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;
562 /* Whether we need to print the host header with braces around host,
563 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
564 "Host: symbolic-name:1234". */
565 int squares_around_host = 0;
568 /* initialize ssl_ctx on first run */
571 uerr_t err = init_ssl (&ssl_ctx);
576 case SSLERRCTXCREATE:
578 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
582 /* try without certfile */
583 logprintf (LOG_NOTQUIET,
584 _("Failed to load certificates from %s\n"),
587 logprintf (LOG_NOTQUIET,
588 _("Trying without the specified certificate\n"));
591 logprintf (LOG_NOTQUIET,
592 _("Failed to get certificate key from %s\n"),
595 logprintf (LOG_NOTQUIET,
596 _("Trying without the specified certificate\n"));
603 #endif /* HAVE_SSL */
605 if (!(*dt & HEAD_ONLY))
606 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
607 know the local filename so we can save to it. */
608 assert (*hs->local_file != NULL);
611 auth_tried_already = 0;
613 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
616 /* We need to come back here when the initial attempt to retrieve
617 without authorization header fails. (Expected to happen at least
618 for the Digest authorization scheme.) */
621 http_keep_alive_1 = http_keep_alive_2 = 0;
623 /* Initialize certain elements of struct http_stat. */
628 hs->remote_time = NULL;
631 /* If we're using a proxy, we will be connecting to the proxy
633 conn = proxy ? proxy : u;
635 /* First: establish the connection. */
636 if (inhibit_keep_alive
639 !persistent_available_p (conn->host, conn->port)
641 !persistent_available_p (conn->host, conn->port,
642 u->scheme == SCHEME_HTTPS)
643 #endif /* HAVE_SSL */
646 struct address_list *al = lookup_host (conn->host, 0);
649 set_connection_host_name (conn->host);
650 sock = connect_to_many (al, conn->port, 0);
651 set_connection_host_name (NULL);
652 address_list_release (al);
655 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
658 if (conn->scheme == SCHEME_HTTPS)
659 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
661 logputs (LOG_VERBOSE, "\n");
662 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
667 #endif /* HAVE_SSL */
671 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
672 conn->host, conn->port);
673 /* #### pc_last_fd should be accessed through an accessor
678 #endif /* HAVE_SSL */
679 DEBUGP (("Reusing fd %d.\n", sock));
682 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
687 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
688 sprintf (referer, "Referer: %s\r\n", hs->referer);
691 if (*dt & SEND_NOCACHE)
692 pragma_h = "Pragma: no-cache\r\n";
698 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
699 /* Gag me! Some servers (e.g. WebSitePro) have been known to
700 respond to the following `Range' format by generating a
701 multipart/x-byte-ranges MIME document! This MIME type was
702 present in an old draft of the byteranges specification.
703 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
704 only if multiple non-overlapping ranges are requested --
705 which Wget never does. */
706 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
711 STRDUP_ALLOCA (useragent, opt.useragent);
714 useragent = (char *)alloca (10 + strlen (version_string));
715 sprintf (useragent, "Wget/%s", version_string);
717 /* Construct the authentication, if userid is present. */
720 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
721 user = user ? user : opt.http_user;
722 passwd = passwd ? passwd : opt.http_passwd;
729 /* We have the username and the password, but haven't tried
730 any authorization yet. Let's see if the "Basic" method
731 works. If not, we'll come back here and construct a
732 proper authorization method with the right challenges.
734 If we didn't employ this kind of logic, every URL that
735 requires authorization would have to be processed twice,
736 which is very suboptimal and generates a bunch of false
737 "unauthorized" errors in the server log.
739 #### But this logic also has a serious problem when used
740 with stronger authentications: we *first* transmit the
741 username and the password in clear text, and *then*
742 attempt a stronger authentication scheme. That cannot be
743 right! We are only fortunate that almost everyone still
744 uses the `Basic' scheme anyway.
