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"
71 extern char *version_string;
82 static int cookies_loaded_p;
84 #define TEXTHTML_S "text/html"
85 #define HTTP_ACCEPT "*/*"
87 /* Some status code validation macros: */
88 #define H_20X(x) (((x) >= 200) && ((x) < 300))
89 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
90 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
91 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
93 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
95 #define HTTP_STATUS_OK 200
96 #define HTTP_STATUS_CREATED 201
97 #define HTTP_STATUS_ACCEPTED 202
98 #define HTTP_STATUS_NO_CONTENT 204
99 #define HTTP_STATUS_PARTIAL_CONTENTS 206
101 /* Redirection 3xx. */
102 #define HTTP_STATUS_MULTIPLE_CHOICES 300
103 #define HTTP_STATUS_MOVED_PERMANENTLY 301
104 #define HTTP_STATUS_MOVED_TEMPORARILY 302
105 #define HTTP_STATUS_NOT_MODIFIED 304
107 /* Client error 4xx. */
108 #define HTTP_STATUS_BAD_REQUEST 400
109 #define HTTP_STATUS_UNAUTHORIZED 401
110 #define HTTP_STATUS_FORBIDDEN 403
111 #define HTTP_STATUS_NOT_FOUND 404
113 /* Server errors 5xx. */
114 #define HTTP_STATUS_INTERNAL 500
115 #define HTTP_STATUS_NOT_IMPLEMENTED 501
116 #define HTTP_STATUS_BAD_GATEWAY 502
117 #define HTTP_STATUS_UNAVAILABLE 503
120 /* Parse the HTTP status line, which is of format:
122 HTTP-Version SP Status-Code SP Reason-Phrase
124 The function returns the status-code, or -1 if the status line is
125 malformed. The pointer to reason-phrase is returned in RP. */
127 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
129 /* (the variables must not be named `major' and `minor', because
130 that breaks compilation with SunOS4 cc.) */
131 int mjr, mnr, statcode;
134 *reason_phrase_ptr = NULL;
136 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
137 major version, and Y is minor version. */
138 if (strncmp (line, "HTTP/", 5) != 0)
142 /* Calculate major HTTP version. */
144 for (mjr = 0; ISDIGIT (*line); line++)
145 mjr = 10 * mjr + (*line - '0');
146 if (*line != '.' || p == line)
150 /* Calculate minor HTTP version. */
152 for (mnr = 0; ISDIGIT (*line); line++)
153 mnr = 10 * mnr + (*line - '0');
154 if (*line != ' ' || p == line)
156 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
157 minor version can be safely ignored. */
162 /* Calculate status code. */
163 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
165 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
167 /* Set up the reason phrase pointer. */
169 /* RFC2068 requires SPC here, but we allow the string to finish
170 here, in case no reason-phrase is present. */
174 *reason_phrase_ptr = line;
179 *reason_phrase_ptr = line + 1;
184 /* Functions to be used as arguments to header_process(): */
186 struct http_process_range_closure {
192 /* Parse the `Content-Range' header and extract the information it
193 contains. Returns 1 if successful, -1 otherwise. */
195 http_process_range (const char *hdr, void *arg)
197 struct http_process_range_closure *closure
198 = (struct http_process_range_closure *)arg;
201 /* Certain versions of Nutscape proxy server send out
202 `Content-Length' without "bytes" specifier, which is a breach of
203 RFC2068 (as well as the HTTP/1.1 draft which was current at the
204 time). But hell, I must support it... */
205 if (!strncasecmp (hdr, "bytes", 5))
208 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
212 hdr += skip_lws (hdr);
218 for (num = 0; ISDIGIT (*hdr); hdr++)
219 num = 10 * num + (*hdr - '0');
220 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
222 closure->first_byte_pos = num;
224 for (num = 0; ISDIGIT (*hdr); hdr++)
225 num = 10 * num + (*hdr - '0');
226 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
228 closure->last_byte_pos = num;
230 for (num = 0; ISDIGIT (*hdr); hdr++)
231 num = 10 * num + (*hdr - '0');
232 closure->entity_length = num;
236 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
237 Used for `Accept-Ranges'. */
239 http_process_none (const char *hdr, void *arg)
241 int *where = (int *)arg;
243 if (strstr (hdr, "none"))
250 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
252 http_process_type (const char *hdr, void *arg)
254 char **result = (char **)arg;
255 /* Locate P on `;' or the terminating zero, whichever comes first. */
256 const char *p = strchr (hdr, ';');
258 p = hdr + strlen (hdr);
259 while (p > hdr && ISSPACE (*(p - 1)))
261 *result = strdupdelim (hdr, p);
265 /* Check whether the `Connection' header is set to "keep-alive". */
267 http_process_connection (const char *hdr, void *arg)
269 int *flag = (int *)arg;
270 if (!strcasecmp (hdr, "Keep-Alive"))
275 /* Persistent connections. Currently, we cache the most recently used
276 connection as persistent, provided that the HTTP server agrees to
277 make it such. The persistence data is stored in the variables
278 below. Ideally, it would be in a structure, and it should be
279 possible to cache an arbitrary fixed number of these connections.
281 I think the code is quite easy to extend in that direction. */
283 /* Whether a persistent connection is active. */
284 static int pc_active_p;
285 /* Host and port of currently active persistent connection. */
286 static struct address_list *pc_last_host_ip;
287 static unsigned short pc_last_port;
289 /* File descriptor of the currently active persistent connection. */
290 static int pc_last_fd;
293 /* Whether a ssl handshake has occoured on this connection */
294 static int pc_active_ssl;
295 /* SSL connection of the currently active persistent connection. */
296 static SSL *pc_last_ssl;
297 #endif /* HAVE_SSL */
299 /* Mark the persistent connection as invalid. This is used by the
300 CLOSE_* macros after they forcefully close a registered persistent
301 connection. This does not close the file descriptor -- it is left
302 to the caller to do that. (Maybe it should, though.) */
305 invalidate_persistent (void)
310 #endif /* HAVE_SSL */
311 if (pc_last_host_ip != NULL)
313 address_list_release (pc_last_host_ip);
314 pc_last_host_ip = NULL;
316 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
319 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
320 persistent. This will enable someone to use the same connection
321 later. In the context of HTTP, this must be called only AFTER the
322 response has been received and the server has promised that the
323 connection will remain alive.
