2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
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. */
25 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
58 # include "gen_sslfunc.h"
65 extern char *version_string;
71 static int cookies_loaded_p;
73 #define TEXTHTML_S "text/html"
74 #define HTTP_ACCEPT "*/*"
76 /* Some status code validation macros: */
77 #define H_20X(x) (((x) >= 200) && ((x) < 300))
78 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
79 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
80 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
82 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
84 #define HTTP_STATUS_OK 200
85 #define HTTP_STATUS_CREATED 201
86 #define HTTP_STATUS_ACCEPTED 202
87 #define HTTP_STATUS_NO_CONTENT 204
88 #define HTTP_STATUS_PARTIAL_CONTENTS 206
90 /* Redirection 3xx. */
91 #define HTTP_STATUS_MULTIPLE_CHOICES 300
92 #define HTTP_STATUS_MOVED_PERMANENTLY 301
93 #define HTTP_STATUS_MOVED_TEMPORARILY 302
94 #define HTTP_STATUS_NOT_MODIFIED 304
96 /* Client error 4xx. */
97 #define HTTP_STATUS_BAD_REQUEST 400
98 #define HTTP_STATUS_UNAUTHORIZED 401
99 #define HTTP_STATUS_FORBIDDEN 403
100 #define HTTP_STATUS_NOT_FOUND 404
102 /* Server errors 5xx. */
103 #define HTTP_STATUS_INTERNAL 500
104 #define HTTP_STATUS_NOT_IMPLEMENTED 501
105 #define HTTP_STATUS_BAD_GATEWAY 502
106 #define HTTP_STATUS_UNAVAILABLE 503
109 /* Parse the HTTP status line, which is of format:
111 HTTP-Version SP Status-Code SP Reason-Phrase
113 The function returns the status-code, or -1 if the status line is
114 malformed. The pointer to reason-phrase is returned in RP. */
116 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
118 /* (the variables must not be named `major' and `minor', because
119 that breaks compilation with SunOS4 cc.) */
120 int mjr, mnr, statcode;
123 *reason_phrase_ptr = NULL;
125 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
126 major version, and Y is minor version. */
127 if (strncmp (line, "HTTP/", 5) != 0)
131 /* Calculate major HTTP version. */
133 for (mjr = 0; ISDIGIT (*line); line++)
134 mjr = 10 * mjr + (*line - '0');
135 if (*line != '.' || p == line)
139 /* Calculate minor HTTP version. */
141 for (mnr = 0; ISDIGIT (*line); line++)
142 mnr = 10 * mnr + (*line - '0');
143 if (*line != ' ' || p == line)
145 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
146 minor version can be safely ignored. */
151 /* Calculate status code. */
152 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
154 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
156 /* Set up the reason phrase pointer. */
158 /* RFC2068 requires SPC here, but we allow the string to finish
159 here, in case no reason-phrase is present. */
163 *reason_phrase_ptr = line;
168 *reason_phrase_ptr = line + 1;
173 /* Functions to be used as arguments to header_process(): */
175 struct http_process_range_closure {
181 /* Parse the `Content-Range' header and extract the information it
182 contains. Returns 1 if successful, -1 otherwise. */
184 http_process_range (const char *hdr, void *arg)
186 struct http_process_range_closure *closure
187 = (struct http_process_range_closure *)arg;
190 /* Certain versions of Nutscape proxy server send out
191 `Content-Length' without "bytes" specifier, which is a breach of
192 RFC2068 (as well as the HTTP/1.1 draft which was current at the
193 time). But hell, I must support it... */
194 if (!strncasecmp (hdr, "bytes", 5))
197 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
201 hdr += skip_lws (hdr);
207 for (num = 0; ISDIGIT (*hdr); hdr++)
208 num = 10 * num + (*hdr - '0');
209 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
211 closure->first_byte_pos = num;
213 for (num = 0; ISDIGIT (*hdr); hdr++)
214 num = 10 * num + (*hdr - '0');
215 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
217 closure->last_byte_pos = num;
219 for (num = 0; ISDIGIT (*hdr); hdr++)
220 num = 10 * num + (*hdr - '0');
221 closure->entity_length = num;
225 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
226 Used for `Accept-Ranges'. */
228 http_process_none (const char *hdr, void *arg)
230 int *where = (int *)arg;
232 if (strstr (hdr, "none"))
239 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
241 http_process_type (const char *hdr, void *arg)
243 char **result = (char **)arg;
244 /* Locate P on `;' or the terminating zero, whichever comes first. */
245 const char *p = strchr (hdr, ';');
247 p = hdr + strlen (hdr);
248 while (p > hdr && ISSPACE (*(p - 1)))
250 *result = strdupdelim (hdr, p);
254 /* Check whether the `Connection' header is set to "keep-alive". */
256 http_process_connection (const char *hdr, void *arg)
258 int *flag = (int *)arg;
259 if (!strcasecmp (hdr, "Keep-Alive"))
264 /* Persistent connections. Currently, we cache the most recently used
265 connection as persistent, provided that the HTTP server agrees to
266 make it such. The persistence data is stored in the variables
267 below. Ideally, it would be in a structure, and it should be
268 possible to cache an arbitrary fixed number of these connections.
270 I think the code is quite easy to extend in that direction. */
272 /* Whether a persistent connection is active. */
273 static int pc_active_p;
274 /* Host and port of currently active persistent connection. */
275 static struct address_list *pc_last_host_ip;
276 static unsigned short pc_last_port;
278 /* File descriptor of the currently active persistent connection. */
279 static int pc_last_fd;
282 /* Whether a ssl handshake has occoured on this connection */
283 static int pc_active_ssl;
284 /* SSL connection of the currently active persistent connection. */
285 static SSL *pc_last_ssl;
286 #endif /* HAVE_SSL */
288 /* Mark the persistent connection as invalid. This is used by the
289 CLOSE_* macros after they forcefully close a registered persistent
290 connection. This does not close the file descriptor -- it is left
291 to the caller to do that. (Maybe it should, though.) */
294 invalidate_persistent (void)
299 #endif /* HAVE_SSL */
300 if (pc_last_host_ip != NULL)
302 address_list_release (pc_last_host_ip);
303 pc_last_host_ip = NULL;
305 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
308 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
309 persistent. This will enable someone to use the same connection
310 later. In the context of HTTP, this must be called only AFTER the
311 response has been received and the server has promised that the
312 connection will remain alive.