746 There should be an option to prevent this from happening,
747 for those who use strong authentication schemes and value
749 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
753 /* Use the full path, i.e. one that includes the leading
754 slash and the query string, but is independent of proxy
756 char *pth = url_full_path (u);
757 wwwauth = create_authorization_line (authenticate_h, user, passwd,
766 char *proxy_user, *proxy_passwd;
767 /* For normal username and password, URL components override
768 command-line/wgetrc parameters. With proxy authentication,
769 it's the reverse, because proxy URLs are normally the
770 "permanent" ones, so command-line args should take
772 if (opt.proxy_user && opt.proxy_passwd)
774 proxy_user = opt.proxy_user;
775 proxy_passwd = opt.proxy_passwd;
779 proxy_user = proxy->user;
780 proxy_passwd = proxy->passwd;
782 /* #### This does not appear right. Can't the proxy request,
783 say, `Digest' authentication? */
784 if (proxy_user && proxy_passwd)
785 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
786 "Proxy-Authorization");
789 /* String of the form :PORT. Used only for non-standard ports. */
791 if (u->port != scheme_default_port (u->scheme))
793 port_maybe = (char *)alloca (numdigit (u->port) + 2);
794 sprintf (port_maybe, ":%d", u->port);
797 if (!inhibit_keep_alive)
798 request_keep_alive = "Connection: Keep-Alive\r\n";
800 request_keep_alive = NULL;
803 cookies = build_cookies_request (u->host, u->port, u->path,
805 u->scheme == SCHEME_HTTPS
812 full_path = xstrdup (u->url);
814 /* Use the full path, i.e. one that includes the leading slash and
815 the query string. E.g. if u->path is "foo/bar" and u->query is
816 "param=value", full_path will be "/foo/bar?param=value". */
817 full_path = url_full_path (u);
819 if (strchr (u->host, ':'))
820 squares_around_host = 1;
822 /* Allocate the memory for the request. */
823 request = (char *)alloca (strlen (command)
827 + (port_maybe ? strlen (port_maybe) : 0)
828 + strlen (HTTP_ACCEPT)
829 + (request_keep_alive
830 ? strlen (request_keep_alive) : 0)
831 + (referer ? strlen (referer) : 0)
832 + (cookies ? strlen (cookies) : 0)
833 + (wwwauth ? strlen (wwwauth) : 0)
834 + (proxyauth ? strlen (proxyauth) : 0)
835 + (range ? strlen (range) : 0)
837 + (opt.user_header ? strlen (opt.user_header) : 0)
839 /* Construct the request. */
845 %s%s%s%s%s%s%s%s\r\n",
848 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
849 port_maybe ? port_maybe : "",
851 request_keep_alive ? request_keep_alive : "",
852 referer ? referer : "",
853 cookies ? cookies : "",
854 wwwauth ? wwwauth : "",
855 proxyauth ? proxyauth : "",
858 opt.user_header ? opt.user_header : "");
859 DEBUGP (("---request begin---\n%s---request end---\n", request));
861 /* Free the temporary memory. */
862 FREE_MAYBE (wwwauth);
863 FREE_MAYBE (proxyauth);
864 FREE_MAYBE (cookies);
867 /* Send the request to server. */
869 if (conn->scheme == SCHEME_HTTPS)
870 num_written = ssl_iwrite (ssl, request, strlen (request));
872 #endif /* HAVE_SSL */
873 num_written = iwrite (sock, request, strlen (request));
877 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
879 CLOSE_INVALIDATE (sock);
882 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
883 proxy ? "Proxy" : "HTTP");
884 contlen = contrange = -1;
889 /* Before reading anything, initialize the rbuf. */
890 rbuf_initialize (&rbuf, sock);
892 if (conn->scheme == SCHEME_HTTPS)
896 #endif /* HAVE_SSL */
899 /* Header-fetching loop. */
907 /* Get the header. */
908 status = header_get (&rbuf, &hdr,
909 /* Disallow continuations for status line. */
910 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
912 /* Check for errors. */
913 if (status == HG_EOF && *hdr)
915 /* This used to be an unconditional error, but that was
916 somewhat controversial, because of a large number of
917 broken CGI's that happily "forget" to send the second EOL
918 before closing the connection of a HEAD request.
