2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
25 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
58 # include "gen_sslfunc.h"
65 extern char *version_string;
71 static int cookies_loaded_p;
73 #define TEXTHTML_S "text/html"
74 #define HTTP_ACCEPT "*/*"
76 /* Some status code validation macros: */
77 #define H_20X(x) (((x) >= 200) && ((x) < 300))
78 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
79 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
80 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
82 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
84 #define HTTP_STATUS_OK 200
85 #define HTTP_STATUS_CREATED 201
86 #define HTTP_STATUS_ACCEPTED 202
87 #define HTTP_STATUS_NO_CONTENT 204
88 #define HTTP_STATUS_PARTIAL_CONTENTS 206
90 /* Redirection 3xx. */
91 #define HTTP_STATUS_MULTIPLE_CHOICES 300
92 #define HTTP_STATUS_MOVED_PERMANENTLY 301
93 #define HTTP_STATUS_MOVED_TEMPORARILY 302
94 #define HTTP_STATUS_NOT_MODIFIED 304
96 /* Client error 4xx. */
97 #define HTTP_STATUS_BAD_REQUEST 400
98 #define HTTP_STATUS_UNAUTHORIZED 401
99 #define HTTP_STATUS_FORBIDDEN 403
100 #define HTTP_STATUS_NOT_FOUND 404
102 /* Server errors 5xx. */
103 #define HTTP_STATUS_INTERNAL 500
104 #define HTTP_STATUS_NOT_IMPLEMENTED 501
105 #define HTTP_STATUS_BAD_GATEWAY 502
106 #define HTTP_STATUS_UNAVAILABLE 503
109 /* Parse the HTTP status line, which is of format:
111 HTTP-Version SP Status-Code SP Reason-Phrase
113 The function returns the status-code, or -1 if the status line is
114 malformed. The pointer to reason-phrase is returned in RP. */
116 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
118 /* (the variables must not be named `major' and `minor', because
119 that breaks compilation with SunOS4 cc.) */
120 int mjr, mnr, statcode;
123 *reason_phrase_ptr = NULL;
125 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
126 major version, and Y is minor version. */
127 if (strncmp (line, "HTTP/", 5) != 0)
131 /* Calculate major HTTP version. */
133 for (mjr = 0; ISDIGIT (*line); line++)
134 mjr = 10 * mjr + (*line - '0');
135 if (*line != '.' || p == line)
139 /* Calculate minor HTTP version. */
141 for (mnr = 0; ISDIGIT (*line); line++)
142 mnr = 10 * mnr + (*line - '0');
143 if (*line != ' ' || p == line)
145 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
146 minor version can be safely ignored. */
151 /* Calculate status code. */
152 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
154 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
156 /* Set up the reason phrase pointer. */
158 /* RFC2068 requires SPC here, but we allow the string to finish
159 here, in case no reason-phrase is present. */
163 *reason_phrase_ptr = line;
168 *reason_phrase_ptr = line + 1;
173 /* Functions to be used as arguments to header_process(): */
175 struct http_process_range_closure {
181 /* Parse the `Content-Range' header and extract the information it
182 contains. Returns 1 if successful, -1 otherwise. */
184 http_process_range (const char *hdr, void *arg)
186 struct http_process_range_closure *closure
187 = (struct http_process_range_closure *)arg;
190 /* Certain versions of Nutscape proxy server send out
191 `Content-Length' without "bytes" specifier, which is a breach of
192 RFC2068 (as well as the HTTP/1.1 draft which was current at the
193 time). But hell, I must support it... */
194 if (!strncasecmp (hdr, "bytes", 5))
197 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
201 hdr += skip_lws (hdr);
207 for (num = 0; ISDIGIT (*hdr); hdr++)
208 num = 10 * num + (*hdr - '0');
209 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
211 closure->first_byte_pos = num;
213 for (num = 0; ISDIGIT (*hdr); hdr++)
214 num = 10 * num + (*hdr - '0');
215 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
217 closure->last_byte_pos = num;
219 for (num = 0; ISDIGIT (*hdr); hdr++)
220 num = 10 * num + (*hdr - '0');
221 closure->entity_length = num;
225 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
226 Used for `Accept-Ranges'. */
228 http_process_none (const char *hdr, void *arg)
230 int *where = (int *)arg;
232 if (strstr (hdr, "none"))
239 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
241 http_process_type (const char *hdr, void *arg)
243 char **result = (char **)arg;
244 /* Locate P on `;' or the terminating zero, whichever comes first. */
245 const char *p = strchr (hdr, ';');
247 p = hdr + strlen (hdr);
248 while (p > hdr && ISSPACE (*(p - 1)))
250 *result = strdupdelim (hdr, p);
254 /* Check whether the `Connection' header is set to "keep-alive". */
256 http_process_connection (const char *hdr, void *arg)
258 int *flag = (int *)arg;
259 if (!strcasecmp (hdr, "Keep-Alive"))
264 /* Persistent connections. Currently, we cache the most recently used
265 connection as persistent, provided that the HTTP server agrees to
266 make it such. The persistence data is stored in the variables
267 below. Ideally, it would be in a structure, and it should be
268 possible to cache an arbitrary fixed number of these connections.
270 I think the code is quite easy to extend in that direction. */
272 /* Whether a persistent connection is active. */
273 static int pc_active_p;
274 /* Host and port of currently active persistent connection. */
275 static struct address_list *pc_last_host_ip;
276 static unsigned short pc_last_port;
278 /* File descriptor of the currently active persistent connection. */
279 static int pc_last_fd;
282 /* Whether a ssl handshake has occoured on this connection */
283 static int pc_active_ssl;
284 /* SSL connection of the currently active persistent connection. */
285 static SSL *pc_last_ssl;
286 #endif /* HAVE_SSL */
288 /* Mark the persistent connection as invalid. This is used by the
289 CLOSE_* macros after they forcefully close a registered persistent
290 connection. This does not close the file descriptor -- it is left
291 to the caller to do that. (Maybe it should, though.) */
294 invalidate_persistent (void)
299 #endif /* HAVE_SSL */
300 if (pc_last_host_ip != NULL)
302 address_list_release (pc_last_host_ip);
303 pc_last_host_ip = NULL;
305 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
308 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
309 persistent. This will enable someone to use the same connection
310 later. In the context of HTTP, this must be called only AFTER the
311 response has been received and the server has promised that the
312 connection will remain alive.
