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>
50 # include <netdb.h> /* for h_errno */
64 # include "gen_sslfunc.h"
68 extern char *version_string;
79 static int cookies_loaded_p;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 hdr += skip_lws (hdr);
211 for (num = 0; ISDIGIT (*hdr); hdr++)
212 num = 10 * num + (*hdr - '0');
213 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
215 closure->first_byte_pos = num;
217 for (num = 0; ISDIGIT (*hdr); hdr++)
218 num = 10 * num + (*hdr - '0');
219 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
221 closure->last_byte_pos = num;
223 for (num = 0; ISDIGIT (*hdr); hdr++)
224 num = 10 * num + (*hdr - '0');
225 closure->entity_length = num;
229 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
230 Used for `Accept-Ranges'. */
232 http_process_none (const char *hdr, void *arg)
234 int *where = (int *)arg;
236 if (strstr (hdr, "none"))
243 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
245 http_process_type (const char *hdr, void *arg)
247 char **result = (char **)arg;
248 /* Locate P on `;' or the terminating zero, whichever comes first. */
249 const char *p = strchr (hdr, ';');
251 p = hdr + strlen (hdr);
252 while (p > hdr && ISSPACE (*(p - 1)))
254 *result = strdupdelim (hdr, p);
258 /* Check whether the `Connection' header is set to "keep-alive". */
260 http_process_connection (const char *hdr, void *arg)
262 int *flag = (int *)arg;
263 if (!strcasecmp (hdr, "Keep-Alive"))
268 /* Persistent connections. Currently, we cache the most recently used
269 connection as persistent, provided that the HTTP server agrees to
270 make it such. The persistence data is stored in the variables
271 below. Ideally, it would be in a structure, and it should be
272 possible to cache an arbitrary fixed number of these connections.
274 I think the code is quite easy to extend in that direction. */
276 /* Whether a persistent connection is active. */
277 static int pc_active_p;
278 /* Host and port of currently active persistent connection. */
279 static struct address_list *pc_last_host_ip;
280 static unsigned short pc_last_port;
282 /* File descriptor of the currently active persistent connection. */
283 static int pc_last_fd;
286 /* Whether a ssl handshake has occoured on this connection */
287 static int pc_active_ssl;
288 /* SSL connection of the currently active persistent connection. */
289 static SSL *pc_last_ssl;
290 #endif /* HAVE_SSL */
292 /* Mark the persistent connection as invalid. This is used by the
293 CLOSE_* macros after they forcefully close a registered persistent
294 connection. This does not close the file descriptor -- it is left
295 to the caller to do that. (Maybe it should, though.) */
298 invalidate_persistent (void)
303 #endif /* HAVE_SSL */
304 if (pc_last_host_ip != NULL)
306 address_list_release (pc_last_host_ip);
307 pc_last_host_ip = NULL;
309 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
312 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
313 persistent. This will enable someone to use the same connection
314 later. In the context of HTTP, this must be called only AFTER the
315 response has been received and the server has promised that the
316 connection will remain alive.
318 If a previous connection was persistent, it is closed. */
321 register_persistent (const char *host, unsigned short port, int fd
329 if (pc_last_fd == fd)
331 /* The connection FD is already registered. Nothing to
337 /* The old persistent connection is still active; let's
338 close it first. This situation arises whenever a
339 persistent connection exists, but we then connect to a
340 different host, and try to register a persistent
341 connection to that one. */
343 /* The ssl disconnect has to take place before the closing
346 shutdown_ssl(pc_last_ssl);
349 invalidate_persistent ();
353 assert (pc_last_host_ip == NULL);
355 /* This lookup_host cannot fail, because it has the results in the
357 pc_last_host_ip = lookup_host (host, 1);
358 assert (pc_last_host_ip != NULL);
365 pc_active_ssl = ssl ? 1 : 0;
367 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
370 /* Return non-zero if a persistent connection is available for
371 connecting to HOST:PORT. */
374 persistent_available_p (const char *host, unsigned short port
381 struct address_list *this_host_ip;
383 /* First, check whether a persistent connection is active at all. */
386 /* Second, check if the active connection pertains to the correct
387 (HOST, PORT) ordered pair. */
388 if (port != pc_last_port)
392 /* Second, a): check if current connection is (not) ssl, too. This
393 test is unlikely to fail because HTTP and HTTPS typicaly use
394 different ports. Yet it is possible, or so I [Christian
395 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
397 if (ssl != pc_active_ssl)
399 #endif /* HAVE_SSL */
401 this_host_ip = lookup_host (host, 1);
405 /* To equate the two host names for the purposes of persistent
406 connections, they need to share all the IP addresses in the
408 success = address_list_match_all (pc_last_host_ip, this_host_ip);
409 address_list_release (this_host_ip);
413 /* Third: check whether the connection is still open. This is
414 important because most server implement a liberal (short) timeout
415 on persistent connections. Wget can of course always reconnect
416 if the connection doesn't work out, but it's nicer to know in
417 advance. This test is a logical followup of the first test, but
418 is "expensive" and therefore placed at the end of the list. */
419 if (!test_socket_open (pc_last_fd))
421 /* Oops, the socket is no longer open. Now that we know that,
422 let's invalidate the persistent connection before returning
425 invalidate_persistent ();
432 # define SHUTDOWN_SSL(ssl) do { \
434 shutdown_ssl (ssl); \
437 # define SHUTDOWN_SSL(ssl)
440 /* The idea behind these two CLOSE macros is to distinguish between
441 two cases: one when the job we've been doing is finished, and we
442 want to close the connection and leave, and two when something is
443 seriously wrong and we're closing the connection as part of
446 In case of keep_alive, CLOSE_FINISH should leave the connection
447 open, while CLOSE_INVALIDATE should still close it.
