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 /* This is somewhat evil, but works in practice: if the address
442 that pconn.socket is connected to is one of the IP addresses
443 HOST resolves to, we don't need to reconnect. #### Is it
444 correct to do this by default? */
447 struct address_list *al;
449 if (!socket_ip_address (pconn.socket, &ip, 0))
451 /* Can't get the peer's address -- something must be wrong
452 with the connection. */
453 invalidate_persistent ();
456 al = lookup_host (host, 0);
459 *host_lookup_failed = 1;
463 found = address_list_find (al, &ip);
464 address_list_release (al);
469 /* HOST resolves to an address pconn.sock is connected to -- no
470 need to reconnect. */
473 /* Finally, check whether the connection is still open. This is
474 important because most server implement a liberal (short) timeout
475 on persistent connections. Wget can of course always reconnect
476 if the connection doesn't work out, but it's nicer to know in
477 advance. This test is a logical followup of the first test, but
478 is "expensive" and therefore placed at the end of the list. */
480 if (!test_socket_open (pconn.socket))
482 /* Oops, the socket is no longer open. Now that we know that,
483 let's invalidate the persistent connection before returning
485 invalidate_persistent ();
492 /* The idea behind these two CLOSE macros is to distinguish between
493 two cases: one when the job we've been doing is finished, and we
494 want to close the connection and leave, and two when something is
495 seriously wrong and we're closing the connection as part of
498 In case of keep_alive, CLOSE_FINISH should leave the connection
499 open, while CLOSE_INVALIDATE should still close it.
501 Note that the semantics of the flag `keep_alive' is "this
502 connection *will* be reused (the server has promised not to close
503 the connection once we're done)", while the semantics of
504 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
505 active, registered connection". */
507 #define CLOSE_FINISH(fd) do { \
510 if (pconn_active && (fd) == pconn.socket) \
511 invalidate_persistent (); \
517 #define CLOSE_INVALIDATE(fd) do { \
518 if (pconn_active && (fd) == pconn.socket) \
519 invalidate_persistent (); \
526 long len; /* received length */
527 long contlen; /* expected length */
528 long restval; /* the restart value */
529 int res; /* the result of last read */
530 char *newloc; /* new location (redirection) */
531 char *remote_time; /* remote time-stamp string */
532 char *error; /* textual HTTP error */
533 int statcode; /* status code */
534 double dltime; /* time of the download in msecs */
535 int no_truncate; /* whether truncating the file is
537 const char *referer; /* value of the referer header. */
538 char **local_file; /* local file. */
542 free_hstat (struct http_stat *hs)
544 xfree_null (hs->newloc);
545 xfree_null (hs->remote_time);
546 xfree_null (hs->error);
548 /* Guard against being called twice. */
550 hs->remote_time = NULL;
554 static char *create_authorization_line PARAMS ((const char *, const char *,
555 const char *, const char *,
557 static char *basic_authentication_encode PARAMS ((const char *, const char *,
559 static int known_authentication_scheme_p PARAMS ((const char *));
561 time_t http_atotm PARAMS ((const char *));
563 #define BEGINS_WITH(line, string_constant) \
564 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
565 && (ISSPACE (line[sizeof (string_constant) - 1]) \
566 || !line[sizeof (string_constant) - 1]))
568 /* Retrieve a document through HTTP protocol. It recognizes status
569 code, and correctly handles redirections. It closes the network
570 socket. If it receives an error from the functions below it, it
571 will print it if there is enough information to do so (almost
572 always), returning the error to the caller (i.e. http_loop).
574 Various HTTP parameters are stored to hs. Although it parses the
575 response code correctly, it is not used in a sane way. The caller
578 If PROXY is non-NULL, the connection will be made to the proxy
579 server, and u->url will be requested. */
581 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
583 char *request, *type, *command, *full_path;
585 char *pragma_h, *referer, *useragent, *range, *wwwauth;
586 char *authenticate_h;
590 char *request_keep_alive;
591 int sock, hcount, all_length, statcode;
593 long contlen, contrange;
596 int auth_tried_already;
599 char *cookies = NULL;
601 /* Whether this connection will be kept alive after the HTTP request
605 /* Flags that detect the two ways of specifying HTTP keep-alive
607 int http_keep_alive_1, http_keep_alive_2;
609 /* Whether keep-alive should be inhibited. */
610 int inhibit_keep_alive;
612 /* Whether we need to print the host header with braces around host,
613 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
614 "Host: symbolic-name:1234". */
615 int squares_around_host = 0;
617 /* Headers sent when using POST. */
618 char *post_content_type, *post_content_length;
619 long post_data_size = 0;
621 int host_lookup_failed;
624 /* Initialize the SSL context. After the first run, this is a
628 case SSLERRCTXCREATE:
630 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
631 return SSLERRCTXCREATE;
633 /* try without certfile */
634 logprintf (LOG_NOTQUIET,
635 _("Failed to load certificates from %s\n"),
637 logprintf (LOG_NOTQUIET,
638 _("Trying without the specified certificate\n"));
641 logprintf (LOG_NOTQUIET,
642 _("Failed to get certificate key from %s\n"),
644 logprintf (LOG_NOTQUIET,
645 _("Trying without the specified certificate\n"));
650 #endif /* HAVE_SSL */
652 if (!(*dt & HEAD_ONLY))
653 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
654 know the local filename so we can save to it. */
655 assert (*hs->local_file != NULL);
658 auth_tried_already = 0;
660 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
663 /* We need to come back here when the initial attempt to retrieve
664 without authorization header fails. (Expected to happen at least
665 for the Digest authorization scheme.) */
668 http_keep_alive_1 = http_keep_alive_2 = 0;
670 post_content_type = NULL;
671 post_content_length = NULL;
673 /* Initialize certain elements of struct http_stat. */
678 hs->remote_time = NULL;
681 /* If we're using a proxy, we will be connecting to the proxy
683 conn = proxy ? proxy : u;
685 host_lookup_failed = 0;
687 /* First: establish the connection. */
688 if (inhibit_keep_alive
689 || !persistent_available_p (conn->host, conn->port,
691 u->scheme == SCHEME_HTTPS
695 , &host_lookup_failed))
697 /* In its current implementation, persistent_available_p will
698 look up conn->host in some cases. If that lookup failed, we
699 don't need to bother with connect_to_host. */
700 if (host_lookup_failed)
703 sock = connect_to_host (conn->host, conn->port);
707 return CONNECT_ERROR (errno);
710 if (conn->scheme == SCHEME_HTTPS)
712 if (!ssl_connect (sock))
714 logputs (LOG_VERBOSE, "\n");
715 logprintf (LOG_NOTQUIET,
716 _("Unable to establish SSL connection.\n"));
722 #endif /* HAVE_SSL */
726 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
727 conn->host, conn->port);
728 /* #### pc_last_fd should be accessed through an accessor
731 using_ssl = pconn.ssl;
732 DEBUGP (("Reusing fd %d.\n", sock));
737 else if (opt.post_file_name || opt.post_data)
745 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
746 sprintf (referer, "Referer: %s\r\n", hs->referer);
749 if (*dt & SEND_NOCACHE)
750 pragma_h = "Pragma: no-cache\r\n";
756 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
757 /* Gag me! Some servers (e.g. WebSitePro) have been known to
758 respond to the following `Range' format by generating a
759 multipart/x-byte-ranges MIME document! This MIME type was
760 present in an old draft of the byteranges specification.
