HTTP, this must be called only AFTER the
381 response has been received and the server has promised that the
382 connection will remain alive.
384 If a previous connection was persistent, it is closed. */
387 register_persistent (const char *host, int port, int fd, int ssl)
391 if (pconn.socket == fd)
393 /* The connection FD is already registered. */
398 /* The old persistent connection is still active; close it
399 first. This situation arises whenever a persistent
400 connection exists, but we then connect to a different
401 host, and try to register a persistent connection to that
403 invalidate_persistent ();
409 pconn.host = xstrdup (host);
413 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
416 /* Return non-zero if a persistent connection is available for
417 connecting to HOST:PORT. */
420 persistent_available_p (const char *host, int port, int ssl,
421 int *host_lookup_failed)
423 /* First, check whether a persistent connection is active at all. */
427 /* If we want SSL and the last connection wasn't or vice versa,
428 don't use it. Checking for host and port is not enough because
429 HTTP and HTTPS can apparently coexist on the same port. */
430 if (ssl != pconn.ssl)
433 /* If we're not connecting to the same port, we're not interested. */
434 if (port != pconn.port)
437 /* If the host is the same, we're in business. If not, there is
438 still hope -- read below. */
439 if (0 != strcasecmp (host, pconn.host))
441 /* If pconn.socket is already talking to HOST, we needn't
442 reconnect. This happens often when both sites are virtual
443 hosts distinguished only by name and served by the same
444 network interface, and hence the same web server (possibly
445 set up by the ISP and serving many different web sites).
446 This admittedly non-standard optimization does not contradict
447 HTTP and works well with popular server software. */
451 struct address_list *al;
454 /* Don't try to talk to two different SSL sites over the same
455 secure connection! (Besides, it's not clear if name-based
456 virtual hosting is even possible with SSL.) */
459 /* If pconn.socket's peer is one of the IP addresses HOST
460 resolves to, pconn.socket is for all intents and purposes
461 already talking to HOST. */
463 if (!socket_ip_address (pconn.socket, &ip, 0))
465 /* Can't get the peer's address -- something must be very
466 wrong with the connection. */
467 invalidate_persistent ();
470 al = lookup_host (host, 0);
473 *host_lookup_failed = 1;
477 found = address_list_find (al, &ip);
478 address_list_release (al);
483 /* The persistent connection's peer address was found among the
484 addresses HOST resolved to; therefore, pconn.sock is in fact
485 already talking to HOST -- no need to reconnect. */
488 /* Finally, check whether the connection is still open. This is
489 important because most server implement a liberal (short) timeout
490 on persistent connections. Wget can of course always reconnect
491 if the connection doesn't work out, but it's nicer to know in
492 advance. This test is a logical followup of the first test, but
493 is "expensive" and therefore placed at the end of the list. */
495 if (!test_socket_open (pconn.socket))
497 /* Oops, the socket is no longer open. Now that we know that,
498 let's invalidate the persistent connection before returning
500 invalidate_persistent ();
507 /* The idea behind these two CLOSE macros is to distinguish between
508 two cases: one when the job we've been doing is finished, and we
509 want to close the connection and leave, and two when something is
510 seriously wrong and we're closing the connection as part of
513 In case of keep_alive, CLOSE_FINISH should leave the connection
514 open, while CLOSE_INVALIDATE should still close it.
516 Note that the semantics of the flag `keep_alive' is "this
517 connection *will* be reused (the server has promised not to close
518 the connection once we're done)", while the semantics of
519 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
520 active, registered connection". */
522 #define CLOSE_FINISH(fd) do { \
525 if (pconn_active && (fd) == pconn.socket) \
526 invalidate_persistent (); \
532 #define CLOSE_INVALIDATE(fd) do { \
533 if (pconn_active && (fd) == pconn.socket) \
534 invalidate_persistent (); \
541 long len; /* received length */
542 long contlen; /* expected length */
543 long restval; /* the restart value */
544 int res; /* the result of last read */
545 char *newloc; /* new location (redirection) */
546 char *remote_time; /* remote time-stamp string */
547 char *error; /* textual HTTP error */
548 int statcode; /* status code */
549 double dltime; /* time of the download in msecs */
550 int no_truncate; /* whether truncating the file is
552 const char *referer; /* value of the referer header. */
553 char **local_file; /* local file. */
557 free_hstat (struct http_stat *hs)
559 xfree_null (hs->newloc);
560 xfree_null (hs->remote_time);
561 xfree_null (hs->error);
563 /* Guard against being called twice. */
565 hs->remote_time = NULL;
569 static char *create_authorization_line PARAMS ((const char *, const char *,
570 const char *, const char *,
572 static char *basic_authentication_encode PARAMS ((const char *, const char *,
574 static int known_authentication_scheme_p PARAMS ((const char *));
576 time_t http_atotm PARAMS ((const char *));
578 #define BEGINS_WITH(line, string_constant) \
579 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
580 && (ISSPACE (line[sizeof (string_constant) - 1]) \
581 || !line[sizeof (string_constant) - 1]))
583 /* Retrieve a document through HTTP protocol. It recognizes status
584 code, and correctly handles redirections. It closes the network
585 socket. If it receives an error from the functions below it, it
586 will print it if there is enough information to do so (almost
587 always), returning the error to the caller (i.e. http_loop).