314 If a previous connection was persistent, it is closed. */
318 register_persistent (const char *host, unsigned short port, int fd, SSL *ssl)
322 register_persistent (const char *host, unsigned short port, int fd)
327 if (pc_last_fd == fd)
329 /* The connection FD is already registered. Nothing to
335 /* The old persistent connection is still active; let's
336 close it first. This situation arises whenever a
337 persistent connection exists, but we then connect to a
338 different host, and try to register a persistent
339 connection to that one. */
341 /* The ssl disconnect has to take place before the closing
344 shutdown_ssl(pc_last_ssl);
347 invalidate_persistent ();
351 assert (pc_last_host_ip == NULL);
353 /* This lookup_host cannot fail, because it has the results in the
355 pc_last_host_ip = lookup_host (host, 1);
356 assert (pc_last_host_ip != NULL);
363 pc_active_ssl = ssl ? 1 : 0;
365 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
369 # define SHUTDOWN_SSL(ssl) do { \
371 shutdown_ssl (ssl); \
374 # define SHUTDOWN_SSL(ssl)
377 /* Return non-zero if a persistent connection is available for
378 connecting to HOST:PORT. */
382 persistent_available_p (const char *host, unsigned short port, int ssl)
386 persistent_available_p (const char *host, unsigned short port)
390 struct address_list *this_host_ip;
392 /* First, check whether a persistent connection is active at all. */
395 /* Second, check if the active connection pertains to the correct
396 (HOST, PORT) ordered pair. */
397 if (port != pc_last_port)
401 /* Second, a): check if current connection is (not) ssl, too. This
402 test is unlikely to fail because HTTP and HTTPS typicaly use
403 different ports. Yet it is possible, or so I [Christian
404 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
406 if (ssl != pc_active_ssl)
408 #endif /* HAVE_SSL */
410 this_host_ip = lookup_host (host, 1);
414 /* To equate the two host names for the purposes of persistent
415 connections, they need to share all the IP addresses in the
417 success = address_list_match_all (pc_last_host_ip, this_host_ip);
418 address_list_release (this_host_ip);
422 /* Third: check whether the connection is still open. This is
423 important because most server implement a liberal (short) timeout
424 on persistent connections. Wget can of course always reconnect
425 if the connection doesn't work out, but it's nicer to know in
426 advance. This test is a logical followup of the first test, but
427 is "expensive" and therefore placed at the end of the list. */
428 if (!test_socket_open (pc_last_fd))
430 /* Oops, the socket is no longer open. Now that we know that,
431 let's invalidate the persistent connection before returning
435 SHUTDOWN_SSL (pc_last_ssl);
438 invalidate_persistent ();
444 /* The idea behind these two CLOSE macros is to distinguish between
445 two cases: one when the job we've been doing is finished, and we
446 want to close the connection and leave, and two when something is
447 seriously wrong and we're closing the connection as part of
450 In case of keep_alive, CLOSE_FINISH should leave the connection
451 open, while CLOSE_INVALIDATE should still close it.
453 Note that the semantics of the flag `keep_alive' is "this
454 connection *will* be reused (the server has promised not to close
455 the connection once we're done)", while the semantics of
456 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
457 active, registered connection". */
459 #define CLOSE_FINISH(fd) do { \
462 SHUTDOWN_SSL (ssl); \
464 if (pc_active_p && (fd) == pc_last_fd) \
465 invalidate_persistent (); \
469 #define CLOSE_INVALIDATE(fd) do { \
470 SHUTDOWN_SSL (ssl); \
472 if (pc_active_p && (fd) == pc_last_fd) \
473 invalidate_persistent (); \
478 long len; /* received length */
479 long contlen; /* expected length */
480 long restval; /* the restart value */
481 int res; /* the result of last read */
482 char *newloc; /* new location (redirection) */
483 char *remote_time; /* remote time-stamp string */
484 char *error; /* textual HTTP error */
485 int statcode; /* status code */
486 long dltime; /* time of the download */
487 int no_truncate; /* whether truncating the file is
489 const char *referer; /* value of the referer header. */
490 char **local_file; /* local file. */
494 free_hstat (struct http_stat *hs)
496 FREE_MAYBE (hs->newloc);
497 FREE_MAYBE (hs->remote_time);
498 FREE_MAYBE (hs->error);
500 /* Guard against being called twice. */
502 hs->remote_time = NULL;
506 static char *create_authorization_line PARAMS ((const char *, const char *,
507 const char *, const char *,
509 static char *basic_authentication_encode PARAMS ((const char *, const char *,
511 static int known_authentication_scheme_p PARAMS ((const char *));
513 time_t http_atotm PARAMS ((char *));
515 #define BEGINS_WITH(line, string_constant) \
516 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
517 && (ISSPACE (line[sizeof (string_constant) - 1]) \
518 || !line[sizeof (string_constant) - 1]))
520 /* Retrieve a document through HTTP protocol. It recognizes status
521 code, and correctly handles redirections. It closes the network
522 socket. If it receives an error from the functions below it, it
523 will print it if there is enough information to do so (almost
524 always), returning the error to the caller (i.e. http_loop).
526 Various HTTP parameters are stored to hs. Although it parses the
527 response code correctly, it is not used in a sane way. The caller
530 If PROXY is non-NULL, the connection will be made to the proxy
531 server, and u->url will be requested. */
533 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
535 char *request, *type, *command, *full_path;
537 char *pragma_h, *referer, *useragent, *range, *wwwauth;
538 char *authenticate_h;
542 char *request_keep_alive;
543 int sock, hcount, num_written, all_length, statcode;
544 long contlen, contrange;
547 int auth_tried_already;
550 static SSL_CTX *ssl_ctx = NULL;
552 #endif /* HAVE_SSL */
553 char *cookies = NULL;
555 /* Whether this connection will be kept alive after the HTTP request
559 /* Flags that detect the two ways of specifying HTTP keep-alive
561 int http_keep_alive_1, http_keep_alive_2;
563 /* Whether keep-alive should be inhibited. */
564 int inhibit_keep_alive;
566 /* Whether we need to print the host header with braces around host,
567 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
568 "Host: symbolic-name:1234". */
569 int squares_around_host = 0;
572 /* initialize ssl_ctx on first run */
575 uerr_t err = init_ssl (&ssl_ctx);
580 case SSLERRCTXCREATE:
582 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
586 /* try without certfile */
587 logprintf (LOG_NOTQUIET,
588 _("Failed to load certificates from %s\n"),
591 logprintf (LOG_NOTQUIET,
592 _("Trying without the specified certificate\n"));
595 logprintf (LOG_NOTQUIET,
596 _("Failed to get certificate key from %s\n"),
599 logprintf (LOG_NOTQUIET,
600 _("Trying without the specified certificate\n"));
607 #endif /* HAVE_SSL */
609 if (!(*dt & HEAD_ONLY))
610 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
611 know the local filename so we can save to it. */
612 assert (*hs->local_file != NULL);
615 auth_tried_already = 0;
617 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
620 /* We need to come back here when the initial attempt to retrieve
621 without authorization header fails. (Expected to happen at least
622 for the Digest authorization scheme.) */
625 http_keep_alive_1 = http_keep_alive_2 = 0;
627 /* Initialize certain elements of struct http_stat. */
632 hs->remote_time = NULL;
635 /* If we're using a proxy, we will be connecting to the proxy
637 conn = proxy ? proxy : u;
639 /* First: establish the connection. */
640 if (inhibit_keep_alive
643 !persistent_available_p (conn->host, conn->port)
645 !persistent_available_p (conn->host, conn->port,
646 u->scheme == SCHEME_HTTPS)
647 #endif /* HAVE_SSL */
650 struct address_list *al = lookup_host (conn->host, 0);
653 set_connection_host_name (conn->host);
654 sock = connect_to_many (al, conn->port, 0);
655 set_connection_host_name (NULL);
656 address_list_release (al);
659 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
662 if (conn->scheme == SCHEME_HTTPS)
663 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
665 logputs (LOG_VERBOSE, "\n");
666 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
671 #endif /* HAVE_SSL */
675 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
676 conn->host, conn->port);
677 /* #### pc_last_fd should be accessed through an accessor
682 #endif /* HAVE_SSL */
683 DEBUGP (("Reusing fd %d.\n", sock));
686 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
691 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
692 sprintf (referer, "Referer: %s\r\n", hs->referer);
695 if (*dt & SEND_NOCACHE)
696 pragma_h = "Pragma: no-cache\r\n";
702 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
703 /* Gag me! Some servers (e.g. WebSitePro) have been known to
704 respond to the following `Range' format by generating a
705 multipart/x-byte-ranges MIME document! This MIME type was
706 present in an old draft of the byteranges specification.