325 If a previous connection was persistent, it is closed. */
328 register_persistent (const char *host, unsigned short port, int fd
336 if (pc_last_fd == fd)
338 /* The connection FD is already registered. Nothing to
344 /* The old persistent connection is still active; let's
345 close it first. This situation arises whenever a
346 persistent connection exists, but we then connect to a
347 different host, and try to register a persistent
348 connection to that one. */
350 /* The ssl disconnect has to take place before the closing
353 shutdown_ssl(pc_last_ssl);
356 invalidate_persistent ();
360 assert (pc_last_host_ip == NULL);
362 /* This lookup_host cannot fail, because it has the results in the
364 pc_last_host_ip = lookup_host (host, 1);
365 assert (pc_last_host_ip != NULL);
372 pc_active_ssl = ssl ? 1 : 0;
374 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
378 # define SHUTDOWN_SSL(ssl) do { \
380 shutdown_ssl (ssl); \
383 # define SHUTDOWN_SSL(ssl)
386 /* Return non-zero if a persistent connection is available for
387 connecting to HOST:PORT. */
390 persistent_available_p (const char *host, unsigned short port
397 struct address_list *this_host_ip;
399 /* First, check whether a persistent connection is active at all. */
402 /* Second, check if the active connection pertains to the correct
403 (HOST, PORT) ordered pair. */
404 if (port != pc_last_port)
408 /* Second, a): check if current connection is (not) ssl, too. This
409 test is unlikely to fail because HTTP and HTTPS typicaly use
410 different ports. Yet it is possible, or so I [Christian
411 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
413 if (ssl != pc_active_ssl)
415 #endif /* HAVE_SSL */
417 this_host_ip = lookup_host (host, 1);
421 /* To equate the two host names for the purposes of persistent
422 connections, they need to share all the IP addresses in the
424 success = address_list_match_all (pc_last_host_ip, this_host_ip);
425 address_list_release (this_host_ip);
429 /* Third: check whether the connection is still open. This is
430 important because most server implement a liberal (short) timeout
431 on persistent connections. Wget can of course always reconnect
432 if the connection doesn't work out, but it's nicer to know in
433 advance. This test is a logical followup of the first test, but
434 is "expensive" and therefore placed at the end of the list. */
435 if (!test_socket_open (pc_last_fd))
437 /* Oops, the socket is no longer open. Now that we know that,
438 let's invalidate the persistent connection before returning
442 SHUTDOWN_SSL (pc_last_ssl);
445 invalidate_persistent ();
451 /* The idea behind these two CLOSE macros is to distinguish between
452 two cases: one when the job we've been doing is finished, and we
453 want to close the connection and leave, and two when something is
454 seriously wrong and we're closing the connection as part of
457 In case of keep_alive, CLOSE_FINISH should leave the connection
458 open, while CLOSE_INVALIDATE should still close it.
460 Note that the semantics of the flag `keep_alive' is "this
461 connection *will* be reused (the server has promised not to close
462 the connection once we're done)", while the semantics of
463 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
464 active, registered connection". */
466 #define CLOSE_FINISH(fd) do { \
469 SHUTDOWN_SSL (ssl); \
471 if (pc_active_p && (fd) == pc_last_fd) \
472 invalidate_persistent (); \
476 #define CLOSE_INVALIDATE(fd) do { \
477 SHUTDOWN_SSL (ssl); \
479 if (pc_active_p && (fd) == pc_last_fd) \
480 invalidate_persistent (); \
485 long len; /* received length */
486 long contlen; /* expected length */
487 long restval; /* the restart value */
488 int res; /* the result of last read */
489 char *newloc; /* new location (redirection) */
490 char *remote_time; /* remote time-stamp string */
491 char *error; /* textual HTTP error */
492 int statcode; /* status code */
493 long dltime; /* time of the download */
494 int no_truncate; /* whether truncating the file is
496 const char *referer; /* value of the referer header. */
497 char **local_file; /* local file. */
501 free_hstat (struct http_stat *hs)
503 FREE_MAYBE (hs->newloc);
504 FREE_MAYBE (hs->remote_time);
505 FREE_MAYBE (hs->error);
507 /* Guard against being called twice. */
509 hs->remote_time = NULL;
513 static char *create_authorization_line PARAMS ((const char *, const char *,
514 const char *, const char *,
516 static char *basic_authentication_encode PARAMS ((const char *, const char *,
518 static int known_authentication_scheme_p PARAMS ((const char *));
520 time_t http_atotm PARAMS ((char *));
522 #define BEGINS_WITH(line, string_constant) \
523 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
524 && (ISSPACE (line[sizeof (string_constant) - 1]) \
525 || !line[sizeof (string_constant) - 1]))
527 /* Retrieve a document through HTTP protocol. It recognizes status
528 code, and correctly handles redirections. It closes the network
529 socket. If it receives an error from the functions below it, it
530 will print it if there is enough information to do so (almost
531 always), returning the error to the caller (i.e. http_loop).
533 Various HTTP parameters are stored to hs. Although it parses the
534 response code correctly, it is not used in a sane way. The caller
537 If PROXY is non-NULL, the connection will be made to the proxy
538 server, and u->url will be requested. */
540 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
542 char *request, *type, *command, *full_path;
544 char *pragma_h, *referer, *useragent, *range, *wwwauth;
545 char *authenticate_h;
549 char *request_keep_alive;
550 int sock, hcount, num_written, all_length, statcode;
551 long contlen, contrange;
554 int auth_tried_already;
557 static SSL_CTX *ssl_ctx = NULL;
559 #endif /* HAVE_SSL */
560 char *cookies = NULL;
562 /* Whether this connection will be kept alive after the HTTP request
566 /* Flags that detect the two ways of specifying HTTP keep-alive
568 int http_keep_alive_1, http_keep_alive_2;
570 /* Whether keep-alive should be inhibited. */
571 int inhibit_keep_alive;
574 /* initialize ssl_ctx on first run */
577 uerr_t err = init_ssl (&ssl_ctx);
582 case SSLERRCTXCREATE:
584 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
588 /* try without certfile */
589 logprintf (LOG_NOTQUIET,
590 _("Failed to load certificates from %s\n"),
593 logprintf (LOG_NOTQUIET,
594 _("Trying without the specified certificate\n"));
597 logprintf (LOG_NOTQUIET,
598 _("Failed to get certificate key from %s\n"),
601 logprintf (LOG_NOTQUIET,
602 _("Trying without the specified certificate\n"));
609 #endif /* HAVE_SSL */
611 if (!(*dt & HEAD_ONLY))
612 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
613 know the local filename so we can save to it. */
614 assert (*hs->local_file != NULL);
617 auth_tried_already = 0;
619 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
622 /* We need to come back here when the initial attempt to retrieve
623 without authorization header fails. (Expected to happen at least
624 for the Digest authorization scheme.) */
627 http_keep_alive_1 = http_keep_alive_2 = 0;
629 /* Initialize certain elements of struct http_stat. */
634 hs->remote_time = NULL;
637 /* If we're using a proxy, we will be connecting to the proxy
639 conn = proxy ? proxy : u;
641 /* First: establish the connection. */
642 if (inhibit_keep_alive
645 !persistent_available_p (conn->host, conn->port)
647 !persistent_available_p (conn->host, conn->port,
648 u->scheme == SCHEME_HTTPS)
649 #endif /* HAVE_SSL */
652 struct address_list *al = lookup_host (conn->host, 0);
655 set_connection_host_name (conn->host);
656 sock = connect_to_many (al, conn->port, 0);
657 set_connection_host_name (NULL);
658 address_list_release (al);
661 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
664 if (conn->scheme == SCHEME_HTTPS)
665 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
667 logputs (LOG_VERBOSE, "\n");
668 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
673 #endif /* HAVE_SSL */
677 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
678 conn->host, conn->port);
679 /* #### pc_last_fd should be accessed through an accessor
684 #endif /* HAVE_SSL */
685 DEBUGP (("Reusing fd %d.\n", sock));
688 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
693 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
694 sprintf (referer, "Referer: %s\r\n", hs->referer);
697 if (*dt & SEND_NOCACHE)
698 pragma_h = "Pragma: no-cache\r\n";
704 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
705 /* Gag me! Some servers (e.g. WebSitePro) have been known to
706 respond to the following `Range' format by generating a
707 multipart/x-byte-ranges MIME document! This MIME type was
708 present in an old draft of the byteranges specification.