920 So, the deal is to check whether the header is empty
921 (*hdr is zero if it is); if yes, it means that the
922 previous header was fully retrieved, and that -- most
923 probably -- the request is complete. "...be liberal in
924 what you accept." Oh boy. */
925 logputs (LOG_VERBOSE, "\n");
926 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
929 FREE_MAYBE (all_headers);
930 CLOSE_INVALIDATE (sock);
933 else if (status == HG_ERROR)
935 logputs (LOG_VERBOSE, "\n");
936 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
940 FREE_MAYBE (all_headers);
941 CLOSE_INVALIDATE (sock);
945 /* If the headers are to be saved to a file later, save them to
947 if (opt.save_headers)
949 int lh = strlen (hdr);
950 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
951 memcpy (all_headers + all_length, hdr, lh);
953 all_headers[all_length++] = '\n';
954 all_headers[all_length] = '\0';
957 /* Check for status line. */
961 /* Parse the first line of server response. */
962 statcode = parse_http_status_line (hdr, &error);
963 hs->statcode = statcode;
964 /* Store the descriptive response. */
965 if (statcode == -1) /* malformed response */
967 /* A common reason for "malformed response" error is the
968 case when no data was actually received. Handle this
971 hs->error = xstrdup (_("No data received"));
973 hs->error = xstrdup (_("Malformed status line"));
978 hs->error = xstrdup (_("(no description)"));
980 hs->error = xstrdup (error);
988 if (opt.server_response)
989 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
991 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
997 /* Exit on empty header. */
1004 /* Print the header if requested. */
1005 if (opt.server_response && hcount != 1)
1006 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1008 /* Try getting content-length. */
1009 if (contlen == -1 && !opt.ignore_length)
1010 if (header_process (hdr, "Content-Length", header_extract_number,
1013 /* Try getting content-type. */
1015 if (header_process (hdr, "Content-Type", http_process_type, &type))
1017 /* Try getting location. */
1019 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1021 /* Try getting last-modified. */
1022 if (!hs->remote_time)
1023 if (header_process (hdr, "Last-Modified", header_strdup,
1026 /* Try getting cookies. */
1028 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1030 /* Try getting www-authentication. */
1031 if (!authenticate_h)
1032 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1035 /* Check for accept-ranges header. If it contains the word
1036 `none', disable the ranges. */
1037 if (*dt & ACCEPTRANGES)
1040 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1043 *dt &= ~ACCEPTRANGES;
1047 /* Try getting content-range. */
1048 if (contrange == -1)
1050 struct http_process_range_closure closure;
1051 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1053 contrange = closure.first_byte_pos;
1057 /* Check for keep-alive related responses. */
1058 if (!inhibit_keep_alive)
1060 /* Check for the `Keep-Alive' header. */
1061 if (!http_keep_alive_1)
1063 if (header_process (hdr, "Keep-Alive", header_exists,
1064 &http_keep_alive_1))
1067 /* Check for `Connection: Keep-Alive'. */
1068 if (!http_keep_alive_2)
1070 if (header_process (hdr, "Connection", http_process_connection,
1071 &http_keep_alive_2))
1079 logputs (LOG_VERBOSE, "\n");
1082 && (http_keep_alive_1 || http_keep_alive_2))
1084 assert (inhibit_keep_alive == 0);
1088 /* The server has promised that it will not close the connection
1089 when we're done. This means that we can register it. */
1091 register_persistent (conn->host, conn->port, sock);
1093 register_persistent (conn->host, conn->port, sock, ssl);
1094 #endif /* HAVE_SSL */
1096 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1099 /* Authorization is required. */
1103 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1104 might be more bytes in the body. */
1105 if (auth_tried_already)
1107 /* If we have tried it already, then there is not point
1110 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1111 xfree (authenticate_h);
1114 else if (!known_authentication_scheme_p (authenticate_h))
1116 xfree (authenticate_h);
1117 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1120 else if (BEGINS_WITH (authenticate_h, "Basic"))
1122 /* The authentication scheme is basic, the one we try by
1123 default, and it failed. There's no sense in trying
1129 auth_tried_already = 1;
1133 /* We do not need this anymore. */
1136 xfree (authenticate_h);
1137 authenticate_h = NULL;
1140 /* 20x responses are counted among successful by default. */
1141 if (H_20X (statcode))
1144 /* Return if redirected. */
1145 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1147 /* RFC2068 says that in case of the 300 (multiple choices)
1148 response, the server can output a preferred URL through
1149 `Location' header; otherwise, the request should be treated
1150 like GET. So, if the location is set, it will be a
1151 redirection; otherwise, just proceed normally. */
1152 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1156 logprintf (LOG_VERBOSE,
1157 _("Location: %s%s\n"),
1158 hs->newloc ? hs->newloc : _("unspecified"),
1159 hs->newloc ? _(" [following]") : "");
1160 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1161 might be more bytes in the body. */
1163 FREE_MAYBE (all_headers);
1168 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1171 /* We don't assume text/html by default. */
1174 if (opt.html_extension && (*dt & TEXTHTML))
1175 /* -E / --html-extension / html_extension = on was specified, and this is a
1176 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1177 already the file's suffix, tack on ".html". */
1179 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1181 if (last_period_in_local_filename == NULL ||
1182 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1183 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1185 size_t local_filename_len = strlen(*hs->local_file);
1187 *hs->local_file = xrealloc(*hs->local_file,
1188 local_filename_len + sizeof(".html"));
1189 strcpy(*hs->local_file + local_filename_len, ".html");
1191 *dt |= ADDED_HTML_EXTENSION;
1195 if (contrange == -1)
1197 /* We did not get a content-range header. This means that the
1198 server did not honor our `Range' request. Normally, this
1199 means we should reset hs->restval and continue normally. */
1201 /* However, if `-c' is used, we need to be a bit more careful:
1203 1. If `-c' is specified and the file already existed when
1204 Wget was started, it would be a bad idea for us to start
1205 downloading it from scratch, effectively truncating it. I
1206 believe this cannot happen unless `-c' was specified.