707 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
708 only if multiple non-overlapping ranges are requested --
709 which Wget never does. */
710 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
715 STRDUP_ALLOCA (useragent, opt.useragent);
718 useragent = (char *)alloca (10 + strlen (version_string));
719 sprintf (useragent, "Wget/%s", version_string);
721 /* Construct the authentication, if userid is present. */
724 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
725 user = user ? user : opt.http_user;
726 passwd = passwd ? passwd : opt.http_passwd;
733 /* We have the username and the password, but haven't tried
734 any authorization yet. Let's see if the "Basic" method
735 works. If not, we'll come back here and construct a
736 proper authorization method with the right challenges.
738 If we didn't employ this kind of logic, every URL that
739 requires authorization would have to be processed twice,
740 which is very suboptimal and generates a bunch of false
741 "unauthorized" errors in the server log.
743 #### But this logic also has a serious problem when used
744 with stronger authentications: we *first* transmit the
745 username and the password in clear text, and *then*
746 attempt a stronger authentication scheme. That cannot be
747 right! We are only fortunate that almost everyone still
748 uses the `Basic' scheme anyway.
750 There should be an option to prevent this from happening,
751 for those who use strong authentication schemes and value
753 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
757 /* Use the full path, i.e. one that includes the leading
758 slash and the query string, but is independent of proxy
760 char *pth = url_full_path (u);
761 wwwauth = create_authorization_line (authenticate_h, user, passwd,
770 char *proxy_user, *proxy_passwd;
771 /* For normal username and password, URL components override
772 command-line/wgetrc parameters. With proxy authentication,
773 it's the reverse, because proxy URLs are normally the
774 "permanent" ones, so command-line args should take
776 if (opt.proxy_user && opt.proxy_passwd)
778 proxy_user = opt.proxy_user;
779 proxy_passwd = opt.proxy_passwd;
783 proxy_user = proxy->user;
784 proxy_passwd = proxy->passwd;
786 /* #### This does not appear right. Can't the proxy request,
787 say, `Digest' authentication? */
788 if (proxy_user && proxy_passwd)
789 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
790 "Proxy-Authorization");
793 /* String of the form :PORT. Used only for non-standard ports. */
795 if (u->port != scheme_default_port (u->scheme))
797 port_maybe = (char *)alloca (numdigit (u->port) + 2);
798 sprintf (port_maybe, ":%d", u->port);
801 if (!inhibit_keep_alive)
802 request_keep_alive = "Connection: Keep-Alive\r\n";
804 request_keep_alive = NULL;
807 cookies = build_cookies_request (u->host, u->port, u->path,
809 u->scheme == SCHEME_HTTPS
816 full_path = xstrdup (u->url);
818 /* Use the full path, i.e. one that includes the leading slash and
819 the query string. E.g. if u->path is "foo/bar" and u->query is
820 "param=value", full_path will be "/foo/bar?param=value". */
821 full_path = url_full_path (u);
823 if (strchr (u->host, ':'))
824 squares_around_host = 1;
826 /* Allocate the memory for the request. */
827 request = (char *)alloca (strlen (command)
831 + (port_maybe ? strlen (port_maybe) : 0)
832 + strlen (HTTP_ACCEPT)
833 + (request_keep_alive
834 ? strlen (request_keep_alive) : 0)
835 + (referer ? strlen (referer) : 0)
836 + (cookies ? strlen (cookies) : 0)
837 + (wwwauth ? strlen (wwwauth) : 0)
838 + (proxyauth ? strlen (proxyauth) : 0)
839 + (range ? strlen (range) : 0)
841 + (opt.user_header ? strlen (opt.user_header) : 0)
843 /* Construct the request. */
849 %s%s%s%s%s%s%s%s\r\n",
852 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
853 port_maybe ? port_maybe : "",
855 request_keep_alive ? request_keep_alive : "",
856 referer ? referer : "",
857 cookies ? cookies : "",
858 wwwauth ? wwwauth : "",
859 proxyauth ? proxyauth : "",
862 opt.user_header ? opt.user_header : "");
863 DEBUGP (("---request begin---\n%s---request end---\n", request));
865 /* Free the temporary memory. */
866 FREE_MAYBE (wwwauth);
867 FREE_MAYBE (proxyauth);
868 FREE_MAYBE (cookies);
871 /* Send the request to server. */
873 if (conn->scheme == SCHEME_HTTPS)
874 num_written = ssl_iwrite (ssl, request, strlen (request));
876 #endif /* HAVE_SSL */
877 num_written = iwrite (sock, request, strlen (request));
881 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
883 CLOSE_INVALIDATE (sock);
886 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
887 proxy ? "Proxy" : "HTTP");
888 contlen = contrange = -1;
893 /* Before reading anything, initialize the rbuf. */
894 rbuf_initialize (&rbuf, sock);
896 if (conn->scheme == SCHEME_HTTPS)
900 #endif /* HAVE_SSL */
903 /* Header-fetching loop. */
911 /* Get the header. */
912 status = header_get (&rbuf, &hdr,
913 /* Disallow continuations for status line. */
914 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
916 /* Check for errors. */
917 if (status == HG_EOF && *hdr)
919 /* This used to be an unconditional error, but that was
920 somewhat controversial, because of a large number of
921 broken CGI's that happily "forget" to send the second EOL
922 before closing the connection of a HEAD request.
