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;
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;
833 char *proxy_user, *proxy_passwd;
834 /* For normal username and password, URL components override
835 command-line/wgetrc parameters. With proxy
836 authentication, it's the reverse, because proxy URLs are
837 normally the "permanent" ones, so command-line args
838 should take precedence. */
839 if (opt.proxy_user && opt.proxy_passwd)
841 proxy_user = opt.proxy_user;
842 proxy_passwd = opt.proxy_passwd;
846 proxy_user = proxy->user;
847 proxy_passwd = proxy->passwd;
849 /* #### This does not appear right. Can't the proxy request,
850 say, `Digest' authentication? */
851 if (proxy_user && proxy_passwd)
852 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
853 "Proxy-Authorization");
855 /* If we're using a proxy, we will be connecting to the proxy
860 host_lookup_failed = 0;
863 /* First: establish the connection. */
865 if (!inhibit_keep_alive)
867 /* Look for a persistent connection to target host, unless a
868 proxy is used. The exception is when SSL is in use, in which
869 case the proxy is nothing but a passthrough to the target
870 host, registered as a connection to the latter. */
871 struct url *relevant = conn;
873 if (u->scheme == SCHEME_HTTPS)
877 if (persistent_available_p (relevant->host, relevant->port,
879 relevant->scheme == SCHEME_HTTPS,
883 &host_lookup_failed))
886 using_ssl = pconn.ssl;
887 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
888 pconn.host, pconn.port);
889 DEBUGP (("Reusing fd %d.\n", sock));
895 /* In its current implementation, persistent_available_p will
896 look up conn->host in some cases. If that lookup failed, we
897 don't need to bother with connect_to_host. */
898 if (host_lookup_failed)
901 sock = connect_to_host (conn->host, conn->port);
905 return (retryable_socket_connect_error (errno)
906 ? CONERROR : CONIMPOSSIBLE);
909 if (proxy && u->scheme == SCHEME_HTTPS)
911 /* When requesting SSL URLs through proxies, use the
912 CONNECT method to request passthrough. */
916 + (proxyauth ? strlen (proxyauth) : 0));
917 sprintf (connect, "CONNECT %s:%d HTTP/1.0\r\n%s\r\n",
918 u->host, u->port, proxyauth ? proxyauth : "");
919 /* Now that PROXYAUTH is part of the CONNECT request, zero
920 it out so we don't send proxy authorization with the
921 regular request below. */
924 DEBUGP (("Writing to proxy: [%s]\n", connect));
925 write_error = fd_write (sock, connect, strlen (connect), -1);
928 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
930 CLOSE_INVALIDATE (sock);
934 head = fd_read_http_head (sock);
937 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
939 CLOSE_INVALIDATE (sock);
948 DEBUGP (("proxy responded with: [%s]\n", head));
950 resp = response_new (head);
951 statcode = response_status (resp, &message);
952 response_free (resp);
956 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
957 message ? message : "?");
958 xfree_null (message);
963 /* SOCK is now *really* connected to u->host, so update CONN
964 to reflect this. That way register_persistent will
965 register SOCK as being connected to u->host:u->port. */
969 if (conn->scheme == SCHEME_HTTPS)
971 if (!ssl_connect (sock))
978 #endif /* HAVE_SSL */
983 else if (opt.post_file_name || opt.post_data)
991 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
992 sprintf (referer, "Referer: %s\r\n", hs->referer);
995 if (*dt & SEND_NOCACHE)
996 pragma_h = "Pragma: no-cache\r\n";
1002 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
1003 /* Gag me! Some servers (e.g. WebSitePro) have been known to
1004 respond to the following `Range' format by generating a
1005 multipart/x-byte-ranges MIME document! This MIME type was
1006 present in an old draft of the byteranges specification.
1007 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
1008 only if multiple non-overlapping ranges are requested --
1009 which Wget never does. */
1010 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
1015 STRDUP_ALLOCA (useragent, opt.useragent);
1018 useragent = (char *)alloca (10 + strlen (version_string));
1019 sprintf (useragent, "Wget/%s", version_string);
1021 /* Construct the authentication, if userid is present. */
1024 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1025 user = user ? user : opt.http_user;
1026 passwd = passwd ? passwd : opt.http_passwd;
1031 if (!authenticate_h)
1033 /* We have the username and the password, but haven't tried
1034 any authorization yet. Let's see if the "Basic" method
1035 works. If not, we'll come back here and construct a
1036 proper authorization method with the right challenges.
1038 If we didn't employ this kind of logic, every URL that
1039 requires authorization would have to be processed twice,
1040 which is very suboptimal and generates a bunch of false
1041 "unauthorized" errors in the server log.
1043 #### But this logic also has a serious problem when used
1044 with stronger authentications: we *first* transmit the
1045 username and the password in clear text, and *then*
1046 attempt a stronger authentication scheme. That cannot be
1047 right! We are only fortunate that almost everyone still
1048 uses the `Basic' scheme anyway.