449 Note that the semantics of the flag `keep_alive' is "this
450 connection *will* be reused (the server has promised not to close
451 the connection once we're done)", while the semantics of
452 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
453 active, registered connection". */
455 #define CLOSE_FINISH(fd) do { \
458 SHUTDOWN_SSL (ssl); \
460 if (pc_active_p && (fd) == pc_last_fd) \
461 invalidate_persistent (); \
465 #define CLOSE_INVALIDATE(fd) do { \
466 SHUTDOWN_SSL (ssl); \
468 if (pc_active_p && (fd) == pc_last_fd) \
469 invalidate_persistent (); \
474 long len; /* received length */
475 long contlen; /* expected length */
476 long restval; /* the restart value */
477 int res; /* the result of last read */
478 char *newloc; /* new location (redirection) */
479 char *remote_time; /* remote time-stamp string */
480 char *error; /* textual HTTP error */
481 int statcode; /* status code */
482 long dltime; /* time of the download */
483 int no_truncate; /* whether truncating the file is
485 const char *referer; /* value of the referer header. */
486 char **local_file; /* local file. */
490 free_hstat (struct http_stat *hs)
492 FREE_MAYBE (hs->newloc);
493 FREE_MAYBE (hs->remote_time);
494 FREE_MAYBE (hs->error);
496 /* Guard against being called twice. */
498 hs->remote_time = NULL;
502 static char *create_authorization_line PARAMS ((const char *, const char *,
503 const char *, const char *,
505 static char *basic_authentication_encode PARAMS ((const char *, const char *,
507 static int known_authentication_scheme_p PARAMS ((const char *));
509 time_t http_atotm PARAMS ((char *));
511 #define BEGINS_WITH(line, string_constant) \
512 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
513 && (ISSPACE (line[sizeof (string_constant) - 1]) \
514 || !line[sizeof (string_constant) - 1]))
516 /* Retrieve a document through HTTP protocol. It recognizes status
517 code, and correctly handles redirections. It closes the network
518 socket. If it receives an error from the functions below it, it
519 will print it if there is enough information to do so (almost
520 always), returning the error to the caller (i.e. http_loop).
522 Various HTTP parameters are stored to hs. Although it parses the
523 response code correctly, it is not used in a sane way. The caller
526 If PROXY is non-NULL, the connection will be made to the proxy
527 server, and u->url will be requested. */
529 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
531 char *request, *type, *command, *full_path;
533 char *pragma_h, *referer, *useragent, *range, *wwwauth;
534 char *authenticate_h;
538 char *request_keep_alive;
539 int sock, hcount, num_written, all_length, statcode;
540 long contlen, contrange;
543 int auth_tried_already;
546 static SSL_CTX *ssl_ctx = NULL;
548 #endif /* HAVE_SSL */
549 struct wget_timer *timer;
550 char *cookies = NULL;
552 /* Whether this connection will be kept alive after the HTTP request
556 /* Flags that detect the two ways of specifying HTTP keep-alive
558 int http_keep_alive_1, http_keep_alive_2;
560 /* Whether keep-alive should be inhibited. */
561 int inhibit_keep_alive;
564 /* initialize ssl_ctx on first run */
567 err = init_ssl (&ssl_ctx);
572 case SSLERRCTXCREATE:
574 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
578 /* try without certfile */
579 logprintf (LOG_NOTQUIET,
580 _("Failed to load certificates from %s\n"),
583 logprintf (LOG_NOTQUIET,
584 _("Trying without the specified certificate\n"));
587 logprintf (LOG_NOTQUIET,
588 _("Failed to get certificate key from %s\n"),
591 logprintf (LOG_NOTQUIET,
592 _("Trying without the specified certificate\n"));
599 #endif /* HAVE_SSL */
601 if (!(*dt & HEAD_ONLY))
602 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
603 know the local filename so we can save to it. */
604 assert (*hs->local_file != NULL);
607 auth_tried_already = 0;
609 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
612 /* We need to come back here when the initial attempt to retrieve
613 without authorization header fails. (Expected to happen at least
614 for the Digest authorization scheme.) */
617 http_keep_alive_1 = http_keep_alive_2 = 0;
619 /* Initialize certain elements of struct http_stat. */
624 hs->remote_time = NULL;
627 /* If we're using a proxy, we will be connecting to the proxy
629 conn = proxy ? proxy : u;
631 /* First: establish the connection. */
632 if (inhibit_keep_alive
635 !persistent_available_p (conn->host, conn->port)
637 !persistent_available_p (conn->host, conn->port,
638 u->scheme == SCHEME_HTTPS)
639 #endif /* HAVE_SSL */
642 struct address_list *al = lookup_host (conn->host, 0);
645 set_connection_host_name (conn->host);
646 sock = connect_to_many (al, conn->port, 0);
647 set_connection_host_name (NULL);
648 address_list_release (al);
651 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
654 if (conn->scheme == SCHEME_HTTPS)
655 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
657 logputs (LOG_VERBOSE, "\n");
658 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
663 #endif /* HAVE_SSL */
667 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
668 conn->host, conn->port);
669 /* #### pc_last_fd should be accessed through an accessor
674 #endif /* HAVE_SSL */
675 DEBUGP (("Reusing fd %d.\n", sock));
678 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
683 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
684 sprintf (referer, "Referer: %s\r\n", hs->referer);
687 if (*dt & SEND_NOCACHE)
688 pragma_h = "Pragma: no-cache\r\n";
694 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
695 /* Gag me! Some servers (e.g. WebSitePro) have been known to
696 respond to the following `Range' format by generating a
697 multipart/x-byte-ranges MIME document! This MIME type was
698 present in an old draft of the byteranges specification.
699 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
700 only if multiple non-overlapping ranges are requested --
701 which Wget never does. */
702 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
707 STRDUP_ALLOCA (useragent, opt.useragent);
710 useragent = (char *)alloca (10 + strlen (version_string));
711 sprintf (useragent, "Wget/%s", version_string);
713 /* Construct the authentication, if userid is present. */
716 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
717 user = user ? user : opt.http_user;
718 passwd = passwd ? passwd : opt.http_passwd;
725 /* We have the username and the password, but haven't tried
726 any authorization yet. Let's see if the "Basic" method
727 works. If not, we'll come back here and construct a
728 proper authorization method with the right challenges.
730 If we didn't employ this kind of logic, every URL that
731 requires authorization would have to be processed twice,
732 which is very suboptimal and generates a bunch of false
733 "unauthorized" errors in the server log.
735 #### But this logic also has a serious problem when used
736 with stronger authentications: we *first* transmit the
737 username and the password in clear text, and *then*
738 attempt a stronger authentication scheme. That cannot be
739 right! We are only fortunate that almost everyone still
740 uses the `Basic' scheme anyway.
