2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001, 2002
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 In addition, as a special exception, the Free Software Foundation
22 gives permission to link the code of its release of Wget with the
23 OpenSSL project's "OpenSSL" library (or with modified versions of it
24 that use the same license as the "OpenSSL" library), and distribute
25 the linked executables. You must obey the GNU General Public License
26 in all respects for all of the code used other than "OpenSSL". If you
27 modify this file, you may extend this exception to your version of the
28 file, but you are not obligated to do so. If you do not wish to do
29 so, delete this exception statement from your version. */
35 #include <sys/types.h>
46 #if TIME_WITH_SYS_TIME
47 # include <sys/time.h>
51 # include <sys/time.h>
68 # include "gen_sslfunc.h"
76 extern char *version_string;
77 extern LARGE_INT total_downloaded_bytes;
80 # define MIN(x, y) ((x) > (y) ? (y) : (x))
84 static int cookies_loaded_p;
85 struct cookie_jar *wget_cookie_jar;
87 #define TEXTHTML_S "text/html"
88 #define TEXTXHTML_S "application/xhtml+xml"
89 #define HTTP_ACCEPT "*/*"
91 /* Some status code validation macros: */
92 #define H_20X(x) (((x) >= 200) && ((x) < 300))
93 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
94 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
95 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
96 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
98 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
100 #define HTTP_STATUS_OK 200
101 #define HTTP_STATUS_CREATED 201
102 #define HTTP_STATUS_ACCEPTED 202
103 #define HTTP_STATUS_NO_CONTENT 204
104 #define HTTP_STATUS_PARTIAL_CONTENTS 206
106 /* Redirection 3xx. */
107 #define HTTP_STATUS_MULTIPLE_CHOICES 300
108 #define HTTP_STATUS_MOVED_PERMANENTLY 301
109 #define HTTP_STATUS_MOVED_TEMPORARILY 302
110 #define HTTP_STATUS_NOT_MODIFIED 304
111 #define HTTP_STATUS_TEMPORARY_REDIRECT 307
113 /* Client error 4xx. */
114 #define HTTP_STATUS_BAD_REQUEST 400
115 #define HTTP_STATUS_UNAUTHORIZED 401
116 #define HTTP_STATUS_FORBIDDEN 403
117 #define HTTP_STATUS_NOT_FOUND 404
119 /* Server errors 5xx. */
120 #define HTTP_STATUS_INTERNAL 500
121 #define HTTP_STATUS_NOT_IMPLEMENTED 501
122 #define HTTP_STATUS_BAD_GATEWAY 502
123 #define HTTP_STATUS_UNAVAILABLE 503
126 head_terminator (const char *hunk, int oldlen, int peeklen)
128 const char *start, *end;
130 /* If at first peek, verify whether HUNK starts with "HTTP". If
131 not, this is a HTTP/0.9 request and we must bail out without
133 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
139 start = hunk + oldlen - 4;
140 end = hunk + oldlen + peeklen;
142 for (; start < end - 1; start++)
149 if (start[1] == '\n')
155 /* Read the HTTP request head from FD and return it. The error
156 conditions are the same as with fd_read_hunk.
158 To support HTTP/0.9 responses, this function tries to make sure
159 that the data begins with "HTTP". If this is not the case, no data
160 is read and an empty request is returned, so that the remaining
161 data can be treated as body. */
164 fd_read_http_head (int fd)
166 return fd_read_hunk (fd, head_terminator, 512);
170 /* The response data. */
173 /* The array of pointers that indicate where each header starts.
174 For example, given three headers "foo", "bar", and "baz":
175 foo: value\r\nbar: value\r\nbaz: value\r\n\r\n
177 I.e. headers[0] points to the beginning of foo, headers[1] points
178 to the end of foo and the beginning of bar, etc. */
179 const char **headers;
182 static struct response *
183 response_new (const char *head)
188 struct response *resp = xnew0 (struct response);
193 /* Empty head means that we're dealing with a headerless
194 (HTTP/0.9) response. In that case, don't set HEADERS at
199 /* Split HEAD into header lines, so that response_header_* functions
200 don't need to do this over and over again. */
206 DO_REALLOC (resp->headers, size, count + 1, const char *);
207 resp->headers[count++] = hdr;
209 /* Break upon encountering an empty line. */
210 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
213 /* Find the end of HDR, including continuations. */
216 const char *end = strchr (hdr, '\n');
222 while (*hdr == ' ' || *hdr == '\t');
224 DO_REALLOC (resp->headers, size, count + 1, const char *);
225 resp->headers[count++] = NULL;
231 response_header_bounds (const struct response *resp, const char *name,
232 const char **begptr, const char **endptr)
235 const char **headers = resp->headers;
238 if (!headers || !headers[1])
241 name_len = strlen (name);
243 for (i = 1; headers[i + 1]; i++)
245 const char *b = headers[i];
246 const char *e = headers[i + 1];
248 && b[name_len] == ':'
249 && 0 == strncasecmp (b, name, name_len))
252 while (b < e && ISSPACE (*b))
254 while (b < e && ISSPACE (e[-1]))
265 response_header_copy (const struct response *resp, const char *name,
266 char *buf, int bufsize)
269 if (!response_header_bounds (resp, name, &b, &e))
273 int len = MIN (e - b, bufsize);
274 strncpy (buf, b, len);
281 response_header_strdup (const struct response *resp, const char *name)
284 if (!response_header_bounds (resp, name, &b, &e))
286 return strdupdelim (b, e);
289 /* Parse the HTTP status line, which is of format:
291 HTTP-Version SP Status-Code SP Reason-Phrase
293 The function returns the status-code, or -1 if the status line
294 appears malformed. The pointer to "reason-phrase" message is
295 returned in *MESSAGE. */
298 response_status (const struct response *resp, char **message)
305 /* For a HTTP/0.9 response, always assume 200 response. */
307 *message = xstrdup ("OK");
311 p = resp->headers[0];
312 end = resp->headers[1];
318 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
322 /* "/x.x" (optional because some Gnutella servers have been reported
323 as not sending the "/x.x" part. */
324 if (p < end && *p == '/')
327 while (p < end && ISDIGIT (*p))
329 if (p < end && *p == '.')