709 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
710 only if multiple non-overlapping ranges are requested --
711 which Wget never does. */
712 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
717 STRDUP_ALLOCA (useragent, opt.useragent);
720 useragent = (char *)alloca (10 + strlen (version_string));
721 sprintf (useragent, "Wget/%s", version_string);
723 /* Construct the authentication, if userid is present. */
726 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
727 user = user ? user : opt.http_user;
728 passwd = passwd ? passwd : opt.http_passwd;
735 /* We have the username and the password, but haven't tried
736 any authorization yet. Let's see if the "Basic" method
737 works. If not, we'll come back here and construct a
738 proper authorization method with the right challenges.
740 If we didn't employ this kind of logic, every URL that
741 requires authorization would have to be processed twice,
742 which is very suboptimal and generates a bunch of false
743 "unauthorized" errors in the server log.
745 #### But this logic also has a serious problem when used
746 with stronger authentications: we *first* transmit the
747 username and the password in clear text, and *then*
748 attempt a stronger authentication scheme. That cannot be
749 right! We are only fortunate that almost everyone still
750 uses the `Basic' scheme anyway.
752 There should be an option to prevent this from happening,
753 for those who use strong authentication schemes and value
755 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
759 /* Use the full path, i.e. one that includes the leading
760 slash and the query string, but is independent of proxy
762 char *pth = url_full_path (u);
763 wwwauth = create_authorization_line (authenticate_h, user, passwd,
772 char *proxy_user, *proxy_passwd;
773 /* For normal username and password, URL components override
774 command-line/wgetrc parameters. With proxy authentication,
775 it's the reverse, because proxy URLs are normally the
776 "permanent" ones, so command-line args should take
778 if (opt.proxy_user && opt.proxy_passwd)
780 proxy_user = opt.proxy_user;
781 proxy_passwd = opt.proxy_passwd;
785 proxy_user = proxy->user;
786 proxy_passwd = proxy->passwd;
788 /* #### This does not appear right. Can't the proxy request,
789 say, `Digest' authentication? */
790 if (proxy_user && proxy_passwd)
791 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
792 "Proxy-Authorization");
795 /* String of the form :PORT. Used only for non-standard ports. */
797 if (u->port != scheme_default_port (u->scheme))
799 port_maybe = (char *)alloca (numdigit (u->port) + 2);
800 sprintf (port_maybe, ":%d", u->port);
803 if (!inhibit_keep_alive)
804 request_keep_alive = "Connection: Keep-Alive\r\n";
806 request_keep_alive = NULL;
809 cookies = build_cookies_request (u->host, u->port, u->path,
811 u->scheme == SCHEME_HTTPS
818 full_path = xstrdup (u->url);
820 /* Use the full path, i.e. one that includes the leading slash and
821 the query string. E.g. if u->path is "foo/bar" and u->query is
822 "param=value", full_path will be "/foo/bar?param=value". */
823 full_path = url_full_path (u);
825 /* Allocate the memory for the request. */
826 request = (char *)alloca (strlen (command)
830 + (port_maybe ? strlen (port_maybe) : 0)
831 + strlen (HTTP_ACCEPT)
832 + (request_keep_alive
833 ? strlen (request_keep_alive) : 0)
834 + (referer ? strlen (referer) : 0)
835 + (cookies ? strlen (cookies) : 0)
836 + (wwwauth ? strlen (wwwauth) : 0)
837 + (proxyauth ? strlen (proxyauth) : 0)
838 + (range ? strlen (range) : 0)
840 + (opt.user_header ? strlen (opt.user_header) : 0)
842 /* Construct the request. */
848 %s%s%s%s%s%s%s%s\r\n",
851 port_maybe ? port_maybe : "",
853 request_keep_alive ? request_keep_alive : "",
854 referer ? referer : "",
855 cookies ? cookies : "",
856 wwwauth ? wwwauth : "",
857 proxyauth ? proxyauth : "",
860 opt.user_header ? opt.user_header : "");
861 DEBUGP (("---request begin---\n%s---request end---\n", request));
863 /* Free the temporary memory. */
864 FREE_MAYBE (wwwauth);
865 FREE_MAYBE (proxyauth);
866 FREE_MAYBE (cookies);
869 /* Send the request to server. */
871 if (u->scheme == SCHEME_HTTPS)
872 num_written = ssl_iwrite (ssl, request, strlen (request));
874 #endif /* HAVE_SSL */
875 num_written = iwrite (sock, request, strlen (request));
879 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
881 CLOSE_INVALIDATE (sock);
884 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
885 proxy ? "Proxy" : "HTTP");
886 contlen = contrange = -1;
891 /* Before reading anything, initialize the rbuf. */
892 rbuf_initialize (&rbuf, sock);
894 if (u->scheme == SCHEME_HTTPS)
898 #endif /* HAVE_SSL */
901 /* Header-fetching loop. */
909 /* Get the header. */
910 status = header_get (&rbuf, &hdr,
911 /* Disallow continuations for status line. */
912 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
914 /* Check for errors. */
915 if (status == HG_EOF && *hdr)
917 /* This used to be an unconditional error, but that was
918 somewhat controversial, because of a large number of
919 broken CGI's that happily "forget" to send the second EOL
920 before closing the connection of a HEAD request.