742 There should be an option to prevent this from happening,
743 for those who use strong authentication schemes and value
745 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
749 wwwauth = create_authorization_line (authenticate_h, user, passwd,
757 char *proxy_user, *proxy_passwd;
758 /* For normal username and password, URL components override
759 command-line/wgetrc parameters. With proxy authentication,
760 it's the reverse, because proxy URLs are normally the
761 "permanent" ones, so command-line args should take
763 if (opt.proxy_user && opt.proxy_passwd)
765 proxy_user = opt.proxy_user;
766 proxy_passwd = opt.proxy_passwd;
770 proxy_user = proxy->user;
771 proxy_passwd = proxy->passwd;
773 /* #### This does not appear right. Can't the proxy request,
774 say, `Digest' authentication? */
775 if (proxy_user && proxy_passwd)
776 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
777 "Proxy-Authorization");
780 /* String of the form :PORT. Used only for non-standard ports. */
782 if (u->port != scheme_default_port (u->scheme))
784 port_maybe = (char *)alloca (numdigit (u->port) + 2);
785 sprintf (port_maybe, ":%d", u->port);
788 if (!inhibit_keep_alive)
789 request_keep_alive = "Connection: Keep-Alive\r\n";
791 request_keep_alive = NULL;
794 cookies = build_cookies_request (u->host, u->port, u->path,
796 u->scheme == SCHEME_HTTPS
803 full_path = xstrdup (u->url);
805 full_path = url_full_path (u);
807 /* Allocate the memory for the request. */
808 request = (char *)alloca (strlen (command)
812 + (port_maybe ? strlen (port_maybe) : 0)
813 + strlen (HTTP_ACCEPT)
814 + (request_keep_alive
815 ? strlen (request_keep_alive) : 0)
816 + (referer ? strlen (referer) : 0)
817 + (cookies ? strlen (cookies) : 0)
818 + (wwwauth ? strlen (wwwauth) : 0)
819 + (proxyauth ? strlen (proxyauth) : 0)
820 + (range ? strlen (range) : 0)
822 + (opt.user_header ? strlen (opt.user_header) : 0)
824 /* Construct the request. */
830 %s%s%s%s%s%s%s%s\r\n",
833 port_maybe ? port_maybe : "",
835 request_keep_alive ? request_keep_alive : "",
836 referer ? referer : "",
837 cookies ? cookies : "",
838 wwwauth ? wwwauth : "",
839 proxyauth ? proxyauth : "",
842 opt.user_header ? opt.user_header : "");
843 DEBUGP (("---request begin---\n%s---request end---\n", request));
845 /* Free the temporary memory. */
846 FREE_MAYBE (wwwauth);
847 FREE_MAYBE (proxyauth);
848 FREE_MAYBE (cookies);
851 /* Send the request to server. */
853 if (u->scheme == SCHEME_HTTPS)
854 num_written = ssl_iwrite (ssl, request, strlen (request));
856 #endif /* HAVE_SSL */
857 num_written = iwrite (sock, request, strlen (request));
861 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
863 CLOSE_INVALIDATE (sock);
866 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
867 proxy ? "Proxy" : "HTTP");
868 contlen = contrange = -1;
873 /* Before reading anything, initialize the rbuf. */
874 rbuf_initialize (&rbuf, sock);
876 if (u->scheme == SCHEME_HTTPS)
880 #endif /* HAVE_SSL */
883 /* Header-fetching loop. */
891 /* Get the header. */
892 status = header_get (&rbuf, &hdr,
893 /* Disallow continuations for status line. */
894 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
896 /* Check for errors. */
897 if (status == HG_EOF && *hdr)
899 /* This used to be an unconditional error, but that was
900 somewhat controversial, because of a large number of
901 broken CGI's that happily "forget" to send the second EOL
902 before closing the connection of a HEAD request.
904 So, the deal is to check whether the header is empty
905 (*hdr is zero if it is); if yes, it means that the
906 previous header was fully retrieved, and that -- most
907 probably -- the request is complete. "...be liberal in
908 what you accept." Oh boy. */
909 logputs (LOG_VERBOSE, "\n");
910 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
913 FREE_MAYBE (all_headers);
914 CLOSE_INVALIDATE (sock);
917 else if (status == HG_ERROR)
919 logputs (LOG_VERBOSE, "\n");
920 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
924 FREE_MAYBE (all_headers);
925 CLOSE_INVALIDATE (sock);
929 /* If the headers are to be saved to a file later, save them to
931 if (opt.save_headers)
933 int lh = strlen (hdr);
934 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
935 memcpy (all_headers + all_length, hdr, lh);
937 all_headers[all_length++] = '\n';
938 all_headers[all_length] = '\0';
941 /* Check for status line. */
945 /* Parse the first line of server response. */
946 statcode = parse_http_status_line (hdr, &error);
947 hs->statcode = statcode;
948 /* Store the descriptive response. */
949 if (statcode == -1) /* malformed response */
951 /* A common reason for "malformed response" error is the
952 case when no data was actually received. Handle this
955 hs->error = xstrdup (_("No data received"));
957 hs->error = xstrdup (_("Malformed status line"));
962 hs->error = xstrdup (_("(no description)"));
964 hs->error = xstrdup (error);
972 if (opt.server_response)
973 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
975 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
981 /* Exit on empty header. */
988 /* Print the header if requested. */
989 if (opt.server_response && hcount != 1)
990 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
992 /* Try getting content-length. */
993 if (contlen == -1 && !opt.ignore_length)
994 if (header_process (hdr, "Content-Length", header_extract_number,
997 /* Try getting content-type. */
999 if (header_process (hdr, "Content-Type", http_process_type, &type))
1001 /* Try getting location. */
1003 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1005 /* Try getting last-modified. */
1006 if (!hs->remote_time)
1007 if (header_process (hdr, "Last-Modified", header_strdup,
1010 /* Try getting cookies. */
1012 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1014 /* Try getting www-authentication. */
1015 if (!authenticate_h)
1016 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1019 /* Check for accept-ranges header. If it contains the word
1020 `none', disable the ranges. */
1021 if (*dt & ACCEPTRANGES)
1024 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1027 *dt &= ~ACCEPTRANGES;
1031 /* Try getting content-range. */
1032 if (contrange == -1)
1034 struct http_process_range_closure closure;
1035 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1037 contrange = closure.first_byte_pos;
1041 /* Check for keep-alive related responses. */
1042 if (!inhibit_keep_alive)
1044 /* Check for the `Keep-Alive' header. */
1045 if (!http_keep_alive_1)
1047 if (header_process (hdr, "Keep-Alive", header_exists,