589 Various HTTP parameters are stored to hs. Although it parses the
590 response code correctly, it is not used in a sane way. The caller
593 If PROXY is non-NULL, the connection will be made to the proxy
594 server, and u->url will be requested. */
596 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
598 char *request, *type, *command, *full_path;
600 char *pragma_h, *referer, *useragent, *range, *wwwauth;
601 char *authenticate_h;
605 char *request_keep_alive;
606 int sock, hcount, all_length, statcode;
608 long contlen, contrange;
611 int auth_tried_already;
614 char *cookies = NULL;
616 /* Whether this connection will be kept alive after the HTTP request
620 /* Flags that detect the two ways of specifying HTTP keep-alive
622 int http_keep_alive_1, http_keep_alive_2;
624 /* Whether keep-alive should be inhibited. */
625 int inhibit_keep_alive;
627 /* Whether we need to print the host header with braces around host,
628 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
629 "Host: symbolic-name:1234". */
630 int squares_around_host = 0;
632 /* Headers sent when using POST. */
633 char *post_content_type, *post_content_length;
634 long post_data_size = 0;
636 int host_lookup_failed;
639 /* Initialize the SSL context. After the first run, this is a
643 case SSLERRCTXCREATE:
645 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
646 return SSLERRCTXCREATE;
648 /* try without certfile */
649 logprintf (LOG_NOTQUIET,
650 _("Failed to load certificates from %s\n"),
652 logprintf (LOG_NOTQUIET,
653 _("Trying without the specified certificate\n"));
656 logprintf (LOG_NOTQUIET,
657 _("Failed to get certificate key from %s\n"),
659 logprintf (LOG_NOTQUIET,
660 _("Trying without the specified certificate\n"));
665 #endif /* HAVE_SSL */
667 if (!(*dt & HEAD_ONLY))
668 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
669 know the local filename so we can save to it. */
670 assert (*hs->local_file != NULL);
673 auth_tried_already = 0;
675 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
678 /* We need to come back here when the initial attempt to retrieve
679 without authorization header fails. (Expected to happen at least
680 for the Digest authorization scheme.) */
683 http_keep_alive_1 = http_keep_alive_2 = 0;
685 post_content_type = NULL;
686 post_content_length = NULL;
688 /* Initialize certain elements of struct http_stat. */
693 hs->remote_time = NULL;
696 /* If we're using a proxy, we will be connecting to the proxy
698 conn = proxy ? proxy : u;
700 host_lookup_failed = 0;
702 /* First: establish the connection. */
703 if (inhibit_keep_alive
704 || !persistent_available_p (conn->host, conn->port,
706 u->scheme == SCHEME_HTTPS
710 , &host_lookup_failed))
712 /* In its current implementation, persistent_available_p will
713 look up conn->host in some cases. If that lookup failed, we
714 don't need to bother with connect_to_host. */
715 if (host_lookup_failed)
718 sock = connect_to_host (conn->host, conn->port);
722 return (retryable_socket_connect_error (errno)
723 ? CONERROR : CONIMPOSSIBLE);
726 if (conn->scheme == SCHEME_HTTPS)
728 if (!ssl_connect (sock))
730 logputs (LOG_VERBOSE, "\n");
731 logprintf (LOG_NOTQUIET,
732 _("Unable to establish SSL connection.\n"));
738 #endif /* HAVE_SSL */
742 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
743 conn->host, conn->port);
744 /* #### pc_last_fd should be accessed through an accessor
747 using_ssl = pconn.ssl;
748 DEBUGP (("Reusing fd %d.\n", sock));
753 else if (opt.post_file_name || opt.post_data)
761 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
762 sprintf (referer, "Referer: %s\r\n", hs->referer);
765 if (*dt & SEND_NOCACHE)
766 pragma_h = "Pragma: no-cache\r\n";
772 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
773 /* Gag me! Some servers (e.g. WebSitePro) have been known to
774 respond to the following `Range' format by generating a
775 multipart/x-byte-ranges MIME document! This MIME type was
776 present in an old draft of the byteranges specification.
777 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
778 only if multiple non-overlapping ranges are requested --
779 which Wget never does. */
780 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
785 STRDUP_ALLOCA (useragent, opt.useragent);
788 useragent = (char *)alloca (10 + strlen (version_string));
789 sprintf (useragent, "Wget/%s", version_string);
791 /* Construct the authentication, if userid is present. */
794 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
795 user = user ? user : opt.http_user;
796 passwd = passwd ? passwd : opt.http_passwd;
803 /* We have the username and the password, but haven't tried
804 any authorization yet. Let's see if the "Basic" method
805 works. If not, we'll come back here and construct a
806 proper authorization method with the right challenges.
808 If we didn't employ this kind of logic, every URL that
809 requires authorization would have to be processed twice,
810 which is very suboptimal and generates a bunch of false
811 "unauthorized" errors in the server log.
813 #### But this logic also has a serious problem when used
814 with stronger authentications: we *first* transmit the
815 username and the password in clear text, and *then*
816 attempt a stronger authentication scheme. That cannot be
817 right! We are only fortunate that almost everyone still
818 uses the `Basic' scheme anyway.
820 There should be an option to prevent this from happening,
821 for those who use strong authentication schemes and value
823 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
827 /* Use the full path, i.e. one that includes the leading
828 slash and the query string, but is independent of proxy
830 char *pth = url_full_path (u);
831 wwwauth = create_authorization_line (authenticate_h, user, passwd,
840 char *proxy_user, *proxy_passwd;
841 /* For normal username and password, URL components override
842 command-line/wgetrc parameters. With proxy authentication,
843 it's the reverse, because proxy URLs are normally the
844 "permanent" ones, so command-line args should take
846 if (opt.proxy_user && opt.proxy_passwd)
848 proxy_user = opt.proxy_user;
849 proxy_passwd = opt.proxy_passwd;
853 proxy_user = proxy->user;
854 proxy_passwd = proxy->passwd;
856 /* #### This does not appear right. Can't the proxy request,
857 say, `Digest' authentication? */
858 if (proxy_user && proxy_passwd)
859 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
860 "Proxy-Authorization");
863 /* String of the form :PORT. Used only for non-standard ports. */
865 if (u->port != scheme_default_port (u->scheme))
867 port_maybe = (char *)alloca (numdigit (u->port) + 2);
868 sprintf (port_maybe, ":%d", u->port);
871 if (!inhibit_keep_alive)
872 request_keep_alive = "Connection: Keep-Alive\r\n";
874 request_keep_alive = NULL;