924 So, the deal is to check whether the header is empty
925 (*hdr is zero if it is); if yes, it means that the
926 previous header was fully retrieved, and that -- most
927 probably -- the request is complete. "...be liberal in
928 what you accept." Oh boy. */
929 logputs (LOG_VERBOSE, "\n");
930 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
933 FREE_MAYBE (all_headers);
934 CLOSE_INVALIDATE (sock);
937 else if (status == HG_ERROR)
939 logputs (LOG_VERBOSE, "\n");
940 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
944 FREE_MAYBE (all_headers);
945 CLOSE_INVALIDATE (sock);
949 /* If the headers are to be saved to a file later, save them to
951 if (opt.save_headers)
953 int lh = strlen (hdr);
954 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
955 memcpy (all_headers + all_length, hdr, lh);
957 all_headers[all_length++] = '\n';
958 all_headers[all_length] = '\0';
961 /* Check for status line. */
965 /* Parse the first line of server response. */
966 statcode = parse_http_status_line (hdr, &error);
967 hs->statcode = statcode;
968 /* Store the descriptive response. */
969 if (statcode == -1) /* malformed response */
971 /* A common reason for "malformed response" error is the
972 case when no data was actually received. Handle this
975 hs->error = xstrdup (_("No data received"));
977 hs->error = xstrdup (_("Malformed status line"));
982 hs->error = xstrdup (_("(no description)"));
984 hs->error = xstrdup (error);
992 if (opt.server_response)
993 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
995 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1001 /* Exit on empty header. */
1008 /* Print the header if requested. */
1009 if (opt.server_response && hcount != 1)
1010 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1012 /* Try getting content-length. */
1013 if (contlen == -1 && !opt.ignore_length)
1014 if (header_process (hdr, "Content-Length", header_extract_number,
1017 /* Try getting content-type. */
1019 if (header_process (hdr, "Content-Type", http_process_type, &type))
1021 /* Try getting location. */
1023 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1025 /* Try getting last-modified. */
1026 if (!hs->remote_time)
1027 if (header_process (hdr, "Last-Modified", header_strdup,
1030 /* Try getting cookies. */
1032 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1034 /* Try getting www-authentication. */
1035 if (!authenticate_h)
1036 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1039 /* Check for accept-ranges header. If it contains the word
1040 `none', disable the ranges. */
1041 if (*dt & ACCEPTRANGES)
1044 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1047 *dt &= ~ACCEPTRANGES;
1051 /* Try getting content-range. */
1052 if (contrange == -1)
1054 struct http_process_range_closure closure;
1055 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1057 contrange = closure.first_byte_pos;
1061 /* Check for keep-alive related responses. */
1062 if (!inhibit_keep_alive)
1064 /* Check for the `Keep-Alive' header. */
1065 if (!http_keep_alive_1)
1067 if (header_process (hdr, "Keep-Alive", header_exists,
1068 &http_keep_alive_1))
1071 /* Check for `Connection: Keep-Alive'. */
1072 if (!http_keep_alive_2)
1074 if (header_process (hdr, "Connection", http_process_connection,
1075 &http_keep_alive_2))
1083 logputs (LOG_VERBOSE, "\n");
1086 && (http_keep_alive_1 || http_keep_alive_2))
1088 assert (inhibit_keep_alive == 0);
1092 /* The server has promised that it will not close the connection
1093 when we're done. This means that we can register it. */
1095 register_persistent (conn->host, conn->port, sock);
1097 register_persistent (conn->host, conn->port, sock, ssl);
1098 #endif /* HAVE_SSL */
1100 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1103 /* Authorization is required. */
1107 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1108 might be more bytes in the body. */
1109 if (auth_tried_already)
1111 /* If we have tried it already, then there is not point
1114 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1115 xfree (authenticate_h);
1118 else if (!known_authentication_scheme_p (authenticate_h))
1120 xfree (authenticate_h);
1121 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1124 else if (BEGINS_WITH (authenticate_h, "Basic"))
1126 /* The authentication scheme is basic, the one we try by
1127 default, and it failed. There's no sense in trying
1133 auth_tried_already = 1;
1137 /* We do not need this anymore. */
1140 xfree (authenticate_h);
1141 authenticate_h = NULL;
1144 /* 20x responses are counted among successful by default. */
1145 if (H_20X (statcode))
1148 /* Return if redirected. */
1149 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1151 /* RFC2068 says that in case of the 300 (multiple choices)
1152 response, the server can output a preferred URL through
1153 `Location' header; otherwise, the request should be treated
1154 like GET. So, if the location is set, it will be a
1155 redirection; otherwise, just proceed normally. */
1156 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1160 logprintf (LOG_VERBOSE,
1161 _("Location: %s%s\n"),
1162 hs->newloc ? hs->newloc : _("unspecified"),
1163 hs->newloc ? _(" [following]") : "");
1164 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1165 might be more bytes in the body. */
1167 FREE_MAYBE (all_headers);
1172 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1175 /* We don't assume text/html by default. */
1178 if (opt.html_extension && (*dt & TEXTHTML))
1179 /* -E / --html-extension / html_extension = on was specified, and this is a
1180 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1181 already the file's suffix, tack on ".html". */
1183 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1185 if (last_period_in_local_filename == NULL ||
1186 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1187 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1189 size_t local_filename_len = strlen(*hs->local_file);
1191 *hs->local_file = xrealloc(*hs->local_file,
1192 local_filename_len + sizeof(".html"));
1193 strcpy(*hs->local_file + local_filename_len, ".html");
1195 *dt |= ADDED_HTML_EXTENSION;
1199 if (contrange == -1)
1201 /* We did not get a content-range header. This means that the
1202 server did not honor our `Range' request. Normally, this
1203 means we should reset hs->restval and continue normally. */
1205 /* However, if `-c' is used, we need to be a bit more careful:
1207 1. If `-c' is specified and the file already existed when
1208 Wget was started, it would be a bad idea for us to start
1209 downloading it from scratch, effectively truncating it. I
1210 believe this cannot happen unless `-c' was specified.