314 If a previous connection was persistent, it is closed. */
317 register_persistent (const char *host, unsigned short port, int fd
325 if (pc_last_fd == fd)
327 /* The connection FD is already registered. Nothing to
333 /* The old persistent connection is still active; let's
334 close it first. This situation arises whenever a
335 persistent connection exists, but we then connect to a
336 different host, and try to register a persistent
337 connection to that one. */
339 /* The ssl disconnect has to take place before the closing
342 shutdown_ssl(pc_last_ssl);
345 invalidate_persistent ();
349 assert (pc_last_host_ip == NULL);
351 /* This lookup_host cannot fail, because it has the results in the
353 pc_last_host_ip = lookup_host (host, 1);
354 assert (pc_last_host_ip != NULL);
361 pc_active_ssl = ssl ? 1 : 0;
363 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
367 # define SHUTDOWN_SSL(ssl) do { \
369 shutdown_ssl (ssl); \
372 # define SHUTDOWN_SSL(ssl)
375 /* Return non-zero if a persistent connection is available for
376 connecting to HOST:PORT. */
379 persistent_available_p (const char *host, unsigned short port
386 struct address_list *this_host_ip;
388 /* First, check whether a persistent connection is active at all. */
391 /* Second, check if the active connection pertains to the correct
392 (HOST, PORT) ordered pair. */
393 if (port != pc_last_port)
397 /* Second, a): check if current connection is (not) ssl, too. This
398 test is unlikely to fail because HTTP and HTTPS typicaly use
399 different ports. Yet it is possible, or so I [Christian
400 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
402 if (ssl != pc_active_ssl)
404 #endif /* HAVE_SSL */
406 this_host_ip = lookup_host (host, 1);
410 /* To equate the two host names for the purposes of persistent
411 connections, they need to share all the IP addresses in the
413 success = address_list_match_all (pc_last_host_ip, this_host_ip);
414 address_list_release (this_host_ip);
418 /* Third: check whether the connection is still open. This is
419 important because most server implement a liberal (short) timeout
420 on persistent connections. Wget can of course always reconnect
421 if the connection doesn't work out, but it's nicer to know in
422 advance. This test is a logical followup of the first test, but
423 is "expensive" and therefore placed at the end of the list. */
424 if (!test_socket_open (pc_last_fd))
426 /* Oops, the socket is no longer open. Now that we know that,
427 let's invalidate the persistent connection before returning
431 SHUTDOWN_SSL (pc_last_ssl);
434 invalidate_persistent ();
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 char *cookies = NULL;
551 /* Whether this connection will be kept alive after the HTTP request
555 /* Flags that detect the two ways of specifying HTTP keep-alive
557 int http_keep_alive_1, http_keep_alive_2;
559 /* Whether keep-alive should be inhibited. */
560 int inhibit_keep_alive;
563 /* initialize ssl_ctx on first run */
566 uerr_t err = init_ssl (&ssl_ctx);
571 case SSLERRCTXCREATE:
573 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
577 /* try without certfile */
578 logprintf (LOG_NOTQUIET,
579 _("Failed to load certificates from %s\n"),
582 logprintf (LOG_NOTQUIET,
583 _("Trying without the specified certificate\n"));
586 logprintf (LOG_NOTQUIET,
587 _("Failed to get certificate key from %s\n"),
590 logprintf (LOG_NOTQUIET,
591 _("Trying without the specified certificate\n"));
598 #endif /* HAVE_SSL */
600 if (!(*dt & HEAD_ONLY))
601 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
602 know the local filename so we can save to it. */
603 assert (*hs->local_file != NULL);
606 auth_tried_already = 0;
608 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
611 /* We need to come back here when the initial attempt to retrieve
612 without authorization header fails. (Expected to happen at least
613 for the Digest authorization scheme.) */
616 http_keep_alive_1 = http_keep_alive_2 = 0;
618 /* Initialize certain elements of struct http_stat. */
623 hs->remote_time = NULL;
626 /* If we're using a proxy, we will be connecting to the proxy
628 conn = proxy ? proxy : u;
630 /* First: establish the connection. */
631 if (inhibit_keep_alive
634 !persistent_available_p (conn->host, conn->port)
636 !persistent_available_p (conn->host, conn->port,
637 u->scheme == SCHEME_HTTPS)
638 #endif /* HAVE_SSL */
641 struct address_list *al = lookup_host (conn->host, 0);
644 set_connection_host_name (conn->host);
645 sock = connect_to_many (al, conn->port, 0);
646 set_connection_host_name (NULL);
647 address_list_release (al);
650 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
653 if (conn->scheme == SCHEME_HTTPS)
654 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
656 logputs (LOG_VERBOSE, "\n");
657 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
662 #endif /* HAVE_SSL */
666 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
667 conn->host, conn->port);
668 /* #### pc_last_fd should be accessed through an accessor
673 #endif /* HAVE_SSL */
674 DEBUGP (("Reusing fd %d.\n", sock));
677 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
682 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
683 sprintf (referer, "Referer: %s\r\n", hs->referer);
686 if (*dt & SEND_NOCACHE)
687 pragma_h = "Pragma: no-cache\r\n";
693 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
694 /* Gag me! Some servers (e.g. WebSitePro) have been known to
695 respond to the following `Range' format by generating a
696 multipart/x-byte-ranges MIME document! This MIME type was
697 present in an old draft of the byteranges specification.
698 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
699 only if multiple non-overlapping ranges are requested --
700 which Wget never does. */
701 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
706 STRDUP_ALLOCA (useragent, opt.useragent);
709 useragent = (char *)alloca (10 + strlen (version_string));
710 sprintf (useragent, "Wget/%s", version_string);
712 /* Construct the authentication, if userid is present. */
715 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
716 user = user ? user : opt.http_user;
717 passwd = passwd ? passwd : opt.http_passwd;
724 /* We have the username and the password, but haven't tried
725 any authorization yet. Let's see if the "Basic" method
726 works. If not, we'll come back here and construct a
727 proper authorization method with the right challenges.
729 If we didn't employ this kind of logic, every URL that
730 requires authorization would have to be processed twice,
731 which is very suboptimal and generates a bunch of false
732 "unauthorized" errors in the server log.