877 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
879 u->scheme == SCHEME_HTTPS
885 if (opt.post_data || opt.post_file_name)
887 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
889 post_data_size = strlen (opt.post_data);
892 post_data_size = file_size (opt.post_file_name);
893 if (post_data_size == -1)
895 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
900 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
901 sprintf (post_content_length,
902 "Content-Length: %ld\r\n", post_data_size);
906 full_path = xstrdup (u->url);
908 /* Use the full path, i.e. one that includes the leading slash and
909 the query string. E.g. if u->path is "foo/bar" and u->query is
910 "param=value", full_path will be "/foo/bar?param=value". */
911 full_path = url_full_path (u);
913 if (strchr (u->host, ':'))
914 squares_around_host = 1;
916 /* Allocate the memory for the request. */
917 request = (char *)alloca (strlen (command)
921 + (port_maybe ? strlen (port_maybe) : 0)
922 + strlen (HTTP_ACCEPT)
923 + (request_keep_alive
924 ? strlen (request_keep_alive) : 0)
925 + (referer ? strlen (referer) : 0)
926 + (cookies ? strlen (cookies) : 0)
927 + (wwwauth ? strlen (wwwauth) : 0)
928 + (proxyauth ? strlen (proxyauth) : 0)
929 + (range ? strlen (range) : 0)
932 ? strlen (post_content_type) : 0)
933 + (post_content_length
934 ? strlen (post_content_length) : 0)
935 + (opt.user_header ? strlen (opt.user_header) : 0)
937 /* Construct the request. */
943 %s%s%s%s%s%s%s%s%s%s\r\n",
946 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
947 port_maybe ? port_maybe : "",
949 request_keep_alive ? request_keep_alive : "",
950 referer ? referer : "",
951 cookies ? cookies : "",
952 wwwauth ? wwwauth : "",
953 proxyauth ? proxyauth : "",
956 post_content_type ? post_content_type : "",
957 post_content_length ? post_content_length : "",
958 opt.user_header ? opt.user_header : "");
959 DEBUGP (("\n---request begin---\n%s", request));
961 /* Free the temporary memory. */
962 xfree_null (wwwauth);
963 xfree_null (proxyauth);
964 xfree_null (cookies);
967 /* Send the request to server. */
968 write_error = xwrite (sock, request, strlen (request), -1);
970 if (write_error >= 0)
974 DEBUGP (("[POST data: %s]\n", opt.post_data));
975 write_error = xwrite (sock, opt.post_data, post_data_size, -1);
977 else if (opt.post_file_name && post_data_size != 0)
978 write_error = post_file (sock, opt.post_file_name, post_data_size);
980 DEBUGP (("---request end---\n"));
984 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
986 CLOSE_INVALIDATE (sock);
989 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
990 proxy ? "Proxy" : "HTTP");
991 contlen = contrange = -1;
996 /* Before reading anything, initialize the rbuf. */
997 rbuf_initialize (&rbuf, sock);
1001 DEBUGP (("\n---response begin---\n"));
1003 /* Header-fetching loop. */
1011 /* Get the header. */
1012 status = header_get (&rbuf, &hdr,
1013 /* Disallow continuations for status line. */
1014 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
1016 /* Check for errors. */
1017 if (status == HG_EOF && *hdr)
1019 /* This used to be an unconditional error, but that was
1020 somewhat controversial, because of a large number of
1021 broken CGI's that happily "forget" to send the second EOL
1022 before closing the connection of a HEAD request.
1024 So, the deal is to check whether the header is empty
1025 (*hdr is zero if it is); if yes, it means that the
1026 previous header was fully retrieved, and that -- most
1027 probably -- the request is complete. "...be liberal in
1028 what you accept." Oh boy. */
1029 logputs (LOG_VERBOSE, "\n");
1030 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1033 xfree_null (all_headers);
1034 CLOSE_INVALIDATE (sock);
1037 else if (status == HG_ERROR)
1039 logputs (LOG_VERBOSE, "\n");
1040 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1044 xfree_null (all_headers);
1045 CLOSE_INVALIDATE (sock);
1049 /* If the headers are to be saved to a file later, save them to
1051 if (opt.save_headers)
1053 int lh = strlen (hdr);
1054 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1055 memcpy (all_headers + all_length, hdr, lh);
1057 all_headers[all_length++] = '\n';
1058 all_headers[all_length] = '\0';
1061 /* Check for status line. */
1065 /* Parse the first line of server response. */
1066 statcode = parse_http_status_line (hdr, &error);
1067 hs->statcode = statcode;
1068 /* Store the descriptive response. */
1069 if (statcode == -1) /* malformed response */
1071 /* A common reason for "malformed response" error is the
1072 case when no data was actually received. Handle this
1075 hs->error = xstrdup (_("No data received"));
1077 hs->error = xstrdup (_("Malformed status line"));
1082 hs->error = xstrdup (_("(no description)"));
1084 hs->error = xstrdup (error);
1086 if ((statcode != -1)
1092 if (opt.server_response)
1093 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1095 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1101 /* Exit on empty header. */
1108 /* Print the header if requested. */
1109 if (opt.server_response && hcount != 1)
1110 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1112 /* Try getting content-length. */
1113 if (contlen == -1 && !opt.ignore_length)
1114 if (header_process (hdr, "Content-Length", header_extract_number,
1117 /* Try getting content-type. */
1119 if (header_process (hdr, "Content-Type", http_process_type, &type))
1121 /* Try getting location. */
1123 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1125 /* Try getting last-modified. */
1126 if (!hs->remote_time)
1127 if (header_process (hdr, "Last-Modified", header_strdup,
1130 /* Try getting cookies. */
1132 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1134 /* Try getting www-authentication. */
1135 if (!authenticate_h)
1136 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1139 /* Check for accept-ranges header. If it contains the word
1140 `none', disable the ranges. */
1141 if (*dt & ACCEPTRANGES)
1144 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1147 *dt &= ~ACCEPTRANGES;
1151 /* Try getting content-range. */
1152 if (contrange == -1)
1154 struct http_process_range_closure closure;
1155 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1157 contrange = closure.first_byte_pos;
1161 /* Check for keep-alive related responses. */
1162 if (!inhibit_keep_alive)
1164 /* Check for the `Keep-Alive' header. */
1165 if (!http_keep_alive_1)
1167 if (header_process (hdr, "Keep-Alive", header_exists,
1168 &http_keep_alive_1))
1171 /* Check for `Connection: Keep-Alive'. */
1172 if (!http_keep_alive_2)
1174 if (header_process (hdr, "Connection", http_process_connection,
1175 &http_keep_alive_2))
1182 DEBUGP (("---response end---\n"));
1184 logputs (LOG_VERBOSE, "\n");
1187 && (http_keep_alive_1 || http_keep_alive_2))