1212 2. If `-c' is used on a file that is already fully
1213 downloaded, we're requesting bytes after the end of file,
1214 which can result in server not honoring `Range'. If this is
1215 the case, `Content-Length' will be equal to the length of the
1217 if (opt.always_rest)
1219 /* Check for condition #2. */
1220 if (hs->restval > 0 /* restart was requested. */
1221 && contlen != -1 /* we got content-length. */
1222 && hs->restval >= contlen /* file fully downloaded
1226 logputs (LOG_VERBOSE, _("\
1227 \n The file is already fully retrieved; nothing to do.\n\n"));
1228 /* In case the caller inspects. */
1231 /* Mark as successfully retrieved. */
1234 FREE_MAYBE (all_headers);
1235 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1236 might be more bytes in the body. */
1237 return RETRUNNEEDED;
1240 /* Check for condition #1. */
1241 if (hs->no_truncate)
1243 logprintf (LOG_NOTQUIET,
1246 Continued download failed on this file, which conflicts with `-c'.\n\
1247 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1249 FREE_MAYBE (all_headers);
1250 CLOSE_INVALIDATE (sock);
1251 return CONTNOTSUPPORTED;
1259 else if (contrange != hs->restval ||
1260 (H_PARTIAL (statcode) && contrange == -1))
1262 /* This means the whole request was somehow misunderstood by the
1263 server. Bail out. */
1265 FREE_MAYBE (all_headers);
1266 CLOSE_INVALIDATE (sock);
1273 contlen += contrange;
1275 contrange = -1; /* If conent-length was not sent,
1276 content-range will be ignored. */
1278 hs->contlen = contlen;
1282 if ((*dt & RETROKF) && !opt.server_response)
1284 /* No need to print this output if the body won't be
1285 downloaded at all, or if the original server response is
1287 logputs (LOG_VERBOSE, _("Length: "));
1290 logputs (LOG_VERBOSE, legible (contlen));
1291 if (contrange != -1)
1292 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1293 legible (contlen - contrange));
1296 logputs (LOG_VERBOSE,
1297 opt.ignore_length ? _("ignored") : _("unspecified"));
1299 logprintf (LOG_VERBOSE, " [%s]\n", type);
1301 logputs (LOG_VERBOSE, "\n");
1305 type = NULL; /* We don't need it any more. */
1307 /* Return if we have no intention of further downloading. */
1308 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1310 /* In case the caller cares to look... */
1314 FREE_MAYBE (all_headers);
1315 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1316 might be more bytes in the body. */
1317 return RETRFINISHED;
1320 /* Open the local file. */
1323 mkalldirs (*hs->local_file);
1325 rotate_backups (*hs->local_file);
1326 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1329 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1330 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1331 might be more bytes in the body. */
1332 FREE_MAYBE (all_headers);
1338 extern int global_download_count;
1340 /* To ensure that repeated "from scratch" downloads work for -O
1341 files, we rewind the file pointer, unless restval is
1342 non-zero. (This works only when -O is used on regular files,
1343 but it's still a valuable feature.)
1345 However, this loses when more than one URL is specified on
1346 the command line the second rewinds eradicates the contents
1347 of the first download. Thus we disable the above trick for
1348 all the downloads except the very first one.
1350 #### A possible solution to this would be to remember the
1351 file position in the output document and to seek to that
1352 position, instead of rewinding. */
1353 if (!hs->restval && global_download_count == 0)
1355 /* This will silently fail for streams that don't correspond
1356 to regular files, but that's OK. */
1358 /* ftruncate is needed because opt.dfp is opened in append
1359 mode if opt.always_rest is set. */
1360 ftruncate (fileno (fp), 0);
1365 /* #### This confuses the code that checks for file size. There
1366 should be some overhead information. */
1367 if (opt.save_headers)
1368 fwrite (all_headers, 1, all_length, fp);
1370 /* Get the contents of the document. */
1371 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1372 (contlen != -1 ? contlen : 0),
1373 &rbuf, keep_alive, &hs->dltime);
1376 CLOSE_FINISH (sock);
1378 CLOSE_INVALIDATE (sock);
1381 /* Close or flush the file. We have to be careful to check for
1382 error here. Checking the result of fwrite() is not enough --
1383 errors could go unnoticed! */
1386 flush_res = fclose (fp);
1388 flush_res = fflush (fp);
1389 if (flush_res == EOF)
1392 FREE_MAYBE (all_headers);
1395 return RETRFINISHED;
1398 /* The genuine HTTP loop! This is the part where the retrieval is
1399 retried, and retried, and retried, and... */
1401 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1402 int *dt, struct url *proxy)
1405 int use_ts, got_head = 0; /* time-stamping info */
1406 char *filename_plus_orig_suffix;
1407 char *local_filename = NULL;
1408 char *tms, *locf, *tmrate;
1410 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1411 long local_size = 0; /* the size of the local file */
1412 size_t filename_len;
1413 struct http_stat hstat; /* HTTP status */
1417 /* This used to be done in main(), but it's a better idea to do it
1418 here so that we don't go through the hoops if we're just using
1420 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1422 load_cookies (opt.cookies_input);
1423 cookies_loaded_p = 1;
1428 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1429 has_wildcards_p because it would also warn on `?', and we know that
1430 shows up in CGI paths a *lot*. */
1431 if (strchr (u->url, '*'))
1432 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1434 /* Determine the local filename. */
1435 if (local_file && *local_file)
1436 hstat.local_file = local_file;
1437 else if (local_file)
1439 *local_file = url_filename (u);
1440 hstat.local_file = local_file;
1444 dummy = url_filename (u);
1445 hstat.local_file = &dummy;
1448 if (!opt.output_document)
1449 locf = *hstat.local_file;
1451 locf = opt.output_document;
1453 hstat.referer = referer;
1455 filename_len = strlen (*hstat.local_file);
1456 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1458 if (opt.noclobber && file_exists_p (*hstat.local_file))
1460 /* If opt.noclobber is turned on and file already exists, do not
1461 retrieve the file */
1462 logprintf (LOG_VERBOSE, _("\
1463 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1464 /* If the file is there, we suppose it's retrieved OK. */
1467 /* #### Bogusness alert. */
1468 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1469 if (has_html_suffix_p (*hstat.local_file))
1477 if (opt.timestamping)
1479 boolean local_dot_orig_file_exists = FALSE;
1481 if (opt.backup_converted)
1482 /* If -K is specified, we'll act on the assumption that it was specified
1483 last time these files were downloaded as well, and instead of just
1484 comparing local file X against server file X, we'll compare local
1485 file X.orig (if extant, else X) against server file X. If -K
1486 _wasn't_ specified last time, or the server contains files called
1487 *.orig, -N will be back to not operating correctly with -k. */
1489 /* Would a single s[n]printf() call be faster? --dan
1491 Definitely not. sprintf() is horribly slow. It's a
1492 different question whether the difference between the two
1493 affects a program. Usually I'd say "no", but at one
1494 point I profiled Wget, and found that a measurable and
1495 non-negligible amount of time was lost calling sprintf()
1496 in url.c. Replacing sprintf with inline calls to
1497 strcpy() and long_to_string() made a difference.