761 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
762 only if multiple non-overlapping ranges are requested --
763 which Wget never does. */
764 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
769 STRDUP_ALLOCA (useragent, opt.useragent);
772 useragent = (char *)alloca (10 + strlen (version_string));
773 sprintf (useragent, "Wget/%s", version_string);
775 /* Construct the authentication, if userid is present. */
778 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
779 user = user ? user : opt.http_user;
780 passwd = passwd ? passwd : opt.http_passwd;
787 /* We have the username and the password, but haven't tried
788 any authorization yet. Let's see if the "Basic" method
789 works. If not, we'll come back here and construct a
790 proper authorization method with the right challenges.
792 If we didn't employ this kind of logic, every URL that
793 requires authorization would have to be processed twice,
794 which is very suboptimal and generates a bunch of false
795 "unauthorized" errors in the server log.
797 #### But this logic also has a serious problem when used
798 with stronger authentications: we *first* transmit the
799 username and the password in clear text, and *then*
800 attempt a stronger authentication scheme. That cannot be
801 right! We are only fortunate that almost everyone still
802 uses the `Basic' scheme anyway.
804 There should be an option to prevent this from happening,
805 for those who use strong authentication schemes and value
807 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
811 /* Use the full path, i.e. one that includes the leading
812 slash and the query string, but is independent of proxy
814 char *pth = url_full_path (u);
815 wwwauth = create_authorization_line (authenticate_h, user, passwd,
824 char *proxy_user, *proxy_passwd;
825 /* For normal username and password, URL components override
826 command-line/wgetrc parameters. With proxy authentication,
827 it's the reverse, because proxy URLs are normally the
828 "permanent" ones, so command-line args should take
830 if (opt.proxy_user && opt.proxy_passwd)
832 proxy_user = opt.proxy_user;
833 proxy_passwd = opt.proxy_passwd;
837 proxy_user = proxy->user;
838 proxy_passwd = proxy->passwd;
840 /* #### This does not appear right. Can't the proxy request,
841 say, `Digest' authentication? */
842 if (proxy_user && proxy_passwd)
843 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
844 "Proxy-Authorization");
847 /* String of the form :PORT. Used only for non-standard ports. */
849 if (u->port != scheme_default_port (u->scheme))
851 port_maybe = (char *)alloca (numdigit (u->port) + 2);
852 sprintf (port_maybe, ":%d", u->port);
855 if (!inhibit_keep_alive)
856 request_keep_alive = "Connection: Keep-Alive\r\n";
858 request_keep_alive = NULL;
861 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
863 u->scheme == SCHEME_HTTPS
869 if (opt.post_data || opt.post_file_name)
871 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
873 post_data_size = strlen (opt.post_data);
876 post_data_size = file_size (opt.post_file_name);
877 if (post_data_size == -1)
879 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
884 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
885 sprintf (post_content_length,
886 "Content-Length: %ld\r\n", post_data_size);
890 full_path = xstrdup (u->url);
892 /* Use the full path, i.e. one that includes the leading slash and
893 the query string. E.g. if u->path is "foo/bar" and u->query is
894 "param=value", full_path will be "/foo/bar?param=value". */
895 full_path = url_full_path (u);
897 if (strchr (u->host, ':'))
898 squares_around_host = 1;
900 /* Allocate the memory for the request. */
901 request = (char *)alloca (strlen (command)
905 + (port_maybe ? strlen (port_maybe) : 0)
906 + strlen (HTTP_ACCEPT)
907 + (request_keep_alive
908 ? strlen (request_keep_alive) : 0)
909 + (referer ? strlen (referer) : 0)
910 + (cookies ? strlen (cookies) : 0)
911 + (wwwauth ? strlen (wwwauth) : 0)
912 + (proxyauth ? strlen (proxyauth) : 0)
913 + (range ? strlen (range) : 0)
916 ? strlen (post_content_type) : 0)
917 + (post_content_length
918 ? strlen (post_content_length) : 0)
919 + (opt.user_header ? strlen (opt.user_header) : 0)
921 /* Construct the request. */
927 %s%s%s%s%s%s%s%s%s%s\r\n",
930 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
931 port_maybe ? port_maybe : "",
933 request_keep_alive ? request_keep_alive : "",
934 referer ? referer : "",
935 cookies ? cookies : "",
936 wwwauth ? wwwauth : "",
937 proxyauth ? proxyauth : "",
940 post_content_type ? post_content_type : "",
941 post_content_length ? post_content_length : "",
942 opt.user_header ? opt.user_header : "");
943 DEBUGP (("---request begin---\n%s", request));
945 /* Free the temporary memory. */
946 xfree_null (wwwauth);
947 xfree_null (proxyauth);
948 xfree_null (cookies);
951 /* Send the request to server. */
952 write_error = xwrite (sock, request, strlen (request), -1);
954 if (write_error >= 0)
958 DEBUGP (("[POST data: %s]\n", opt.post_data));
959 write_error = xwrite (sock, opt.post_data, post_data_size, -1);
961 else if (opt.post_file_name && post_data_size != 0)
962 write_error = post_file (sock, opt.post_file_name, post_data_size);
964 DEBUGP (("---request end---\n"));
968 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
970 CLOSE_INVALIDATE (sock);
973 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
974 proxy ? "Proxy" : "HTTP");
975 contlen = contrange = -1;
980 /* Before reading anything, initialize the rbuf. */
981 rbuf_initialize (&rbuf, sock);
984 /* Header-fetching loop. */
992 /* Get the header. */
993 status = header_get (&rbuf, &hdr,
994 /* Disallow continuations for status line. */
995 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
997 /* Check for errors. */
998 if (status == HG_EOF && *hdr)
1000 /* This used to be an unconditional error, but that was
1001 somewhat controversial, because of a large number of
1002 broken CGI's that happily "forget" to send the second EOL
1003 before closing the connection of a HEAD request.