331 while (p < end && ISDIGIT (*p))
335 while (p < end && ISSPACE (*p))
337 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
340 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
345 while (p < end && ISSPACE (*p))
347 while (p < end && ISSPACE (end[-1]))
349 *message = strdupdelim (p, end);
356 response_free (struct response *resp)
358 xfree_null (resp->headers);
363 print_server_response_1 (const char *b, const char *e)
366 if (b < e && e[-1] == '\n')
368 if (b < e && e[-1] == '\r')
370 BOUNDED_TO_ALLOCA (b, e, ln);
371 logprintf (LOG_VERBOSE, " %s\n", ln);
375 print_server_response (const struct response *resp)
380 for (i = 0; resp->headers[i + 1]; i++)
381 print_server_response_1 (resp->headers[i], resp->headers[i + 1]);
384 /* Parse the `Content-Range' header and extract the information it
385 contains. Returns 1 if successful, -1 otherwise. */
387 parse_content_range (const char *hdr, long *first_byte_ptr,
388 long *last_byte_ptr, long *entity_length_ptr)
392 /* Ancient versions of Netscape proxy server, presumably predating
393 rfc2068, sent out `Content-Range' without the "bytes"
395 if (!strncasecmp (hdr, "bytes", 5))
398 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
402 while (ISSPACE (*hdr))
409 for (num = 0; ISDIGIT (*hdr); hdr++)
410 num = 10 * num + (*hdr - '0');
411 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
413 *first_byte_ptr = num;
415 for (num = 0; ISDIGIT (*hdr); hdr++)
416 num = 10 * num + (*hdr - '0');
417 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
419 *last_byte_ptr = num;
421 for (num = 0; ISDIGIT (*hdr); hdr++)
422 num = 10 * num + (*hdr - '0');
423 *entity_length_ptr = num;
427 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
428 PROMISED_SIZE bytes are sent over the wire -- if the file is
429 longer, read only that much; if the file is shorter, report an error. */
432 post_file (int sock, const char *file_name, long promised_size)
434 static char chunk[8192];
439 DEBUGP (("[writing POST file %s ... ", file_name));
441 fp = fopen (file_name, "rb");
444 while (!feof (fp) && written < promised_size)
447 int length = fread (chunk, 1, sizeof (chunk), fp);
450 towrite = MIN (promised_size - written, length);
451 write_error = fd_write (sock, chunk, towrite, -1);
461 /* If we've written less than was promised, report a (probably
462 nonsensical) error rather than break the promise. */
463 if (written < promised_size)
469 assert (written == promised_size);
470 DEBUGP (("done]\n"));
474 /* Persistent connections. Currently, we cache the most recently used
475 connection as persistent, provided that the HTTP server agrees to
476 make it such. The persistence data is stored in the variables
477 below. Ideally, it should be possible to cache an arbitrary fixed
478 number of these connections. */
480 /* Whether a persistent connection is active. */
481 static int pconn_active;
484 /* The socket of the connection. */
487 /* Host and port of the currently active persistent connection. */
491 /* Whether a ssl handshake has occoured on this connection. */
495 /* Mark the persistent connection as invalid and free the resources it
496 uses. This is used by the CLOSE_* macros after they forcefully
497 close a registered persistent connection. */
500 invalidate_persistent (void)
502 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
504 fd_close (pconn.socket);
509 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
510 persistent. This will enable someone to use the same connection
511 later. In the context of HTTP, this must be called only AFTER the
512 response has been received and the server has promised that the
513 connection will remain alive.
515 If a previous connection was persistent, it is closed. */
518 register_persistent (const char *host, int port, int fd, int ssl)
522 if (pconn.socket == fd)
524 /* The connection FD is already registered. */
529 /* The old persistent connection is still active; close it
530 first. This situation arises whenever a persistent
531 connection exists, but we then connect to a different
532 host, and try to register a persistent connection to that
534 invalidate_persistent ();
540 pconn.host = xstrdup (host);
544 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
547 /* Return non-zero if a persistent connection is available for
548 connecting to HOST:PORT. */
551 persistent_available_p (const char *host, int port, int ssl,
552 int *host_lookup_failed)
554 /* First, check whether a persistent connection is active at all. */
558 /* If we want SSL and the last connection wasn't or vice versa,
559 don't use it. Checking for host and port is not enough because
560 HTTP and HTTPS can apparently coexist on the same port. */
561 if (ssl != pconn.ssl)
564 /* If we're not connecting to the same port, we're not interested. */
565 if (port != pconn.port)
568 /* If the host is the same, we're in business. If not, there is
569 still hope -- read below. */
570 if (0 != strcasecmp (host, pconn.host))
572 /* If pconn.socket is already talking to HOST, we needn't
573 reconnect. This happens often when both sites are virtual
574 hosts distinguished only by name and served by the same
575 network interface, and hence the same web server (possibly
576 set up by the ISP and serving many different web sites).
577 This admittedly non-standard optimization does not contradict
578 HTTP and works well with popular server software. */
582 struct address_list *al;
585 /* Don't try to talk to two different SSL sites over the same
586 secure connection! (Besides, it's not clear if name-based
587 virtual hosting is even possible with SSL.) */
590 /* If pconn.socket's peer is one of the IP addresses HOST
591 resolves to, pconn.socket is for all intents and purposes
592 already talking to HOST. */
594 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
596 /* Can't get the peer's address -- something must be very
597 wrong with the connection. */
598 invalidate_persistent ();
601 al = lookup_host (host, 0);
604 *host_lookup_failed = 1;
608 found = address_list_contains (al, &ip);
609 address_list_release (al);
614 /* The persistent connection's peer address was found among the
615 addresses HOST resolved to; therefore, pconn.sock is in fact
616 already talking to HOST -- no need to reconnect. */
619 /* Finally, check whether the connection is still open. This is
620 important because most server implement a liberal (short) timeout
621 on persistent connections. Wget can of course always reconnect
622 if the connection doesn't work out, but it's nicer to know in
623 advance. This test is a logical followup of the first test, but
624 is "expensive" and therefore placed at the end of the list. */
626 if (!test_socket_open (pconn.socket))
628 /* Oops, the socket is no longer open. Now that we know that,
629 let's invalidate the persistent connection before returning
631 invalidate_persistent ();
638 /* The idea behind these two CLOSE macros is to distinguish between
639 two cases: one when the job we've been doing is finished, and we
640 want to close the connection and leave, and two when something is
641 seriously wrong and we're closing the connection as part of
644 In case of keep_alive, CLOSE_FINISH should leave the connection
645 open, while CLOSE_INVALIDATE should still close it.
647 Note that the semantics of the flag `keep_alive' is "this
648 connection *will* be reused (the server has promised not to close
649 the connection once we're done)", while the semantics of
650 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
651 active, registered connection". */
653 #define CLOSE_FINISH(fd) do { \
656 if (pconn_active && (fd) == pconn.socket) \
657 invalidate_persistent (); \
663 #define CLOSE_INVALIDATE(fd) do { \
664 if (pconn_active && (fd) == pconn.socket) \
665 invalidate_persistent (); \
672 long len; /* received length */
673 long contlen; /* expected length */
674 long restval; /* the restart value */
675 int res; /* the result of last read */
676 char *newloc; /* new location (redirection) */
677 char *remote_time; /* remote time-stamp string */
678 char *error; /* textual HTTP error */
679 int statcode; /* status code */
680 double dltime; /* time of the download in msecs */
681 int no_truncate; /* whether truncating the file is
683 const char *referer; /* value of the referer header. */
684 char **local_file; /* local file. */
688 free_hstat (struct http_stat *hs)
690 xfree_null (hs->newloc);
691 xfree_null (hs->remote_time);
692 xfree_null (hs->error);
694 /* Guard against being called twice. */
696 hs->remote_time = NULL;
700 static char *create_authorization_line PARAMS ((const char *, const char *,
701 const char *, const char *,
703 static char *basic_authentication_encode PARAMS ((const char *, const char *,
705 static int known_authentication_scheme_p PARAMS ((const char *));
707 time_t http_atotm PARAMS ((const char *));
709 #define BEGINS_WITH(line, string_constant) \
710 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
711 && (ISSPACE (line[sizeof (string_constant) - 1]) \
712 || !line[sizeof (string_constant) - 1]))
714 /* Retrieve a document through HTTP protocol. It recognizes status
715 code, and correctly handles redirections. It closes the network
716 socket. If it receives an error from the functions below it, it
717 will print it if there is enough information to do so (almost
718 always), returning the error to the caller (i.e. http_loop).