734 #### But this logic also has a serious problem when used
735 with stronger authentications: we *first* transmit the
736 username and the password in clear text, and *then*
737 attempt a stronger authentication scheme. That cannot be
738 right! We are only fortunate that almost everyone still
739 uses the `Basic' scheme anyway.
741 There should be an option to prevent this from happening,
742 for those who use strong authentication schemes and value
744 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
748 /* Use the full path, i.e. one that includes the leading
749 slash and the query string, but is independent of proxy
751 char *pth = url_full_path (u);
752 wwwauth = create_authorization_line (authenticate_h, user, passwd,
761 char *proxy_user, *proxy_passwd;
762 /* For normal username and password, URL components override
763 command-line/wgetrc parameters. With proxy authentication,
764 it's the reverse, because proxy URLs are normally the
765 "permanent" ones, so command-line args should take
767 if (opt.proxy_user && opt.proxy_passwd)
769 proxy_user = opt.proxy_user;
770 proxy_passwd = opt.proxy_passwd;
774 proxy_user = proxy->user;
775 proxy_passwd = proxy->passwd;
777 /* #### This does not appear right. Can't the proxy request,
778 say, `Digest' authentication? */
779 if (proxy_user && proxy_passwd)
780 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
781 "Proxy-Authorization");
784 /* String of the form :PORT. Used only for non-standard ports. */
786 if (u->port != scheme_default_port (u->scheme))
788 port_maybe = (char *)alloca (numdigit (u->port) + 2);
789 sprintf (port_maybe, ":%d", u->port);
792 if (!inhibit_keep_alive)
793 request_keep_alive = "Connection: Keep-Alive\r\n";
795 request_keep_alive = NULL;
798 cookies = build_cookies_request (u->host, u->port, u->path,
800 u->scheme == SCHEME_HTTPS
807 full_path = xstrdup (u->url);
809 /* Use the full path, i.e. one that includes the leading slash and
810 the query string. E.g. if u->path is "foo/bar" and u->query is
811 "param=value", full_path will be "/foo/bar?param=value". */
812 full_path = url_full_path (u);
814 /* Allocate the memory for the request. */
815 request = (char *)alloca (strlen (command)
819 + (port_maybe ? strlen (port_maybe) : 0)
820 + strlen (HTTP_ACCEPT)
821 + (request_keep_alive
822 ? strlen (request_keep_alive) : 0)
823 + (referer ? strlen (referer) : 0)
824 + (cookies ? strlen (cookies) : 0)
825 + (wwwauth ? strlen (wwwauth) : 0)
826 + (proxyauth ? strlen (proxyauth) : 0)
827 + (range ? strlen (range) : 0)
829 + (opt.user_header ? strlen (opt.user_header) : 0)
831 /* Construct the request. */
837 %s%s%s%s%s%s%s%s\r\n",
840 port_maybe ? port_maybe : "",
842 request_keep_alive ? request_keep_alive : "",
843 referer ? referer : "",
844 cookies ? cookies : "",
845 wwwauth ? wwwauth : "",
846 proxyauth ? proxyauth : "",
849 opt.user_header ? opt.user_header : "");
850 DEBUGP (("---request begin---\n%s---request end---\n", request));
852 /* Free the temporary memory. */
853 FREE_MAYBE (wwwauth);
854 FREE_MAYBE (proxyauth);
855 FREE_MAYBE (cookies);
858 /* Send the request to server. */
860 if (conn->scheme == SCHEME_HTTPS)
861 num_written = ssl_iwrite (ssl, request, strlen (request));
863 #endif /* HAVE_SSL */
864 num_written = iwrite (sock, request, strlen (request));
868 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
870 CLOSE_INVALIDATE (sock);
873 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
874 proxy ? "Proxy" : "HTTP");
875 contlen = contrange = -1;
880 /* Before reading anything, initialize the rbuf. */
881 rbuf_initialize (&rbuf, sock);
883 if (conn->scheme == SCHEME_HTTPS)
887 #endif /* HAVE_SSL */
890 /* Header-fetching loop. */
898 /* Get the header. */
899 status = header_get (&rbuf, &hdr,
900 /* Disallow continuations for status line. */
901 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
903 /* Check for errors. */
904 if (status == HG_EOF && *hdr)
906 /* This used to be an unconditional error, but that was
907 somewhat controversial, because of a large number of
908 broken CGI's that happily "forget" to send the second EOL
909 before closing the connection of a HEAD request.
911 So, the deal is to check whether the header is empty
912 (*hdr is zero if it is); if yes, it means that the
913 previous header was fully retrieved, and that -- most
914 probably -- the request is complete. "...be liberal in
915 what you accept." Oh boy. */
916 logputs (LOG_VERBOSE, "\n");
917 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
920 FREE_MAYBE (all_headers);
921 CLOSE_INVALIDATE (sock);
924 else if (status == HG_ERROR)
926 logputs (LOG_VERBOSE, "\n");
927 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
931 FREE_MAYBE (all_headers);
932 CLOSE_INVALIDATE (sock);
936 /* If the headers are to be saved to a file later, save them to
938 if (opt.save_headers)
940 int lh = strlen (hdr);
941 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
942 memcpy (all_headers + all_length, hdr, lh);
944 all_headers[all_length++] = '\n';
945 all_headers[all_length] = '\0';
948 /* Check for status line. */
952 /* Parse the first line of server response. */
953 statcode = parse_http_status_line (hdr, &error);
954 hs->statcode = statcode;
955 /* Store the descriptive response. */
956 if (statcode == -1) /* malformed response */
958 /* A common reason for "malformed response" error is the
959 case when no data was actually received. Handle this
962 hs->error = xstrdup (_("No data received"));
964 hs->error = xstrdup (_("Malformed status line"));
969 hs->error = xstrdup (_("(no description)"));
971 hs->error = xstrdup (error);
979 if (opt.server_response)
980 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
982 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
988 /* Exit on empty header. */
995 /* Print the header if requested. */
996 if (opt.server_response && hcount != 1)
997 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
999 /* Try getting content-length. */
1000 if (contlen == -1 && !opt.ignore_length)
1001 if (header_process (hdr, "Content-Length", header_extract_number,
1004 /* Try getting content-type. */
1006 if (header_process (hdr, "Content-Type", http_process_type, &type))
1008 /* Try getting location. */
1010 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1012 /* Try getting last-modified. */
1013 if (!hs->remote_time)
1014 if (header_process (hdr, "Last-Modified", header_strdup,
1017 /* Try getting cookies. */
1019 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1021 /* Try getting www-authentication. */
1022 if (!authenticate_h)
1023 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1026 /* Check for accept-ranges header. If it contains the word
1027 `none', disable the ranges. */
1028 if (*dt & ACCEPTRANGES)
1031 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1034 *dt &= ~ACCEPTRANGES;
1038 /* Try getting content-range. */
1039 if (contrange == -1)