922 So, the deal is to check whether the header is empty
923 (*hdr is zero if it is); if yes, it means that the
924 previous header was fully retrieved, and that -- most
925 probably -- the request is complete. "...be liberal in
926 what you accept." Oh boy. */
927 logputs (LOG_VERBOSE, "\n");
928 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
931 FREE_MAYBE (all_headers);
932 CLOSE_INVALIDATE (sock);
935 else if (status == HG_ERROR)
937 logputs (LOG_VERBOSE, "\n");
938 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
942 FREE_MAYBE (all_headers);
943 CLOSE_INVALIDATE (sock);
947 /* If the headers are to be saved to a file later, save them to
949 if (opt.save_headers)
951 int lh = strlen (hdr);
952 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
953 memcpy (all_headers + all_length, hdr, lh);
955 all_headers[all_length++] = '\n';
956 all_headers[all_length] = '\0';
959 /* Check for status line. */
963 /* Parse the first line of server response. */
964 statcode = parse_http_status_line (hdr, &error);
965 hs->statcode = statcode;
966 /* Store the descriptive response. */
967 if (statcode == -1) /* malformed response */
969 /* A common reason for "malformed response" error is the
970 case when no data was actually received. Handle this
973 hs->error = xstrdup (_("No data received"));
975 hs->error = xstrdup (_("Malformed status line"));
980 hs->error = xstrdup (_("(no description)"));
982 hs->error = xstrdup (error);
990 if (opt.server_response)
991 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
993 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
999 /* Exit on empty header. */
1006 /* Print the header if requested. */
1007 if (opt.server_response && hcount != 1)
1008 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1010 /* Try getting content-length. */
1011 if (contlen == -1 && !opt.ignore_length)
1012 if (header_process (hdr, "Content-Length", header_extract_number,
1015 /* Try getting content-type. */
1017 if (header_process (hdr, "Content-Type", http_process_type, &type))
1019 /* Try getting location. */
1021 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1023 /* Try getting last-modified. */
1024 if (!hs->remote_time)
1025 if (header_process (hdr, "Last-Modified", header_strdup,
1028 /* Try getting cookies. */
1030 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1032 /* Try getting www-authentication. */
1033 if (!authenticate_h)
1034 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1037 /* Check for accept-ranges header. If it contains the word
1038 `none', disable the ranges. */
1039 if (*dt & ACCEPTRANGES)
1042 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1045 *dt &= ~ACCEPTRANGES;
1049 /* Try getting content-range. */
1050 if (contrange == -1)
1052 struct http_process_range_closure closure;
1053 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1055 contrange = closure.first_byte_pos;
1059 /* Check for keep-alive related responses. */
1060 if (!inhibit_keep_alive)
1062 /* Check for the `Keep-Alive' header. */
1063 if (!http_keep_alive_1)
1065 if (header_process (hdr, "Keep-Alive", header_exists,
1066 &http_keep_alive_1))
1069 /* Check for `Connection: Keep-Alive'. */
1070 if (!http_keep_alive_2)
1072 if (header_process (hdr, "Connection", http_process_connection,
1073 &http_keep_alive_2))
1081 logputs (LOG_VERBOSE, "\n");
1084 && (http_keep_alive_1 || http_keep_alive_2))
1086 assert (inhibit_keep_alive == 0);
1090 /* The server has promised that it will not close the connection
1091 when we're done. This means that we can register it. */
1093 register_persistent (conn->host, conn->port, sock);
1095 register_persistent (conn->host, conn->port, sock, ssl);
1096 #endif /* HAVE_SSL */
1098 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1101 /* Authorization is required. */
1105 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1106 might be more bytes in the body. */
1107 if (auth_tried_already)
1109 /* If we have tried it already, then there is not point
1112 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1113 xfree (authenticate_h);
1116 else if (!known_authentication_scheme_p (authenticate_h))
1118 xfree (authenticate_h);
1119 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1122 else if (BEGINS_WITH (authenticate_h, "Basic"))
1124 /* The authentication scheme is basic, the one we try by
1125 default, and it failed. There's no sense in trying
1131 auth_tried_already = 1;
1135 /* We do not need this anymore. */
1138 xfree (authenticate_h);
1139 authenticate_h = NULL;
1142 /* 20x responses are counted among successful by default. */
1143 if (H_20X (statcode))
1146 /* Return if redirected. */
1147 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1149 /* RFC2068 says that in case of the 300 (multiple choices)
1150 response, the server can output a preferred URL through
1151 `Location' header; otherwise, the request should be treated
1152 like GET. So, if the location is set, it will be a
1153 redirection; otherwise, just proceed normally. */
1154 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1158 logprintf (LOG_VERBOSE,
1159 _("Location: %s%s\n"),
1160 hs->newloc ? hs->newloc : _("unspecified"),
1161 hs->newloc ? _(" [following]") : "");
1162 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1163 might be more bytes in the body. */
1165 FREE_MAYBE (all_headers);
1170 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1173 /* We don't assume text/html by default. */
1176 if (opt.html_extension && (*dt & TEXTHTML))
1177 /* -E / --html-extension / html_extension = on was specified, and this is a
1178 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1179 already the file's suffix, tack on ".html". */
1181 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1183 if (last_period_in_local_filename == NULL ||
1184 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1185 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1187 size_t local_filename_len = strlen(*hs->local_file);
1189 *hs->local_file = xrealloc(*hs->local_file,
1190 local_filename_len + sizeof(".html"));
1191 strcpy(*hs->local_file + local_filename_len, ".html");
1193 *dt |= ADDED_HTML_EXTENSION;
1197 if (contrange == -1)
1199 /* We did not get a content-range header. This means that the
1200 server did not honor our `Range' request. Normally, this
1201 means we should reset hs->restval and continue normally. */
1203 /* However, if `-c' is used, we need to be a bit more careful:
1205 1. If `-c' is specified and the file already existed when
1206 Wget was started, it would be a bad idea for us to start
1207 downloading it from scratch, effectively truncating it. I
1208 believe this cannot happen unless `-c' was specified.