720 Various HTTP parameters are stored to hs.
722 If PROXY is non-NULL, the connection will be made to the proxy
723 server, and u->url will be requested. */
725 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
727 char *request, *type, *command, *full_path;
729 char *pragma_h, *referer, *useragent, *range, *wwwauth;
730 char *authenticate_h;
733 char *request_keep_alive;
736 long contlen, contrange;
739 int auth_tried_already;
741 char *cookies = NULL;
744 struct response *resp;
749 /* Whether this connection will be kept alive after the HTTP request
753 /* Flag that detects having received a keep-alive response. */
754 int keep_alive_confirmed;
756 /* Whether keep-alive should be inhibited. */
757 int inhibit_keep_alive;
759 /* Whether we need to print the host header with braces around host,
760 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
761 "Host: symbolic-name:1234". */
762 int squares_around_host = 0;
764 /* Headers sent when using POST. */
765 char *post_content_type, *post_content_length;
766 long post_data_size = 0;
768 int host_lookup_failed;
771 /* Initialize the SSL context. After the first run, this is a
775 case SSLERRCTXCREATE:
777 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
778 return SSLERRCTXCREATE;
780 /* try without certfile */
781 logprintf (LOG_NOTQUIET,
782 _("Failed to load certificates from %s\n"),
784 logprintf (LOG_NOTQUIET,
785 _("Trying without the specified certificate\n"));
788 logprintf (LOG_NOTQUIET,
789 _("Failed to get certificate key from %s\n"),
791 logprintf (LOG_NOTQUIET,
792 _("Trying without the specified certificate\n"));
797 #endif /* HAVE_SSL */
799 if (!(*dt & HEAD_ONLY))
800 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
801 know the local filename so we can save to it. */
802 assert (*hs->local_file != NULL);
804 authenticate_h = NULL;
805 auth_tried_already = 0;
807 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
810 /* We need to come back here when the initial attempt to retrieve
811 without authorization header fails. (Expected to happen at least
812 for the Digest authorization scheme.) */
815 keep_alive_confirmed = 0;
817 post_content_type = NULL;
818 post_content_length = NULL;
820 /* Initialize certain elements of struct http_stat. */
825 hs->remote_time = NULL;
828 /* If we're using a proxy, we will be connecting to the proxy
830 conn = proxy ? proxy : u;
832 host_lookup_failed = 0;
834 /* First: establish the connection. */
835 if (inhibit_keep_alive
836 || !persistent_available_p (conn->host, conn->port,
838 u->scheme == SCHEME_HTTPS
842 , &host_lookup_failed))
844 /* In its current implementation, persistent_available_p will
845 look up conn->host in some cases. If that lookup failed, we
846 don't need to bother with connect_to_host. */
847 if (host_lookup_failed)
850 sock = connect_to_host (conn->host, conn->port);
854 return (retryable_socket_connect_error (errno)
855 ? CONERROR : CONIMPOSSIBLE);
858 if (conn->scheme == SCHEME_HTTPS)
860 if (!ssl_connect (sock))
862 logputs (LOG_VERBOSE, "\n");
863 logprintf (LOG_NOTQUIET,
864 _("Unable to establish SSL connection.\n"));
870 #endif /* HAVE_SSL */
874 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
875 pconn.host, pconn.port);
877 using_ssl = pconn.ssl;
878 DEBUGP (("Reusing fd %d.\n", sock));
883 else if (opt.post_file_name || opt.post_data)
891 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
892 sprintf (referer, "Referer: %s\r\n", hs->referer);
895 if (*dt & SEND_NOCACHE)
896 pragma_h = "Pragma: no-cache\r\n";
902 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
903 /* Gag me! Some servers (e.g. WebSitePro) have been known to
904 respond to the following `Range' format by generating a
905 multipart/x-byte-ranges MIME document! This MIME type was
906 present in an old draft of the byteranges specification.
907 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
908 only if multiple non-overlapping ranges are requested --
909 which Wget never does. */
910 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
915 STRDUP_ALLOCA (useragent, opt.useragent);
918 useragent = (char *)alloca (10 + strlen (version_string));
919 sprintf (useragent, "Wget/%s", version_string);
921 /* Construct the authentication, if userid is present. */
924 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
925 user = user ? user : opt.http_user;
926 passwd = passwd ? passwd : opt.http_passwd;
933 /* We have the username and the password, but haven't tried
934 any authorization yet. Let's see if the "Basic" method
935 works. If not, we'll come back here and construct a
936 proper authorization method with the right challenges.
938 If we didn't employ this kind of logic, every URL that
939 requires authorization would have to be processed twice,
940 which is very suboptimal and generates a bunch of false
941 "unauthorized" errors in the server log.
943 #### But this logic also has a serious problem when used
944 with stronger authentications: we *first* transmit the
945 username and the password in clear text, and *then*
946 attempt a stronger authentication scheme. That cannot be
947 right! We are only fortunate that almost everyone still
948 uses the `Basic' scheme anyway.