1041 struct http_process_range_closure closure;
1042 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1044 contrange = closure.first_byte_pos;
1048 /* Check for keep-alive related responses. */
1049 if (!inhibit_keep_alive)
1051 /* Check for the `Keep-Alive' header. */
1052 if (!http_keep_alive_1)
1054 if (header_process (hdr, "Keep-Alive", header_exists,
1055 &http_keep_alive_1))
1058 /* Check for `Connection: Keep-Alive'. */
1059 if (!http_keep_alive_2)
1061 if (header_process (hdr, "Connection", http_process_connection,
1062 &http_keep_alive_2))
1070 logputs (LOG_VERBOSE, "\n");
1073 && (http_keep_alive_1 || http_keep_alive_2))
1075 assert (inhibit_keep_alive == 0);
1079 /* The server has promised that it will not close the connection
1080 when we're done. This means that we can register it. */
1082 register_persistent (conn->host, conn->port, sock);
1084 register_persistent (conn->host, conn->port, sock, ssl);
1085 #endif /* HAVE_SSL */
1087 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1090 /* Authorization is required. */
1094 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1095 might be more bytes in the body. */
1096 if (auth_tried_already)
1098 /* If we have tried it already, then there is not point
1101 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1102 xfree (authenticate_h);
1105 else if (!known_authentication_scheme_p (authenticate_h))
1107 xfree (authenticate_h);
1108 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1111 else if (BEGINS_WITH (authenticate_h, "Basic"))
1113 /* The authentication scheme is basic, the one we try by
1114 default, and it failed. There's no sense in trying
1120 auth_tried_already = 1;
1124 /* We do not need this anymore. */
1127 xfree (authenticate_h);
1128 authenticate_h = NULL;
1131 /* 20x responses are counted among successful by default. */
1132 if (H_20X (statcode))
1135 /* Return if redirected. */
1136 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1138 /* RFC2068 says that in case of the 300 (multiple choices)
1139 response, the server can output a preferred URL through
1140 `Location' header; otherwise, the request should be treated
1141 like GET. So, if the location is set, it will be a
1142 redirection; otherwise, just proceed normally. */
1143 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1147 logprintf (LOG_VERBOSE,
1148 _("Location: %s%s\n"),
1149 hs->newloc ? hs->newloc : _("unspecified"),
1150 hs->newloc ? _(" [following]") : "");
1151 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1152 might be more bytes in the body. */
1154 FREE_MAYBE (all_headers);
1159 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1162 /* We don't assume text/html by default. */
1165 if (opt.html_extension && (*dt & TEXTHTML))
1166 /* -E / --html-extension / html_extension = on was specified, and this is a
1167 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1168 already the file's suffix, tack on ".html". */
1170 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1172 if (last_period_in_local_filename == NULL ||
1173 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1174 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1176 size_t local_filename_len = strlen(*hs->local_file);
1178 *hs->local_file = xrealloc(*hs->local_file,
1179 local_filename_len + sizeof(".html"));
1180 strcpy(*hs->local_file + local_filename_len, ".html");
1182 *dt |= ADDED_HTML_EXTENSION;
1186 if (contrange == -1)
1188 /* We did not get a content-range header. This means that the
1189 server did not honor our `Range' request. Normally, this
1190 means we should reset hs->restval and continue normally. */
1192 /* However, if `-c' is used, we need to be a bit more careful:
1194 1. If `-c' is specified and the file already existed when
1195 Wget was started, it would be a bad idea for us to start
1196 downloading it from scratch, effectively truncating it. I
1197 believe this cannot happen unless `-c' was specified.
1199 2. If `-c' is used on a file that is already fully
1200 downloaded, we're requesting bytes after the end of file,
1201 which can result in server not honoring `Range'. If this is
1202 the case, `Content-Length' will be equal to the length of the
1204 if (opt.always_rest)
1206 /* Check for condition #2. */
1207 if (hs->restval > 0 /* restart was requested. */
1208 && contlen != -1 /* we got content-length. */
1209 && hs->restval >= contlen /* file fully downloaded
1213 logputs (LOG_VERBOSE, _("\
1214 \n The file is already fully retrieved; nothing to do.\n\n"));
1215 /* In case the caller inspects. */
1219 FREE_MAYBE (all_headers);
1220 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1221 might be more bytes in the body. */
1222 return RETRUNNEEDED;
1225 /* Check for condition #1. */
1226 if (hs->no_truncate)
1228 logprintf (LOG_NOTQUIET,
1231 Continued download failed on this file, which conflicts with `-c'.\n\
1232 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1234 FREE_MAYBE (all_headers);
1235 CLOSE_INVALIDATE (sock);
1236 return CONTNOTSUPPORTED;
1244 else if (contrange != hs->restval ||
1245 (H_PARTIAL (statcode) && contrange == -1))
1247 /* This means the whole request was somehow misunderstood by the
1248 server. Bail out. */
1250 FREE_MAYBE (all_headers);
1251 CLOSE_INVALIDATE (sock);
1258 contlen += contrange;
1260 contrange = -1; /* If conent-length was not sent,
1261 content-range will be ignored. */
1263 hs->contlen = contlen;
1267 if ((*dt & RETROKF) && !opt.server_response)
1269 /* No need to print this output if the body won't be
1270 downloaded at all, or if the original server response is
1272 logputs (LOG_VERBOSE, _("Length: "));
1275 logputs (LOG_VERBOSE, legible (contlen));
1276 if (contrange != -1)
1277 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1278 legible (contlen - contrange));
1281 logputs (LOG_VERBOSE,
1282 opt.ignore_length ? _("ignored") : _("unspecified"));
1284 logprintf (LOG_VERBOSE, " [%s]\n", type);
1286 logputs (LOG_VERBOSE, "\n");
1290 type = NULL; /* We don't need it any more. */
1292 /* Return if we have no intention of further downloading. */
1293 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1295 /* In case the caller cares to look... */
1299 FREE_MAYBE (all_headers);
1300 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1301 might be more bytes in the body. */
1302 return RETRFINISHED;
1305 /* Open the local file. */
1308 mkalldirs (*hs->local_file);
1310 rotate_backups (*hs->local_file);
1311 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1314 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1315 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1316 might be more bytes in the body. */
1317 FREE_MAYBE (all_headers);
1323 extern int global_download_count;
1325 /* To ensure that repeated "from scratch" downloads work for -O
1326 files, we rewind the file pointer, unless restval is
1327 non-zero. (This works only when -O is used on regular files,
1328 but it's still a valuable feature.)