1189 assert (inhibit_keep_alive == 0);
1193 /* The server has promised that it will not close the connection
1194 when we're done. This means that we can register it. */
1195 register_persistent (conn->host, conn->port, sock, using_ssl);
1197 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1200 /* Authorization is required. */
1204 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1205 might be more bytes in the body. */
1206 if (auth_tried_already)
1208 /* If we have tried it already, then there is not point
1211 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1212 xfree (authenticate_h);
1215 else if (!known_authentication_scheme_p (authenticate_h))
1217 xfree (authenticate_h);
1218 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1221 else if (BEGINS_WITH (authenticate_h, "Basic"))
1223 /* The authentication scheme is basic, the one we try by
1224 default, and it failed. There's no sense in trying
1230 auth_tried_already = 1;
1234 /* We do not need this anymore. */
1237 xfree (authenticate_h);
1238 authenticate_h = NULL;
1241 /* 20x responses are counted among successful by default. */
1242 if (H_20X (statcode))
1245 /* Return if redirected. */
1246 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1248 /* RFC2068 says that in case of the 300 (multiple choices)
1249 response, the server can output a preferred URL through
1250 `Location' header; otherwise, the request should be treated
1251 like GET. So, if the location is set, it will be a
1252 redirection; otherwise, just proceed normally. */
1253 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1257 logprintf (LOG_VERBOSE,
1258 _("Location: %s%s\n"),
1259 hs->newloc ? hs->newloc : _("unspecified"),
1260 hs->newloc ? _(" [following]") : "");
1261 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1262 might be more bytes in the body. */
1264 xfree_null (all_headers);
1269 /* If content-type is not given, assume text/html. This is because
1270 of the multitude of broken CGI's that "forget" to generate the
1273 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1274 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1279 if (opt.html_extension && (*dt & TEXTHTML))
1280 /* -E / --html-extension / html_extension = on was specified, and this is a
1281 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1282 already the file's suffix, tack on ".html". */
1284 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1286 if (last_period_in_local_filename == NULL
1287 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1288 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1290 size_t local_filename_len = strlen(*hs->local_file);
1292 *hs->local_file = xrealloc(*hs->local_file,
1293 local_filename_len + sizeof(".html"));
1294 strcpy(*hs->local_file + local_filename_len, ".html");
1296 *dt |= ADDED_HTML_EXTENSION;
1300 if (contrange == -1)
1302 /* We did not get a content-range header. This means that the
1303 server did not honor our `Range' request. Normally, this
1304 means we should reset hs->restval and continue normally. */
1306 /* However, if `-c' is used, we need to be a bit more careful:
1308 1. If `-c' is specified and the file already existed when
1309 Wget was started, it would be a bad idea for us to start
1310 downloading it from scratch, effectively truncating it. I
1311 believe this cannot happen unless `-c' was specified.
1313 2. If `-c' is used on a file that is already fully
1314 downloaded, we're requesting bytes after the end of file,
1315 which can result in server not honoring `Range'. If this is
1316 the case, `Content-Length' will be equal to the length of the
1318 if (opt.always_rest)
1320 /* Check for condition #2. */
1321 if (hs->restval > 0 /* restart was requested. */
1322 && contlen != -1 /* we got content-length. */
1323 && hs->restval >= contlen /* file fully downloaded
1327 logputs (LOG_VERBOSE, _("\
1328 \n The file is already fully retrieved; nothing to do.\n\n"));
1329 /* In case the caller inspects. */
1332 /* Mark as successfully retrieved. */
1335 xfree_null (all_headers);
1336 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1337 might be more bytes in the body. */
1338 return RETRUNNEEDED;
1341 /* Check for condition #1. */
1342 if (hs->no_truncate)
1344 logprintf (LOG_NOTQUIET,
1347 Continued download failed on this file, which conflicts with `-c'.\n\
1348 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1350 xfree_null (all_headers);
1351 CLOSE_INVALIDATE (sock);
1352 return CONTNOTSUPPORTED;
1360 else if (contrange != hs->restval ||
1361 (H_PARTIAL (statcode) && contrange == -1))
1363 /* This means the whole request was somehow misunderstood by the
1364 server. Bail out. */
1366 xfree_null (all_headers);
1367 CLOSE_INVALIDATE (sock);
1374 contlen += contrange;
1376 contrange = -1; /* If conent-length was not sent,
1377 content-range will be ignored. */
1379 hs->contlen = contlen;
1383 if ((*dt & RETROKF) && !opt.server_response)
1385 /* No need to print this output if the body won't be
1386 downloaded at all, or if the original server response is
1388 logputs (LOG_VERBOSE, _("Length: "));
1391 logputs (LOG_VERBOSE, legible (contlen));
1392 if (contrange != -1)
1393 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1394 legible (contlen - contrange));
1397 logputs (LOG_VERBOSE,
1398 opt.ignore_length ? _("ignored") : _("unspecified"));
1400 logprintf (LOG_VERBOSE, " [%s]\n", type);
1402 logputs (LOG_VERBOSE, "\n");
1406 type = NULL; /* We don't need it any more. */
1408 /* Return if we have no intention of further downloading. */
1409 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1411 /* In case the caller cares to look... */
1415 xfree_null (all_headers);
1416 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1417 might be more bytes in the body. */
1418 return RETRFINISHED;
1421 /* Open the local file. */
1424 mkalldirs (*hs->local_file);
1426 rotate_backups (*hs->local_file);
1427 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1430 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1431 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1432 might be more bytes in the body. */
1433 xfree_null (all_headers);
1439 extern int global_download_count;
1441 /* To ensure that repeated "from scratch" downloads work for -O
1442 files, we rewind the file pointer, unless restval is
1443 non-zero. (This works only when -O is used on regular files,
1444 but it's still a valuable feature.)
1446 However, this loses when more than one URL is specified on
1447 the command line the second rewinds eradicates the contents
1448 of the first download. Thus we disable the above trick for
1449 all the downloads except the very first one.
1451 #### A possible solution to this would be to remember the
1452 file position in the output document and to seek to that
1453 position, instead of rewinding.