1050 There should be an option to prevent this from happening,
1051 for those who use strong authentication schemes and value
1053 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
1057 /* Use the full path, i.e. one that includes the leading
1058 slash and the query string, but is independent of proxy
1060 char *pth = url_full_path (u);
1061 wwwauth = create_authorization_line (authenticate_h, user, passwd,
1067 /* String of the form :PORT. Used only for non-standard ports. */
1069 if (u->port != scheme_default_port (u->scheme))
1071 port_maybe = (char *)alloca (numdigit (u->port) + 2);
1072 sprintf (port_maybe, ":%d", u->port);
1075 if (!inhibit_keep_alive)
1076 request_keep_alive = "Connection: Keep-Alive\r\n";
1078 request_keep_alive = NULL;
1081 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
1083 u->scheme == SCHEME_HTTPS
1089 if (opt.post_data || opt.post_file_name)
1091 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
1093 post_data_size = strlen (opt.post_data);
1096 post_data_size = file_size (opt.post_file_name);
1097 if (post_data_size == -1)
1099 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1100 opt.post_file_name);
1104 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
1105 sprintf (post_content_length,
1106 "Content-Length: %ld\r\n", post_data_size);
1110 full_path = xstrdup (u->url);
1112 /* Use the full path, i.e. one that includes the leading slash and
1113 the query string. E.g. if u->path is "foo/bar" and u->query is
1114 "param=value", full_path will be "/foo/bar?param=value". */
1115 full_path = url_full_path (u);
1117 if (strchr (u->host, ':'))
1118 squares_around_host = 1;
1120 /* Allocate the memory for the request. */
1121 request = (char *)alloca (strlen (command)
1122 + strlen (full_path)
1123 + strlen (useragent)
1125 + (port_maybe ? strlen (port_maybe) : 0)
1126 + strlen (HTTP_ACCEPT)
1127 + (request_keep_alive
1128 ? strlen (request_keep_alive) : 0)
1129 + (referer ? strlen (referer) : 0)
1130 + (cookies ? strlen (cookies) : 0)
1131 + (wwwauth ? strlen (wwwauth) : 0)
1132 + (proxyauth ? strlen (proxyauth) : 0)
1133 + (range ? strlen (range) : 0)
1135 + (post_content_type
1136 ? strlen (post_content_type) : 0)
1137 + (post_content_length
1138 ? strlen (post_content_length) : 0)
1139 + (opt.user_header ? strlen (opt.user_header) : 0)
1141 /* Construct the request. */
1142 sprintf (request, "\
1147 %s%s%s%s%s%s%s%s%s%s\r\n",
1150 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
1151 port_maybe ? port_maybe : "",
1153 request_keep_alive ? request_keep_alive : "",
1154 referer ? referer : "",
1155 cookies ? cookies : "",
1156 wwwauth ? wwwauth : "",
1157 proxyauth ? proxyauth : "",
1160 post_content_type ? post_content_type : "",
1161 post_content_length ? post_content_length : "",
1162 opt.user_header ? opt.user_header : "");
1163 DEBUGP (("\n---request begin---\n%s", request));
1165 /* Free the temporary memory. */
1166 xfree_null (wwwauth);
1167 xfree_null (proxyauth);
1168 xfree_null (cookies);
1171 /* Send the request to server. */
1172 write_error = fd_write (sock, request, strlen (request), -1);
1174 if (write_error >= 0)
1178 DEBUGP (("[POST data: %s]\n", opt.post_data));
1179 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1181 else if (opt.post_file_name && post_data_size != 0)
1182 write_error = post_file (sock, opt.post_file_name, post_data_size);
1184 DEBUGP (("---request end---\n"));
1186 if (write_error < 0)
1188 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1190 CLOSE_INVALIDATE (sock);
1193 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1194 proxy ? "Proxy" : "HTTP");
1195 contlen = contrange = -1;
1200 head = fd_read_http_head (sock);
1203 logputs (LOG_VERBOSE, "\n");
1206 logputs (LOG_NOTQUIET, _("No data received.\n"));
1207 CLOSE_INVALIDATE (sock);
1212 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1214 CLOSE_INVALIDATE (sock);
1218 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1220 resp = response_new (head);
1222 /* Check for status line. */
1224 statcode = response_status (resp, &message);
1225 if (!opt.server_response)
1226 logprintf (LOG_VERBOSE, "%2d %s\n", statcode, message ? message : "");
1229 logprintf (LOG_VERBOSE, "\n");
1230 print_server_response (resp);
1233 hs->statcode = statcode;
1235 hs->error = xstrdup (_("Malformed status line"));
1237 hs->error = xstrdup (_("(no description)"));
1239 hs->error = xstrdup (message);
1241 if (response_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1242 contlen = strtol (hdrval, NULL, 10);
1243 type = response_header_strdup (resp, "Content-Type");
1246 char *tmp = strchr (type, ';');
1249 while (tmp > type && ISSPACE (tmp[-1]))
1254 hs->newloc = response_header_strdup (resp, "Location");
1255 hs->remote_time = response_header_strdup (resp, "Last-Modified");
1256 set_cookie = response_header_strdup (resp, "Set-Cookie");
1259 /* The jar should have been created by now. */
1260 assert (wget_cookie_jar != NULL);
1261 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1265 authenticate_h = response_header_strdup (resp, "WWW-Authenticate");
1266 if (response_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1268 long first_byte_pos, last_byte_pos, entity_length;
1269 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1271 contrange = first_byte_pos;
1274 /* Check for keep-alive related responses. */
1275 if (!inhibit_keep_alive && contlen != -1)
1277 if (response_header_copy (resp, "Keep-Alive", NULL, 0))
1279 else if (response_header_copy (resp, "Connection", hdrval,
1282 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1286 response_free (resp);
1289 /* The server has promised that it will not close the connection
1290 when we're done. This means that we can register it. */
1291 register_persistent (conn->host, conn->port, sock, using_ssl);
1293 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1296 /* Authorization is required. */