1499 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1500 memcpy (filename_plus_orig_suffix + filename_len,
1501 ".orig", sizeof (".orig"));
1503 /* Try to stat() the .orig file. */
1504 if (stat (filename_plus_orig_suffix, &st) == 0)
1506 local_dot_orig_file_exists = TRUE;
1507 local_filename = filename_plus_orig_suffix;
1511 if (!local_dot_orig_file_exists)
1512 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1513 if (stat (*hstat.local_file, &st) == 0)
1514 local_filename = *hstat.local_file;
1516 if (local_filename != NULL)
1517 /* There was a local file, so we'll check later to see if the version
1518 the server has is the same version we already have, allowing us to
1524 /* Modification time granularity is 2 seconds for Windows, so
1525 increase local time by 1 second for later comparison. */
1528 local_size = st.st_size;
1532 /* Reset the counter. */
1534 *dt = 0 | ACCEPTRANGES;
1538 /* Increment the pass counter. */
1540 sleep_between_retrievals (count);
1541 /* Get the current time string. */
1542 tms = time_str (NULL);
1543 /* Print fetch message, if opt.verbose. */
1546 char *hurl = url_string (u, 1);
1550 sprintf (tmp, _("(try:%2d)"), count);
1551 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1552 tms, hurl, tmp, locf);
1554 ws_changetitle (hurl, 1);
1559 /* Default document type is empty. However, if spider mode is
1560 on or time-stamping is employed, HEAD_ONLY commands is
1561 encoded within *dt. */
1562 if (opt.spider || (use_ts && !got_head))
1566 /* Assume no restarting. */
1568 /* Decide whether or not to restart. */
1569 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1570 /* #### this calls access() and then stat(); could be optimized. */
1571 && file_exists_p (locf))
1572 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1573 hstat.restval = st.st_size;
1575 /* In `-c' is used and the file is existing and non-empty,
1576 refuse to truncate it if the server doesn't support continued
1578 hstat.no_truncate = 0;
1579 if (opt.always_rest && hstat.restval)
1580 hstat.no_truncate = 1;
1582 /* Decide whether to send the no-cache directive. We send it in
1584 a) we're using a proxy, and we're past our first retrieval.
1585 Some proxies are notorious for caching incomplete data, so
1586 we require a fresh get.
1587 b) caching is explicitly inhibited. */
1588 if ((proxy && count > 1) /* a */
1589 || !opt.allow_cache /* b */
1591 *dt |= SEND_NOCACHE;
1593 *dt &= ~SEND_NOCACHE;
1595 /* Try fetching the document, or at least its head. */
1596 err = gethttp (u, &hstat, dt, proxy);
1598 /* It's unfortunate that wget determines the local filename before finding
1599 out the Content-Type of the file. Barring a major restructuring of the
1600 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1601 *hstat.local_file to tack on ".html". */
1602 if (!opt.output_document)
1603 locf = *hstat.local_file;
1605 locf = opt.output_document;
1608 tms = time_str (NULL);
1609 /* Get the new location (with or without the redirection). */
1611 *newloc = xstrdup (hstat.newloc);
1614 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1615 case CONERROR: case READERR: case WRITEFAILED:
1617 /* Non-fatal errors continue executing the loop, which will
1618 bring them to "while" statement at the end, to judge
1619 whether the number of tries was exceeded. */
1620 free_hstat (&hstat);
1621 printwhat (count, opt.ntry);
1624 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1625 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1626 /* Fatal errors just return from the function. */
1627 free_hstat (&hstat);
1631 case FWRITEERR: case FOPENERR:
1632 /* Another fatal error. */
1633 logputs (LOG_VERBOSE, "\n");
1634 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1635 *hstat.local_file, strerror (errno));
1636 free_hstat (&hstat);
1641 /* Another fatal error. */
1642 logputs (LOG_VERBOSE, "\n");
1643 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1644 free_hstat (&hstat);
1649 /* Return the new location to the caller. */
1652 logprintf (LOG_NOTQUIET,
1653 _("ERROR: Redirection (%d) without location.\n"),
1655 free_hstat (&hstat);
1659 free_hstat (&hstat);
1664 /* The file was already fully retrieved. */
1665 free_hstat (&hstat);
1670 /* Deal with you later. */
1673 /* All possibilities should have been exhausted. */
1676 if (!(*dt & RETROKF))
1680 /* #### Ugly ugly ugly! */
1681 char *hurl = url_string (u, 1);
1682 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1685 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1686 tms, hstat.statcode, hstat.error);
1687 logputs (LOG_VERBOSE, "\n");
1688 free_hstat (&hstat);
1693 /* Did we get the time-stamp? */
1696 if (opt.timestamping && !hstat.remote_time)
1698 logputs (LOG_NOTQUIET, _("\
1699 Last-modified header missing -- time-stamps turned off.\n"));
1701 else if (hstat.remote_time)
1703 /* Convert the date-string into struct tm. */
1704 tmr = http_atotm (hstat.remote_time);
1705 if (tmr == (time_t) (-1))
1706 logputs (LOG_VERBOSE, _("\
1707 Last-modified header invalid -- time-stamp ignored.\n"));
1711 /* The time-stamping section. */
1716 use_ts = 0; /* no more time-stamping */
1717 count = 0; /* the retrieve count for HEAD is
1719 if (hstat.remote_time && tmr != (time_t) (-1))
1721 /* Now time-stamping can be used validly. Time-stamping
1722 means that if the sizes of the local and remote file
1723 match, and local file is newer than the remote file,
1724 it will not be retrieved. Otherwise, the normal
1725 download procedure is resumed. */
1727 (hstat.contlen == -1 || local_size == hstat.contlen))
1729 logprintf (LOG_VERBOSE, _("\
1730 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1732 free_hstat (&hstat);
1736 else if (tml >= tmr)
1737 logprintf (LOG_VERBOSE, _("\
1738 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1740 logputs (LOG_VERBOSE,
1741 _("Remote file is newer, retrieving.\n"));
1743 free_hstat (&hstat);
1746 if ((tmr != (time_t) (-1))
1748 && ((hstat.len == hstat.contlen) ||
1749 ((hstat.res == 0) &&
1750 ((hstat.contlen == -1) ||
1751 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1753 /* #### This code repeats in http.c and ftp.c. Move it to a
1755 const char *fl = NULL;
1756 if (opt.output_document)
1758 if (opt.od_known_regular)
1759 fl = opt.output_document;
1762 fl = *hstat.local_file;
1766 /* End of time-stamping section. */
1770 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1775 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1777 if (hstat.len == hstat.contlen)
1781 logprintf (LOG_VERBOSE,
1782 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1783 tms, tmrate, locf, hstat.len, hstat.contlen);
1784 logprintf (LOG_NONVERBOSE,
1785 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1786 tms, u->url, hstat.len, hstat.contlen, locf, count);
1789 downloaded_increase (hstat.len);
1791 /* Remember that we downloaded the file for later ".orig" code. */
1792 if (*dt & ADDED_HTML_EXTENSION)
1793 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1795 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1797 free_hstat (&hstat);
1801 else if (hstat.res == 0) /* No read error */
1803 if (hstat.contlen == -1) /* We don't know how much we were supposed
1804 to get, so assume we succeeded. */