1455 We don't truncate stdout, since that breaks
1456 "wget -O - [...] >> foo".
1458 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1460 /* This will silently fail for streams that don't correspond
1461 to regular files, but that's OK. */
1463 /* ftruncate is needed because opt.dfp is opened in append
1464 mode if opt.always_rest is set. */
1465 ftruncate (fileno (fp), 0);
1470 /* #### This confuses the code that checks for file size. There
1471 should be some overhead information. */
1472 if (opt.save_headers)
1473 fwrite (all_headers, 1, all_length, fp);
1475 /* Get the contents of the document. */
1476 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1477 (contlen != -1 ? contlen : 0),
1478 &rbuf, keep_alive, &hs->dltime);
1481 CLOSE_FINISH (sock);
1483 CLOSE_INVALIDATE (sock);
1486 /* Close or flush the file. We have to be careful to check for
1487 error here. Checking the result of fwrite() is not enough --
1488 errors could go unnoticed! */
1491 flush_res = fclose (fp);
1493 flush_res = fflush (fp);
1494 if (flush_res == EOF)
1497 xfree_null (all_headers);
1500 return RETRFINISHED;
1503 /* The genuine HTTP loop! This is the part where the retrieval is
1504 retried, and retried, and retried, and... */
1506 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1507 int *dt, struct url *proxy)
1510 int use_ts, got_head = 0; /* time-stamping info */
1511 char *filename_plus_orig_suffix;
1512 char *local_filename = NULL;
1513 char *tms, *locf, *tmrate;
1515 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1516 long local_size = 0; /* the size of the local file */
1517 size_t filename_len;
1518 struct http_stat hstat; /* HTTP status */
1522 /* This used to be done in main(), but it's a better idea to do it
1523 here so that we don't go through the hoops if we're just using
1527 if (!wget_cookie_jar)
1528 wget_cookie_jar = cookie_jar_new ();
1529 if (opt.cookies_input && !cookies_loaded_p)
1531 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1532 cookies_loaded_p = 1;
1538 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1539 has_wildcards_p because it would also warn on `?', and we know that
1540 shows up in CGI paths a *lot*. */
1541 if (strchr (u->url, '*'))
1542 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1544 /* Determine the local filename. */
1545 if (local_file && *local_file)
1546 hstat.local_file = local_file;
1547 else if (local_file)
1549 *local_file = url_file_name (u);
1550 hstat.local_file = local_file;
1554 dummy = url_file_name (u);
1555 hstat.local_file = &dummy;
1558 if (!opt.output_document)
1559 locf = *hstat.local_file;
1561 locf = opt.output_document;
1563 hstat.referer = referer;
1565 filename_len = strlen (*hstat.local_file);
1566 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1568 if (opt.noclobber && file_exists_p (*hstat.local_file))
1570 /* If opt.noclobber is turned on and file already exists, do not
1571 retrieve the file */
1572 logprintf (LOG_VERBOSE, _("\
1573 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1574 /* If the file is there, we suppose it's retrieved OK. */
1577 /* #### Bogusness alert. */
1578 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1579 if (has_html_suffix_p (*hstat.local_file))
1587 if (opt.timestamping)
1589 int local_dot_orig_file_exists = 0;
1591 if (opt.backup_converted)
1592 /* If -K is specified, we'll act on the assumption that it was specified
1593 last time these files were downloaded as well, and instead of just
1594 comparing local file X against server file X, we'll compare local
1595 file X.orig (if extant, else X) against server file X. If -K
1596 _wasn't_ specified last time, or the server contains files called
1597 *.orig, -N will be back to not operating correctly with -k. */
1599 /* Would a single s[n]printf() call be faster? --dan
1601 Definitely not. sprintf() is horribly slow. It's a
1602 different question whether the difference between the two
1603 affects a program. Usually I'd say "no", but at one
1604 point I profiled Wget, and found that a measurable and
1605 non-negligible amount of time was lost calling sprintf()
1606 in url.c. Replacing sprintf with inline calls to
1607 strcpy() and long_to_string() made a difference.
1609 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1610 memcpy (filename_plus_orig_suffix + filename_len,
1611 ".orig", sizeof (".orig"));
1613 /* Try to stat() the .orig file. */
1614 if (stat (filename_plus_orig_suffix, &st) == 0)
1616 local_dot_orig_file_exists = 1;
1617 local_filename = filename_plus_orig_suffix;
1621 if (!local_dot_orig_file_exists)
1622 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1623 if (stat (*hstat.local_file, &st) == 0)
1624 local_filename = *hstat.local_file;
1626 if (local_filename != NULL)
1627 /* There was a local file, so we'll check later to see if the version
1628 the server has is the same version we already have, allowing us to
1634 /* Modification time granularity is 2 seconds for Windows, so
1635 increase local time by 1 second for later comparison. */
1638 local_size = st.st_size;
1642 /* Reset the counter. */
1644 *dt = 0 | ACCEPTRANGES;
1648 /* Increment the pass counter. */
1650 sleep_between_retrievals (count);
1651 /* Get the current time string. */
1652 tms = time_str (NULL);
1653 /* Print fetch message, if opt.verbose. */
1656 char *hurl = url_string (u, 1);
1660 sprintf (tmp, _("(try:%2d)"), count);
1661 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1662 tms, hurl, tmp, locf);
1664 ws_changetitle (hurl, 1);
1669 /* Default document type is empty. However, if spider mode is
1670 on or time-stamping is employed, HEAD_ONLY commands is
1671 encoded within *dt. */
1672 if (opt.spider || (use_ts && !got_head))
1676 /* Assume no restarting. */
1678 /* Decide whether or not to restart. */
1679 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1680 /* #### this calls access() and then stat(); could be optimized. */
1681 && file_exists_p (locf))
1682 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1683 hstat.restval = st.st_size;
1685 /* In `-c' is used and the file is existing and non-empty,
1686 refuse to truncate it if the server doesn't support continued
1688 hstat.no_truncate = 0;
1689 if (opt.always_rest && hstat.restval)
1690 hstat.no_truncate = 1;
1692 /* Decide whether to send the no-cache directive. We send it in
1694 a) we're using a proxy, and we're past our first retrieval.