1300 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1301 might be more bytes in the body. */
1302 if (auth_tried_already)
1304 /* If we have tried it already, then there is not point
1307 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1308 xfree (authenticate_h);
1311 else if (!known_authentication_scheme_p (authenticate_h))
1313 xfree (authenticate_h);
1314 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1317 else if (BEGINS_WITH (authenticate_h, "Basic"))
1319 /* The authentication scheme is basic, the one we try by
1320 default, and it failed. There's no sense in trying
1326 auth_tried_already = 1;
1330 /* We do not need this anymore. */
1333 xfree (authenticate_h);
1334 authenticate_h = NULL;
1337 /* 20x responses are counted among successful by default. */
1338 if (H_20X (statcode))
1341 /* Return if redirected. */
1342 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1344 /* RFC2068 says that in case of the 300 (multiple choices)
1345 response, the server can output a preferred URL through
1346 `Location' header; otherwise, the request should be treated
1347 like GET. So, if the location is set, it will be a
1348 redirection; otherwise, just proceed normally. */
1349 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1353 logprintf (LOG_VERBOSE,
1354 _("Location: %s%s\n"),
1355 hs->newloc ? hs->newloc : _("unspecified"),
1356 hs->newloc ? _(" [following]") : "");
1357 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1358 might be more bytes in the body. */
1364 /* If content-type is not given, assume text/html. This is because
1365 of the multitude of broken CGI's that "forget" to generate the
1368 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1369 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1374 if (opt.html_extension && (*dt & TEXTHTML))
1375 /* -E / --html-extension / html_extension = on was specified, and this is a
1376 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1377 already the file's suffix, tack on ".html". */
1379 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1381 if (last_period_in_local_filename == NULL
1382 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1383 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1385 size_t local_filename_len = strlen(*hs->local_file);
1387 *hs->local_file = xrealloc(*hs->local_file,
1388 local_filename_len + sizeof(".html"));
1389 strcpy(*hs->local_file + local_filename_len, ".html");
1391 *dt |= ADDED_HTML_EXTENSION;
1395 if (contrange == -1)
1397 /* We did not get a content-range header. This means that the
1398 server did not honor our `Range' request. Normally, this
1399 means we should reset hs->restval and continue normally. */
1401 /* However, if `-c' is used, we need to be a bit more careful:
1403 1. If `-c' is specified and the file already existed when
1404 Wget was started, it would be a bad idea for us to start
1405 downloading it from scratch, effectively truncating it. I
1406 believe this cannot happen unless `-c' was specified.
1408 2. If `-c' is used on a file that is already fully
1409 downloaded, we're requesting bytes after the end of file,
1410 which can result in server not honoring `Range'. If this is
1411 the case, `Content-Length' will be equal to the length of the
1413 if (opt.always_rest)
1415 /* Check for condition #2. */
1416 if (hs->restval > 0 /* restart was requested. */
1417 && contlen != -1 /* we got content-length. */
1418 && hs->restval >= contlen /* file fully downloaded
1422 logputs (LOG_VERBOSE, _("\
1423 \n The file is already fully retrieved; nothing to do.\n\n"));
1424 /* In case the caller inspects. */
1427 /* Mark as successfully retrieved. */
1430 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1431 might be more bytes in the body. */
1432 return RETRUNNEEDED;
1435 /* Check for condition #1. */
1436 if (hs->no_truncate)
1438 logprintf (LOG_NOTQUIET,
1441 Continued download failed on this file, which conflicts with `-c'.\n\
1442 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1444 CLOSE_INVALIDATE (sock);
1445 return CONTNOTSUPPORTED;
1453 else if (contrange != hs->restval ||
1454 (H_PARTIAL (statcode) && contrange == -1))
1456 /* This means the whole request was somehow misunderstood by the
1457 server. Bail out. */
1459 CLOSE_INVALIDATE (sock);
1466 contlen += contrange;
1468 contrange = -1; /* If conent-length was not sent,
1469 content-range will be ignored. */
1471 hs->contlen = contlen;
1475 if ((*dt & RETROKF) && !opt.server_response)
1477 /* No need to print this output if the body won't be
1478 downloaded at all, or if the original server response is
1480 logputs (LOG_VERBOSE, _("Length: "));
1483 logputs (LOG_VERBOSE, legible (contlen));
1484 if (contrange != -1)
1485 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1486 legible (contlen - contrange));
1489 logputs (LOG_VERBOSE,
1490 opt.ignore_length ? _("ignored") : _("unspecified"));
1492 logprintf (LOG_VERBOSE, " [%s]\n", type);
1494 logputs (LOG_VERBOSE, "\n");
1498 type = NULL; /* We don't need it any more. */
1500 /* Return if we have no intention of further downloading. */
1501 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1503 /* In case the caller cares to look... */
1507 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1508 might be more bytes in the body. */
1509 return RETRFINISHED;
1512 /* Open the local file. */
1515 mkalldirs (*hs->local_file);
1517 rotate_backups (*hs->local_file);
1518 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1521 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1522 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1523 might be more bytes in the body. */
1529 extern int global_download_count;
1531 /* To ensure that repeated "from scratch" downloads work for -O
1532 files, we rewind the file pointer, unless restval is
1533 non-zero. (This works only when -O is used on regular files,
1534 but it's still a valuable feature.)