1005 So, the deal is to check whether the header is empty
1006 (*hdr is zero if it is); if yes, it means that the
1007 previous header was fully retrieved, and that -- most
1008 probably -- the request is complete. "...be liberal in
1009 what you accept." Oh boy. */
1010 logputs (LOG_VERBOSE, "\n");
1011 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1014 xfree_null (all_headers);
1015 CLOSE_INVALIDATE (sock);
1018 else if (status == HG_ERROR)
1020 logputs (LOG_VERBOSE, "\n");
1021 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1025 xfree_null (all_headers);
1026 CLOSE_INVALIDATE (sock);
1030 /* If the headers are to be saved to a file later, save them to
1032 if (opt.save_headers)
1034 int lh = strlen (hdr);
1035 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1036 memcpy (all_headers + all_length, hdr, lh);
1038 all_headers[all_length++] = '\n';
1039 all_headers[all_length] = '\0';
1042 /* Check for status line. */
1046 /* Parse the first line of server response. */
1047 statcode = parse_http_status_line (hdr, &error);
1048 hs->statcode = statcode;
1049 /* Store the descriptive response. */
1050 if (statcode == -1) /* malformed response */
1052 /* A common reason for "malformed response" error is the
1053 case when no data was actually received. Handle this
1056 hs->error = xstrdup (_("No data received"));
1058 hs->error = xstrdup (_("Malformed status line"));
1063 hs->error = xstrdup (_("(no description)"));
1065 hs->error = xstrdup (error);
1067 if ((statcode != -1)
1073 if (opt.server_response)
1074 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1076 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1082 /* Exit on empty header. */
1089 /* Print the header if requested. */
1090 if (opt.server_response && hcount != 1)
1091 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1093 /* Try getting content-length. */
1094 if (contlen == -1 && !opt.ignore_length)
1095 if (header_process (hdr, "Content-Length", header_extract_number,
1098 /* Try getting content-type. */
1100 if (header_process (hdr, "Content-Type", http_process_type, &type))
1102 /* Try getting location. */
1104 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1106 /* Try getting last-modified. */
1107 if (!hs->remote_time)
1108 if (header_process (hdr, "Last-Modified", header_strdup,
1111 /* Try getting cookies. */
1113 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1115 /* Try getting www-authentication. */
1116 if (!authenticate_h)
1117 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1120 /* Check for accept-ranges header. If it contains the word
1121 `none', disable the ranges. */
1122 if (*dt & ACCEPTRANGES)
1125 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1128 *dt &= ~ACCEPTRANGES;
1132 /* Try getting content-range. */
1133 if (contrange == -1)
1135 struct http_process_range_closure closure;
1136 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1138 contrange = closure.first_byte_pos;
1142 /* Check for keep-alive related responses. */
1143 if (!inhibit_keep_alive)
1145 /* Check for the `Keep-Alive' header. */
1146 if (!http_keep_alive_1)
1148 if (header_process (hdr, "Keep-Alive", header_exists,
1149 &http_keep_alive_1))
1152 /* Check for `Connection: Keep-Alive'. */
1153 if (!http_keep_alive_2)
1155 if (header_process (hdr, "Connection", http_process_connection,
1156 &http_keep_alive_2))
1164 logputs (LOG_VERBOSE, "\n");
1167 && (http_keep_alive_1 || http_keep_alive_2))
1169 assert (inhibit_keep_alive == 0);
1173 /* The server has promised that it will not close the connection
1174 when we're done. This means that we can register it. */
1175 register_persistent (conn->host, conn->port, sock, using_ssl);
1177 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1180 /* Authorization is required. */
1184 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1185 might be more bytes in the body. */
1186 if (auth_tried_already)
1188 /* If we have tried it already, then there is not point
1191 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1192 xfree (authenticate_h);
1195 else if (!known_authentication_scheme_p (authenticate_h))
1197 xfree (authenticate_h);
1198 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1201 else if (BEGINS_WITH (authenticate_h, "Basic"))
1203 /* The authentication scheme is basic, the one we try by
1204 default, and it failed. There's no sense in trying
1210 auth_tried_already = 1;
1214 /* We do not need this anymore. */
1217 xfree (authenticate_h);
1218 authenticate_h = NULL;
1221 /* 20x responses are counted among successful by default. */
1222 if (H_20X (statcode))
1225 /* Return if redirected. */
1226 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1228 /* RFC2068 says that in case of the 300 (multiple choices)
1229 response, the server can output a preferred URL through
1230 `Location' header; otherwise, the request should be treated
1231 like GET. So, if the location is set, it will be a
1232 redirection; otherwise, just proceed normally. */
1233 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1237 logprintf (LOG_VERBOSE,
1238 _("Location: %s%s\n"),
1239 hs->newloc ? hs->newloc : _("unspecified"),
1240 hs->newloc ? _(" [following]") : "");
1241 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1242 might be more bytes in the body. */
1244 xfree_null (all_headers);
1249 /* If content-type is not given, assume text/html. This is because
1250 of the multitude of broken CGI's that "forget" to generate the
1253 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1254 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1259 if (opt.html_extension && (*dt & TEXTHTML))
1260 /* -E / --html-extension / html_extension = on was specified, and this is a
1261 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1262 already the file's suffix, tack on ".html". */
1264 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1266 if (last_period_in_local_filename == NULL
1267 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1268 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1270 size_t local_filename_len = strlen(*hs->local_file);
1272 *hs->local_file = xrealloc(*hs->local_file,
1273 local_filename_len + sizeof(".html"));
1274 strcpy(*hs->local_file + local_filename_len, ".html");
1276 *dt |= ADDED_HTML_EXTENSION;
1280 if (contrange == -1)
1282 /* We did not get a content-range header. This means that the
1283 server did not honor our `Range' request. Normally, this
1284 means we should reset hs->restval and continue normally. */
1286 /* However, if `-c' is used, we need to be a bit more careful:
1288 1. If `-c' is specified and the file already existed when
1289 Wget was started, it would be a bad idea for us to start
1290 downloading it from scratch, effectively truncating it. I
1291 believe this cannot happen unless `-c' was specified.