1048 &http_keep_alive_1))
1051 /* Check for `Connection: Keep-Alive'. */
1052 if (!http_keep_alive_2)
1054 if (header_process (hdr, "Connection", http_process_connection,
1055 &http_keep_alive_2))
1063 logputs (LOG_VERBOSE, "\n");
1066 && (http_keep_alive_1 || http_keep_alive_2))
1068 assert (inhibit_keep_alive == 0);
1072 /* The server has promised that it will not close the connection
1073 when we're done. This means that we can register it. */
1075 register_persistent (conn->host, conn->port, sock);
1077 register_persistent (conn->host, conn->port, sock, ssl);
1078 #endif /* HAVE_SSL */
1080 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1083 /* Authorization is required. */
1087 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1088 might be more bytes in the body. */
1089 if (auth_tried_already)
1091 /* If we have tried it already, then there is not point
1094 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1095 xfree (authenticate_h);
1098 else if (!known_authentication_scheme_p (authenticate_h))
1100 xfree (authenticate_h);
1101 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1104 else if (BEGINS_WITH (authenticate_h, "Basic"))
1106 /* The authentication scheme is basic, the one we try by
1107 default, and it failed. There's no sense in trying
1113 auth_tried_already = 1;
1117 /* We do not need this anymore. */
1120 xfree (authenticate_h);
1121 authenticate_h = NULL;
1124 /* 20x responses are counted among successful by default. */
1125 if (H_20X (statcode))
1128 /* Return if redirected. */
1129 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1131 /* RFC2068 says that in case of the 300 (multiple choices)
1132 response, the server can output a preferred URL through
1133 `Location' header; otherwise, the request should be treated
1134 like GET. So, if the location is set, it will be a
1135 redirection; otherwise, just proceed normally. */
1136 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1140 logprintf (LOG_VERBOSE,
1141 _("Location: %s%s\n"),
1142 hs->newloc ? hs->newloc : _("unspecified"),
1143 hs->newloc ? _(" [following]") : "");
1144 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1145 might be more bytes in the body. */
1147 FREE_MAYBE (all_headers);
1152 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1155 /* We don't assume text/html by default. */
1158 if (opt.html_extension && (*dt & TEXTHTML))
1159 /* -E / --html-extension / html_extension = on was specified, and this is a
1160 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1161 already the file's suffix, tack on ".html". */
1163 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1165 if (last_period_in_local_filename == NULL ||
1166 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1167 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1169 size_t local_filename_len = strlen(*hs->local_file);
1171 *hs->local_file = xrealloc(*hs->local_file,
1172 local_filename_len + sizeof(".html"));
1173 strcpy(*hs->local_file + local_filename_len, ".html");
1175 *dt |= ADDED_HTML_EXTENSION;
1179 if (contrange == -1)
1181 /* We did not get a content-range header. This means that the
1182 server did not honor our `Range' request. Normally, this
1183 means we should reset hs->restval and continue normally. */
1185 /* However, if `-c' is used, we need to be a bit more careful:
1187 1. If `-c' is specified and the file already existed when
1188 Wget was started, it would be a bad idea for us to start
1189 downloading it from scratch, effectively truncating it. I
1190 believe this cannot happen unless `-c' was specified.
1192 2. If `-c' is used on a file that is already fully
1193 downloaded, we're requesting bytes after the end of file,
1194 which can result in server not honoring `Range'. If this is
1195 the case, `Content-Length' will be equal to the length of the
1197 if (opt.always_rest)
1199 /* Check for condition #2. */
1200 if (hs->restval > 0 /* restart was requested. */
1201 && contlen != -1 /* we got content-length. */
1202 && hs->restval >= contlen /* file fully downloaded
1206 logputs (LOG_VERBOSE, _("\
1207 \n The file is already fully retrieved; nothing to do.\n\n"));
1208 /* In case the caller inspects. */
1212 FREE_MAYBE (all_headers);
1213 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1214 might be more bytes in the body. */
1215 return RETRUNNEEDED;
1218 /* Check for condition #1. */
1219 if (hs->no_truncate)
1221 logprintf (LOG_NOTQUIET,
1224 Continued download failed on this file, which conflicts with `-c'.\n\
1225 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1227 FREE_MAYBE (all_headers);
1228 CLOSE_INVALIDATE (sock);
1229 return CONTNOTSUPPORTED;
1237 else if (contrange != hs->restval ||
1238 (H_PARTIAL (statcode) && contrange == -1))
1240 /* This means the whole request was somehow misunderstood by the
1241 server. Bail out. */
1243 FREE_MAYBE (all_headers);
1244 CLOSE_INVALIDATE (sock);
1251 contlen += contrange;
1253 contrange = -1; /* If conent-length was not sent,
1254 content-range will be ignored. */
1256 hs->contlen = contlen;
1260 if ((*dt & RETROKF) && !opt.server_response)
1262 /* No need to print this output if the body won't be
1263 downloaded at all, or if the original server response is
1265 logputs (LOG_VERBOSE, _("Length: "));
1268 logputs (LOG_VERBOSE, legible (contlen));
1269 if (contrange != -1)
1270 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1271 legible (contlen - contrange));
1274 logputs (LOG_VERBOSE,
1275 opt.ignore_length ? _("ignored") : _("unspecified"));
1277 logprintf (LOG_VERBOSE, " [%s]\n", type);
1279 logputs (LOG_VERBOSE, "\n");
1283 type = NULL; /* We don't need it any more. */
1285 /* Return if we have no intention of further downloading. */
1286 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1288 /* In case the caller cares to look... */
1292 FREE_MAYBE (all_headers);
1293 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1294 might be more bytes in the body. */
1295 return RETRFINISHED;
1298 /* Open the local file. */
1301 mkalldirs (*hs->local_file);
1303 rotate_backups (*hs->local_file);
1304 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1307 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1308 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1309 might be more bytes in the body. */
1310 FREE_MAYBE (all_headers);
1316 extern int global_download_count;
1318 /* To ensure that repeated "from scratch" downloads work for -O
1319 files, we rewind the file pointer, unless restval is
1320 non-zero. (This works only when -O is used on regular files,
1321 but it's still a valuable feature.)