1808 logprintf (LOG_VERBOSE,
1809 _("%s (%s) - `%s' saved [%ld]\n\n"),
1810 tms, tmrate, locf, hstat.len);
1811 logprintf (LOG_NONVERBOSE,
1812 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1813 tms, u->url, hstat.len, locf, count);
1816 downloaded_increase (hstat.len);
1818 /* Remember that we downloaded the file for later ".orig" code. */
1819 if (*dt & ADDED_HTML_EXTENSION)
1820 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1822 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1824 free_hstat (&hstat);
1828 else if (hstat.len < hstat.contlen) /* meaning we lost the
1829 connection too soon */
1831 logprintf (LOG_VERBOSE,
1832 _("%s (%s) - Connection closed at byte %ld. "),
1833 tms, tmrate, hstat.len);
1834 printwhat (count, opt.ntry);
1835 free_hstat (&hstat);
1838 else if (!opt.kill_longer) /* meaning we got more than expected */
1840 logprintf (LOG_VERBOSE,
1841 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1842 tms, tmrate, locf, hstat.len, hstat.contlen);
1843 logprintf (LOG_NONVERBOSE,
1844 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1845 tms, u->url, hstat.len, hstat.contlen, locf, count);
1847 downloaded_increase (hstat.len);
1849 /* Remember that we downloaded the file for later ".orig" code. */
1850 if (*dt & ADDED_HTML_EXTENSION)
1851 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1853 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1855 free_hstat (&hstat);
1859 else /* the same, but not accepted */
1861 logprintf (LOG_VERBOSE,
1862 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1863 tms, tmrate, hstat.len, hstat.contlen);
1864 printwhat (count, opt.ntry);
1865 free_hstat (&hstat);
1869 else /* now hstat.res can only be -1 */
1871 if (hstat.contlen == -1)
1873 logprintf (LOG_VERBOSE,
1874 _("%s (%s) - Read error at byte %ld (%s)."),
1875 tms, tmrate, hstat.len, strerror (errno));
1876 printwhat (count, opt.ntry);
1877 free_hstat (&hstat);
1880 else /* hstat.res == -1 and contlen is given */
1882 logprintf (LOG_VERBOSE,
1883 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1884 tms, tmrate, hstat.len, hstat.contlen,
1886 printwhat (count, opt.ntry);
1887 free_hstat (&hstat);
1894 while (!opt.ntry || (count < opt.ntry));
1898 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1899 than local timezone.
1901 mktime is similar but assumes struct tm, also known as the
1902 "broken-down" form of time, is in local time zone. mktime_from_utc
1903 uses mktime to make the conversion understanding that an offset
1904 will be introduced by the local time assumption.
1906 mktime_from_utc then measures the introduced offset by applying
1907 gmtime to the initial result and applying mktime to the resulting
1908 "broken-down" form. The difference between the two mktime results
1909 is the measured offset which is then subtracted from the initial
1910 mktime result to yield a calendar time which is the value returned.
1912 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1913 consistent offset (the non DST offset) since tm and tm+o might be
1914 on opposite sides of a DST change.
1916 Some implementations of mktime return -1 for the nonexistent
1917 localtime hour at the beginning of DST. In this event, use
1918 mktime(tm - 1hr) + 3600.
1922 gmtime(t+o) --> tm+o
1923 mktime(tm+o) --> t+2o
1924 t+o - (t+2o - t+o) = t
1926 Note that glibc contains a function of the same purpose named
1927 `timegm' (reverse of gmtime). But obviously, it is not universally
1928 available, and unfortunately it is not straightforwardly
1929 extractable for use here. Perhaps configure should detect timegm
1930 and use it where available.
1932 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1933 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1934 Further improved by Roger with assistance from Edward J. Sabol
1935 based on input by Jamie Zawinski. */
1938 mktime_from_utc (struct tm *t)
1949 return -1; /* can't deal with output from strptime */
1960 return -1; /* can't deal with output from gmtime */
1963 return (tl - (tb - tl));
1966 /* Check whether the result of strptime() indicates success.
1967 strptime() returns the pointer to how far it got to in the string.
1968 The processing has been successful if the string is at `GMT' or
1969 `+X', or at the end of the string.
1971 In extended regexp parlance, the function returns 1 if P matches
1972 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1973 can return) is considered a failure and 0 is returned. */
1975 check_end (const char *p)
1979 while (ISSPACE (*p))
1982 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1983 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1989 /* Convert the textual specification of time in TIME_STRING to the
1990 number of seconds since the Epoch.
1992 TIME_STRING can be in any of the three formats RFC2068 allows the
1993 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1994 Timezones are ignored, and should be GMT.
1996 Return the computed time_t representation, or -1 if the conversion
1999 This function uses strptime with various string formats for parsing
2000 TIME_STRING. This results in a parser that is not as lenient in
2001 interpreting TIME_STRING as I would like it to be. Being based on
2002 strptime, it always allows shortened months, one-digit days, etc.,
2003 but due to the multitude of formats in which time can be
2004 represented, an ideal HTTP time parser would be even more
2005 forgiving. It should completely ignore things like week days and
2006 concentrate only on the various forms of representing years,
2007 months, days, hours, minutes, and seconds. For example, it would
2008 be nice if it accepted ISO 8601 out of the box.
2010 I've investigated free and PD code for this purpose, but none was
2011 usable. getdate was big and unwieldy, and had potential copyright
2012 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2013 distributed with phttpd, is excellent, but we cannot use it because
2014 it is not assigned to the FSF. So I stuck it with strptime. */
2017 http_atotm (char *time_string)
2019 /* NOTE: Solaris strptime man page claims that %n and %t match white
2020 space, but that's not universally available. Instead, we simply
2021 use ` ' to mean "skip all WS", which works under all strptime
2022 implementations I've tested. */
2024 static const char *time_formats[] = {
2025 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2026 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2027 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2028 (google.com uses this for their cookies.) */
2029 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2035 /* According to Roger Beeman, we need to initialize tm_isdst, since
2036 strptime won't do it. */
2039 /* Note that under foreign locales Solaris strptime() fails to
2040 recognize English dates, which renders this function useless. We
2041 solve this by being careful not to affect LC_TIME when
2042 initializing locale.
2044 Another solution would be to temporarily set locale to C, invoke
2045 strptime(), and restore it back. This is slow and dirty,
2046 however, and locale support other than LC_MESSAGES can mess other
2047 things, so I rather chose to stick with just setting LC_MESSAGES.