1210 2. If `-c' is used on a file that is already fully
1211 downloaded, we're requesting bytes after the end of file,
1212 which can result in server not honoring `Range'. If this is
1213 the case, `Content-Length' will be equal to the length of the
1215 if (opt.always_rest)
1217 /* Check for condition #2. */
1218 if (hs->restval > 0 /* restart was requested. */
1219 && contlen != -1 /* we got content-length. */
1220 && hs->restval >= contlen /* file fully downloaded
1224 logputs (LOG_VERBOSE, _("\
1225 \n The file is already fully retrieved; nothing to do.\n\n"));
1226 /* In case the caller inspects. */
1230 FREE_MAYBE (all_headers);
1231 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1232 might be more bytes in the body. */
1233 return RETRUNNEEDED;
1236 /* Check for condition #1. */
1237 if (hs->no_truncate)
1239 logprintf (LOG_NOTQUIET,
1242 Continued download failed on this file, which conflicts with `-c'.\n\
1243 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1245 FREE_MAYBE (all_headers);
1246 CLOSE_INVALIDATE (sock);
1247 return CONTNOTSUPPORTED;
1255 else if (contrange != hs->restval ||
1256 (H_PARTIAL (statcode) && contrange == -1))
1258 /* This means the whole request was somehow misunderstood by the
1259 server. Bail out. */
1261 FREE_MAYBE (all_headers);
1262 CLOSE_INVALIDATE (sock);
1269 contlen += contrange;
1271 contrange = -1; /* If conent-length was not sent,
1272 content-range will be ignored. */
1274 hs->contlen = contlen;
1278 if ((*dt & RETROKF) && !opt.server_response)
1280 /* No need to print this output if the body won't be
1281 downloaded at all, or if the original server response is
1283 logputs (LOG_VERBOSE, _("Length: "));
1286 logputs (LOG_VERBOSE, legible (contlen));
1287 if (contrange != -1)
1288 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1289 legible (contlen - contrange));
1292 logputs (LOG_VERBOSE,
1293 opt.ignore_length ? _("ignored") : _("unspecified"));
1295 logprintf (LOG_VERBOSE, " [%s]\n", type);
1297 logputs (LOG_VERBOSE, "\n");
1301 type = NULL; /* We don't need it any more. */
1303 /* Return if we have no intention of further downloading. */
1304 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1306 /* In case the caller cares to look... */
1310 FREE_MAYBE (all_headers);
1311 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1312 might be more bytes in the body. */
1313 return RETRFINISHED;
1316 /* Open the local file. */
1319 mkalldirs (*hs->local_file);
1321 rotate_backups (*hs->local_file);
1322 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1325 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1326 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1327 might be more bytes in the body. */
1328 FREE_MAYBE (all_headers);
1334 extern int global_download_count;
1336 /* To ensure that repeated "from scratch" downloads work for -O
1337 files, we rewind the file pointer, unless restval is
1338 non-zero. (This works only when -O is used on regular files,
1339 but it's still a valuable feature.)
1341 However, this loses when more than one URL is specified on
1342 the command line the second rewinds eradicates the contents
1343 of the first download. Thus we disable the above trick for
1344 all the downloads except the very first one.
1346 #### A possible solution to this would be to remember the
1347 file position in the output document and to seek to that
1348 position, instead of rewinding. */
1349 if (!hs->restval && global_download_count == 0)
1351 /* This will silently fail for streams that don't correspond
1352 to regular files, but that's OK. */
1354 /* ftruncate is needed because opt.dfp is opened in append
1355 mode if opt.always_rest is set. */
1356 ftruncate (fileno (fp), 0);
1361 /* #### This confuses the code that checks for file size. There
1362 should be some overhead information. */
1363 if (opt.save_headers)
1364 fwrite (all_headers, 1, all_length, fp);
1366 /* Get the contents of the document. */
1367 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1368 (contlen != -1 ? contlen : 0),
1369 &rbuf, keep_alive, &hs->dltime);
1372 /* Close or flush the file. We have to be careful to check for
1373 error here. Checking the result of fwrite() is not enough --
1374 errors could go unnoticed! */
1377 flush_res = fclose (fp);
1379 flush_res = fflush (fp);
1380 if (flush_res == EOF)
1383 FREE_MAYBE (all_headers);
1384 CLOSE_FINISH (sock);
1387 return RETRFINISHED;
1390 /* The genuine HTTP loop! This is the part where the retrieval is
1391 retried, and retried, and retried, and... */
1393 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1394 int *dt, struct url *proxy)
1397 int use_ts, got_head = 0; /* time-stamping info */
1398 char *filename_plus_orig_suffix;
1399 char *local_filename = NULL;
1400 char *tms, *suf, *locf, *tmrate;
1402 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1403 long local_size = 0; /* the size of the local file */
1404 size_t filename_len;
1405 struct http_stat hstat; /* HTTP status */
1409 /* This used to be done in main(), but it's a better idea to do it
1410 here so that we don't go through the hoops if we're just using
1412 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1414 load_cookies (opt.cookies_input);
1415 cookies_loaded_p = 1;
1420 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1421 has_wildcards_p because it would also warn on `?', and we know that
1422 shows up in CGI paths a *lot*. */
1423 if (strchr (u->url, '*'))
1424 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1426 /* Determine the local filename. */
1427 if (local_file && *local_file)
1428 hstat.local_file = local_file;
1429 else if (local_file)
1431 *local_file = url_filename (u);
1432 hstat.local_file = local_file;
1436 dummy = url_filename (u);
1437 hstat.local_file = &dummy;
1440 if (!opt.output_document)
1441 locf = *hstat.local_file;
1443 locf = opt.output_document;
1445 hstat.referer = referer;
1447 filename_len = strlen (*hstat.local_file);
1448 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1450 if (opt.noclobber && file_exists_p (*hstat.local_file))
1452 /* If opt.noclobber is turned on and file already exists, do not
1453 retrieve the file */
1454 logprintf (LOG_VERBOSE, _("\
1455 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1456 /* If the file is there, we suppose it's retrieved OK. */
1459 /* #### Bogusness alert. */
1460 /* If its suffix is "html" or "htm", assume text/html. */
1461 if (((suf = suffix (*hstat.local_file)) != NULL)
1462 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1470 if (opt.timestamping)
1472 boolean local_dot_orig_file_exists = FALSE;
1474 if (opt.backup_converted)
1475 /* If -K is specified, we'll act on the assumption that it was specified
1476 last time these files were downloaded as well, and instead of just
1477 comparing local file X against server file X, we'll compare local
1478 file X.orig (if extant, else X) against server file X. If -K
1479 _wasn't_ specified last time, or the server contains files called
1480 *.orig, -N will be back to not operating correctly with -k. */
1482 /* Would a single s[n]printf() call be faster? --dan
1484 Definitely not. sprintf() is horribly slow. It's a
1485 different question whether the difference between the two
1486 affects a program. Usually I'd say "no", but at one
1487 point I profiled Wget, and found that a measurable and
1488 non-negligible amount of time was lost calling sprintf()
1489 in url.c. Replacing sprintf with inline calls to
1490 strcpy() and long_to_string() made a difference.