1330 However, this loses when more than one URL is specified on
1331 the command line the second rewinds eradicates the contents
1332 of the first download. Thus we disable the above trick for
1333 all the downloads except the very first one.
1335 #### A possible solution to this would be to remember the
1336 file position in the output document and to seek to that
1337 position, instead of rewinding. */
1338 if (!hs->restval && global_download_count == 0)
1340 /* This will silently fail for streams that don't correspond
1341 to regular files, but that's OK. */
1343 /* ftruncate is needed because opt.dfp is opened in append
1344 mode if opt.always_rest is set. */
1345 ftruncate (fileno (fp), 0);
1350 /* #### This confuses the code that checks for file size. There
1351 should be some overhead information. */
1352 if (opt.save_headers)
1353 fwrite (all_headers, 1, all_length, fp);
1355 /* Get the contents of the document. */
1356 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1357 (contlen != -1 ? contlen : 0),
1358 &rbuf, keep_alive, &hs->dltime);
1361 CLOSE_FINISH (sock);
1363 CLOSE_INVALIDATE (sock);
1366 /* Close or flush the file. We have to be careful to check for
1367 error here. Checking the result of fwrite() is not enough --
1368 errors could go unnoticed! */
1371 flush_res = fclose (fp);
1373 flush_res = fflush (fp);
1374 if (flush_res == EOF)
1377 FREE_MAYBE (all_headers);
1380 return RETRFINISHED;
1383 /* The genuine HTTP loop! This is the part where the retrieval is
1384 retried, and retried, and retried, and... */
1386 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1387 int *dt, struct url *proxy)
1390 int use_ts, got_head = 0; /* time-stamping info */
1391 char *filename_plus_orig_suffix;
1392 char *local_filename = NULL;
1393 char *tms, *suf, *locf, *tmrate;
1395 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1396 long local_size = 0; /* the size of the local file */
1397 size_t filename_len;
1398 struct http_stat hstat; /* HTTP status */
1402 /* This used to be done in main(), but it's a better idea to do it
1403 here so that we don't go through the hoops if we're just using
1405 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1407 load_cookies (opt.cookies_input);
1408 cookies_loaded_p = 1;
1413 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1414 has_wildcards_p because it would also warn on `?', and we know that
1415 shows up in CGI paths a *lot*. */
1416 if (strchr (u->url, '*'))
1417 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1419 /* Determine the local filename. */
1420 if (local_file && *local_file)
1421 hstat.local_file = local_file;
1422 else if (local_file)
1424 *local_file = url_filename (u);
1425 hstat.local_file = local_file;
1429 dummy = url_filename (u);
1430 hstat.local_file = &dummy;
1433 if (!opt.output_document)
1434 locf = *hstat.local_file;
1436 locf = opt.output_document;
1438 hstat.referer = referer;
1440 filename_len = strlen (*hstat.local_file);
1441 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1443 if (opt.noclobber && file_exists_p (*hstat.local_file))
1445 /* If opt.noclobber is turned on and file already exists, do not
1446 retrieve the file */
1447 logprintf (LOG_VERBOSE, _("\
1448 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1449 /* If the file is there, we suppose it's retrieved OK. */
1452 /* #### Bogusness alert. */
1453 /* If its suffix is "html" or "htm", assume text/html. */
1454 if (((suf = suffix (*hstat.local_file)) != NULL)
1455 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1463 if (opt.timestamping)
1465 boolean local_dot_orig_file_exists = FALSE;
1467 if (opt.backup_converted)
1468 /* If -K is specified, we'll act on the assumption that it was specified
1469 last time these files were downloaded as well, and instead of just
1470 comparing local file X against server file X, we'll compare local
1471 file X.orig (if extant, else X) against server file X. If -K
1472 _wasn't_ specified last time, or the server contains files called
1473 *.orig, -N will be back to not operating correctly with -k. */
1475 /* Would a single s[n]printf() call be faster? --dan
1477 Definitely not. sprintf() is horribly slow. It's a
1478 different question whether the difference between the two
1479 affects a program. Usually I'd say "no", but at one
1480 point I profiled Wget, and found that a measurable and
1481 non-negligible amount of time was lost calling sprintf()
1482 in url.c. Replacing sprintf with inline calls to
1483 strcpy() and long_to_string() made a difference.
1485 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1486 memcpy (filename_plus_orig_suffix + filename_len,
1487 ".orig", sizeof (".orig"));
1489 /* Try to stat() the .orig file. */
1490 if (stat (filename_plus_orig_suffix, &st) == 0)
1492 local_dot_orig_file_exists = TRUE;
1493 local_filename = filename_plus_orig_suffix;
1497 if (!local_dot_orig_file_exists)
1498 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1499 if (stat (*hstat.local_file, &st) == 0)
1500 local_filename = *hstat.local_file;
1502 if (local_filename != NULL)
1503 /* There was a local file, so we'll check later to see if the version
1504 the server has is the same version we already have, allowing us to
1509 local_size = st.st_size;
1513 /* Reset the counter. */
1515 *dt = 0 | ACCEPTRANGES;
1519 /* Increment the pass counter. */
1521 sleep_between_retrievals (count);
1522 /* Get the current time string. */
1523 tms = time_str (NULL);
1524 /* Print fetch message, if opt.verbose. */
1527 char *hurl = url_string (u, 1);
1531 sprintf (tmp, _("(try:%2d)"), count);
1532 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1533 tms, hurl, tmp, locf);
1535 ws_changetitle (hurl, 1);
1540 /* Default document type is empty. However, if spider mode is
1541 on or time-stamping is employed, HEAD_ONLY commands is
1542 encoded within *dt. */
1543 if (opt.spider || (use_ts && !got_head))
1547 /* Assume no restarting. */
1549 /* Decide whether or not to restart. */
1550 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1551 /* #### this calls access() and then stat(); could be optimized. */
1552 && file_exists_p (locf))
1553 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1554 hstat.restval = st.st_size;
1556 /* In `-c' is used and the file is existing and non-empty,
1557 refuse to truncate it if the server doesn't support continued
1559 hstat.no_truncate = 0;
1560 if (opt.always_rest && hstat.restval)
1561 hstat.no_truncate = 1;
1563 /* Decide whether to send the no-cache directive. We send it in
1565 a) we're using a proxy, and we're past our first retrieval.