1293 2. If `-c' is used on a file that is already fully
1294 downloaded, we're requesting bytes after the end of file,
1295 which can result in server not honoring `Range'. If this is
1296 the case, `Content-Length' will be equal to the length of the
1298 if (opt.always_rest)
1300 /* Check for condition #2. */
1301 if (hs->restval > 0 /* restart was requested. */
1302 && contlen != -1 /* we got content-length. */
1303 && hs->restval >= contlen /* file fully downloaded
1307 logputs (LOG_VERBOSE, _("\
1308 \n The file is already fully retrieved; nothing to do.\n\n"));
1309 /* In case the caller inspects. */
1312 /* Mark as successfully retrieved. */
1315 xfree_null (all_headers);
1316 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1317 might be more bytes in the body. */
1318 return RETRUNNEEDED;
1321 /* Check for condition #1. */
1322 if (hs->no_truncate)
1324 logprintf (LOG_NOTQUIET,
1327 Continued download failed on this file, which conflicts with `-c'.\n\
1328 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1330 xfree_null (all_headers);
1331 CLOSE_INVALIDATE (sock);
1332 return CONTNOTSUPPORTED;
1340 else if (contrange != hs->restval ||
1341 (H_PARTIAL (statcode) && contrange == -1))
1343 /* This means the whole request was somehow misunderstood by the
1344 server. Bail out. */
1346 xfree_null (all_headers);
1347 CLOSE_INVALIDATE (sock);
1354 contlen += contrange;
1356 contrange = -1; /* If conent-length was not sent,
1357 content-range will be ignored. */
1359 hs->contlen = contlen;
1363 if ((*dt & RETROKF) && !opt.server_response)
1365 /* No need to print this output if the body won't be
1366 downloaded at all, or if the original server response is
1368 logputs (LOG_VERBOSE, _("Length: "));
1371 logputs (LOG_VERBOSE, legible (contlen));
1372 if (contrange != -1)
1373 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1374 legible (contlen - contrange));
1377 logputs (LOG_VERBOSE,
1378 opt.ignore_length ? _("ignored") : _("unspecified"));
1380 logprintf (LOG_VERBOSE, " [%s]\n", type);
1382 logputs (LOG_VERBOSE, "\n");
1386 type = NULL; /* We don't need it any more. */
1388 /* Return if we have no intention of further downloading. */
1389 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1391 /* In case the caller cares to look... */
1395 xfree_null (all_headers);
1396 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1397 might be more bytes in the body. */
1398 return RETRFINISHED;
1401 /* Open the local file. */
1404 mkalldirs (*hs->local_file);
1406 rotate_backups (*hs->local_file);
1407 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1410 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1411 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1412 might be more bytes in the body. */
1413 xfree_null (all_headers);
1419 extern int global_download_count;
1421 /* To ensure that repeated "from scratch" downloads work for -O
1422 files, we rewind the file pointer, unless restval is
1423 non-zero. (This works only when -O is used on regular files,
1424 but it's still a valuable feature.)
1426 However, this loses when more than one URL is specified on
1427 the command line the second rewinds eradicates the contents
1428 of the first download. Thus we disable the above trick for
1429 all the downloads except the very first one.
1431 #### A possible solution to this would be to remember the
1432 file position in the output document and to seek to that
1433 position, instead of rewinding.
1435 We don't truncate stdout, since that breaks
1436 "wget -O - [...] >> foo".