950 There should be an option to prevent this from happening,
951 for those who use strong authentication schemes and value
953 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
957 /* Use the full path, i.e. one that includes the leading
958 slash and the query string, but is independent of proxy
960 char *pth = url_full_path (u);
961 wwwauth = create_authorization_line (authenticate_h, user, passwd,
970 char *proxy_user, *proxy_passwd;
971 /* For normal username and password, URL components override
972 command-line/wgetrc parameters. With proxy authentication,
973 it's the reverse, because proxy URLs are normally the
974 "permanent" ones, so command-line args should take
976 if (opt.proxy_user && opt.proxy_passwd)
978 proxy_user = opt.proxy_user;
979 proxy_passwd = opt.proxy_passwd;
983 proxy_user = proxy->user;
984 proxy_passwd = proxy->passwd;
986 /* #### This does not appear right. Can't the proxy request,
987 say, `Digest' authentication? */
988 if (proxy_user && proxy_passwd)
989 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
990 "Proxy-Authorization");
993 /* String of the form :PORT. Used only for non-standard ports. */
995 if (u->port != scheme_default_port (u->scheme))
997 port_maybe = (char *)alloca (numdigit (u->port) + 2);
998 sprintf (port_maybe, ":%d", u->port);
1001 if (!inhibit_keep_alive)
1002 request_keep_alive = "Connection: Keep-Alive\r\n";
1004 request_keep_alive = NULL;
1007 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
1009 u->scheme == SCHEME_HTTPS
1015 if (opt.post_data || opt.post_file_name)
1017 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
1019 post_data_size = strlen (opt.post_data);
1022 post_data_size = file_size (opt.post_file_name);
1023 if (post_data_size == -1)
1025 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1026 opt.post_file_name);
1030 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
1031 sprintf (post_content_length,
1032 "Content-Length: %ld\r\n", post_data_size);
1036 full_path = xstrdup (u->url);
1038 /* Use the full path, i.e. one that includes the leading slash and
1039 the query string. E.g. if u->path is "foo/bar" and u->query is
1040 "param=value", full_path will be "/foo/bar?param=value". */
1041 full_path = url_full_path (u);
1043 if (strchr (u->host, ':'))
1044 squares_around_host = 1;
1046 /* Allocate the memory for the request. */
1047 request = (char *)alloca (strlen (command)
1048 + strlen (full_path)
1049 + strlen (useragent)
1051 + (port_maybe ? strlen (port_maybe) : 0)
1052 + strlen (HTTP_ACCEPT)
1053 + (request_keep_alive
1054 ? strlen (request_keep_alive) : 0)
1055 + (referer ? strlen (referer) : 0)
1056 + (cookies ? strlen (cookies) : 0)
1057 + (wwwauth ? strlen (wwwauth) : 0)
1058 + (proxyauth ? strlen (proxyauth) : 0)
1059 + (range ? strlen (range) : 0)
1061 + (post_content_type
1062 ? strlen (post_content_type) : 0)
1063 + (post_content_length
1064 ? strlen (post_content_length) : 0)
1065 + (opt.user_header ? strlen (opt.user_header) : 0)
1067 /* Construct the request. */
1068 sprintf (request, "\
1073 %s%s%s%s%s%s%s%s%s%s\r\n",
1076 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
1077 port_maybe ? port_maybe : "",
1079 request_keep_alive ? request_keep_alive : "",
1080 referer ? referer : "",
1081 cookies ? cookies : "",
1082 wwwauth ? wwwauth : "",
1083 proxyauth ? proxyauth : "",
1086 post_content_type ? post_content_type : "",
1087 post_content_length ? post_content_length : "",
1088 opt.user_header ? opt.user_header : "");
1089 DEBUGP (("\n---request begin---\n%s", request));
1091 /* Free the temporary memory. */
1092 xfree_null (wwwauth);
1093 xfree_null (proxyauth);
1094 xfree_null (cookies);
1097 /* Send the request to server. */
1098 write_error = fd_write (sock, request, strlen (request), -1);
1100 if (write_error >= 0)
1104 DEBUGP (("[POST data: %s]\n", opt.post_data));
1105 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1107 else if (opt.post_file_name && post_data_size != 0)
1108 write_error = post_file (sock, opt.post_file_name, post_data_size);
1110 DEBUGP (("---request end---\n"));
1112 if (write_error < 0)
1114 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1116 CLOSE_INVALIDATE (sock);
1119 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1120 proxy ? "Proxy" : "HTTP");
1121 contlen = contrange = -1;
1126 head = fd_read_http_head (sock);
1129 logputs (LOG_VERBOSE, "\n");
1132 logputs (LOG_NOTQUIET, _("No data received.\n"));
1133 CLOSE_INVALIDATE (sock);
1138 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1140 CLOSE_INVALIDATE (sock);
1145 DEBUGP (("\n---response begin---\n"));
1146 DEBUGP (("%s", head));
1147 DEBUGP (("---response end---\n"));
1149 resp = response_new (head);
1151 /* Check for status line. */
1153 statcode = response_status (resp, &message);
1154 if (!opt.server_response)
1155 logprintf (LOG_VERBOSE, "%2d %s\n", statcode, message ? message : "");
1158 logprintf (LOG_VERBOSE, "\n");
1159 print_server_response (resp);
1162 hs->statcode = statcode;
1164 hs->error = xstrdup (_("Malformed status line"));
1166 hs->error = xstrdup (_("(no description)"));
1168 hs->error = xstrdup (message);
1170 if (response_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1171 contlen = strtol (hdrval, NULL, 10);
1172 type = response_header_strdup (resp, "Content-Type");
1175 char *tmp = strchr (type, ';');
1178 while (tmp > type && ISSPACE (tmp[-1]))
1183 hs->newloc = response_header_strdup (resp, "Location");
1184 hs->remote_time = response_header_strdup (resp, "Last-Modified");
1185 set_cookie = response_header_strdup (resp, "Set-Cookie");
1188 /* The jar should have been created by now. */
1189 assert (wget_cookie_jar != NULL);
1190 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1194 authenticate_h = response_header_strdup (resp, "WWW-Authenticate");
1195 if (response_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1197 long first_byte_pos, last_byte_pos, entity_length;
1198 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1200 contrange = first_byte_pos;
1203 /* Check for keep-alive related responses. */
1204 if (!inhibit_keep_alive && contlen != -1)
1206 if (response_header_copy (resp, "Keep-Alive", NULL, 0))
1208 else if (response_header_copy (resp, "Connection", hdrval,
1211 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1215 response_free (resp);
1218 /* The server has promised that it will not close the connection
1219 when we're done. This means that we can register it. */
1220 register_persistent (conn->host, conn->port, sock, using_ssl);
1222 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1225 /* Authorization is required. */
1229 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1230 might be more bytes in the body. */
1231 if (auth_tried_already)
1233 /* If we have tried it already, then there is not point
1236 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1237 xfree (authenticate_h);
1240 else if (!known_authentication_scheme_p (authenticate_h))
1242 xfree (authenticate_h);
1243 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1246 else if (BEGINS_WITH (authenticate_h, "Basic"))
1248 /* The authentication scheme is basic, the one we try by
1249 default, and it failed. There's no sense in trying
1255 auth_tried_already = 1;
1259 /* We do not need this anymore. */
1262 xfree (authenticate_h);
1263 authenticate_h = NULL;
1266 /* 20x responses are counted among successful by default. */
1267 if (H_20X (statcode))
1270 /* Return if redirected. */
1271 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1273 /* RFC2068 says that in case of the 300 (multiple choices)
1274 response, the server can output a preferred URL through
1275 `Location' header; otherwise, the request should be treated
1276 like GET. So, if the location is set, it will be a
1277 redirection; otherwise, just proceed normally. */
1278 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1282 logprintf (LOG_VERBOSE,
1283 _("Location: %s%s\n"),
1284 hs->newloc ? hs->newloc : _("unspecified"),
1285 hs->newloc ? _(" [following]") : "");
1286 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1287 might be more bytes in the body. */
1293 /* If content-type is not given, assume text/html. This is because
1294 of the multitude of broken CGI's that "forget" to generate the
1297 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1298 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1303 if (opt.html_extension && (*dt & TEXTHTML))
1304 /* -E / --html-extension / html_extension = on was specified, and this is a
1305 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1306 already the file's suffix, tack on ".html". */
1308 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1310 if (last_period_in_local_filename == NULL
1311 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1312 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1314 size_t local_filename_len = strlen(*hs->local_file);
1316 *hs->local_file = xrealloc(*hs->local_file,
1317 local_filename_len + sizeof(".html"));
1318 strcpy(*hs->local_file + local_filename_len, ".html");
1320 *dt |= ADDED_HTML_EXTENSION;
1324 if (contrange == -1)
1326 /* We did not get a content-range header. This means that the
1327 server did not honor our `Range' request. Normally, this
1328 means we should reset hs->restval and continue normally. */
1330 /* However, if `-c' is used, we need to be a bit more careful:
1332 1. If `-c' is specified and the file already existed when
1333 Wget was started, it would be a bad idea for us to start
1334 downloading it from scratch, effectively truncating it. I
1335 believe this cannot happen unless `-c' was specified.