1695 Some proxies are notorious for caching incomplete data, so
1696 we require a fresh get.
1697 b) caching is explicitly inhibited. */
1698 if ((proxy && count > 1) /* a */
1699 || !opt.allow_cache /* b */
1701 *dt |= SEND_NOCACHE;
1703 *dt &= ~SEND_NOCACHE;
1705 /* Try fetching the document, or at least its head. */
1706 err = gethttp (u, &hstat, dt, proxy);
1708 /* It's unfortunate that wget determines the local filename before finding
1709 out the Content-Type of the file. Barring a major restructuring of the
1710 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1711 *hstat.local_file to tack on ".html". */
1712 if (!opt.output_document)
1713 locf = *hstat.local_file;
1715 locf = opt.output_document;
1718 tms = time_str (NULL);
1719 /* Get the new location (with or without the redirection). */
1721 *newloc = xstrdup (hstat.newloc);
1724 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1725 case CONERROR: case READERR: case WRITEFAILED:
1727 /* Non-fatal errors continue executing the loop, which will
1728 bring them to "while" statement at the end, to judge
1729 whether the number of tries was exceeded. */
1730 free_hstat (&hstat);
1731 printwhat (count, opt.ntry);
1734 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1735 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1736 /* Fatal errors just return from the function. */
1737 free_hstat (&hstat);
1741 case FWRITEERR: case FOPENERR:
1742 /* Another fatal error. */
1743 logputs (LOG_VERBOSE, "\n");
1744 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1745 *hstat.local_file, strerror (errno));
1746 free_hstat (&hstat);
1751 /* Another fatal error. */
1752 logputs (LOG_VERBOSE, "\n");
1753 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1754 free_hstat (&hstat);
1759 /* Return the new location to the caller. */
1762 logprintf (LOG_NOTQUIET,
1763 _("ERROR: Redirection (%d) without location.\n"),
1765 free_hstat (&hstat);
1769 free_hstat (&hstat);
1774 /* The file was already fully retrieved. */
1775 free_hstat (&hstat);
1780 /* Deal with you later. */
1783 /* All possibilities should have been exhausted. */
1786 if (!(*dt & RETROKF))
1790 /* #### Ugly ugly ugly! */
1791 char *hurl = url_string (u, 1);
1792 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1795 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1796 tms, hstat.statcode, hstat.error);
1797 logputs (LOG_VERBOSE, "\n");
1798 free_hstat (&hstat);
1803 /* Did we get the time-stamp? */
1806 if (opt.timestamping && !hstat.remote_time)
1808 logputs (LOG_NOTQUIET, _("\
1809 Last-modified header missing -- time-stamps turned off.\n"));
1811 else if (hstat.remote_time)
1813 /* Convert the date-string into struct tm. */
1814 tmr = http_atotm (hstat.remote_time);
1815 if (tmr == (time_t) (-1))
1816 logputs (LOG_VERBOSE, _("\
1817 Last-modified header invalid -- time-stamp ignored.\n"));
1821 /* The time-stamping section. */
1826 use_ts = 0; /* no more time-stamping */
1827 count = 0; /* the retrieve count for HEAD is
1829 if (hstat.remote_time && tmr != (time_t) (-1))
1831 /* Now time-stamping can be used validly. Time-stamping
1832 means that if the sizes of the local and remote file
1833 match, and local file is newer than the remote file,
1834 it will not be retrieved. Otherwise, the normal
1835 download procedure is resumed. */
1837 (hstat.contlen == -1 || local_size == hstat.contlen))
1839 logprintf (LOG_VERBOSE, _("\
1840 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1842 free_hstat (&hstat);
1846 else if (tml >= tmr)
1847 logprintf (LOG_VERBOSE, _("\
1848 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1850 logputs (LOG_VERBOSE,
1851 _("Remote file is newer, retrieving.\n"));
1853 free_hstat (&hstat);
1856 if ((tmr != (time_t) (-1))
1858 && ((hstat.len == hstat.contlen) ||
1859 ((hstat.res == 0) &&
1860 ((hstat.contlen == -1) ||
1861 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1863 /* #### This code repeats in http.c and ftp.c. Move it to a
1865 const char *fl = NULL;
1866 if (opt.output_document)
1868 if (opt.od_known_regular)
1869 fl = opt.output_document;
1872 fl = *hstat.local_file;
1876 /* End of time-stamping section. */
1880 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1885 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1887 if (hstat.len == hstat.contlen)
1891 logprintf (LOG_VERBOSE,
1892 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1893 tms, tmrate, locf, hstat.len, hstat.contlen);
1894 logprintf (LOG_NONVERBOSE,
1895 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1896 tms, u->url, hstat.len, hstat.contlen, locf, count);
1899 total_downloaded_bytes += hstat.len;
1901 /* Remember that we downloaded the file for later ".orig" code. */
1902 if (*dt & ADDED_HTML_EXTENSION)
1903 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1905 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1907 free_hstat (&hstat);
1911 else if (hstat.res == 0) /* No read error */
1913 if (hstat.contlen == -1) /* We don't know how much we were supposed
1914 to get, so assume we succeeded. */
1918 logprintf (LOG_VERBOSE,
1919 _("%s (%s) - `%s' saved [%ld]\n\n"),
1920 tms, tmrate, locf, hstat.len);
1921 logprintf (LOG_NONVERBOSE,
1922 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1923 tms, u->url, hstat.len, locf, count);
1926 total_downloaded_bytes += hstat.len;
1928 /* Remember that we downloaded the file for later ".orig" code. */
1929 if (*dt & ADDED_HTML_EXTENSION)
1930 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1932 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1934 free_hstat (&hstat);
1938 else if (hstat.len < hstat.contlen) /* meaning we lost the
1939 connection too soon */
1941 logprintf (LOG_VERBOSE,
1942 _("%s (%s) - Connection closed at byte %ld. "),
1943 tms, tmrate, hstat.len);
1944 printwhat (count, opt.ntry);
1945 free_hstat (&hstat);
1948 else if (!opt.kill_longer) /* meaning we got more than expected */
1950 logprintf (LOG_VERBOSE,
1951 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1952 tms, tmrate, locf, hstat.len, hstat.contlen);
1953 logprintf (LOG_NONVERBOSE,
1954 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1955 tms, u->url, hstat.len, hstat.contlen, locf, count);
1957 total_downloaded_bytes += hstat.len;
1959 /* Remember that we downloaded the file for later ".orig" code. */
1960 if (*dt & ADDED_HTML_EXTENSION)
1961 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1963 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1965 free_hstat (&hstat);
1969 else /* the same, but not accepted */
1971 logprintf (LOG_VERBOSE,
1972 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1973 tms, tmrate, hstat.len, hstat.contlen);
1974 printwhat (count, opt.ntry);
1975 free_hstat (&hstat);
1979 else /* now hstat.res can only be -1 */
1981 if (hstat.contlen == -1)
1983 logprintf (LOG_VERBOSE,
1984 _("%s (%s) - Read error at byte %ld (%s)."),
1985 tms, tmrate, hstat.len, strerror (errno));
1986 printwhat (count, opt.ntry);
1987 free_hstat (&hstat);
1990 else /* hstat.res == -1 and contlen is given */
1992 logprintf (LOG_VERBOSE,
1993 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1994 tms, tmrate, hstat.len, hstat.contlen,
1996 printwhat (count, opt.ntry);
1997 free_hstat (&hstat);
2004 while (!opt.ntry || (count < opt.ntry));
2008 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2009 than local timezone.