1566 Some proxies are notorious for caching incomplete data, so
1567 we require a fresh get.
1568 b) caching is explicitly inhibited. */
1569 if ((proxy && count > 1) /* a */
1570 || !opt.allow_cache /* b */
1572 *dt |= SEND_NOCACHE;
1574 *dt &= ~SEND_NOCACHE;
1576 /* Try fetching the document, or at least its head. */
1577 err = gethttp (u, &hstat, dt, proxy);
1579 /* It's unfortunate that wget determines the local filename before finding
1580 out the Content-Type of the file. Barring a major restructuring of the
1581 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1582 *hstat.local_file to tack on ".html". */
1583 if (!opt.output_document)
1584 locf = *hstat.local_file;
1586 locf = opt.output_document;
1589 tms = time_str (NULL);
1590 /* Get the new location (with or without the redirection). */
1592 *newloc = xstrdup (hstat.newloc);
1595 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1596 case CONERROR: case READERR: case WRITEFAILED:
1598 /* Non-fatal errors continue executing the loop, which will
1599 bring them to "while" statement at the end, to judge
1600 whether the number of tries was exceeded. */
1601 free_hstat (&hstat);
1602 printwhat (count, opt.ntry);
1605 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1606 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1607 /* Fatal errors just return from the function. */
1608 free_hstat (&hstat);
1612 case FWRITEERR: case FOPENERR:
1613 /* Another fatal error. */
1614 logputs (LOG_VERBOSE, "\n");
1615 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1616 *hstat.local_file, strerror (errno));
1617 free_hstat (&hstat);
1622 /* Another fatal error. */
1623 logputs (LOG_VERBOSE, "\n");
1624 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1625 free_hstat (&hstat);
1630 /* Return the new location to the caller. */
1633 logprintf (LOG_NOTQUIET,
1634 _("ERROR: Redirection (%d) without location.\n"),
1636 free_hstat (&hstat);
1640 free_hstat (&hstat);
1645 /* The file was already fully retrieved. */
1646 free_hstat (&hstat);
1651 /* Deal with you later. */
1654 /* All possibilities should have been exhausted. */
1657 if (!(*dt & RETROKF))
1661 /* #### Ugly ugly ugly! */
1662 char *hurl = url_string (u, 1);
1663 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1666 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1667 tms, hstat.statcode, hstat.error);
1668 logputs (LOG_VERBOSE, "\n");
1669 free_hstat (&hstat);
1674 /* Did we get the time-stamp? */
1677 if (opt.timestamping && !hstat.remote_time)
1679 logputs (LOG_NOTQUIET, _("\
1680 Last-modified header missing -- time-stamps turned off.\n"));
1682 else if (hstat.remote_time)
1684 /* Convert the date-string into struct tm. */
1685 tmr = http_atotm (hstat.remote_time);
1686 if (tmr == (time_t) (-1))
1687 logputs (LOG_VERBOSE, _("\
1688 Last-modified header invalid -- time-stamp ignored.\n"));
1692 /* The time-stamping section. */
1697 use_ts = 0; /* no more time-stamping */
1698 count = 0; /* the retrieve count for HEAD is
1700 if (hstat.remote_time && tmr != (time_t) (-1))
1702 /* Now time-stamping can be used validly. Time-stamping
1703 means that if the sizes of the local and remote file
1704 match, and local file is newer than the remote file,
1705 it will not be retrieved. Otherwise, the normal
1706 download procedure is resumed. */
1708 (hstat.contlen == -1 || local_size == hstat.contlen))
1710 logprintf (LOG_VERBOSE, _("\
1711 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1713 free_hstat (&hstat);
1717 else if (tml >= tmr)
1718 logprintf (LOG_VERBOSE, _("\
1719 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1721 logputs (LOG_VERBOSE,
1722 _("Remote file is newer, retrieving.\n"));
1724 free_hstat (&hstat);
1727 if ((tmr != (time_t) (-1))
1729 && ((hstat.len == hstat.contlen) ||
1730 ((hstat.res == 0) &&
1731 ((hstat.contlen == -1) ||
1732 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1734 /* #### This code repeats in http.c and ftp.c. Move it to a
1736 const char *fl = NULL;
1737 if (opt.output_document)
1739 if (opt.od_known_regular)
1740 fl = opt.output_document;
1743 fl = *hstat.local_file;
1747 /* End of time-stamping section. */
1751 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1756 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1758 if (hstat.len == hstat.contlen)
1762 logprintf (LOG_VERBOSE,
1763 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1764 tms, tmrate, locf, hstat.len, hstat.contlen);
1765 logprintf (LOG_NONVERBOSE,
1766 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1767 tms, u->url, hstat.len, hstat.contlen, locf, count);
1770 downloaded_increase (hstat.len);
1772 /* Remember that we downloaded the file for later ".orig" code. */
1773 if (*dt & ADDED_HTML_EXTENSION)
1774 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1776 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1778 free_hstat (&hstat);
1782 else if (hstat.res == 0) /* No read error */
1784 if (hstat.contlen == -1) /* We don't know how much we were supposed
1785 to get, so assume we succeeded. */
1789 logprintf (LOG_VERBOSE,
1790 _("%s (%s) - `%s' saved [%ld]\n\n"),
1791 tms, tmrate, locf, hstat.len);
1792 logprintf (LOG_NONVERBOSE,
1793 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1794 tms, u->url, hstat.len, locf, count);
1797 downloaded_increase (hstat.len);
1799 /* Remember that we downloaded the file for later ".orig" code. */
1800 if (*dt & ADDED_HTML_EXTENSION)
1801 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1803 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1805 free_hstat (&hstat);
1809 else if (hstat.len < hstat.contlen) /* meaning we lost the
1810 connection too soon */
1812 logprintf (LOG_VERBOSE,
1813 _("%s (%s) - Connection closed at byte %ld. "),
1814 tms, tmrate, hstat.len);
1815 printwhat (count, opt.ntry);
1816 free_hstat (&hstat);
1819 else if (!opt.kill_longer) /* meaning we got more than expected */
1821 logprintf (LOG_VERBOSE,
1822 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1823 tms, tmrate, locf, hstat.len, hstat.contlen);
1824 logprintf (LOG_NONVERBOSE,
1825 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1826 tms, u->url, hstat.len, hstat.contlen, locf, count);
1828 downloaded_increase (hstat.len);
1830 /* Remember that we downloaded the file for later ".orig" code. */
1831 if (*dt & ADDED_HTML_EXTENSION)
1832 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1834 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1836 free_hstat (&hstat);
1840 else /* the same, but not accepted */
1842 logprintf (LOG_VERBOSE,
1843 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1844 tms, tmrate, hstat.len, hstat.contlen);
1845 printwhat (count, opt.ntry);
1846 free_hstat (&hstat);
1850 else /* now hstat.res can only be -1 */
1852 if (hstat.contlen == -1)
1854 logprintf (LOG_VERBOSE,
1855 _("%s (%s) - Read error at byte %ld (%s)."),
1856 tms, tmrate, hstat.len, strerror (errno));
1857 printwhat (count, opt.ntry);
1858 free_hstat (&hstat);
1861 else /* hstat.res == -1 and contlen is given */
1863 logprintf (LOG_VERBOSE,
1864 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1865 tms, tmrate, hstat.len, hstat.contlen,
1867 printwhat (count, opt.ntry);
1868 free_hstat (&hstat);
1875 while (!opt.ntry || (count < opt.ntry));
1879 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1880 than local timezone.