1323 However, this loses when more than one URL is specified on
1324 the command line the second rewinds eradicates the contents
1325 of the first download. Thus we disable the above trick for
1326 all the downloads except the very first one.
1328 #### A possible solution to this would be to remember the
1329 file position in the output document and to seek to that
1330 position, instead of rewinding. */
1331 if (!hs->restval && global_download_count == 0)
1333 /* This will silently fail for streams that don't correspond
1334 to regular files, but that's OK. */
1336 /* ftruncate is needed because opt.dfp is opened in append
1337 mode if opt.always_rest is set. */
1338 ftruncate (fileno (fp), 0);
1343 /* #### This confuses the code that checks for file size. There
1344 should be some overhead information. */
1345 if (opt.save_headers)
1346 fwrite (all_headers, 1, all_length, fp);
1347 timer = wtimer_new ();
1348 /* Get the contents of the document. */
1349 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1350 (contlen != -1 ? contlen : 0),
1352 hs->dltime = wtimer_elapsed (timer);
1353 wtimer_delete (timer);
1355 /* Close or flush the file. We have to be careful to check for
1356 error here. Checking the result of fwrite() is not enough --
1357 errors could go unnoticed! */
1360 flush_res = fclose (fp);
1362 flush_res = fflush (fp);
1363 if (flush_res == EOF)
1366 FREE_MAYBE (all_headers);
1367 CLOSE_FINISH (sock);
1370 return RETRFINISHED;
1373 /* The genuine HTTP loop! This is the part where the retrieval is
1374 retried, and retried, and retried, and... */
1376 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1377 int *dt, struct url *proxy)
1380 int use_ts, got_head = 0; /* time-stamping info */
1381 char *filename_plus_orig_suffix;
1382 char *local_filename = NULL;
1383 char *tms, *suf, *locf, *tmrate;
1385 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1386 long local_size = 0; /* the size of the local file */
1387 size_t filename_len;
1388 struct http_stat hstat; /* HTTP status */
1392 /* This used to be done in main(), but it's a better idea to do it
1393 here so that we don't go through the hoops if we're just using
1395 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1397 load_cookies (opt.cookies_input);
1398 cookies_loaded_p = 1;
1403 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1404 has_wildcards_p because it would also warn on `?', and we know that
1405 shows up in CGI paths a *lot*. */
1406 if (strchr (u->url, '*'))
1407 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1409 /* Determine the local filename. */
1410 if (local_file && *local_file)
1411 hstat.local_file = local_file;
1412 else if (local_file)
1414 *local_file = url_filename (u);
1415 hstat.local_file = local_file;
1419 dummy = url_filename (u);
1420 hstat.local_file = &dummy;
1423 if (!opt.output_document)
1424 locf = *hstat.local_file;
1426 locf = opt.output_document;
1428 hstat.referer = referer;
1430 filename_len = strlen (*hstat.local_file);
1431 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1433 if (opt.noclobber && file_exists_p (*hstat.local_file))
1435 /* If opt.noclobber is turned on and file already exists, do not
1436 retrieve the file */
1437 logprintf (LOG_VERBOSE, _("\
1438 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1439 /* If the file is there, we suppose it's retrieved OK. */
1442 /* #### Bogusness alert. */
1443 /* If its suffix is "html" or "htm", assume text/html. */
1444 if (((suf = suffix (*hstat.local_file)) != NULL)
1445 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1453 if (opt.timestamping)
1455 boolean local_dot_orig_file_exists = FALSE;
1457 if (opt.backup_converted)
1458 /* If -K is specified, we'll act on the assumption that it was specified
1459 last time these files were downloaded as well, and instead of just
1460 comparing local file X against server file X, we'll compare local
1461 file X.orig (if extant, else X) against server file X. If -K
1462 _wasn't_ specified last time, or the server contains files called
1463 *.orig, -N will be back to not operating correctly with -k. */
1465 /* Would a single s[n]printf() call be faster? --dan
1467 Definitely not. sprintf() is horribly slow. It's a
1468 different question whether the difference between the two
1469 affects a program. Usually I'd say "no", but at one
1470 point I profiled Wget, and found that a measurable and
1471 non-negligible amount of time was lost calling sprintf()
1472 in url.c. Replacing sprintf with inline calls to
1473 strcpy() and long_to_string() made a difference.
1475 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1476 memcpy (filename_plus_orig_suffix + filename_len,
1477 ".orig", sizeof (".orig"));
1479 /* Try to stat() the .orig file. */
1480 if (stat (filename_plus_orig_suffix, &st) == 0)
1482 local_dot_orig_file_exists = TRUE;
1483 local_filename = filename_plus_orig_suffix;
1487 if (!local_dot_orig_file_exists)
1488 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1489 if (stat (*hstat.local_file, &st) == 0)
1490 local_filename = *hstat.local_file;
1492 if (local_filename != NULL)
1493 /* There was a local file, so we'll check later to see if the version
1494 the server has is the same version we already have, allowing us to
1499 local_size = st.st_size;
1503 /* Reset the counter. */
1505 *dt = 0 | ACCEPTRANGES;
1509 /* Increment the pass counter. */
1511 sleep_between_retrievals (count);
1512 /* Get the current time string. */
1513 tms = time_str (NULL);
1514 /* Print fetch message, if opt.verbose. */
1517 char *hurl = url_string (u, 1);
1521 sprintf (tmp, _("(try:%2d)"), count);
1522 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1523 tms, hurl, tmp, locf);
1525 ws_changetitle (hurl, 1);
1530 /* Default document type is empty. However, if spider mode is
1531 on or time-stamping is employed, HEAD_ONLY commands is
1532 encoded within *dt. */
1533 if (opt.spider || (use_ts && !got_head))
1537 /* Assume no restarting. */
1539 /* Decide whether or not to restart. */
1540 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1541 /* #### this calls access() and then stat(); could be optimized. */
1542 && file_exists_p (locf))
1543 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1544 hstat.restval = st.st_size;
1546 /* In `-c' is used and the file is existing and non-empty,
1547 refuse to truncate it if the server doesn't support continued
1549 hstat.no_truncate = 0;
1550 if (opt.always_rest && hstat.restval)
1551 hstat.no_truncate = 1;
1553 /* Decide whether to send the no-cache directive. We send it in
1555 a) we're using a proxy, and we're past our first retrieval.