1337 2. If `-c' is used on a file that is already fully
1338 downloaded, we're requesting bytes after the end of file,
1339 which can result in server not honoring `Range'. If this is
1340 the case, `Content-Length' will be equal to the length of the
1342 if (opt.always_rest)
1344 /* Check for condition #2. */
1345 if (hs->restval > 0 /* restart was requested. */
1346 && contlen != -1 /* we got content-length. */
1347 && hs->restval >= contlen /* file fully downloaded
1351 logputs (LOG_VERBOSE, _("\
1352 \n The file is already fully retrieved; nothing to do.\n\n"));
1353 /* In case the caller inspects. */
1356 /* Mark as successfully retrieved. */
1359 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1360 might be more bytes in the body. */
1361 return RETRUNNEEDED;
1364 /* Check for condition #1. */
1365 if (hs->no_truncate)
1367 logprintf (LOG_NOTQUIET,
1370 Continued download failed on this file, which conflicts with `-c'.\n\
1371 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1373 CLOSE_INVALIDATE (sock);
1374 return CONTNOTSUPPORTED;
1382 else if (contrange != hs->restval ||
1383 (H_PARTIAL (statcode) && contrange == -1))
1385 /* This means the whole request was somehow misunderstood by the
1386 server. Bail out. */
1388 CLOSE_INVALIDATE (sock);
1395 contlen += contrange;
1397 contrange = -1; /* If conent-length was not sent,
1398 content-range will be ignored. */
1400 hs->contlen = contlen;
1404 if ((*dt & RETROKF) && !opt.server_response)
1406 /* No need to print this output if the body won't be
1407 downloaded at all, or if the original server response is
1409 logputs (LOG_VERBOSE, _("Length: "));
1412 logputs (LOG_VERBOSE, legible (contlen));
1413 if (contrange != -1)
1414 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1415 legible (contlen - contrange));
1418 logputs (LOG_VERBOSE,
1419 opt.ignore_length ? _("ignored") : _("unspecified"));
1421 logprintf (LOG_VERBOSE, " [%s]\n", type);
1423 logputs (LOG_VERBOSE, "\n");
1427 type = NULL; /* We don't need it any more. */
1429 /* Return if we have no intention of further downloading. */
1430 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1432 /* In case the caller cares to look... */
1436 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1437 might be more bytes in the body. */
1438 return RETRFINISHED;
1441 /* Open the local file. */
1444 mkalldirs (*hs->local_file);
1446 rotate_backups (*hs->local_file);
1447 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1450 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1451 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1452 might be more bytes in the body. */
1458 extern int global_download_count;
1460 /* To ensure that repeated "from scratch" downloads work for -O
1461 files, we rewind the file pointer, unless restval is
1462 non-zero. (This works only when -O is used on regular files,
1463 but it's still a valuable feature.)
1465 However, this loses when more than one URL is specified on
1466 the command line the second rewinds eradicates the contents
1467 of the first download. Thus we disable the above trick for
1468 all the downloads except the very first one.
1470 #### A possible solution to this would be to remember the
1471 file position in the output document and to seek to that
1472 position, instead of rewinding.
1474 We don't truncate stdout, since that breaks
1475 "wget -O - [...] >> foo".
1477 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1479 /* This will silently fail for streams that don't correspond
1480 to regular files, but that's OK. */
1482 /* ftruncate is needed because opt.dfp is opened in append
1483 mode if opt.always_rest is set. */
1484 ftruncate (fileno (fp), 0);
1489 /* #### This confuses the code that checks for file size. There
1490 should be some overhead information. */
1491 if (opt.save_headers)
1492 fwrite (head, 1, strlen (head), fp);
1494 /* Get the contents of the document. */
1495 hs->res = fd_read_body (sock, fp, &hs->len, hs->restval,
1496 (contlen != -1 ? contlen : 0),
1497 keep_alive, &hs->dltime);
1500 CLOSE_FINISH (sock);
1502 CLOSE_INVALIDATE (sock);
1505 /* Close or flush the file. We have to be careful to check for
1506 error here. Checking the result of fwrite() is not enough --
1507 errors could go unnoticed! */
1510 flush_res = fclose (fp);
1512 flush_res = fflush (fp);
1513 if (flush_res == EOF)
1518 return RETRFINISHED;
1521 /* The genuine HTTP loop! This is the part where the retrieval is
1522 retried, and retried, and retried, and... */
1524 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1525 int *dt, struct url *proxy)
1528 int use_ts, got_head = 0; /* time-stamping info */
1529 char *filename_plus_orig_suffix;
1530 char *local_filename = NULL;
1531 char *tms, *locf, *tmrate;
1533 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1534 long local_size = 0; /* the size of the local file */
1535 size_t filename_len;
1536 struct http_stat hstat; /* HTTP status */
1540 /* This used to be done in main(), but it's a better idea to do it
1541 here so that we don't go through the hoops if we're just using
1545 if (!wget_cookie_jar)
1546 wget_cookie_jar = cookie_jar_new ();
1547 if (opt.cookies_input && !cookies_loaded_p)
1549 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1550 cookies_loaded_p = 1;
1556 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1557 has_wildcards_p because it would also warn on `?', and we know that
1558 shows up in CGI paths a *lot*. */
1559 if (strchr (u->url, '*'))
1560 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1562 /* Determine the local filename. */
1563 if (local_file && *local_file)
1564 hstat.local_file = local_file;
1565 else if (local_file)
1567 *local_file = url_file_name (u);
1568 hstat.local_file = local_file;
1572 dummy = url_file_name (u);
1573 hstat.local_file = &dummy;
1576 if (!opt.output_document)
1577 locf = *hstat.local_file;
1579 locf = opt.output_document;
1581 hstat.referer = referer;
1583 filename_len = strlen (*hstat.local_file);
1584 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1586 if (opt.noclobber && file_exists_p (*hstat.local_file))
1588 /* If opt.noclobber is turned on and file already exists, do not
1589 retrieve the file */
1590 logprintf (LOG_VERBOSE, _("\
1591 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1592 /* If the file is there, we suppose it's retrieved OK. */
1595 /* #### Bogusness alert. */
1596 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1597 if (has_html_suffix_p (*hstat.local_file))
1605 if (opt.timestamping)
1607 int local_dot_orig_file_exists = 0;
1609 if (opt.backup_converted)
1610 /* If -K is specified, we'll act on the assumption that it was specified
1611 last time these files were downloaded as well, and instead of just
1612 comparing local file X against server file X, we'll compare local
1613 file X.orig (if extant, else X) against server file X. If -K
1614 _wasn't_ specified last time, or the server contains files called
1615 *.orig, -N will be back to not operating correctly with -k. */
1617 /* Would a single s[n]printf() call be faster? --dan
1619 Definitely not. sprintf() is horribly slow. It's a
1620 different question whether the difference between the two
1621 affects a program. Usually I'd say "no", but at one
1622 point I profiled Wget, and found that a measurable and
1623 non-negligible amount of time was lost calling sprintf()
1624 in url.c. Replacing sprintf with inline calls to
1625 strcpy() and long_to_string() made a difference.