2011 mktime is similar but assumes struct tm, also known as the
2012 "broken-down" form of time, is in local time zone. mktime_from_utc
2013 uses mktime to make the conversion understanding that an offset
2014 will be introduced by the local time assumption.
2016 mktime_from_utc then measures the introduced offset by applying
2017 gmtime to the initial result and applying mktime to the resulting
2018 "broken-down" form. The difference between the two mktime results
2019 is the measured offset which is then subtracted from the initial
2020 mktime result to yield a calendar time which is the value returned.
2022 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2023 consistent offset (the non DST offset) since tm and tm+o might be
2024 on opposite sides of a DST change.
2026 Some implementations of mktime return -1 for the nonexistent
2027 localtime hour at the beginning of DST. In this event, use
2028 mktime(tm - 1hr) + 3600.
2032 gmtime(t+o) --> tm+o
2033 mktime(tm+o) --> t+2o
2034 t+o - (t+2o - t+o) = t
2036 Note that glibc contains a function of the same purpose named
2037 `timegm' (reverse of gmtime). But obviously, it is not universally
2038 available, and unfortunately it is not straightforwardly
2039 extractable for use here. Perhaps configure should detect timegm
2040 and use it where available.
2042 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2043 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2044 Further improved by Roger with assistance from Edward J. Sabol
2045 based on input by Jamie Zawinski. */
2048 mktime_from_utc (struct tm *t)
2059 return -1; /* can't deal with output from strptime */
2070 return -1; /* can't deal with output from gmtime */
2073 return (tl - (tb - tl));
2076 /* Check whether the result of strptime() indicates success.
2077 strptime() returns the pointer to how far it got to in the string.
2078 The processing has been successful if the string is at `GMT' or
2079 `+X', or at the end of the string.
2081 In extended regexp parlance, the function returns 1 if P matches
2082 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2083 can return) is considered a failure and 0 is returned. */
2085 check_end (const char *p)
2089 while (ISSPACE (*p))
2092 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2093 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2099 /* Convert the textual specification of time in TIME_STRING to the
2100 number of seconds since the Epoch.
2102 TIME_STRING can be in any of the three formats RFC2068 allows the
2103 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2104 Timezones are ignored, and should be GMT.
2106 Return the computed time_t representation, or -1 if the conversion
2109 This function uses strptime with various string formats for parsing
2110 TIME_STRING. This results in a parser that is not as lenient in
2111 interpreting TIME_STRING as I would like it to be. Being based on
2112 strptime, it always allows shortened months, one-digit days, etc.,
2113 but due to the multitude of formats in which time can be
2114 represented, an ideal HTTP time parser would be even more
2115 forgiving. It should completely ignore things like week days and
2116 concentrate only on the various forms of representing years,
2117 months, days, hours, minutes, and seconds. For example, it would
2118 be nice if it accepted ISO 8601 out of the box.
2120 I've investigated free and PD code for this purpose, but none was
2121 usable. getdate was big and unwieldy, and had potential copyright
2122 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2123 distributed with phttpd, is excellent, but we cannot use it because
2124 it is not assigned to the FSF. So I stuck it with strptime. */
2127 http_atotm (const char *time_string)
2129 /* NOTE: Solaris strptime man page claims that %n and %t match white
2130 space, but that's not universally available. Instead, we simply
2131 use ` ' to mean "skip all WS", which works under all strptime
2132 implementations I've tested. */
2134 static const char *time_formats[] = {
2135 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2136 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2137 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2138 (google.com uses this for their cookies.) */
2139 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2145 /* According to Roger Beeman, we need to initialize tm_isdst, since
2146 strptime won't do it. */
2149 /* Note that under foreign locales Solaris strptime() fails to
2150 recognize English dates, which renders this function useless. We
2151 solve this by being careful not to affect LC_TIME when
2152 initializing locale.
2154 Another solution would be to temporarily set locale to C, invoke
2155 strptime(), and restore it back. This is slow and dirty,
2156 however, and locale support other than LC_MESSAGES can mess other
2157 things, so I rather chose to stick with just setting LC_MESSAGES.