1536 However, this loses when more than one URL is specified on
1537 the command line the second rewinds eradicates the contents
1538 of the first download. Thus we disable the above trick for
1539 all the downloads except the very first one.
1541 #### A possible solution to this would be to remember the
1542 file position in the output document and to seek to that
1543 position, instead of rewinding.
1545 We don't truncate stdout, since that breaks
1546 "wget -O - [...] >> foo".
1548 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1550 /* This will silently fail for streams that don't correspond
1551 to regular files, but that's OK. */
1553 /* ftruncate is needed because opt.dfp is opened in append
1554 mode if opt.always_rest is set. */
1555 ftruncate (fileno (fp), 0);
1560 /* #### This confuses the code that checks for file size. There
1561 should be some overhead information. */
1562 if (opt.save_headers)
1563 fwrite (head, 1, strlen (head), fp);
1565 /* Get the contents of the document. */
1566 hs->res = fd_read_body (sock, fp, &hs->len, hs->restval,
1567 (contlen != -1 ? contlen : 0),
1568 keep_alive, &hs->dltime);
1571 CLOSE_FINISH (sock);
1573 CLOSE_INVALIDATE (sock);
1576 /* Close or flush the file. We have to be careful to check for
1577 error here. Checking the result of fwrite() is not enough --
1578 errors could go unnoticed! */
1581 flush_res = fclose (fp);
1583 flush_res = fflush (fp);
1584 if (flush_res == EOF)
1589 return RETRFINISHED;
1592 /* The genuine HTTP loop! This is the part where the retrieval is
1593 retried, and retried, and retried, and... */
1595 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1596 int *dt, struct url *proxy)
1599 int use_ts, got_head = 0; /* time-stamping info */
1600 char *filename_plus_orig_suffix;
1601 char *local_filename = NULL;
1602 char *tms, *locf, *tmrate;
1604 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1605 long local_size = 0; /* the size of the local file */
1606 size_t filename_len;
1607 struct http_stat hstat; /* HTTP status */
1611 /* This used to be done in main(), but it's a better idea to do it
1612 here so that we don't go through the hoops if we're just using
1616 if (!wget_cookie_jar)
1617 wget_cookie_jar = cookie_jar_new ();
1618 if (opt.cookies_input && !cookies_loaded_p)
1620 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1621 cookies_loaded_p = 1;
1627 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1628 has_wildcards_p because it would also warn on `?', and we know that
1629 shows up in CGI paths a *lot*. */
1630 if (strchr (u->url, '*'))
1631 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1633 /* Determine the local filename. */
1634 if (local_file && *local_file)
1635 hstat.local_file = local_file;
1636 else if (local_file)
1638 *local_file = url_file_name (u);
1639 hstat.local_file = local_file;
1643 dummy = url_file_name (u);
1644 hstat.local_file = &dummy;
1647 if (!opt.output_document)
1648 locf = *hstat.local_file;
1650 locf = opt.output_document;
1652 hstat.referer = referer;
1654 filename_len = strlen (*hstat.local_file);
1655 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1657 if (opt.noclobber && file_exists_p (*hstat.local_file))
1659 /* If opt.noclobber is turned on and file already exists, do not
1660 retrieve the file */
1661 logprintf (LOG_VERBOSE, _("\
1662 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1663 /* If the file is there, we suppose it's retrieved OK. */
1666 /* #### Bogusness alert. */
1667 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1668 if (has_html_suffix_p (*hstat.local_file))
1676 if (opt.timestamping)
1678 int local_dot_orig_file_exists = 0;
1680 if (opt.backup_converted)
1681 /* If -K is specified, we'll act on the assumption that it was specified
1682 last time these files were downloaded as well, and instead of just
1683 comparing local file X against server file X, we'll compare local
1684 file X.orig (if extant, else X) against server file X. If -K
1685 _wasn't_ specified last time, or the server contains files called
1686 *.orig, -N will be back to not operating correctly with -k. */
1688 /* Would a single s[n]printf() call be faster? --dan
1690 Definitely not. sprintf() is horribly slow. It's a
1691 different question whether the difference between the two
1692 affects a program. Usually I'd say "no", but at one
1693 point I profiled Wget, and found that a measurable and
1694 non-negligible amount of time was lost calling sprintf()
1695 in url.c. Replacing sprintf with inline calls to
1696 strcpy() and long_to_string() made a difference.