1627 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1628 memcpy (filename_plus_orig_suffix + filename_len,
1629 ".orig", sizeof (".orig"));
1631 /* Try to stat() the .orig file. */
1632 if (stat (filename_plus_orig_suffix, &st) == 0)
1634 local_dot_orig_file_exists = 1;
1635 local_filename = filename_plus_orig_suffix;
1639 if (!local_dot_orig_file_exists)
1640 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1641 if (stat (*hstat.local_file, &st) == 0)
1642 local_filename = *hstat.local_file;
1644 if (local_filename != NULL)
1645 /* There was a local file, so we'll check later to see if the version
1646 the server has is the same version we already have, allowing us to
1652 /* Modification time granularity is 2 seconds for Windows, so
1653 increase local time by 1 second for later comparison. */
1656 local_size = st.st_size;
1660 /* Reset the counter. */
1662 *dt = 0 | ACCEPTRANGES;
1666 /* Increment the pass counter. */
1668 sleep_between_retrievals (count);
1669 /* Get the current time string. */
1670 tms = time_str (NULL);
1671 /* Print fetch message, if opt.verbose. */
1674 char *hurl = url_string (u, 1);
1678 sprintf (tmp, _("(try:%2d)"), count);
1679 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1680 tms, hurl, tmp, locf);
1682 ws_changetitle (hurl, 1);
1687 /* Default document type is empty. However, if spider mode is
1688 on or time-stamping is employed, HEAD_ONLY commands is
1689 encoded within *dt. */
1690 if (opt.spider || (use_ts && !got_head))
1694 /* Assume no restarting. */
1696 /* Decide whether or not to restart. */
1697 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1698 /* #### this calls access() and then stat(); could be optimized. */
1699 && file_exists_p (locf))
1700 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1701 hstat.restval = st.st_size;
1703 /* In `-c' is used and the file is existing and non-empty,
1704 refuse to truncate it if the server doesn't support continued
1706 hstat.no_truncate = 0;
1707 if (opt.always_rest && hstat.restval)
1708 hstat.no_truncate = 1;
1710 /* Decide whether to send the no-cache directive. We send it in
1712 a) we're using a proxy, and we're past our first retrieval.
1713 Some proxies are notorious for caching incomplete data, so
1714 we require a fresh get.
1715 b) caching is explicitly inhibited. */
1716 if ((proxy && count > 1) /* a */
1717 || !opt.allow_cache /* b */
1719 *dt |= SEND_NOCACHE;
1721 *dt &= ~SEND_NOCACHE;
1723 /* Try fetching the document, or at least its head. */
1724 err = gethttp (u, &hstat, dt, proxy);
1726 /* It's unfortunate that wget determines the local filename before finding
1727 out the Content-Type of the file. Barring a major restructuring of the
1728 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1729 *hstat.local_file to tack on ".html". */
1730 if (!opt.output_document)
1731 locf = *hstat.local_file;
1733 locf = opt.output_document;
1736 tms = time_str (NULL);
1737 /* Get the new location (with or without the redirection). */
1739 *newloc = xstrdup (hstat.newloc);
1742 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1743 case CONERROR: case READERR: case WRITEFAILED:
1745 /* Non-fatal errors continue executing the loop, which will
1746 bring them to "while" statement at the end, to judge
1747 whether the number of tries was exceeded. */
1748 free_hstat (&hstat);
1749 printwhat (count, opt.ntry);
1752 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1753 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1754 /* Fatal errors just return from the function. */
1755 free_hstat (&hstat);
1759 case FWRITEERR: case FOPENERR:
1760 /* Another fatal error. */
1761 logputs (LOG_VERBOSE, "\n");
1762 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1763 *hstat.local_file, strerror (errno));
1764 free_hstat (&hstat);
1769 /* Another fatal error. */
1770 logputs (LOG_VERBOSE, "\n");
1771 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1772 free_hstat (&hstat);
1777 /* Return the new location to the caller. */
1780 logprintf (LOG_NOTQUIET,
1781 _("ERROR: Redirection (%d) without location.\n"),
1783 free_hstat (&hstat);
1787 free_hstat (&hstat);
1792 /* The file was already fully retrieved. */
1793 free_hstat (&hstat);
1798 /* Deal with you later. */
1801 /* All possibilities should have been exhausted. */
1804 if (!(*dt & RETROKF))
1808 /* #### Ugly ugly ugly! */
1809 char *hurl = url_string (u, 1);
1810 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1813 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1814 tms, hstat.statcode, hstat.error);
1815 logputs (LOG_VERBOSE, "\n");
1816 free_hstat (&hstat);
1821 /* Did we get the time-stamp? */
1824 if (opt.timestamping && !hstat.remote_time)
1826 logputs (LOG_NOTQUIET, _("\
1827 Last-modified header missing -- time-stamps turned off.\n"));
1829 else if (hstat.remote_time)
1831 /* Convert the date-string into struct tm. */
1832 tmr = http_atotm (hstat.remote_time);
1833 if (tmr == (time_t) (-1))
1834 logputs (LOG_VERBOSE, _("\
1835 Last-modified header invalid -- time-stamp ignored.\n"));
1839 /* The time-stamping section. */
1844 use_ts = 0; /* no more time-stamping */
1845 count = 0; /* the retrieve count for HEAD is
1847 if (hstat.remote_time && tmr != (time_t) (-1))
1849 /* Now time-stamping can be used validly. Time-stamping
1850 means that if the sizes of the local and remote file
1851 match, and local file is newer than the remote file,
1852 it will not be retrieved. Otherwise, the normal
1853 download procedure is resumed. */
1855 (hstat.contlen == -1 || local_size == hstat.contlen))
1857 logprintf (LOG_VERBOSE, _("\
1858 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1860 free_hstat (&hstat);
1864 else if (tml >= tmr)
1865 logprintf (LOG_VERBOSE, _("\
1866 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1868 logputs (LOG_VERBOSE,
1869 _("Remote file is newer, retrieving.\n"));
1871 free_hstat (&hstat);
1874 if ((tmr != (time_t) (-1))
1876 && ((hstat.len == hstat.contlen) ||
1877 ((hstat.res == 0) &&
1878 ((hstat.contlen == -1) ||
1879 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1881 /* #### This code repeats in http.c and ftp.c. Move it to a
1883 const char *fl = NULL;
1884 if (opt.output_document)
1886 if (opt.od_known_regular)
1887 fl = opt.output_document;
1890 fl = *hstat.local_file;
1894 /* End of time-stamping section. */
1898 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1903 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1905 if (hstat.len == hstat.contlen)
1909 logprintf (LOG_VERBOSE,
1910 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1911 tms, tmrate, locf, hstat.len, hstat.contlen);
1912 logprintf (LOG_NONVERBOSE,
1913 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1914 tms, u->url, hstat.len, hstat.contlen, locf, count);
1917 total_downloaded_bytes += hstat.len;
1919 /* Remember that we downloaded the file for later ".orig" code. */
1920 if (*dt & ADDED_HTML_EXTENSION)
1921 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1923 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1925 free_hstat (&hstat);