2159 GNU strptime does not have this problem because it recognizes
2160 both international and local dates. */
2162 for (i = 0; i < countof (time_formats); i++)
2163 if (check_end (strptime (time_string, time_formats[i], &t)))
2164 return mktime_from_utc (&t);
2166 /* All formats have failed. */
2170 /* Authorization support: We support two authorization schemes:
2172 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2174 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2175 consisting of answering to the server's challenge with the proper
2178 /* How many bytes it will take to store LEN bytes in base64. */
2179 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2181 /* Encode the string S of length LENGTH to base64 format and place it
2182 to STORE. STORE will be 0-terminated, and must point to a writable
2183 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2185 base64_encode (const char *s, char *store, int length)
2187 /* Conversion table. */
2188 static char tbl[64] = {
2189 'A','B','C','D','E','F','G','H',
2190 'I','J','K','L','M','N','O','P',
2191 'Q','R','S','T','U','V','W','X',
2192 'Y','Z','a','b','c','d','e','f',
2193 'g','h','i','j','k','l','m','n',
2194 'o','p','q','r','s','t','u','v',
2195 'w','x','y','z','0','1','2','3',
2196 '4','5','6','7','8','9','+','/'
2199 unsigned char *p = (unsigned char *)store;
2201 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2202 for (i = 0; i < length; i += 3)
2204 *p++ = tbl[s[0] >> 2];
2205 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2206 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2207 *p++ = tbl[s[2] & 0x3f];
2210 /* Pad the result if necessary... */
2211 if (i == length + 1)
2213 else if (i == length + 2)
2214 *(p - 1) = *(p - 2) = '=';
2215 /* ...and zero-terminate it. */
2219 /* Create the authentication header contents for the `Basic' scheme.
2220 This is done by encoding the string `USER:PASS' in base64 and
2221 prepending `HEADER: Basic ' to it. */
2223 basic_authentication_encode (const char *user, const char *passwd,
2226 char *t1, *t2, *res;
2227 int len1 = strlen (user) + 1 + strlen (passwd);
2228 int len2 = BASE64_LENGTH (len1);
2230 t1 = (char *)alloca (len1 + 1);
2231 sprintf (t1, "%s:%s", user, passwd);
2232 t2 = (char *)alloca (1 + len2);
2233 base64_encode (t1, t2, len1);
2234 res = (char *)xmalloc (len2 + 11 + strlen (header));
2235 sprintf (res, "%s: Basic %s\r\n", header, t2);
2241 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2242 of a field in such a header. If the field is the one specified by
2243 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2244 digest authorization code), extract its value in the (char*)
2245 variable pointed by RET. Returns negative on a malformed header,
2246 or number of bytes that have been parsed by this call. */
2248 extract_header_attr (const char *au, const char *attr_name, char **ret)
2250 const char *cp, *ep;
2254 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2256 cp += strlen (attr_name);
2259 cp += skip_lws (cp);
2264 cp += skip_lws (cp);
2269 for (ep = cp; *ep && *ep != '\"'; ep++)
2274 *ret = strdupdelim (cp, ep);
2281 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2282 an array of 16 bytes containing the hash keys, and BUF should be a
2283 buffer of 33 writable characters (32 for hex digits plus one for
2284 zero termination). */
2286 dump_hash (unsigned char *buf, const unsigned char *hash)
2290 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2292 *buf++ = XNUM_TO_digit (*hash >> 4);
2293 *buf++ = XNUM_TO_digit (*hash & 0xf);
2298 /* Take the line apart to find the challenge, and compose a digest
2299 authorization header. See RFC2069 section 2.1.2. */
2301 digest_authentication_encode (const char *au, const char *user,
2302 const char *passwd, const char *method,
2305 static char *realm, *opaque, *nonce;
2310 { "realm", &realm },
2311 { "opaque", &opaque },
2316 realm = opaque = nonce = NULL;
2318 au += 6; /* skip over `Digest' */
2323 au += skip_lws (au);
2324 for (i = 0; i < countof (options); i++)
2326 int skip = extract_header_attr (au, options[i].name,
2327 options[i].variable);
2331 xfree_null (opaque);
2341 if (i == countof (options))
2343 while (*au && *au != '=')
2347 au += skip_lws (au);
2351 while (*au && *au != '\"')
2358 while (*au && *au != ',')
2363 if (!realm || !nonce || !user || !passwd || !path || !method)
2366 xfree_null (opaque);
2371 /* Calculate the digest value. */
2373 ALLOCA_MD5_CONTEXT (ctx);
2374 unsigned char hash[MD5_HASHLEN];
2375 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2376 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2378 /* A1BUF = H(user ":" realm ":" password) */
2380 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2381 gen_md5_update ((unsigned char *)":", 1, ctx);
2382 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2383 gen_md5_update ((unsigned char *)":", 1, ctx);
2384 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2385 gen_md5_finish (ctx, hash);
2386 dump_hash (a1buf, hash);
2388 /* A2BUF = H(method ":" path) */
2390 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2391 gen_md5_update ((unsigned char *)":", 1, ctx);
2392 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2393 gen_md5_finish (ctx, hash);
2394 dump_hash (a2buf, hash);
2396 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2398 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2399 gen_md5_update ((unsigned char *)":", 1, ctx);
2400 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2401 gen_md5_update ((unsigned char *)":", 1, ctx);
2402 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2403 gen_md5_finish (ctx, hash);
2404 dump_hash (response_digest, hash);
2406 res = (char*) xmalloc (strlen (user)
2411 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2412 + (opaque ? strlen (opaque) : 0)
2414 sprintf (res, "Authorization: Digest \
2415 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2416 user, realm, nonce, path, response_digest);
2419 char *p = res + strlen (res);
2420 strcat (p, ", opaque=\"");
2424 strcat (res, "\r\n");
2428 #endif /* USE_DIGEST */
2431 #define BEGINS_WITH(line, string_constant) \
2432 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2433 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2434 || !line[sizeof (string_constant) - 1]))
2437 known_authentication_scheme_p (const char *au)
2439 return BEGINS_WITH (au, "Basic")
2440 || BEGINS_WITH (au, "Digest")
2441 || BEGINS_WITH (au, "NTLM");
2446 /* Create the HTTP authorization request header. When the
2447 `WWW-Authenticate' response header is seen, according to the
2448 authorization scheme specified in that header (`Basic' and `Digest'
2449 are supported by the current implementation), produce an
2450 appropriate HTTP authorization request header. */
2452 create_authorization_line (const char *au, const char *user,
2453 const char *passwd, const char *method,
2456 char *wwwauth = NULL;
2458 if (!strncasecmp (au, "Basic", 5))
2459 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2460 if (!strncasecmp (au, "NTLM", 4))
2461 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2463 else if (!strncasecmp (au, "Digest", 6))
2464 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2465 #endif /* USE_DIGEST */