1698 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1699 memcpy (filename_plus_orig_suffix + filename_len,
1700 ".orig", sizeof (".orig"));
1702 /* Try to stat() the .orig file. */
1703 if (stat (filename_plus_orig_suffix, &st) == 0)
1705 local_dot_orig_file_exists = 1;
1706 local_filename = filename_plus_orig_suffix;
1710 if (!local_dot_orig_file_exists)
1711 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1712 if (stat (*hstat.local_file, &st) == 0)
1713 local_filename = *hstat.local_file;
1715 if (local_filename != NULL)
1716 /* There was a local file, so we'll check later to see if the version
1717 the server has is the same version we already have, allowing us to
1723 /* Modification time granularity is 2 seconds for Windows, so
1724 increase local time by 1 second for later comparison. */
1727 local_size = st.st_size;
1731 /* Reset the counter. */
1733 *dt = 0 | ACCEPTRANGES;
1737 /* Increment the pass counter. */
1739 sleep_between_retrievals (count);
1740 /* Get the current time string. */
1741 tms = time_str (NULL);
1742 /* Print fetch message, if opt.verbose. */
1745 char *hurl = url_string (u, 1);
1749 sprintf (tmp, _("(try:%2d)"), count);
1750 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1751 tms, hurl, tmp, locf);
1753 ws_changetitle (hurl, 1);
1758 /* Default document type is empty. However, if spider mode is
1759 on or time-stamping is employed, HEAD_ONLY commands is
1760 encoded within *dt. */
1761 if (opt.spider || (use_ts && !got_head))
1765 /* Assume no restarting. */
1767 /* Decide whether or not to restart. */
1768 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1769 /* #### this calls access() and then stat(); could be optimized. */
1770 && file_exists_p (locf))
1771 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1772 hstat.restval = st.st_size;
1774 /* In `-c' is used and the file is existing and non-empty,
1775 refuse to truncate it if the server doesn't support continued
1777 hstat.no_truncate = 0;
1778 if (opt.always_rest && hstat.restval)
1779 hstat.no_truncate = 1;
1781 /* Decide whether to send the no-cache directive. We send it in
1783 a) we're using a proxy, and we're past our first retrieval.
1784 Some proxies are notorious for caching incomplete data, so
1785 we require a fresh get.
1786 b) caching is explicitly inhibited. */
1787 if ((proxy && count > 1) /* a */
1788 || !opt.allow_cache /* b */
1790 *dt |= SEND_NOCACHE;
1792 *dt &= ~SEND_NOCACHE;
1794 /* Try fetching the document, or at least its head. */
1795 err = gethttp (u, &hstat, dt, proxy);
1797 /* It's unfortunate that wget determines the local filename before finding
1798 out the Content-Type of the file. Barring a major restructuring of the
1799 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1800 *hstat.local_file to tack on ".html". */
1801 if (!opt.output_document)
1802 locf = *hstat.local_file;
1804 locf = opt.output_document;
1807 tms = time_str (NULL);
1808 /* Get the new location (with or without the redirection). */
1810 *newloc = xstrdup (hstat.newloc);
1813 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1814 case CONERROR: case READERR: case WRITEFAILED:
1816 /* Non-fatal errors continue executing the loop, which will
1817 bring them to "while" statement at the end, to judge
1818 whether the number of tries was exceeded. */
1819 free_hstat (&hstat);
1820 printwhat (count, opt.ntry);
1823 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1824 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1825 /* Fatal errors just return from the function. */
1826 free_hstat (&hstat);
1830 case FWRITEERR: case FOPENERR:
1831 /* Another fatal error. */
1832 logputs (LOG_VERBOSE, "\n");
1833 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1834 *hstat.local_file, strerror (errno));
1835 free_hstat (&hstat);
1840 /* Another fatal error. */
1841 logputs (LOG_VERBOSE, "\n");
1842 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1843 free_hstat (&hstat);
1848 /* Return the new location to the caller. */
1851 logprintf (LOG_NOTQUIET,
1852 _("ERROR: Redirection (%d) without location.\n"),
1854 free_hstat (&hstat);
1858 free_hstat (&hstat);
1863 /* The file was already fully retrieved. */
1864 free_hstat (&hstat);
1869 /* Deal with you later. */
1872 /* All possibilities should have been exhausted. */
1875 if (!(*dt & RETROKF))
1879 /* #### Ugly ugly ugly! */
1880 char *hurl = url_string (u, 1);
1881 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1884 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1885 tms, hstat.statcode, hstat.error);
1886 logputs (LOG_VERBOSE, "\n");
1887 free_hstat (&hstat);
1892 /* Did we get the time-stamp? */
1895 if (opt.timestamping && !hstat.remote_time)
1897 logputs (LOG_NOTQUIET, _("\
1898 Last-modified header missing -- time-stamps turned off.\n"));
1900 else if (hstat.remote_time)
1902 /* Convert the date-string into struct tm. */
1903 tmr = http_atotm (hstat.remote_time);
1904 if (tmr == (time_t) (-1))
1905 logputs (LOG_VERBOSE, _("\
1906 Last-modified header invalid -- time-stamp ignored.\n"));
1910 /* The time-stamping section. */
1915 use_ts = 0; /* no more time-stamping */
1916 count = 0; /* the retrieve count for HEAD is