2049 GNU strptime does not have this problem because it recognizes
2050 both international and local dates. */
2052 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2053 if (check_end (strptime (time_string, time_formats[i], &t)))
2054 return mktime_from_utc (&t);
2056 /* All formats have failed. */
2060 /* Authorization support: We support two authorization schemes:
2062 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2064 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2065 consisting of answering to the server's challenge with the proper
2068 /* How many bytes it will take to store LEN bytes in base64. */
2069 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2071 /* Encode the string S of length LENGTH to base64 format and place it
2072 to STORE. STORE will be 0-terminated, and must point to a writable
2073 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2075 base64_encode (const char *s, char *store, int length)
2077 /* Conversion table. */
2078 static char tbl[64] = {
2079 'A','B','C','D','E','F','G','H',
2080 'I','J','K','L','M','N','O','P',
2081 'Q','R','S','T','U','V','W','X',
2082 'Y','Z','a','b','c','d','e','f',
2083 'g','h','i','j','k','l','m','n',
2084 'o','p','q','r','s','t','u','v',
2085 'w','x','y','z','0','1','2','3',
2086 '4','5','6','7','8','9','+','/'
2089 unsigned char *p = (unsigned char *)store;
2091 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2092 for (i = 0; i < length; i += 3)
2094 *p++ = tbl[s[0] >> 2];
2095 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2096 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2097 *p++ = tbl[s[2] & 0x3f];
2100 /* Pad the result if necessary... */
2101 if (i == length + 1)
2103 else if (i == length + 2)
2104 *(p - 1) = *(p - 2) = '=';
2105 /* ...and zero-terminate it. */
2109 /* Create the authentication header contents for the `Basic' scheme.
2110 This is done by encoding the string `USER:PASS' in base64 and
2111 prepending `HEADER: Basic ' to it. */
2113 basic_authentication_encode (const char *user, const char *passwd,
2116 char *t1, *t2, *res;
2117 int len1 = strlen (user) + 1 + strlen (passwd);
2118 int len2 = BASE64_LENGTH (len1);
2120 t1 = (char *)alloca (len1 + 1);
2121 sprintf (t1, "%s:%s", user, passwd);
2122 t2 = (char *)alloca (1 + len2);
2123 base64_encode (t1, t2, len1);
2124 res = (char *)xmalloc (len2 + 11 + strlen (header));
2125 sprintf (res, "%s: Basic %s\r\n", header, t2);
2131 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2132 of a field in such a header. If the field is the one specified by
2133 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2134 digest authorization code), extract its value in the (char*)
2135 variable pointed by RET. Returns negative on a malformed header,
2136 or number of bytes that have been parsed by this call. */
2138 extract_header_attr (const char *au, const char *attr_name, char **ret)
2140 const char *cp, *ep;
2144 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2146 cp += strlen (attr_name);
2149 cp += skip_lws (cp);
2154 cp += skip_lws (cp);
2159 for (ep = cp; *ep && *ep != '\"'; ep++)
2164 *ret = strdupdelim (cp, ep);
2171 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2172 an array of 16 bytes containing the hash keys, and BUF should be a
2173 buffer of 33 writable characters (32 for hex digits plus one for
2174 zero termination). */
2176 dump_hash (unsigned char *buf, const unsigned char *hash)
2180 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2182 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2183 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2188 /* Take the line apart to find the challenge, and compose a digest
2189 authorization header. See RFC2069 section 2.1.2. */
2191 digest_authentication_encode (const char *au, const char *user,
2192 const char *passwd, const char *method,
2195 static char *realm, *opaque, *nonce;
2200 { "realm", &realm },
2201 { "opaque", &opaque },
2206 realm = opaque = nonce = NULL;
2208 au += 6; /* skip over `Digest' */
2213 au += skip_lws (au);
2214 for (i = 0; i < ARRAY_SIZE (options); i++)
2216 int skip = extract_header_attr (au, options[i].name,
2217 options[i].variable);
2221 FREE_MAYBE (opaque);
2231 if (i == ARRAY_SIZE (options))
2233 while (*au && *au != '=')
2237 au += skip_lws (au);
2241 while (*au && *au != '\"')
2248 while (*au && *au != ',')
2253 if (!realm || !nonce || !user || !passwd || !path || !method)
2256 FREE_MAYBE (opaque);
2261 /* Calculate the digest value. */
2263 ALLOCA_MD5_CONTEXT (ctx);
2264 unsigned char hash[MD5_HASHLEN];
2265 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2266 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2268 /* A1BUF = H(user ":" realm ":" password) */
2270 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2271 gen_md5_update ((unsigned char *)":", 1, ctx);
2272 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2273 gen_md5_update ((unsigned char *)":", 1, ctx);
2274 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2275 gen_md5_finish (ctx, hash);
2276 dump_hash (a1buf, hash);
2278 /* A2BUF = H(method ":" path) */
2280 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2281 gen_md5_update ((unsigned char *)":", 1, ctx);
2282 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2283 gen_md5_finish (ctx, hash);
2284 dump_hash (a2buf, hash);
2286 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2288 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2289 gen_md5_update ((unsigned char *)":", 1, ctx);
2290 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2291 gen_md5_update ((unsigned char *)":", 1, ctx);
2292 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2293 gen_md5_finish (ctx, hash);
2294 dump_hash (response_digest, hash);
2296 res = (char*) xmalloc (strlen (user)
2301 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2302 + (opaque ? strlen (opaque) : 0)
2304 sprintf (res, "Authorization: Digest \
2305 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2306 user, realm, nonce, path, response_digest);
2309 char *p = res + strlen (res);
2310 strcat (p, ", opaque=\"");
2314 strcat (res, "\r\n");
2318 #endif /* USE_DIGEST */
2321 #define BEGINS_WITH(line, string_constant) \
2322 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2323 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2324 || !line[sizeof (string_constant) - 1]))
2327 known_authentication_scheme_p (const char *au)
2329 return BEGINS_WITH (au, "Basic")
2330 || BEGINS_WITH (au, "Digest")
2331 || BEGINS_WITH (au, "NTLM");
2336 /* Create the HTTP authorization request header. When the
2337 `WWW-Authenticate' response header is seen, according to the
2338 authorization scheme specified in that header (`Basic' and `Digest'
2339 are supported by the current implementation), produce an
2340 appropriate HTTP authorization request header. */
2342 create_authorization_line (const char *au, const char *user,
2343 const char *passwd, const char *method,
2346 char *wwwauth = NULL;
2348 if (!strncasecmp (au, "Basic", 5))
2349 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2350 if (!strncasecmp (au, "NTLM", 4))
2351 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2353 else if (!strncasecmp (au, "Digest", 6))
2354 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2355 #endif /* USE_DIGEST */
2362 if (pc_last_host_ip)
2363 address_list_release (pc_last_host_ip);