1492 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1493 memcpy (filename_plus_orig_suffix + filename_len,
1494 ".orig", sizeof (".orig"));
1496 /* Try to stat() the .orig file. */
1497 if (stat (filename_plus_orig_suffix, &st) == 0)
1499 local_dot_orig_file_exists = TRUE;
1500 local_filename = filename_plus_orig_suffix;
1504 if (!local_dot_orig_file_exists)
1505 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1506 if (stat (*hstat.local_file, &st) == 0)
1507 local_filename = *hstat.local_file;
1509 if (local_filename != NULL)
1510 /* There was a local file, so we'll check later to see if the version
1511 the server has is the same version we already have, allowing us to
1516 local_size = st.st_size;
1520 /* Reset the counter. */
1522 *dt = 0 | ACCEPTRANGES;
1526 /* Increment the pass counter. */
1528 sleep_between_retrievals (count);
1529 /* Get the current time string. */
1530 tms = time_str (NULL);
1531 /* Print fetch message, if opt.verbose. */
1534 char *hurl = url_string (u, 1);
1538 sprintf (tmp, _("(try:%2d)"), count);
1539 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1540 tms, hurl, tmp, locf);
1542 ws_changetitle (hurl, 1);
1547 /* Default document type is empty. However, if spider mode is
1548 on or time-stamping is employed, HEAD_ONLY commands is
1549 encoded within *dt. */
1550 if (opt.spider || (use_ts && !got_head))
1554 /* Assume no restarting. */
1556 /* Decide whether or not to restart. */
1557 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1558 /* #### this calls access() and then stat(); could be optimized. */
1559 && file_exists_p (locf))
1560 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1561 hstat.restval = st.st_size;
1563 /* In `-c' is used and the file is existing and non-empty,
1564 refuse to truncate it if the server doesn't support continued
1566 hstat.no_truncate = 0;
1567 if (opt.always_rest && hstat.restval)
1568 hstat.no_truncate = 1;
1570 /* Decide whether to send the no-cache directive. We send it in
1572 a) we're using a proxy, and we're past our first retrieval.
1573 Some proxies are notorious for caching incomplete data, so
1574 we require a fresh get.
1575 b) caching is explicitly inhibited. */
1576 if ((proxy && count > 1) /* a */
1577 || !opt.allow_cache /* b */
1579 *dt |= SEND_NOCACHE;
1581 *dt &= ~SEND_NOCACHE;
1583 /* Try fetching the document, or at least its head. */
1584 err = gethttp (u, &hstat, dt, proxy);
1586 /* It's unfortunate that wget determines the local filename before finding
1587 out the Content-Type of the file. Barring a major restructuring of the
1588 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1589 *hstat.local_file to tack on ".html". */
1590 if (!opt.output_document)
1591 locf = *hstat.local_file;
1593 locf = opt.output_document;
1596 tms = time_str (NULL);
1597 /* Get the new location (with or without the redirection). */
1599 *newloc = xstrdup (hstat.newloc);
1602 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1603 case CONERROR: case READERR: case WRITEFAILED:
1605 /* Non-fatal errors continue executing the loop, which will
1606 bring them to "while" statement at the end, to judge
1607 whether the number of tries was exceeded. */
1608 free_hstat (&hstat);
1609 printwhat (count, opt.ntry);
1612 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1613 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1614 /* Fatal errors just return from the function. */
1615 free_hstat (&hstat);
1619 case FWRITEERR: case FOPENERR:
1620 /* Another fatal error. */
1621 logputs (LOG_VERBOSE, "\n");
1622 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1623 *hstat.local_file, strerror (errno));
1624 free_hstat (&hstat);
1629 /* Another fatal error. */
1630 logputs (LOG_VERBOSE, "\n");
1631 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1632 free_hstat (&hstat);
1637 /* Return the new location to the caller. */
1640 logprintf (LOG_NOTQUIET,
1641 _("ERROR: Redirection (%d) without location.\n"),
1643 free_hstat (&hstat);
1647 free_hstat (&hstat);
1652 /* The file was already fully retrieved. */
1653 free_hstat (&hstat);
1658 /* Deal with you later. */
1661 /* All possibilities should have been exhausted. */
1664 if (!(*dt & RETROKF))
1668 /* #### Ugly ugly ugly! */
1669 char *hurl = url_string (u, 1);
1670 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1673 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1674 tms, hstat.statcode, hstat.error);
1675 logputs (LOG_VERBOSE, "\n");
1676 free_hstat (&hstat);
1681 /* Did we get the time-stamp? */
1684 if (opt.timestamping && !hstat.remote_time)
1686 logputs (LOG_NOTQUIET, _("\
1687 Last-modified header missing -- time-stamps turned off.\n"));
1689 else if (hstat.remote_time)
1691 /* Convert the date-string into struct tm. */
1692 tmr = http_atotm (hstat.remote_time);
1693 if (tmr == (time_t) (-1))
1694 logputs (LOG_VERBOSE, _("\
1695 Last-modified header invalid -- time-stamp ignored.\n"));
1699 /* The time-stamping section. */
1704 use_ts = 0; /* no more time-stamping */
1705 count = 0; /* the retrieve count for HEAD is
1707 if (hstat.remote_time && tmr != (time_t) (-1))
1709 /* Now time-stamping can be used validly. Time-stamping
1710 means that if the sizes of the local and remote file
1711 match, and local file is newer than the remote file,
1712 it will not be retrieved. Otherwise, the normal
1713 download procedure is resumed. */
1715 (hstat.contlen == -1 || local_size == hstat.contlen))
1717 logprintf (LOG_VERBOSE, _("\
1718 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1720 free_hstat (&hstat);
1724 else if (tml >= tmr)
1725 logprintf (LOG_VERBOSE, _("\
1726 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1728 logputs (LOG_VERBOSE,
1729 _("Remote file is newer, retrieving.\n"));
1731 free_hstat (&hstat);
1734 if ((tmr != (time_t) (-1))
1736 && ((hstat.len == hstat.contlen) ||
1737 ((hstat.res == 0) &&
1738 ((hstat.contlen == -1) ||
1739 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1741 /* #### This code repeats in http.c and ftp.c. Move it to a
1743 const char *fl = NULL;
1744 if (opt.output_document)
1746 if (opt.od_known_regular)
1747 fl = opt.output_document;
1750 fl = *hstat.local_file;
1754 /* End of time-stamping section. */
1758 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1763 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1765 if (hstat.len == hstat.contlen)
1769 logprintf (LOG_VERBOSE,
1770 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1771 tms, tmrate, locf, hstat.len, hstat.contlen);
1772 logprintf (LOG_NONVERBOSE,
1773 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1774 tms, u->url, hstat.len, hstat.contlen, locf, count);
1777 downloaded_increase (hstat.len);
1779 /* Remember that we downloaded the file for later ".orig" code. */
1780 if (*dt & ADDED_HTML_EXTENSION)
1781 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1783 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1785 free_hstat (&hstat);
1789 else if (hstat.res == 0) /* No read error */
1791 if (hstat.contlen == -1) /* We don't know how much we were supposed
1792 to get, so assume we succeeded. */
1796 logprintf (LOG_VERBOSE,
1797 _("%s (%s) - `%s' saved [%ld]\n\n"),