1918 if (hstat.remote_time && tmr != (time_t) (-1))
1920 /* Now time-stamping can be used validly. Time-stamping
1921 means that if the sizes of the local and remote file
1922 match, and local file is newer than the remote file,
1923 it will not be retrieved. Otherwise, the normal
1924 download procedure is resumed. */
1926 (hstat.contlen == -1 || local_size == hstat.contlen))
1928 logprintf (LOG_VERBOSE, _("\
1929 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1931 free_hstat (&hstat);
1935 else if (tml >= tmr)
1936 logprintf (LOG_VERBOSE, _("\
1937 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1939 logputs (LOG_VERBOSE,
1940 _("Remote file is newer, retrieving.\n"));
1942 free_hstat (&hstat);
1945 if ((tmr != (time_t) (-1))
1947 && ((hstat.len == hstat.contlen) ||
1948 ((hstat.res == 0) &&
1949 ((hstat.contlen == -1) ||
1950 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1952 /* #### This code repeats in http.c and ftp.c. Move it to a
1954 const char *fl = NULL;
1955 if (opt.output_document)
1957 if (opt.od_known_regular)
1958 fl = opt.output_document;
1961 fl = *hstat.local_file;
1965 /* End of time-stamping section. */
1969 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1974 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1976 if (hstat.len == hstat.contlen)
1980 logprintf (LOG_VERBOSE,
1981 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1982 tms, tmrate, locf, hstat.len, hstat.contlen);
1983 logprintf (LOG_NONVERBOSE,
1984 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1985 tms, u->url, hstat.len, hstat.contlen, locf, count);
1988 total_downloaded_bytes += hstat.len;
1990 /* Remember that we downloaded the file for later ".orig" code. */
1991 if (*dt & ADDED_HTML_EXTENSION)
1992 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1994 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1996 free_hstat (&hstat);
2000 else if (hstat.res == 0) /* No read error */
2002 if (hstat.contlen == -1) /* We don't know how much we were supposed
2003 to get, so assume we succeeded. */
2007 logprintf (LOG_VERBOSE,
2008 _("%s (%s) - `%s' saved [%ld]\n\n"),
2009 tms, tmrate, locf, hstat.len);
2010 logprintf (LOG_NONVERBOSE,
2011 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
2012 tms, u->url, hstat.len, locf, count);
2015 total_downloaded_bytes += hstat.len;
2017 /* Remember that we downloaded the file for later ".orig" code. */
2018 if (*dt & ADDED_HTML_EXTENSION)
2019 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2021 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2023 free_hstat (&hstat);
2027 else if (hstat.len < hstat.contlen) /* meaning we lost the
2028 connection too soon */
2030 logprintf (LOG_VERBOSE,
2031 _("%s (%s) - Connection closed at byte %ld. "),
2032 tms, tmrate, hstat.len);
2033 printwhat (count, opt.ntry);
2034 free_hstat (&hstat);
2037 else if (!opt.kill_longer) /* meaning we got more than expected */
2039 logprintf (LOG_VERBOSE,
2040 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2041 tms, tmrate, locf, hstat.len, hstat.contlen);
2042 logprintf (LOG_NONVERBOSE,
2043 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2044 tms, u->url, hstat.len, hstat.contlen, locf, count);
2046 total_downloaded_bytes += hstat.len;
2048 /* Remember that we downloaded the file for later ".orig" code. */
2049 if (*dt & ADDED_HTML_EXTENSION)
2050 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2052 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2054 free_hstat (&hstat);
2058 else /* the same, but not accepted */
2060 logprintf (LOG_VERBOSE,
2061 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2062 tms, tmrate, hstat.len, hstat.contlen);
2063 printwhat (count, opt.ntry);
2064 free_hstat (&hstat);
2068 else /* now hstat.res can only be -1 */
2070 if (hstat.contlen == -1)
2072 logprintf (LOG_VERBOSE,
2073 _("%s (%s) - Read error at byte %ld (%s)."),
2074 tms, tmrate, hstat.len, strerror (errno));
2075 printwhat (count, opt.ntry);
2076 free_hstat (&hstat);
2079 else /* hstat.res == -1 and contlen is given */
2081 logprintf (LOG_VERBOSE,
2082 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2083 tms, tmrate, hstat.len, hstat.contlen,
2085 printwhat (count, opt.ntry);
2086 free_hstat (&hstat);
2093 while (!opt.ntry || (count < opt.ntry));
2097 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2098 than local timezone.
2100 mktime is similar but assumes struct tm, also known as the
2101 "broken-down" form of time, is in local time zone. mktime_from_utc
2102 uses mktime to make the conversion understanding that an offset
2103 will be introduced by the local time assumption.
2105 mktime_from_utc then measures the introduced offset by applying
2106 gmtime to the initial result and applying mktime to the resulting
2107 "broken-down" form. The difference between the two mktime results
2108 is the measured offset which is then subtracted from the initial
2109 mktime result to yield a calendar time which is the value returned.
2111 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2112 consistent offset (the non DST offset) since tm and tm+o might be
2113 on opposite sides of a DST change.