1929 else if (hstat.res == 0) /* No read error */
1931 if (hstat.contlen == -1) /* We don't know how much we were supposed
1932 to get, so assume we succeeded. */
1936 logprintf (LOG_VERBOSE,
1937 _("%s (%s) - `%s' saved [%ld]\n\n"),
1938 tms, tmrate, locf, hstat.len);
1939 logprintf (LOG_NONVERBOSE,
1940 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1941 tms, u->url, hstat.len, locf, count);
1944 total_downloaded_bytes += hstat.len;
1946 /* Remember that we downloaded the file for later ".orig" code. */
1947 if (*dt & ADDED_HTML_EXTENSION)
1948 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1950 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1952 free_hstat (&hstat);
1956 else if (hstat.len < hstat.contlen) /* meaning we lost the
1957 connection too soon */
1959 logprintf (LOG_VERBOSE,
1960 _("%s (%s) - Connection closed at byte %ld. "),
1961 tms, tmrate, hstat.len);
1962 printwhat (count, opt.ntry);
1963 free_hstat (&hstat);
1966 else if (!opt.kill_longer) /* meaning we got more than expected */
1968 logprintf (LOG_VERBOSE,
1969 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1970 tms, tmrate, locf, hstat.len, hstat.contlen);
1971 logprintf (LOG_NONVERBOSE,
1972 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1973 tms, u->url, hstat.len, hstat.contlen, locf, count);
1975 total_downloaded_bytes += hstat.len;
1977 /* Remember that we downloaded the file for later ".orig" code. */
1978 if (*dt & ADDED_HTML_EXTENSION)
1979 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1981 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1983 free_hstat (&hstat);
1987 else /* the same, but not accepted */
1989 logprintf (LOG_VERBOSE,
1990 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1991 tms, tmrate, hstat.len, hstat.contlen);
1992 printwhat (count, opt.ntry);
1993 free_hstat (&hstat);
1997 else /* now hstat.res can only be -1 */
1999 if (hstat.contlen == -1)
2001 logprintf (LOG_VERBOSE,
2002 _("%s (%s) - Read error at byte %ld (%s)."),
2003 tms, tmrate, hstat.len, strerror (errno));
2004 printwhat (count, opt.ntry);
2005 free_hstat (&hstat);
2008 else /* hstat.res == -1 and contlen is given */
2010 logprintf (LOG_VERBOSE,
2011 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2012 tms, tmrate, hstat.len, hstat.contlen,
2014 printwhat (count, opt.ntry);
2015 free_hstat (&hstat);
2022 while (!opt.ntry || (count < opt.ntry));
2026 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2027 than local timezone.
2029 mktime is similar but assumes struct tm, also known as the
2030 "broken-down" form of time, is in local time zone. mktime_from_utc
2031 uses mktime to make the conversion understanding that an offset
2032 will be introduced by the local time assumption.
2034 mktime_from_utc then measures the introduced offset by applying
2035 gmtime to the initial result and applying mktime to the resulting
2036 "broken-down" form. The difference between the two mktime results
2037 is the measured offset which is then subtracted from the initial
2038 mktime result to yield a calendar time which is the value returned.
2040 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2041 consistent offset (the non DST offset) since tm and tm+o might be
2042 on opposite sides of a DST change.
2044 Some implementations of mktime return -1 for the nonexistent
2045 localtime hour at the beginning of DST. In this event, use
2046 mktime(tm - 1hr) + 3600.
2050 gmtime(t+o) --> tm+o
2051 mktime(tm+o) --> t+2o
2052 t+o - (t+2o - t+o) = t
2054 Note that glibc contains a function of the same purpose named
2055 `timegm' (reverse of gmtime). But obviously, it is not universally
2056 available, and unfortunately it is not straightforwardly
2057 extractable for use here. Perhaps configure should detect timegm
2058 and use it where available.
2060 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2061 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2062 Further improved by Roger with assistance from Edward J. Sabol
2063 based on input by Jamie Zawinski. */
2066 mktime_from_utc (struct tm *t)
2077 return -1; /* can't deal with output from strptime */
2088 return -1; /* can't deal with output from gmtime */
2091 return (tl - (tb - tl));
2094 /* Check whether the result of strptime() indicates success.
2095 strptime() returns the pointer to how far it got to in the string.
2096 The processing has been successful if the string is at `GMT' or
2097 `+X', or at the end of the string.
2099 In extended regexp parlance, the function returns 1 if P matches
2100 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2101 can return) is considered a failure and 0 is returned. */
2103 check_end (const char *p)
2107 while (ISSPACE (*p))
2110 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2111 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2117 /* Convert the textual specification of time in TIME_STRING to the
2118 number of seconds since the Epoch.
2120 TIME_STRING can be in any of the three formats RFC2068 allows the
2121 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2122 Timezones are ignored, and should be GMT.
2124 Return the computed time_t representation, or -1 if the conversion
2127 This function uses strptime with various string formats for parsing
2128 TIME_STRING. This results in a parser that is not as lenient in
2129 interpreting TIME_STRING as I would like it to be. Being based on
2130 strptime, it always allows shortened months, one-digit days, etc.,
2131 but due to the multitude of formats in which time can be
2132 represented, an ideal HTTP time parser would be even more
2133 forgiving. It should completely ignore things like week days and
2134 concentrate only on the various forms of representing years,
2135 months, days, hours, minutes, and seconds. For example, it would
2136 be nice if it accepted ISO 8601 out of the box.
2138 I've investigated free and PD code for this purpose, but none was
2139 usable. getdate was big and unwieldy, and had potential copyright
2140 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2141 distributed with phttpd, is excellent, but we cannot use it because
2142 it is not assigned to the FSF. So I stuck it with strptime. */
2145 http_atotm (const char *time_string)
2147 /* NOTE: Solaris strptime man page claims that %n and %t match white
2148 space, but that's not universally available. Instead, we simply
2149 use ` ' to mean "skip all WS", which works under all strptime
2150 implementations I've tested. */
2152 static const char *time_formats[] = {
2153 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2154 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2155 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2156 (google.com uses this for their cookies.) */
2157 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2163 /* According to Roger Beeman, we need to initialize tm_isdst, since
2164 strptime won't do it. */
2167 /* Note that under foreign locales Solaris strptime() fails to
2168 recognize English dates, which renders this function useless. We
2169 solve this by being careful not to affect LC_TIME when
2170 initializing locale.