1438 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1440 /* This will silently fail for streams that don't correspond
1441 to regular files, but that's OK. */
1443 /* ftruncate is needed because opt.dfp is opened in append
1444 mode if opt.always_rest is set. */
1445 ftruncate (fileno (fp), 0);
1450 /* #### This confuses the code that checks for file size. There
1451 should be some overhead information. */
1452 if (opt.save_headers)
1453 fwrite (all_headers, 1, all_length, fp);
1455 /* Get the contents of the document. */
1456 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1457 (contlen != -1 ? contlen : 0),
1458 &rbuf, keep_alive, &hs->dltime);
1461 CLOSE_FINISH (sock);
1463 CLOSE_INVALIDATE (sock);
1466 /* Close or flush the file. We have to be careful to check for
1467 error here. Checking the result of fwrite() is not enough --
1468 errors could go unnoticed! */
1471 flush_res = fclose (fp);
1473 flush_res = fflush (fp);
1474 if (flush_res == EOF)
1477 xfree_null (all_headers);
1480 return RETRFINISHED;
1483 /* The genuine HTTP loop! This is the part where the retrieval is
1484 retried, and retried, and retried, and... */
1486 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1487 int *dt, struct url *proxy)
1490 int use_ts, got_head = 0; /* time-stamping info */
1491 char *filename_plus_orig_suffix;
1492 char *local_filename = NULL;
1493 char *tms, *locf, *tmrate;
1495 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1496 long local_size = 0; /* the size of the local file */
1497 size_t filename_len;
1498 struct http_stat hstat; /* HTTP status */
1502 /* This used to be done in main(), but it's a better idea to do it
1503 here so that we don't go through the hoops if we're just using
1507 if (!wget_cookie_jar)
1508 wget_cookie_jar = cookie_jar_new ();
1509 if (opt.cookies_input && !cookies_loaded_p)
1511 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1512 cookies_loaded_p = 1;
1518 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1519 has_wildcards_p because it would also warn on `?', and we know that
1520 shows up in CGI paths a *lot*. */
1521 if (strchr (u->url, '*'))
1522 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1524 /* Determine the local filename. */
1525 if (local_file && *local_file)
1526 hstat.local_file = local_file;
1527 else if (local_file)
1529 *local_file = url_file_name (u);
1530 hstat.local_file = local_file;
1534 dummy = url_file_name (u);
1535 hstat.local_file = &dummy;
1538 if (!opt.output_document)
1539 locf = *hstat.local_file;
1541 locf = opt.output_document;
1543 hstat.referer = referer;
1545 filename_len = strlen (*hstat.local_file);
1546 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1548 if (opt.noclobber && file_exists_p (*hstat.local_file))
1550 /* If opt.noclobber is turned on and file already exists, do not
1551 retrieve the file */
1552 logprintf (LOG_VERBOSE, _("\
1553 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1554 /* If the file is there, we suppose it's retrieved OK. */
1557 /* #### Bogusness alert. */
1558 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1559 if (has_html_suffix_p (*hstat.local_file))
1567 if (opt.timestamping)
1569 int local_dot_orig_file_exists = 0;
1571 if (opt.backup_converted)
1572 /* If -K is specified, we'll act on the assumption that it was specified
1573 last time these files were downloaded as well, and instead of just
1574 comparing local file X against server file X, we'll compare local
1575 file X.orig (if extant, else X) against server file X. If -K
1576 _wasn't_ specified last time, or the server contains files called
1577 *.orig, -N will be back to not operating correctly with -k. */
1579 /* Would a single s[n]printf() call be faster? --dan
1581 Definitely not. sprintf() is horribly slow. It's a
1582 different question whether the difference between the two
1583 affects a program. Usually I'd say "no", but at one
1584 point I profiled Wget, and found that a measurable and
1585 non-negligible amount of time was lost calling sprintf()
1586 in url.c. Replacing sprintf with inline calls to
1587 strcpy() and long_to_string() made a difference.
1589 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1590 memcpy (filename_plus_orig_suffix + filename_len,
1591 ".orig", sizeof (".orig"));
1593 /* Try to stat() the .orig file. */
1594 if (stat (filename_plus_orig_suffix, &st) == 0)
1596 local_dot_orig_file_exists = 1;
1597 local_filename = filename_plus_orig_suffix;
1601 if (!local_dot_orig_file_exists)
1602 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1603 if (stat (*hstat.local_file, &st) == 0)
1604 local_filename = *hstat.local_file;
1606 if (local_filename != NULL)
1607 /* There was a local file, so we'll check later to see if the version
1608 the server has is the same version we already have, allowing us to
1614 /* Modification time granularity is 2 seconds for Windows, so
1615 increase local time by 1 second for later comparison. */
1618 local_size = st.st_size;
1622 /* Reset the counter. */
1624 *dt = 0 | ACCEPTRANGES;
1628 /* Increment the pass counter. */
1630 sleep_between_retrievals (count);
1631 /* Get the current time string. */
1632 tms = time_str (NULL);
1633 /* Print fetch message, if opt.verbose. */
1636 char *hurl = url_string (u, 1);
1640 sprintf (tmp, _("(try:%2d)"), count);
1641 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1642 tms, hurl, tmp, locf);
1644 ws_changetitle (hurl, 1);
1649 /* Default document type is empty. However, if spider mode is
1650 on or time-stamping is employed, HEAD_ONLY commands is
1651 encoded within *dt. */
1652 if (opt.spider || (use_ts && !got_head))
1656 /* Assume no restarting. */
1658 /* Decide whether or not to restart. */
1659 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1660 /* #### this calls access() and then stat(); could be optimized. */
1661 && file_exists_p (locf))
1662 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1663 hstat.restval = st.st_size;
1665 /* In `-c' is used and the file is existing and non-empty,
1666 refuse to truncate it if the server doesn't support continued
1668 hstat.no_truncate = 0;
1669 if (opt.always_rest && hstat.restval)
1670 hstat.no_truncate = 1;
1672 /* Decide whether to send the no-cache directive. We send it in
1674 a) we're using a proxy, and we're past our first retrieval.
1675 Some proxies are notorious for caching incomplete data, so
1676 we require a fresh get.