2115 Some implementations of mktime return -1 for the nonexistent
2116 localtime hour at the beginning of DST. In this event, use
2117 mktime(tm - 1hr) + 3600.
2121 gmtime(t+o) --> tm+o
2122 mktime(tm+o) --> t+2o
2123 t+o - (t+2o - t+o) = t
2125 Note that glibc contains a function of the same purpose named
2126 `timegm' (reverse of gmtime). But obviously, it is not universally
2127 available, and unfortunately it is not straightforwardly
2128 extractable for use here. Perhaps configure should detect timegm
2129 and use it where available.
2131 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2132 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2133 Further improved by Roger with assistance from Edward J. Sabol
2134 based on input by Jamie Zawinski. */
2137 mktime_from_utc (struct tm *t)
2148 return -1; /* can't deal with output from strptime */
2159 return -1; /* can't deal with output from gmtime */
2162 return (tl - (tb - tl));
2165 /* Check whether the result of strptime() indicates success.
2166 strptime() returns the pointer to how far it got to in the string.
2167 The processing has been successful if the string is at `GMT' or
2168 `+X', or at the end of the string.
2170 In extended regexp parlance, the function returns 1 if P matches
2171 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2172 can return) is considered a failure and 0 is returned. */
2174 check_end (const char *p)
2178 while (ISSPACE (*p))
2181 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2182 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2188 /* Convert the textual specification of time in TIME_STRING to the
2189 number of seconds since the Epoch.
2191 TIME_STRING can be in any of the three formats RFC2068 allows the
2192 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2193 Timezones are ignored, and should be GMT.
2195 Return the computed time_t representation, or -1 if the conversion
2198 This function uses strptime with various string formats for parsing
2199 TIME_STRING. This results in a parser that is not as lenient in
2200 interpreting TIME_STRING as I would like it to be. Being based on
2201 strptime, it always allows shortened months, one-digit days, etc.,
2202 but due to the multitude of formats in which time can be
2203 represented, an ideal HTTP time parser would be even more
2204 forgiving. It should completely ignore things like week days and
2205 concentrate only on the various forms of representing years,
2206 months, days, hours, minutes, and seconds. For example, it would
2207 be nice if it accepted ISO 8601 out of the box.
2209 I've investigated free and PD code for this purpose, but none was
2210 usable. getdate was big and unwieldy, and had potential copyright
2211 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2212 distributed with phttpd, is excellent, but we cannot use it because
2213 it is not assigned to the FSF. So I stuck it with strptime. */
2216 http_atotm (const char *time_string)
2218 /* NOTE: Solaris strptime man page claims that %n and %t match white
2219 space, but that's not universally available. Instead, we simply
2220 use ` ' to mean "skip all WS", which works under all strptime
2221 implementations I've tested. */
2223 static const char *time_formats[] = {
2224 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2225 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2226 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2227 (google.com uses this for their cookies.) */
2228 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2234 /* According to Roger Beeman, we need to initialize tm_isdst, since
2235 strptime won't do it. */
2238 /* Note that under foreign locales Solaris strptime() fails to
2239 recognize English dates, which renders this function useless. We
2240 solve this by being careful not to affect LC_TIME when
2241 initializing locale.
2243 Another solution would be to temporarily set locale to C, invoke
2244 strptime(), and restore it back. This is slow and dirty,
2245 however, and locale support other than LC_MESSAGES can mess other
2246 things, so I rather chose to stick with just setting LC_MESSAGES.
2248 GNU strptime does not have this problem because it recognizes
2249 both international and local dates. */
2251 for (i = 0; i < countof (time_formats); i++)
2252 if (check_end (strptime (time_string, time_formats[i], &t)))
2253 return mktime_from_utc (&t);
2255 /* All formats have failed. */
2259 /* Authorization support: We support two authorization schemes:
2261 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2263 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2264 consisting of answering to the server's challenge with the proper
2267 /* How many bytes it will take to store LEN bytes in base64. */
2268 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2270 /* Encode the string S of length LENGTH to base64 format and place it
2271 to STORE. STORE will be 0-terminated, and must point to a writable
2272 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2274 base64_encode (const char *s, char *store, int length)
2276 /* Conversion table. */
2277 static char tbl[64] = {
2278 'A','B','C','D','E','F','G','H',
2279 'I','J','K','L','M','N','O','P',
2280 'Q','R','S','T','U','V','W','X',
2281 'Y','Z','a','b','c','d','e','f',
2282 'g','h','i','j','k','l','m','n',
2283 'o','p','q','r','s','t','u','v',
2284 'w','x','y','z','0','1','2','3',
2285 '4','5','6','7','8','9','+','/'
2288 unsigned char *p = (unsigned char *)store;
2290 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2291 for (i = 0; i < length; i += 3)
2293 *p++ = tbl[s[0] >> 2];
2294 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2295 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2296 *p++ = tbl[s[2] & 0x3f];
2299 /* Pad the result if necessary... */
2300 if (i == length + 1)
2302 else if (i == length + 2)
2303 *(p - 1) = *(p - 2) = '=';
2304 /* ...and zero-terminate it. */
2308 /* Create the authentication header contents for the `Basic' scheme.