1798 tms, tmrate, locf, hstat.len);
1799 logprintf (LOG_NONVERBOSE,
1800 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1801 tms, u->url, hstat.len, locf, count);
1804 downloaded_increase (hstat.len);
1806 /* Remember that we downloaded the file for later ".orig" code. */
1807 if (*dt & ADDED_HTML_EXTENSION)
1808 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1810 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1812 free_hstat (&hstat);
1816 else if (hstat.len < hstat.contlen) /* meaning we lost the
1817 connection too soon */
1819 logprintf (LOG_VERBOSE,
1820 _("%s (%s) - Connection closed at byte %ld. "),
1821 tms, tmrate, hstat.len);
1822 printwhat (count, opt.ntry);
1823 free_hstat (&hstat);
1826 else if (!opt.kill_longer) /* meaning we got more than expected */
1828 logprintf (LOG_VERBOSE,
1829 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1830 tms, tmrate, locf, hstat.len, hstat.contlen);
1831 logprintf (LOG_NONVERBOSE,
1832 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1833 tms, u->url, hstat.len, hstat.contlen, locf, count);
1835 downloaded_increase (hstat.len);
1837 /* Remember that we downloaded the file for later ".orig" code. */
1838 if (*dt & ADDED_HTML_EXTENSION)
1839 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1841 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1843 free_hstat (&hstat);
1847 else /* the same, but not accepted */
1849 logprintf (LOG_VERBOSE,
1850 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1851 tms, tmrate, hstat.len, hstat.contlen);
1852 printwhat (count, opt.ntry);
1853 free_hstat (&hstat);
1857 else /* now hstat.res can only be -1 */
1859 if (hstat.contlen == -1)
1861 logprintf (LOG_VERBOSE,
1862 _("%s (%s) - Read error at byte %ld (%s)."),
1863 tms, tmrate, hstat.len, strerror (errno));
1864 printwhat (count, opt.ntry);
1865 free_hstat (&hstat);
1868 else /* hstat.res == -1 and contlen is given */
1870 logprintf (LOG_VERBOSE,
1871 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1872 tms, tmrate, hstat.len, hstat.contlen,
1874 printwhat (count, opt.ntry);
1875 free_hstat (&hstat);
1882 while (!opt.ntry || (count < opt.ntry));
1886 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1887 than local timezone.
1889 mktime is similar but assumes struct tm, also known as the
1890 "broken-down" form of time, is in local time zone. mktime_from_utc
1891 uses mktime to make the conversion understanding that an offset
1892 will be introduced by the local time assumption.
1894 mktime_from_utc then measures the introduced offset by applying
1895 gmtime to the initial result and applying mktime to the resulting
1896 "broken-down" form. The difference between the two mktime results
1897 is the measured offset which is then subtracted from the initial
1898 mktime result to yield a calendar time which is the value returned.
1900 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1901 consistent offset (the non DST offset) since tm and tm+o might be
1902 on opposite sides of a DST change.
1904 Some implementations of mktime return -1 for the nonexistent
1905 localtime hour at the beginning of DST. In this event, use
1906 mktime(tm - 1hr) + 3600.
1910 gmtime(t+o) --> tm+o
1911 mktime(tm+o) --> t+2o
1912 t+o - (t+2o - t+o) = t
1914 Note that glibc contains a function of the same purpose named
1915 `timegm' (reverse of gmtime). But obviously, it is not universally
1916 available, and unfortunately it is not straightforwardly
1917 extractable for use here. Perhaps configure should detect timegm
1918 and use it where available.
1920 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1921 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1922 Further improved by Roger with assistance from Edward J. Sabol
1923 based on input by Jamie Zawinski. */
1926 mktime_from_utc (struct tm *t)
1937 return -1; /* can't deal with output from strptime */
1948 return -1; /* can't deal with output from gmtime */
1951 return (tl - (tb - tl));
1954 /* Check whether the result of strptime() indicates success.
1955 strptime() returns the pointer to how far it got to in the string.
1956 The processing has been successful if the string is at `GMT' or
1957 `+X', or at the end of the string.
1959 In extended regexp parlance, the function returns 1 if P matches
1960 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1961 can return) is considered a failure and 0 is returned. */
1963 check_end (const char *p)
1967 while (ISSPACE (*p))
1970 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1971 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1977 /* Convert the textual specification of time in TIME_STRING to the
1978 number of seconds since the Epoch.
1980 TIME_STRING can be in any of the three formats RFC2068 allows the
1981 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1982 Timezones are ignored, and should be GMT.
1984 Return the computed time_t representation, or -1 if the conversion
1987 This function uses strptime with various string formats for parsing
1988 TIME_STRING. This results in a parser that is not as lenient in
1989 interpreting TIME_STRING as I would like it to be. Being based on
1990 strptime, it always allows shortened months, one-digit days, etc.,
1991 but due to the multitude of formats in which time can be
1992 represented, an ideal HTTP time parser would be even more
1993 forgiving. It should completely ignore things like week days and
1994 concentrate only on the various forms of representing years,
1995 months, days, hours, minutes, and seconds. For example, it would
1996 be nice if it accepted ISO 8601 out of the box.
1998 I've investigated free and PD code for this purpose, but none was
1999 usable. getdate was big and unwieldy, and had potential copyright
2000 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2001 distributed with phttpd, is excellent, but we cannot use it because
2002 it is not assigned to the FSF. So I stuck it with strptime. */
2005 http_atotm (char *time_string)
2007 /* NOTE: Solaris strptime man page claims that %n and %t match white
2008 space, but that's not universally available. Instead, we simply
2009 use ` ' to mean "skip all WS", which works under all strptime
2010 implementations I've tested. */
2012 static const char *time_formats[] = {
2013 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2014 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2015 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2016 (google.com uses this for their cookies.) */
2017 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2023 /* According to Roger Beeman, we need to initialize tm_isdst, since
2024 strptime won't do it. */
2027 /* Note that under foreign locales Solaris strptime() fails to
2028 recognize English dates, which renders this function useless. We
2029 solve this by being careful not to affect LC_TIME when
2030 initializing locale.
2032 Another solution would be to temporarily set locale to C, invoke
2033 strptime(), and restore it back. This is slow and dirty,
2034 however, and locale support other than LC_MESSAGES can mess other
2035 things, so I rather chose to stick with just setting LC_MESSAGES.