1556 Some proxies are notorious for caching incomplete data, so
1557 we require a fresh get.
1558 b) caching is explicitly inhibited. */
1559 if ((proxy && count > 1) /* a */
1560 || !opt.allow_cache /* b */
1562 *dt |= SEND_NOCACHE;
1564 *dt &= ~SEND_NOCACHE;
1566 /* Try fetching the document, or at least its head. */
1567 err = gethttp (u, &hstat, dt, proxy);
1569 /* It's unfortunate that wget determines the local filename before finding
1570 out the Content-Type of the file. Barring a major restructuring of the
1571 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1572 *hstat.local_file to tack on ".html". */
1573 if (!opt.output_document)
1574 locf = *hstat.local_file;
1576 locf = opt.output_document;
1579 tms = time_str (NULL);
1580 /* Get the new location (with or without the redirection). */
1582 *newloc = xstrdup (hstat.newloc);
1585 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1586 case CONERROR: case READERR: case WRITEFAILED:
1588 /* Non-fatal errors continue executing the loop, which will
1589 bring them to "while" statement at the end, to judge
1590 whether the number of tries was exceeded. */
1591 free_hstat (&hstat);
1592 printwhat (count, opt.ntry);
1595 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1596 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1597 /* Fatal errors just return from the function. */
1598 free_hstat (&hstat);
1602 case FWRITEERR: case FOPENERR:
1603 /* Another fatal error. */
1604 logputs (LOG_VERBOSE, "\n");
1605 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1606 *hstat.local_file, strerror (errno));
1607 free_hstat (&hstat);
1612 /* Another fatal error. */
1613 logputs (LOG_VERBOSE, "\n");
1614 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1615 free_hstat (&hstat);
1620 /* Return the new location to the caller. */
1623 logprintf (LOG_NOTQUIET,
1624 _("ERROR: Redirection (%d) without location.\n"),
1626 free_hstat (&hstat);
1630 free_hstat (&hstat);
1635 /* The file was already fully retrieved. */
1636 free_hstat (&hstat);
1641 /* Deal with you later. */
1644 /* All possibilities should have been exhausted. */
1647 if (!(*dt & RETROKF))
1651 /* #### Ugly ugly ugly! */
1652 char *hurl = url_string (u, 1);
1653 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1656 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1657 tms, hstat.statcode, hstat.error);
1658 logputs (LOG_VERBOSE, "\n");
1659 free_hstat (&hstat);
1664 /* Did we get the time-stamp? */
1667 if (opt.timestamping && !hstat.remote_time)
1669 logputs (LOG_NOTQUIET, _("\
1670 Last-modified header missing -- time-stamps turned off.\n"));
1672 else if (hstat.remote_time)
1674 /* Convert the date-string into struct tm. */
1675 tmr = http_atotm (hstat.remote_time);
1676 if (tmr == (time_t) (-1))
1677 logputs (LOG_VERBOSE, _("\
1678 Last-modified header invalid -- time-stamp ignored.\n"));
1682 /* The time-stamping section. */
1687 use_ts = 0; /* no more time-stamping */
1688 count = 0; /* the retrieve count for HEAD is
1690 if (hstat.remote_time && tmr != (time_t) (-1))
1692 /* Now time-stamping can be used validly. Time-stamping
1693 means that if the sizes of the local and remote file
1694 match, and local file is newer than the remote file,
1695 it will not be retrieved. Otherwise, the normal
1696 download procedure is resumed. */
1698 (hstat.contlen == -1 || local_size == hstat.contlen))
1700 logprintf (LOG_VERBOSE, _("\
1701 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1703 free_hstat (&hstat);
1707 else if (tml >= tmr)
1708 logprintf (LOG_VERBOSE, _("\
1709 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1711 logputs (LOG_VERBOSE,
1712 _("Remote file is newer, retrieving.\n"));
1714 free_hstat (&hstat);
1717 if ((tmr != (time_t) (-1))
1719 && ((hstat.len == hstat.contlen) ||
1720 ((hstat.res == 0) &&
1721 ((hstat.contlen == -1) ||
1722 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1724 /* #### This code repeats in http.c and ftp.c. Move it to a
1726 const char *fl = NULL;
1727 if (opt.output_document)
1729 if (opt.od_known_regular)
1730 fl = opt.output_document;
1733 fl = *hstat.local_file;
1737 /* End of time-stamping section. */
1741 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1746 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1748 if (hstat.len == hstat.contlen)
1752 logprintf (LOG_VERBOSE,
1753 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1754 tms, tmrate, locf, hstat.len, hstat.contlen);
1755 logprintf (LOG_NONVERBOSE,
1756 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1757 tms, u->url, hstat.len, hstat.contlen, locf, count);
1760 downloaded_increase (hstat.len);
1762 /* Remember that we downloaded the file for later ".orig" code. */
1763 if (*dt & ADDED_HTML_EXTENSION)
1764 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1766 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1768 free_hstat (&hstat);
1772 else if (hstat.res == 0) /* No read error */
1774 if (hstat.contlen == -1) /* We don't know how much we were supposed
1775 to get, so assume we succeeded. */
1779 logprintf (LOG_VERBOSE,
1780 _("%s (%s) - `%s' saved [%ld]\n\n"),
1781 tms, tmrate, locf, hstat.len);
1782 logprintf (LOG_NONVERBOSE,
1783 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1784 tms, u->url, hstat.len, locf, count);
1787 downloaded_increase (hstat.len);
1789 /* Remember that we downloaded the file for later ".orig" code. */
1790 if (*dt & ADDED_HTML_EXTENSION)
1791 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1793 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1795 free_hstat (&hstat);
1799 else if (hstat.len < hstat.contlen) /* meaning we lost the
1800 connection too soon */
1802 logprintf (LOG_VERBOSE,
1803 _("%s (%s) - Connection closed at byte %ld. "),
1804 tms, tmrate, hstat.len);
1805 printwhat (count, opt.ntry);
1806 free_hstat (&hstat);
1809 else if (!opt.kill_longer) /* meaning we got more than expected */
1811 logprintf (LOG_VERBOSE,
1812 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1813 tms, tmrate, locf, hstat.len, hstat.contlen);
1814 logprintf (LOG_NONVERBOSE,
1815 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1816 tms, u->url, hstat.len, hstat.contlen, locf, count);
1818 downloaded_increase (hstat.len);
1820 /* Remember that we downloaded the file for later ".orig" code. */
1821 if (*dt & ADDED_HTML_EXTENSION)
1822 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1824 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1826 free_hstat (&hstat);
1830 else /* the same, but not accepted */
1832 logprintf (LOG_VERBOSE,
1833 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1834 tms, tmrate, hstat.len, hstat.contlen);
1835 printwhat (count, opt.ntry);
1836 free_hstat (&hstat);
1840 else /* now hstat.res can only be -1 */
1842 if (hstat.contlen == -1)
1844 logprintf (LOG_VERBOSE,
1845 _("%s (%s) - Read error at byte %ld (%s)."),
1846 tms, tmrate, hstat.len, strerror (errno));
1847 printwhat (count, opt.ntry);
1848 free_hstat (&hstat);
1851 else /* hstat.res == -1 and contlen is given */
1853 logprintf (LOG_VERBOSE,
1854 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1855 tms, tmrate, hstat.len, hstat.contlen,
1857 printwhat (count, opt.ntry);
1858 free_hstat (&hstat);
1865 while (!opt.ntry || (count < opt.ntry));
1869 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1870 than local timezone.