2172 Another solution would be to temporarily set locale to C, invoke
2173 strptime(), and restore it back. This is slow and dirty,
2174 however, and locale support other than LC_MESSAGES can mess other
2175 things, so I rather chose to stick with just setting LC_MESSAGES.
2177 GNU strptime does not have this problem because it recognizes
2178 both international and local dates. */
2180 for (i = 0; i < countof (time_formats); i++)
2181 if (check_end (strptime (time_string, time_formats[i], &t)))
2182 return mktime_from_utc (&t);
2184 /* All formats have failed. */
2188 /* Authorization support: We support two authorization schemes:
2190 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2192 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2193 consisting of answering to the server's challenge with the proper
2196 /* How many bytes it will take to store LEN bytes in base64. */
2197 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2199 /* Encode the string S of length LENGTH to base64 format and place it
2200 to STORE. STORE will be 0-terminated, and must point to a writable
2201 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2203 base64_encode (const char *s, char *store, int length)
2205 /* Conversion table. */
2206 static char tbl[64] = {
2207 'A','B','C','D','E','F','G','H',
2208 'I','J','K','L','M','N','O','P',
2209 'Q','R','S','T','U','V','W','X',
2210 'Y','Z','a','b','c','d','e','f',
2211 'g','h','i','j','k','l','m','n',
2212 'o','p','q','r','s','t','u','v',
2213 'w','x','y','z','0','1','2','3',
2214 '4','5','6','7','8','9','+','/'
2217 unsigned char *p = (unsigned char *)store;
2219 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2220 for (i = 0; i < length; i += 3)
2222 *p++ = tbl[s[0] >> 2];
2223 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2224 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2225 *p++ = tbl[s[2] & 0x3f];
2228 /* Pad the result if necessary... */
2229 if (i == length + 1)
2231 else if (i == length + 2)
2232 *(p - 1) = *(p - 2) = '=';
2233 /* ...and zero-terminate it. */
2237 /* Create the authentication header contents for the `Basic' scheme.
2238 This is done by encoding the string `USER:PASS' in base64 and
2239 prepending `HEADER: Basic ' to it. */
2241 basic_authentication_encode (const char *user, const char *passwd,
2244 char *t1, *t2, *res;
2245 int len1 = strlen (user) + 1 + strlen (passwd);
2246 int len2 = BASE64_LENGTH (len1);
2248 t1 = (char *)alloca (len1 + 1);
2249 sprintf (t1, "%s:%s", user, passwd);
2250 t2 = (char *)alloca (1 + len2);
2251 base64_encode (t1, t2, len1);
2252 res = (char *)xmalloc (len2 + 11 + strlen (header));
2253 sprintf (res, "%s: Basic %s\r\n", header, t2);
2258 #define SKIP_WS(x) do { \
2259 while (ISSPACE (*(x))) \
2264 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2265 of a field in such a header. If the field is the one specified by
2266 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2267 digest authorization code), extract its value in the (char*)
2268 variable pointed by RET. Returns negative on a malformed header,
2269 or number of bytes that have been parsed by this call. */
2271 extract_header_attr (const char *au, const char *attr_name, char **ret)
2273 const char *cp, *ep;
2277 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2279 cp += strlen (attr_name);
2292 for (ep = cp; *ep && *ep != '\"'; ep++)
2297 *ret = strdupdelim (cp, ep);
2304 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2305 an array of 16 bytes containing the hash keys, and BUF should be a
2306 buffer of 33 writable characters (32 for hex digits plus one for
2307 zero termination). */
2309 dump_hash (unsigned char *buf, const unsigned char *hash)
2313 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2315 *buf++ = XNUM_TO_digit (*hash >> 4);
2316 *buf++ = XNUM_TO_digit (*hash & 0xf);
2321 /* Take the line apart to find the challenge, and compose a digest
2322 authorization header. See RFC2069 section 2.1.2. */
2324 digest_authentication_encode (const char *au, const char *user,
2325 const char *passwd, const char *method,
2328 static char *realm, *opaque, *nonce;
2333 { "realm", &realm },
2334 { "opaque", &opaque },
2339 realm = opaque = nonce = NULL;
2341 au += 6; /* skip over `Digest' */
2347 for (i = 0; i < countof (options); i++)
2349 int skip = extract_header_attr (au, options[i].name,
2350 options[i].variable);
2354 xfree_null (opaque);
2364 if (i == countof (options))
2366 while (*au && *au != '=')
2374 while (*au && *au != '\"')
2381 while (*au && *au != ',')
2386 if (!realm || !nonce || !user || !passwd || !path || !method)
2389 xfree_null (opaque);
2394 /* Calculate the digest value. */
2396 ALLOCA_MD5_CONTEXT (ctx);
2397 unsigned char hash[MD5_HASHLEN];
2398 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2399 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2401 /* A1BUF = H(user ":" realm ":" password) */
2403 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2404 gen_md5_update ((unsigned char *)":", 1, ctx);
2405 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2406 gen_md5_update ((unsigned char *)":", 1, ctx);
2407 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2408 gen_md5_finish (ctx, hash);
2409 dump_hash (a1buf, hash);
2411 /* A2BUF = H(method ":" path) */
2413 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2414 gen_md5_update ((unsigned char *)":", 1, ctx);
2415 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2416 gen_md5_finish (ctx, hash);
2417 dump_hash (a2buf, hash);
2419 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2421 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2422 gen_md5_update ((unsigned char *)":", 1, ctx);
2423 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2424 gen_md5_update ((unsigned char *)":", 1, ctx);
2425 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2426 gen_md5_finish (ctx, hash);
2427 dump_hash (response_digest, hash);
2429 res = (char*) xmalloc (strlen (user)
2434 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2435 + (opaque ? strlen (opaque) : 0)
2437 sprintf (res, "Authorization: Digest \
2438 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2439 user, realm, nonce, path, response_digest);
2442 char *p = res + strlen (res);
2443 strcat (p, ", opaque=\"");
2447 strcat (res, "\r\n");
2451 #endif /* USE_DIGEST */
2454 #define BEGINS_WITH(line, string_constant) \
2455 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2456 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2457 || !line[sizeof (string_constant) - 1]))
2460 known_authentication_scheme_p (const char *au)
2462 return BEGINS_WITH (au, "Basic")
2463 || BEGINS_WITH (au, "Digest")
2464 || BEGINS_WITH (au, "NTLM");
2469 /* Create the HTTP authorization request header. When the
2470 `WWW-Authenticate' response header is seen, according to the
2471 authorization scheme specified in that header (`Basic' and `Digest'
2472 are supported by the current implementation), produce an
2473 appropriate HTTP authorization request header. */
2475 create_authorization_line (const char *au, const char *user,
2476 const char *passwd, const char *method,
2479 char *wwwauth = NULL;
2481 if (!strncasecmp (au, "Basic", 5))
2482 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2483 if (!strncasecmp (au, "NTLM", 4))
2484 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2486 else if (!strncasecmp (au, "Digest", 6))
2487 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2488 #endif /* USE_DIGEST */