2037 GNU strptime does not have this problem because it recognizes
2038 both international and local dates. */
2040 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2041 if (check_end (strptime (time_string, time_formats[i], &t)))
2042 return mktime_from_utc (&t);
2044 /* All formats have failed. */
2048 /* Authorization support: We support two authorization schemes:
2050 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2052 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2053 consisting of answering to the server's challenge with the proper
2056 /* How many bytes it will take to store LEN bytes in base64. */
2057 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2059 /* Encode the string S of length LENGTH to base64 format and place it
2060 to STORE. STORE will be 0-terminated, and must point to a writable
2061 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2063 base64_encode (const char *s, char *store, int length)
2065 /* Conversion table. */
2066 static char tbl[64] = {
2067 'A','B','C','D','E','F','G','H',
2068 'I','J','K','L','M','N','O','P',
2069 'Q','R','S','T','U','V','W','X',
2070 'Y','Z','a','b','c','d','e','f',
2071 'g','h','i','j','k','l','m','n',
2072 'o','p','q','r','s','t','u','v',
2073 'w','x','y','z','0','1','2','3',
2074 '4','5','6','7','8','9','+','/'
2077 unsigned char *p = (unsigned char *)store;
2079 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2080 for (i = 0; i < length; i += 3)
2082 *p++ = tbl[s[0] >> 2];
2083 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2084 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2085 *p++ = tbl[s[2] & 0x3f];
2088 /* Pad the result if necessary... */
2089 if (i == length + 1)
2091 else if (i == length + 2)
2092 *(p - 1) = *(p - 2) = '=';
2093 /* ...and zero-terminate it. */
2097 /* Create the authentication header contents for the `Basic' scheme.
2098 This is done by encoding the string `USER:PASS' in base64 and
2099 prepending `HEADER: Basic ' to it. */
2101 basic_authentication_encode (const char *user, const char *passwd,
2104 char *t1, *t2, *res;
2105 int len1 = strlen (user) + 1 + strlen (passwd);
2106 int len2 = BASE64_LENGTH (len1);
2108 t1 = (char *)alloca (len1 + 1);
2109 sprintf (t1, "%s:%s", user, passwd);
2110 t2 = (char *)alloca (1 + len2);
2111 base64_encode (t1, t2, len1);
2112 res = (char *)xmalloc (len2 + 11 + strlen (header));
2113 sprintf (res, "%s: Basic %s\r\n", header, t2);
2119 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2120 of a field in such a header. If the field is the one specified by
2121 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2122 digest authorization code), extract its value in the (char*)
2123 variable pointed by RET. Returns negative on a malformed header,
2124 or number of bytes that have been parsed by this call. */
2126 extract_header_attr (const char *au, const char *attr_name, char **ret)
2128 const char *cp, *ep;
2132 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2134 cp += strlen (attr_name);
2137 cp += skip_lws (cp);
2142 cp += skip_lws (cp);
2147 for (ep = cp; *ep && *ep != '\"'; ep++)
2152 *ret = strdupdelim (cp, ep);
2159 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2160 an array of 16 bytes containing the hash keys, and BUF should be a
2161 buffer of 33 writable characters (32 for hex digits plus one for
2162 zero termination). */
2164 dump_hash (unsigned char *buf, const unsigned char *hash)
2168 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2170 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2171 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2176 /* Take the line apart to find the challenge, and compose a digest
2177 authorization header. See RFC2069 section 2.1.2. */
2179 digest_authentication_encode (const char *au, const char *user,
2180 const char *passwd, const char *method,
2183 static char *realm, *opaque, *nonce;
2188 { "realm", &realm },
2189 { "opaque", &opaque },
2194 realm = opaque = nonce = NULL;
2196 au += 6; /* skip over `Digest' */
2201 au += skip_lws (au);
2202 for (i = 0; i < ARRAY_SIZE (options); i++)
2204 int skip = extract_header_attr (au, options[i].name,
2205 options[i].variable);
2209 FREE_MAYBE (opaque);
2219 if (i == ARRAY_SIZE (options))
2221 while (*au && *au != '=')
2225 au += skip_lws (au);
2229 while (*au && *au != '\"')
2236 while (*au && *au != ',')
2241 if (!realm || !nonce || !user || !passwd || !path || !method)
2244 FREE_MAYBE (opaque);
2249 /* Calculate the digest value. */
2251 ALLOCA_MD5_CONTEXT (ctx);
2252 unsigned char hash[MD5_HASHLEN];
2253 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2254 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2256 /* A1BUF = H(user ":" realm ":" password) */
2258 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2259 gen_md5_update ((unsigned char *)":", 1, ctx);
2260 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2261 gen_md5_update ((unsigned char *)":", 1, ctx);
2262 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2263 gen_md5_finish (ctx, hash);
2264 dump_hash (a1buf, hash);
2266 /* A2BUF = H(method ":" path) */
2268 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2269 gen_md5_update ((unsigned char *)":", 1, ctx);
2270 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2271 gen_md5_finish (ctx, hash);
2272 dump_hash (a2buf, hash);
2274 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2276 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2277 gen_md5_update ((unsigned char *)":", 1, ctx);
2278 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2279 gen_md5_update ((unsigned char *)":", 1, ctx);
2280 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2281 gen_md5_finish (ctx, hash);
2282 dump_hash (response_digest, hash);
2284 res = (char*) xmalloc (strlen (user)
2289 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2290 + (opaque ? strlen (opaque) : 0)
2292 sprintf (res, "Authorization: Digest \
2293 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2294 user, realm, nonce, path, response_digest);
2297 char *p = res + strlen (res);
2298 strcat (p, ", opaque=\"");
2302 strcat (res, "\r\n");
2306 #endif /* USE_DIGEST */
2309 #define BEGINS_WITH(line, string_constant) \
2310 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2311 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2312 || !line[sizeof (string_constant) - 1]))
2315 known_authentication_scheme_p (const char *au)
2317 return BEGINS_WITH (au, "Basic")
2318 || BEGINS_WITH (au, "Digest")
2319 || BEGINS_WITH (au, "NTLM");
2324 /* Create the HTTP authorization request header. When the
2325 `WWW-Authenticate' response header is seen, according to the
2326 authorization scheme specified in that header (`Basic' and `Digest'
2327 are supported by the current implementation), produce an
2328 appropriate HTTP authorization request header. */
2330 create_authorization_line (const char *au, const char *user,
2331 const char *passwd, const char *method,
2334 char *wwwauth = NULL;
2336 if (!strncasecmp (au, "Basic", 5))
2337 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2338 if (!strncasecmp (au, "NTLM", 4))
2339 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2341 else if (!strncasecmp (au, "Digest", 6))
2342 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2343 #endif /* USE_DIGEST */
2350 if (pc_last_host_ip)
2351 address_list_release (pc_last_host_ip);