1677 b) caching is explicitly inhibited. */
1678 if ((proxy && count > 1) /* a */
1679 || !opt.allow_cache /* b */
1681 *dt |= SEND_NOCACHE;
1683 *dt &= ~SEND_NOCACHE;
1685 /* Try fetching the document, or at least its head. */
1686 err = gethttp (u, &hstat, dt, proxy);
1688 /* It's unfortunate that wget determines the local filename before finding
1689 out the Content-Type of the file. Barring a major restructuring of the
1690 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1691 *hstat.local_file to tack on ".html". */
1692 if (!opt.output_document)
1693 locf = *hstat.local_file;
1695 locf = opt.output_document;
1698 tms = time_str (NULL);
1699 /* Get the new location (with or without the redirection). */
1701 *newloc = xstrdup (hstat.newloc);
1704 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1705 case CONERROR: case READERR: case WRITEFAILED:
1707 /* Non-fatal errors continue executing the loop, which will
1708 bring them to "while" statement at the end, to judge
1709 whether the number of tries was exceeded. */
1710 free_hstat (&hstat);
1711 printwhat (count, opt.ntry);
1714 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1715 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1716 /* Fatal errors just return from the function. */
1717 free_hstat (&hstat);
1721 case FWRITEERR: case FOPENERR:
1722 /* Another fatal error. */
1723 logputs (LOG_VERBOSE, "\n");
1724 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1725 *hstat.local_file, strerror (errno));
1726 free_hstat (&hstat);
1731 /* Another fatal error. */
1732 logputs (LOG_VERBOSE, "\n");
1733 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1734 free_hstat (&hstat);
1739 /* Return the new location to the caller. */
1742 logprintf (LOG_NOTQUIET,
1743 _("ERROR: Redirection (%d) without location.\n"),
1745 free_hstat (&hstat);
1749 free_hstat (&hstat);
1754 /* The file was already fully retrieved. */
1755 free_hstat (&hstat);
1760 /* Deal with you later. */
1763 /* All possibilities should have been exhausted. */
1766 if (!(*dt & RETROKF))
1770 /* #### Ugly ugly ugly! */
1771 char *hurl = url_string (u, 1);
1772 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1775 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1776 tms, hstat.statcode, hstat.error);
1777 logputs (LOG_VERBOSE, "\n");
1778 free_hstat (&hstat);
1783 /* Did we get the time-stamp? */
1786 if (opt.timestamping && !hstat.remote_time)
1788 logputs (LOG_NOTQUIET, _("\
1789 Last-modified header missing -- time-stamps turned off.\n"));
1791 else if (hstat.remote_time)
1793 /* Convert the date-string into struct tm. */
1794 tmr = http_atotm (hstat.remote_time);
1795 if (tmr == (time_t) (-1))
1796 logputs (LOG_VERBOSE, _("\
1797 Last-modified header invalid -- time-stamp ignored.\n"));
1801 /* The time-stamping section. */
1806 use_ts = 0; /* no more time-stamping */
1807 count = 0; /* the retrieve count for HEAD is
1809 if (hstat.remote_time && tmr != (time_t) (-1))
1811 /* Now time-stamping can be used validly. Time-stamping
1812 means that if the sizes of the local and remote file
1813 match, and local file is newer than the remote file,
1814 it will not be retrieved. Otherwise, the normal
1815 download procedure is resumed. */
1817 (hstat.contlen == -1 || local_size == hstat.contlen))
1819 logprintf (LOG_VERBOSE, _("\
1820 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1822 free_hstat (&hstat);
1826 else if (tml >= tmr)
1827 logprintf (LOG_VERBOSE, _("\
1828 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1830 logputs (LOG_VERBOSE,
1831 _("Remote file is newer, retrieving.\n"));
1833 free_hstat (&hstat);
1836 if ((tmr != (time_t) (-1))
1838 && ((hstat.len == hstat.contlen) ||
1839 ((hstat.res == 0) &&
1840 ((hstat.contlen == -1) ||
1841 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1843 /* #### This code repeats in http.c and ftp.c. Move it to a
1845 const char *fl = NULL;
1846 if (opt.output_document)
1848 if (opt.od_known_regular)
1849 fl = opt.output_document;
1852 fl = *hstat.local_file;
1856 /* End of time-stamping section. */
1860 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1865 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1867 if (hstat.len == hstat.contlen)
1871 logprintf (LOG_VERBOSE,
1872 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1873 tms, tmrate, locf, hstat.len, hstat.contlen);
1874 logprintf (LOG_NONVERBOSE,
1875 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1876 tms, u->url, hstat.len, hstat.contlen, locf, count);
1879 total_downloaded_bytes += hstat.len;
1881 /* Remember that we downloaded the file for later ".orig" code. */
1882 if (*dt & ADDED_HTML_EXTENSION)
1883 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1885 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1887 free_hstat (&hstat);
1891 else if (hstat.res == 0) /* No read error */
1893 if (hstat.contlen == -1) /* We don't know how much we were supposed
1894 to get, so assume we succeeded. */
1898 logprintf (LOG_VERBOSE,
1899 _("%s (%s) - `%s' saved [%ld]\n\n"),
1900 tms, tmrate, locf, hstat.len);
1901 logprintf (LOG_NONVERBOSE,
1902 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1903 tms, u->url, hstat.len, locf, count);
1906 total_downloaded_bytes += hstat.len;
1908 /* Remember that we downloaded the file for later ".orig" code. */
1909 if (*dt & ADDED_HTML_EXTENSION)
1910 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1912 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1914 free_hstat (&hstat);
1918 else if (hstat.len < hstat.contlen) /* meaning we lost the
1919 connection too soon */
1921 logprintf (LOG_VERBOSE,
1922 _("%s (%s) - Connection closed at byte %ld. "),
1923 tms, tmrate, hstat.len);
1924 printwhat (count, opt.ntry);
1925 free_hstat (&hstat);
1928 else if (!opt.kill_longer) /* meaning we got more than expected */
1930 logprintf (LOG_VERBOSE,
1931 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1932 tms, tmrate, locf, hstat.len, hstat.contlen);
1933 logprintf (LOG_NONVERBOSE,
1934 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1935 tms, u->url, hstat.len, hstat.contlen, locf, count);
1937 total_downloaded_bytes += hstat.len;
1939 /* Remember that we downloaded the file for later ".orig" code. */
1940 if (*dt & ADDED_HTML_EXTENSION)
1941 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1943 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1945 free_hstat (&hstat);
1949 else /* the same, but not accepted */
1951 logprintf (LOG_VERBOSE,
1952 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1953 tms, tmrate, hstat.len, hstat.contlen);
1954 printwhat (count, opt.ntry);
1955 free_hstat (&hstat);
1959 else /* now hstat.res can only be -1 */
1961 if (hstat.contlen == -1)
1963 logprintf (LOG_VERBOSE,
1964 _("%s (%s) - Read error at byte %ld (%s)."),
1965 tms, tmrate, hstat.len, strerror (errno));
1966 printwhat (count, opt.ntry);
1967 free_hstat (&hstat);
1970 else /* hstat.res == -1 and contlen is given */
1972 logprintf (LOG_VERBOSE,
1973 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1974 tms, tmrate, hstat.len, hstat.contlen,
1976 printwhat (count, opt.ntry);
1977 free_hstat (&hstat);
1984 while (!opt.ntry || (count < opt.ntry));
1988 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1989 than local timezone.