1882 mktime is similar but assumes struct tm, also known as the
1883 "broken-down" form of time, is in local time zone. mktime_from_utc
1884 uses mktime to make the conversion understanding that an offset
1885 will be introduced by the local time assumption.
1887 mktime_from_utc then measures the introduced offset by applying
1888 gmtime to the initial result and applying mktime to the resulting
1889 "broken-down" form. The difference between the two mktime results
1890 is the measured offset which is then subtracted from the initial
1891 mktime result to yield a calendar time which is the value returned.
1893 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1894 consistent offset (the non DST offset) since tm and tm+o might be
1895 on opposite sides of a DST change.
1897 Some implementations of mktime return -1 for the nonexistent
1898 localtime hour at the beginning of DST. In this event, use
1899 mktime(tm - 1hr) + 3600.
1903 gmtime(t+o) --> tm+o
1904 mktime(tm+o) --> t+2o
1905 t+o - (t+2o - t+o) = t
1907 Note that glibc contains a function of the same purpose named
1908 `timegm' (reverse of gmtime). But obviously, it is not universally
1909 available, and unfortunately it is not straightforwardly
1910 extractable for use here. Perhaps configure should detect timegm
1911 and use it where available.
1913 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1914 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1915 Further improved by Roger with assistance from Edward J. Sabol
1916 based on input by Jamie Zawinski. */
1919 mktime_from_utc (struct tm *t)
1930 return -1; /* can't deal with output from strptime */
1941 return -1; /* can't deal with output from gmtime */
1944 return (tl - (tb - tl));
1947 /* Check whether the result of strptime() indicates success.
1948 strptime() returns the pointer to how far it got to in the string.
1949 The processing has been successful if the string is at `GMT' or
1950 `+X', or at the end of the string.
1952 In extended regexp parlance, the function returns 1 if P matches
1953 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1954 can return) is considered a failure and 0 is returned. */
1956 check_end (const char *p)
1960 while (ISSPACE (*p))
1963 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1964 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1970 /* Convert the textual specification of time in TIME_STRING to the
1971 number of seconds since the Epoch.
1973 TIME_STRING can be in any of the three formats RFC2068 allows the
1974 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1975 Timezones are ignored, and should be GMT.
1977 Return the computed time_t representation, or -1 if the conversion
1980 This function uses strptime with various string formats for parsing
1981 TIME_STRING. This results in a parser that is not as lenient in
1982 interpreting TIME_STRING as I would like it to be. Being based on
1983 strptime, it always allows shortened months, one-digit days, etc.,
1984 but due to the multitude of formats in which time can be
1985 represented, an ideal HTTP time parser would be even more
1986 forgiving. It should completely ignore things like week days and
1987 concentrate only on the various forms of representing years,
1988 months, days, hours, minutes, and seconds. For example, it would
1989 be nice if it accepted ISO 8601 out of the box.
1991 I've investigated free and PD code for this purpose, but none was
1992 usable. getdate was big and unwieldy, and had potential copyright
1993 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1994 distributed with phttpd, is excellent, but we cannot use it because
1995 it is not assigned to the FSF. So I stuck it with strptime. */
1998 http_atotm (char *time_string)
2000 /* NOTE: Solaris strptime man page claims that %n and %t match white
2001 space, but that's not universally available. Instead, we simply
2002 use ` ' to mean "skip all WS", which works under all strptime
2003 implementations I've tested. */
2005 static const char *time_formats[] = {
2006 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2007 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2008 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2009 (google.com uses this for their cookies.) */
2010 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2016 /* According to Roger Beeman, we need to initialize tm_isdst, since
2017 strptime won't do it. */
2020 /* Note that under foreign locales Solaris strptime() fails to
2021 recognize English dates, which renders this function useless. We
2022 solve this by being careful not to affect LC_TIME when
2023 initializing locale.
2025 Another solution would be to temporarily set locale to C, invoke
2026 strptime(), and restore it back. This is slow and dirty,
2027 however, and locale support other than LC_MESSAGES can mess other
2028 things, so I rather chose to stick with just setting LC_MESSAGES.