2309 This is done by encoding the string `USER:PASS' in base64 and
2310 prepending `HEADER: Basic ' to it. */
2312 basic_authentication_encode (const char *user, const char *passwd,
2315 char *t1, *t2, *res;
2316 int len1 = strlen (user) + 1 + strlen (passwd);
2317 int len2 = BASE64_LENGTH (len1);
2319 t1 = (char *)alloca (len1 + 1);
2320 sprintf (t1, "%s:%s", user, passwd);
2321 t2 = (char *)alloca (1 + len2);
2322 base64_encode (t1, t2, len1);
2323 res = (char *)xmalloc (len2 + 11 + strlen (header));
2324 sprintf (res, "%s: Basic %s\r\n", header, t2);
2329 #define SKIP_WS(x) do { \
2330 while (ISSPACE (*(x))) \
2335 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2336 of a field in such a header. If the field is the one specified by
2337 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2338 digest authorization code), extract its value in the (char*)
2339 variable pointed by RET. Returns negative on a malformed header,
2340 or number of bytes that have been parsed by this call. */
2342 extract_header_attr (const char *au, const char *attr_name, char **ret)
2344 const char *cp, *ep;
2348 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2350 cp += strlen (attr_name);
2363 for (ep = cp; *ep && *ep != '\"'; ep++)
2368 *ret = strdupdelim (cp, ep);
2375 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2376 an array of 16 bytes containing the hash keys, and BUF should be a
2377 buffer of 33 writable characters (32 for hex digits plus one for
2378 zero termination). */
2380 dump_hash (unsigned char *buf, const unsigned char *hash)
2384 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2386 *buf++ = XNUM_TO_digit (*hash >> 4);
2387 *buf++ = XNUM_TO_digit (*hash & 0xf);
2392 /* Take the line apart to find the challenge, and compose a digest
2393 authorization header. See RFC2069 section 2.1.2. */
2395 digest_authentication_encode (const char *au, const char *user,
2396 const char *passwd, const char *method,
2399 static char *realm, *opaque, *nonce;
2404 { "realm", &realm },
2405 { "opaque", &opaque },
2410 realm = opaque = nonce = NULL;
2412 au += 6; /* skip over `Digest' */
2418 for (i = 0; i < countof (options); i++)
2420 int skip = extract_header_attr (au, options[i].name,
2421 options[i].variable);
2425 xfree_null (opaque);
2435 if (i == countof (options))
2437 while (*au && *au != '=')
2445 while (*au && *au != '\"')
2452 while (*au && *au != ',')
2457 if (!realm || !nonce || !user || !passwd || !path || !method)
2460 xfree_null (opaque);
2465 /* Calculate the digest value. */
2467 ALLOCA_MD5_CONTEXT (ctx);
2468 unsigned char hash[MD5_HASHLEN];
2469 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2470 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2472 /* A1BUF = H(user ":" realm ":" password) */
2474 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2475 gen_md5_update ((unsigned char *)":", 1, ctx);
2476 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2477 gen_md5_update ((unsigned char *)":", 1, ctx);
2478 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2479 gen_md5_finish (ctx, hash);
2480 dump_hash (a1buf, hash);
2482 /* A2BUF = H(method ":" path) */
2484 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2485 gen_md5_update ((unsigned char *)":", 1, ctx);
2486 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2487 gen_md5_finish (ctx, hash);
2488 dump_hash (a2buf, hash);
2490 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2492 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2493 gen_md5_update ((unsigned char *)":", 1, ctx);
2494 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2495 gen_md5_update ((unsigned char *)":", 1, ctx);
2496 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2497 gen_md5_finish (ctx, hash);
2498 dump_hash (response_digest, hash);
2500 res = (char*) xmalloc (strlen (user)
2505 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2506 + (opaque ? strlen (opaque) : 0)
2508 sprintf (res, "Authorization: Digest \
2509 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2510 user, realm, nonce, path, response_digest);
2513 char *p = res + strlen (res);
2514 strcat (p, ", opaque=\"");
2518 strcat (res, "\r\n");
2522 #endif /* USE_DIGEST */
2525 #define BEGINS_WITH(line, string_constant) \
2526 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2527 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2528 || !line[sizeof (string_constant) - 1]))
2531 known_authentication_scheme_p (const char *au)
2533 return BEGINS_WITH (au, "Basic")
2534 || BEGINS_WITH (au, "Digest")
2535 || BEGINS_WITH (au, "NTLM");
2540 /* Create the HTTP authorization request header. When the
2541 `WWW-Authenticate' response header is seen, according to the
2542 authorization scheme specified in that header (`Basic' and `Digest'
2543 are supported by the current implementation), produce an
2544 appropriate HTTP authorization request header. */
2546 create_authorization_line (const char *au, const char *user,
2547 const char *passwd, const char *method,
2550 char *wwwauth = NULL;
2552 if (!strncasecmp (au, "Basic", 5))
2553 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2554 if (!strncasecmp (au, "NTLM", 4))
2555 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2557 else if (!strncasecmp (au, "Digest", 6))
2558 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2559 #endif /* USE_DIGEST */