1991 mktime is similar but assumes struct tm, also known as the
1992 "broken-down" form of time, is in local time zone. mktime_from_utc
1993 uses mktime to make the conversion understanding that an offset
1994 will be introduced by the local time assumption.
1996 mktime_from_utc then measures the introduced offset by applying
1997 gmtime to the initial result and applying mktime to the resulting
1998 "broken-down" form. The difference between the two mktime results
1999 is the measured offset which is then subtracted from the initial
2000 mktime result to yield a calendar time which is the value returned.
2002 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2003 consistent offset (the non DST offset) since tm and tm+o might be
2004 on opposite sides of a DST change.
2006 Some implementations of mktime return -1 for the nonexistent
2007 localtime hour at the beginning of DST. In this event, use
2008 mktime(tm - 1hr) + 3600.
2012 gmtime(t+o) --> tm+o
2013 mktime(tm+o) --> t+2o
2014 t+o - (t+2o - t+o) = t
2016 Note that glibc contains a function of the same purpose named
2017 `timegm' (reverse of gmtime). But obviously, it is not universally
2018 available, and unfortunately it is not straightforwardly
2019 extractable for use here. Perhaps configure should detect timegm
2020 and use it where available.
2022 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2023 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2024 Further improved by Roger with assistance from Edward J. Sabol
2025 based on input by Jamie Zawinski. */
2028 mktime_from_utc (struct tm *t)
2039 return -1; /* can't deal with output from strptime */
2050 return -1; /* can't deal with output from gmtime */
2053 return (tl - (tb - tl));
2056 /* Check whether the result of strptime() indicates success.
2057 strptime() returns the pointer to how far it got to in the string.
2058 The processing has been successful if the string is at `GMT' or
2059 `+X', or at the end of the string.
2061 In extended regexp parlance, the function returns 1 if P matches
2062 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2063 can return) is considered a failure and 0 is returned. */
2065 check_end (const char *p)
2069 while (ISSPACE (*p))
2072 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2073 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2079 /* Convert the textual specification of time in TIME_STRING to the
2080 number of seconds since the Epoch.
2082 TIME_STRING can be in any of the three formats RFC2068 allows the
2083 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2084 Timezones are ignored, and should be GMT.
2086 Return the computed time_t representation, or -1 if the conversion
2089 This function uses strptime with various string formats for parsing
2090 TIME_STRING. This results in a parser that is not as lenient in
2091 interpreting TIME_STRING as I would like it to be. Being based on
2092 strptime, it always allows shortened months, one-digit days, etc.,
2093 but due to the multitude of formats in which time can be
2094 represented, an ideal HTTP time parser would be even more
2095 forgiving. It should completely ignore things like week days and
2096 concentrate only on the various forms of representing years,
2097 months, days, hours, minutes, and seconds. For example, it would
2098 be nice if it accepted ISO 8601 out of the box.
2100 I've investigated free and PD code for this purpose, but none was
2101 usable. getdate was big and unwieldy, and had potential copyright
2102 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2103 distributed with phttpd, is excellent, but we cannot use it because
2104 it is not assigned to the FSF. So I stuck it with strptime. */
2107 http_atotm (const char *time_string)
2109 /* NOTE: Solaris strptime man page claims that %n and %t match white
2110 space, but that's not universally available. Instead, we simply
2111 use ` ' to mean "skip all WS", which works under all strptime
2112 implementations I've tested. */
2114 static const char *time_formats[] = {
2115 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2116 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2117 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2118 (google.com uses this for their cookies.) */
2119 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2125 /* According to Roger Beeman, we need to initialize tm_isdst, since
2126 strptime won't do it. */
2129 /* Note that under foreign locales Solaris strptime() fails to
2130 recognize English dates, which renders this function useless. We
2131 solve this by being careful not to affect LC_TIME when
2132 initializing locale.
2134 Another solution would be to temporarily set locale to C, invoke
2135 strptime(), and restore it back. This is slow and dirty,
2136 however, and locale support other than LC_MESSAGES can mess other
2137 things, so I rather chose to stick with just setting LC_MESSAGES.
2139 GNU strptime does not have this problem because it recognizes
2140 both international and local dates. */
2142 for (i = 0; i < countof (time_formats); i++)
2143 if (check_end (strptime (time_string, time_formats[i], &t)))
2144 return mktime_from_utc (&t);
2146 /* All formats have failed. */
2150 /* Authorization support: We support two authorization schemes:
2152 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2154 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2155 consisting of answering to the server's challenge with the proper
2158 /* How many bytes it will take to store LEN bytes in base64. */
2159 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2161 /* Encode the string S of length LENGTH to base64 format and place it
2162 to STORE. STORE will be 0-terminated, and must point to a writable
2163 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2165 base64_encode (const char *s, char *store, int length)
2167 /* Conversion table. */
2168 static char tbl[64] = {
2169 'A','B','C','D','E','F','G','H',
2170 'I','J','K','L','M','N','O','P',
2171 'Q','R','S','T','U','V','W','X',
2172 'Y','Z','a','b','c','d','e','f',
2173 'g','h','i','j','k','l','m','n',
2174 'o','p','q','r','s','t','u','v',
2175 'w','x','y','z','0','1','2','3',
2176 '4','5','6','7','8','9','+','/'
2179 unsigned char *p = (unsigned char *)store;
2181 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2182 for (i = 0; i < length; i += 3)
2184 *p++ = tbl[s[0] >> 2];
2185 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2186 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2187 *p++ = tbl[s[2] & 0x3f];
2190 /* Pad the result if necessary... */
2191 if (i == length + 1)
2193 else if (i == length + 2)
2194 *(p - 1) = *(p - 2) = '=';
2195 /* ...and zero-terminate it. */
2199 /* Create the authentication header contents for the `Basic' scheme.