2030 GNU strptime does not have this problem because it recognizes
2031 both international and local dates. */
2033 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2034 if (check_end (strptime (time_string, time_formats[i], &t)))
2035 return mktime_from_utc (&t);
2037 /* All formats have failed. */
2041 /* Authorization support: We support two authorization schemes:
2043 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2045 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2046 consisting of answering to the server's challenge with the proper
2049 /* How many bytes it will take to store LEN bytes in base64. */
2050 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2052 /* Encode the string S of length LENGTH to base64 format and place it
2053 to STORE. STORE will be 0-terminated, and must point to a writable
2054 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2056 base64_encode (const char *s, char *store, int length)
2058 /* Conversion table. */
2059 static char tbl[64] = {
2060 'A','B','C','D','E','F','G','H',
2061 'I','J','K','L','M','N','O','P',
2062 'Q','R','S','T','U','V','W','X',
2063 'Y','Z','a','b','c','d','e','f',
2064 'g','h','i','j','k','l','m','n',
2065 'o','p','q','r','s','t','u','v',
2066 'w','x','y','z','0','1','2','3',
2067 '4','5','6','7','8','9','+','/'
2070 unsigned char *p = (unsigned char *)store;
2072 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2073 for (i = 0; i < length; i += 3)
2075 *p++ = tbl[s[0] >> 2];
2076 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2077 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2078 *p++ = tbl[s[2] & 0x3f];
2081 /* Pad the result if necessary... */
2082 if (i == length + 1)
2084 else if (i == length + 2)
2085 *(p - 1) = *(p - 2) = '=';
2086 /* ...and zero-terminate it. */
2090 /* Create the authentication header contents for the `Basic' scheme.
2091 This is done by encoding the string `USER:PASS' in base64 and
2092 prepending `HEADER: Basic ' to it. */
2094 basic_authentication_encode (const char *user, const char *passwd,
2097 char *t1, *t2, *res;
2098 int len1 = strlen (user) + 1 + strlen (passwd);
2099 int len2 = BASE64_LENGTH (len1);
2101 t1 = (char *)alloca (len1 + 1);
2102 sprintf (t1, "%s:%s", user, passwd);
2103 t2 = (char *)alloca (1 + len2);
2104 base64_encode (t1, t2, len1);
2105 res = (char *)xmalloc (len2 + 11 + strlen (header));
2106 sprintf (res, "%s: Basic %s\r\n", header, t2);
2112 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2113 of a field in such a header. If the field is the one specified by
2114 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2115 digest authorization code), extract its value in the (char*)
2116 variable pointed by RET. Returns negative on a malformed header,
2117 or number of bytes that have been parsed by this call. */
2119 extract_header_attr (const char *au, const char *attr_name, char **ret)
2121 const char *cp, *ep;
2125 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2127 cp += strlen (attr_name);
2130 cp += skip_lws (cp);
2135 cp += skip_lws (cp);
2140 for (ep = cp; *ep && *ep != '\"'; ep++)
2145 *ret = strdupdelim (cp, ep);
2152 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2153 an array of 16 bytes containing the hash keys, and BUF should be a
2154 buffer of 33 writable characters (32 for hex digits plus one for
2155 zero termination). */
2157 dump_hash (unsigned char *buf, const unsigned char *hash)
2161 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2163 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2164 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2169 /* Take the line apart to find the challenge, and compose a digest
2170 authorization header. See RFC2069 section 2.1.2. */
2172 digest_authentication_encode (const char *au, const char *user,
2173 const char *passwd, const char *method,
2176 static char *realm, *opaque, *nonce;
2181 { "realm", &realm },
2182 { "opaque", &opaque },
2187 realm = opaque = nonce = NULL;
2189 au += 6; /* skip over `Digest' */
2194 au += skip_lws (au);
2195 for (i = 0; i < ARRAY_SIZE (options); i++)
2197 int skip = extract_header_attr (au, options[i].name,
2198 options[i].variable);
2202 FREE_MAYBE (opaque);
2212 if (i == ARRAY_SIZE (options))
2214 while (*au && *au != '=')
2218 au += skip_lws (au);
2222 while (*au && *au != '\"')
2229 while (*au && *au != ',')
2234 if (!realm || !nonce || !user || !passwd || !path || !method)
2237 FREE_MAYBE (opaque);
2242 /* Calculate the digest value. */
2244 ALLOCA_MD5_CONTEXT (ctx);
2245 unsigned char hash[MD5_HASHLEN];
2246 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2247 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2249 /* A1BUF = H(user ":" realm ":" password) */
2251 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2252 gen_md5_update ((unsigned char *)":", 1, ctx);
2253 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2254 gen_md5_update ((unsigned char *)":", 1, ctx);
2255 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2256 gen_md5_finish (ctx, hash);
2257 dump_hash (a1buf, hash);
2259 /* A2BUF = H(method ":" path) */
2261 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2262 gen_md5_update ((unsigned char *)":", 1, ctx);
2263 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2264 gen_md5_finish (ctx, hash);
2265 dump_hash (a2buf, hash);
2267 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2269 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2270 gen_md5_update ((unsigned char *)":", 1, ctx);
2271 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2272 gen_md5_update ((unsigned char *)":", 1, ctx);
2273 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2274 gen_md5_finish (ctx, hash);
2275 dump_hash (response_digest, hash);
2277 res = (char*) xmalloc (strlen (user)
2282 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2283 + (opaque ? strlen (opaque) : 0)
2285 sprintf (res, "Authorization: Digest \
2286 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2287 user, realm, nonce, path, response_digest);
2290 char *p = res + strlen (res);
2291 strcat (p, ", opaque=\"");
2295 strcat (res, "\r\n");
2299 #endif /* USE_DIGEST */
2302 #define BEGINS_WITH(line, string_constant) \
2303 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2304 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2305 || !line[sizeof (string_constant) - 1]))
2308 known_authentication_scheme_p (const char *au)
2310 return BEGINS_WITH (au, "Basic")
2311 || BEGINS_WITH (au, "Digest")
2312 || BEGINS_WITH (au, "NTLM");
2317 /* Create the HTTP authorization request header. When the
2318 `WWW-Authenticate' response header is seen, according to the
2319 authorization scheme specified in that header (`Basic' and `Digest'
2320 are supported by the current implementation), produce an
2321 appropriate HTTP authorization request header. */
2323 create_authorization_line (const char *au, const char *user,
2324 const char *passwd, const char *method,
2327 char *wwwauth = NULL;
2329 if (!strncasecmp (au, "Basic", 5))
2330 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2331 if (!strncasecmp (au, "NTLM", 4))
2332 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2334 else if (!strncasecmp (au, "Digest", 6))
2335 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2336 #endif /* USE_DIGEST */
2343 if (pc_last_host_ip)
2344 address_list_release (pc_last_host_ip);