1872 mktime is similar but assumes struct tm, also known as the
1873 "broken-down" form of time, is in local time zone. mktime_from_utc
1874 uses mktime to make the conversion understanding that an offset
1875 will be introduced by the local time assumption.
1877 mktime_from_utc then measures the introduced offset by applying
1878 gmtime to the initial result and applying mktime to the resulting
1879 "broken-down" form. The difference between the two mktime results
1880 is the measured offset which is then subtracted from the initial
1881 mktime result to yield a calendar time which is the value returned.
1883 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1884 consistent offset (the non DST offset) since tm and tm+o might be
1885 on opposite sides of a DST change.
1887 Some implementations of mktime return -1 for the nonexistent
1888 localtime hour at the beginning of DST. In this event, use
1889 mktime(tm - 1hr) + 3600.
1893 gmtime(t+o) --> tm+o
1894 mktime(tm+o) --> t+2o
1895 t+o - (t+2o - t+o) = t
1897 Note that glibc contains a function of the same purpose named
1898 `timegm' (reverse of gmtime). But obviously, it is not universally
1899 available, and unfortunately it is not straightforwardly
1900 extractable for use here. Perhaps configure should detect timegm
1901 and use it where available.
1903 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1904 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1905 Further improved by Roger with assistance from Edward J. Sabol
1906 based on input by Jamie Zawinski. */
1909 mktime_from_utc (struct tm *t)
1920 return -1; /* can't deal with output from strptime */
1931 return -1; /* can't deal with output from gmtime */
1934 return (tl - (tb - tl));
1937 /* Check whether the result of strptime() indicates success.
1938 strptime() returns the pointer to how far it got to in the string.
1939 The processing has been successful if the string is at `GMT' or
1940 `+X', or at the end of the string.
1942 In extended regexp parlance, the function returns 1 if P matches
1943 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1944 can return) is considered a failure and 0 is returned. */
1946 check_end (const char *p)
1950 while (ISSPACE (*p))
1953 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1954 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1960 /* Convert the textual specification of time in TIME_STRING to the
1961 number of seconds since the Epoch.
1963 TIME_STRING can be in any of the three formats RFC2068 allows the
1964 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1965 Timezones are ignored, and should be GMT.
1967 Return the computed time_t representation, or -1 if the conversion
1970 This function uses strptime with various string formats for parsing
1971 TIME_STRING. This results in a parser that is not as lenient in
1972 interpreting TIME_STRING as I would like it to be. Being based on
1973 strptime, it always allows shortened months, one-digit days, etc.,
1974 but due to the multitude of formats in which time can be
1975 represented, an ideal HTTP time parser would be even more
1976 forgiving. It should completely ignore things like week days and
1977 concentrate only on the various forms of representing years,
1978 months, days, hours, minutes, and seconds. For example, it would
1979 be nice if it accepted ISO 8601 out of the box.
1981 I've investigated free and PD code for this purpose, but none was
1982 usable. getdate was big and unwieldy, and had potential copyright
1983 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1984 distributed with phttpd, is excellent, but we cannot use it because
1985 it is not assigned to the FSF. So I stuck it with strptime. */
1988 http_atotm (char *time_string)
1990 /* NOTE: Solaris strptime man page claims that %n and %t match white
1991 space, but that's not universally available. Instead, we simply
1992 use ` ' to mean "skip all WS", which works under all strptime
1993 implementations I've tested. */
1995 static const char *time_formats[] = {
1996 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1997 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1998 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1999 (google.com uses this for their cookies.) */
2000 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2006 /* According to Roger Beeman, we need to initialize tm_isdst, since
2007 strptime won't do it. */
2010 /* Note that under foreign locales Solaris strptime() fails to
2011 recognize English dates, which renders this function useless. We
2012 solve this by being careful not to affect LC_TIME when
2013 initializing locale.
2015 Another solution would be to temporarily set locale to C, invoke
2016 strptime(), and restore it back. This is slow and dirty,
2017 however, and locale support other than LC_MESSAGES can mess other
2018 things, so I rather chose to stick with just setting LC_MESSAGES.