2200 This is done by encoding the string `USER:PASS' in base64 and
2201 prepending `HEADER: Basic ' to it. */
2203 basic_authentication_encode (const char *user, const char *passwd,
2206 char *t1, *t2, *res;
2207 int len1 = strlen (user) + 1 + strlen (passwd);
2208 int len2 = BASE64_LENGTH (len1);
2210 t1 = (char *)alloca (len1 + 1);
2211 sprintf (t1, "%s:%s", user, passwd);
2212 t2 = (char *)alloca (1 + len2);
2213 base64_encode (t1, t2, len1);
2214 res = (char *)xmalloc (len2 + 11 + strlen (header));
2215 sprintf (res, "%s: Basic %s\r\n", header, t2);
2221 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2222 of a field in such a header. If the field is the one specified by
2223 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2224 digest authorization code), extract its value in the (char*)
2225 variable pointed by RET. Returns negative on a malformed header,
2226 or number of bytes that have been parsed by this call. */
2228 extract_header_attr (const char *au, const char *attr_name, char **ret)
2230 const char *cp, *ep;
2234 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2236 cp += strlen (attr_name);
2239 cp += skip_lws (cp);
2244 cp += skip_lws (cp);
2249 for (ep = cp; *ep && *ep != '\"'; ep++)
2254 *ret = strdupdelim (cp, ep);
2261 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2262 an array of 16 bytes containing the hash keys, and BUF should be a
2263 buffer of 33 writable characters (32 for hex digits plus one for
2264 zero termination). */
2266 dump_hash (unsigned char *buf, const unsigned char *hash)
2270 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2272 *buf++ = XNUM_TO_digit (*hash >> 4);
2273 *buf++ = XNUM_TO_digit (*hash & 0xf);
2278 /* Take the line apart to find the challenge, and compose a digest
2279 authorization header. See RFC2069 section 2.1.2. */
2281 digest_authentication_encode (const char *au, const char *user,
2282 const char *passwd, const char *method,
2285 static char *realm, *opaque, *nonce;
2290 { "realm", &realm },
2291 { "opaque", &opaque },
2296 realm = opaque = nonce = NULL;
2298 au += 6; /* skip over `Digest' */
2303 au += skip_lws (au);
2304 for (i = 0; i < countof (options); i++)
2306 int skip = extract_header_attr (au, options[i].name,
2307 options[i].variable);
2311 xfree_null (opaque);
2321 if (i == countof (options))
2323 while (*au && *au != '=')
2327 au += skip_lws (au);
2331 while (*au && *au != '\"')
2338 while (*au && *au != ',')
2343 if (!realm || !nonce || !user || !passwd || !path || !method)
2346 xfree_null (opaque);
2351 /* Calculate the digest value. */
2353 ALLOCA_MD5_CONTEXT (ctx);
2354 unsigned char hash[MD5_HASHLEN];
2355 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2356 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2358 /* A1BUF = H(user ":" realm ":" password) */
2360 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2361 gen_md5_update ((unsigned char *)":", 1, ctx);
2362 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2363 gen_md5_update ((unsigned char *)":", 1, ctx);
2364 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2365 gen_md5_finish (ctx, hash);
2366 dump_hash (a1buf, hash);
2368 /* A2BUF = H(method ":" path) */
2370 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2371 gen_md5_update ((unsigned char *)":", 1, ctx);
2372 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2373 gen_md5_finish (ctx, hash);
2374 dump_hash (a2buf, hash);
2376 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2378 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2379 gen_md5_update ((unsigned char *)":", 1, ctx);
2380 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2381 gen_md5_update ((unsigned char *)":", 1, ctx);
2382 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2383 gen_md5_finish (ctx, hash);
2384 dump_hash (response_digest, hash);
2386 res = (char*) xmalloc (strlen (user)
2391 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2392 + (opaque ? strlen (opaque) : 0)
2394 sprintf (res, "Authorization: Digest \
2395 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2396 user, realm, nonce, path, response_digest);
2399 char *p = res + strlen (res);
2400 strcat (p, ", opaque=\"");
2404 strcat (res, "\r\n");
2408 #endif /* USE_DIGEST */
2411 #define BEGINS_WITH(line, string_constant) \
2412 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2413 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2414 || !line[sizeof (string_constant) - 1]))
2417 known_authentication_scheme_p (const char *au)
2419 return BEGINS_WITH (au, "Basic")
2420 || BEGINS_WITH (au, "Digest")
2421 || BEGINS_WITH (au, "NTLM");
2426 /* Create the HTTP authorization request header. When the
2427 `WWW-Authenticate' response header is seen, according to the
2428 authorization scheme specified in that header (`Basic' and `Digest'
2429 are supported by the current implementation), produce an
2430 appropriate HTTP authorization request header. */
2432 create_authorization_line (const char *au, const char *user,
2433 const char *passwd, const char *method,
2436 char *wwwauth = NULL;
2438 if (!strncasecmp (au, "Basic", 5))
2439 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2440 if (!strncasecmp (au, "NTLM", 4))
2441 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2443 else if (!strncasecmp (au, "Digest", 6))
2444 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2445 #endif /* USE_DIGEST */