2020 GNU strptime does not have this problem because it recognizes
2021 both international and local dates. */
2023 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2024 if (check_end (strptime (time_string, time_formats[i], &t)))
2025 return mktime_from_utc (&t);
2027 /* All formats have failed. */
2031 /* Authorization support: We support two authorization schemes:
2033 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2035 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2036 consisting of answering to the server's challenge with the proper
2039 /* How many bytes it will take to store LEN bytes in base64. */
2040 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2042 /* Encode the string S of length LENGTH to base64 format and place it
2043 to STORE. STORE will be 0-terminated, and must point to a writable
2044 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2046 base64_encode (const char *s, char *store, int length)
2048 /* Conversion table. */
2049 static char tbl[64] = {
2050 'A','B','C','D','E','F','G','H',
2051 'I','J','K','L','M','N','O','P',
2052 'Q','R','S','T','U','V','W','X',
2053 'Y','Z','a','b','c','d','e','f',
2054 'g','h','i','j','k','l','m','n',
2055 'o','p','q','r','s','t','u','v',
2056 'w','x','y','z','0','1','2','3',
2057 '4','5','6','7','8','9','+','/'
2060 unsigned char *p = (unsigned char *)store;
2062 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2063 for (i = 0; i < length; i += 3)
2065 *p++ = tbl[s[0] >> 2];
2066 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2067 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2068 *p++ = tbl[s[2] & 0x3f];
2071 /* Pad the result if necessary... */
2072 if (i == length + 1)
2074 else if (i == length + 2)
2075 *(p - 1) = *(p - 2) = '=';
2076 /* ...and zero-terminate it. */
2080 /* Create the authentication header contents for the `Basic' scheme.
2081 This is done by encoding the string `USER:PASS' in base64 and
2082 prepending `HEADER: Basic ' to it. */
2084 basic_authentication_encode (const char *user, const char *passwd,
2087 char *t1, *t2, *res;
2088 int len1 = strlen (user) + 1 + strlen (passwd);
2089 int len2 = BASE64_LENGTH (len1);
2091 t1 = (char *)alloca (len1 + 1);
2092 sprintf (t1, "%s:%s", user, passwd);
2093 t2 = (char *)alloca (1 + len2);
2094 base64_encode (t1, t2, len1);
2095 res = (char *)xmalloc (len2 + 11 + strlen (header));
2096 sprintf (res, "%s: Basic %s\r\n", header, t2);
2102 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2103 of a field in such a header. If the field is the one specified by
2104 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2105 digest authorization code), extract its value in the (char*)
2106 variable pointed by RET. Returns negative on a malformed header,
2107 or number of bytes that have been parsed by this call. */
2109 extract_header_attr (const char *au, const char *attr_name, char **ret)
2111 const char *cp, *ep;
2115 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2117 cp += strlen (attr_name);
2120 cp += skip_lws (cp);
2125 cp += skip_lws (cp);
2130 for (ep = cp; *ep && *ep != '\"'; ep++)
2135 *ret = strdupdelim (cp, ep);
2142 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2143 an array of 16 bytes containing the hash keys, and BUF should be a
2144 buffer of 33 writable characters (32 for hex digits plus one for
2145 zero termination). */
2147 dump_hash (unsigned char *buf, const unsigned char *hash)
2151 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2153 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2154 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2159 /* Take the line apart to find the challenge, and compose a digest
2160 authorization header. See RFC2069 section 2.1.2. */
2162 digest_authentication_encode (const char *au, const char *user,
2163 const char *passwd, const char *method,
2166 static char *realm, *opaque, *nonce;
2171 { "realm", &realm },
2172 { "opaque", &opaque },
2177 realm = opaque = nonce = NULL;
2179 au += 6; /* skip over `Digest' */
2184 au += skip_lws (au);
2185 for (i = 0; i < ARRAY_SIZE (options); i++)
2187 int skip = extract_header_attr (au, options[i].name,
2188 options[i].variable);
2192 FREE_MAYBE (opaque);
2202 if (i == ARRAY_SIZE (options))
2204 while (*au && *au != '=')
2208 au += skip_lws (au);
2212 while (*au && *au != '\"')
2219 while (*au && *au != ',')
2224 if (!realm || !nonce || !user || !passwd || !path || !method)
2227 FREE_MAYBE (opaque);
2232 /* Calculate the digest value. */
2234 MD5_CONTEXT_TYPE ctx;
2235 unsigned char hash[MD5_HASHLEN];
2236 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2237 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2239 /* A1BUF = H(user ":" realm ":" password) */
2241 MD5_UPDATE (user, strlen (user), &ctx);
2242 MD5_UPDATE (":", 1, &ctx);
2243 MD5_UPDATE (realm, strlen (realm), &ctx);
2244 MD5_UPDATE (":", 1, &ctx);
2245 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2246 MD5_FINISH (&ctx, hash);
2247 dump_hash (a1buf, hash);
2249 /* A2BUF = H(method ":" path) */
2251 MD5_UPDATE (method, strlen (method), &ctx);
2252 MD5_UPDATE (":", 1, &ctx);
2253 MD5_UPDATE (path, strlen (path), &ctx);
2254 MD5_FINISH (&ctx, hash);
2255 dump_hash (a2buf, hash);
2257 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2259 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2260 MD5_UPDATE (":", 1, &ctx);
2261 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2262 MD5_UPDATE (":", 1, &ctx);
2263 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2264 MD5_FINISH (&ctx, hash);
2265 dump_hash (response_digest, hash);
2267 res = (char*) xmalloc (strlen (user)
2272 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2273 + (opaque ? strlen (opaque) : 0)
2275 sprintf (res, "Authorization: Digest \
2276 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2277 user, realm, nonce, path, response_digest);
2280 char *p = res + strlen (res);
2281 strcat (p, ", opaque=\"");
2285 strcat (res, "\r\n");
2289 #endif /* USE_DIGEST */
2292 #define BEGINS_WITH(line, string_constant) \
2293 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2294 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2295 || !line[sizeof (string_constant) - 1]))
2298 known_authentication_scheme_p (const char *au)
2300 return BEGINS_WITH (au, "Basic")
2301 || BEGINS_WITH (au, "Digest")
2302 || BEGINS_WITH (au, "NTLM");
2307 /* Create the HTTP authorization request header. When the
2308 `WWW-Authenticate' response header is seen, according to the
2309 authorization scheme specified in that header (`Basic' and `Digest'
2310 are supported by the current implementation), produce an
2311 appropriate HTTP authorization request header. */
2313 create_authorization_line (const char *au, const char *user,
2314 const char *passwd, const char *method,
2317 char *wwwauth = NULL;
2319 if (!strncasecmp (au, "Basic", 5))
2320 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2321 if (!strncasecmp (au, "NTLM", 4))
2322 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2324 else if (!strncasecmp (au, "Digest", 6))
2325 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2326 #endif /* USE_DIGEST */
2333 if (pc_last_host_ip)
2334 address_list_release (pc_last_host_ip);