1208 2. If `-c' is used on a file that is already fully
1209 downloaded, we're requesting bytes after the end of file,
1210 which can result in server not honoring `Range'. If this is
1211 the case, `Content-Length' will be equal to the length of the
1213 if (opt.always_rest)
1215 /* Check for condition #2. */
1216 if (hs->restval > 0 /* restart was requested. */
1217 && contlen != -1 /* we got content-length. */
1218 && hs->restval >= contlen /* file fully downloaded
1222 logputs (LOG_VERBOSE, _("\
1223 \n The file is already fully retrieved; nothing to do.\n\n"));
1224 /* In case the caller inspects. */
1227 /* Mark as successfully retrieved. */
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_FINISH (sock);
1374 CLOSE_INVALIDATE (sock);
1377 /* Close or flush the file. We have to be careful to check for
1378 error here. Checking the result of fwrite() is not enough --
1379 errors could go unnoticed! */
1382 flush_res = fclose (fp);
1384 flush_res = fflush (fp);
1385 if (flush_res == EOF)
1388 FREE_MAYBE (all_headers);
1391 return RETRFINISHED;
1394 /* The genuine HTTP loop! This is the part where the retrieval is
1395 retried, and retried, and retried, and... */
1397 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1398 int *dt, struct url *proxy)
1401 int use_ts, got_head = 0; /* time-stamping info */
1402 char *filename_plus_orig_suffix;
1403 char *local_filename = NULL;
1404 char *tms, *suf, *locf, *tmrate;
1406 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1407 long local_size = 0; /* the size of the local file */
1408 size_t filename_len;
1409 struct http_stat hstat; /* HTTP status */
1413 /* This used to be done in main(), but it's a better idea to do it
1414 here so that we don't go through the hoops if we're just using
1416 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1418 load_cookies (opt.cookies_input);
1419 cookies_loaded_p = 1;
1424 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1425 has_wildcards_p because it would also warn on `?', and we know that
1426 shows up in CGI paths a *lot*. */
1427 if (strchr (u->url, '*'))
1428 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1430 /* Determine the local filename. */
1431 if (local_file && *local_file)
1432 hstat.local_file = local_file;
1433 else if (local_file)
1435 *local_file = url_filename (u);
1436 hstat.local_file = local_file;
1440 dummy = url_filename (u);
1441 hstat.local_file = &dummy;
1444 if (!opt.output_document)
1445 locf = *hstat.local_file;
1447 locf = opt.output_document;
1449 hstat.referer = referer;
1451 filename_len = strlen (*hstat.local_file);
1452 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1454 if (opt.noclobber && file_exists_p (*hstat.local_file))
1456 /* If opt.noclobber is turned on and file already exists, do not
1457 retrieve the file */
1458 logprintf (LOG_VERBOSE, _("\
1459 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1460 /* If the file is there, we suppose it's retrieved OK. */
1463 /* #### Bogusness alert. */
1464 /* If its suffix is "html" or "htm", assume text/html. */
1465 if (((suf = suffix (*hstat.local_file)) != NULL)
1466 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1474 if (opt.timestamping)
1476 boolean local_dot_orig_file_exists = FALSE;
1478 if (opt.backup_converted)
1479 /* If -K is specified, we'll act on the assumption that it was specified
1480 last time these files were downloaded as well, and instead of just
1481 comparing local file X against server file X, we'll compare local
1482 file X.orig (if extant, else X) against server file X. If -K
1483 _wasn't_ specified last time, or the server contains files called
1484 *.orig, -N will be back to not operating correctly with -k. */
1486 /* Would a single s[n]printf() call be faster? --dan
1488 Definitely not. sprintf() is horribly slow. It's a
1489 different question whether the difference between the two
1490 affects a program. Usually I'd say "no", but at one
1491 point I profiled Wget, and found that a measurable and
1492 non-negligible amount of time was lost calling sprintf()
1493 in url.c. Replacing sprintf with inline calls to
1494 strcpy() and long_to_string() made a difference.
1496 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1497 memcpy (filename_plus_orig_suffix + filename_len,
1498 ".orig", sizeof (".orig"));
1500 /* Try to stat() the .orig file. */
1501 if (stat (filename_plus_orig_suffix, &st) == 0)
1503 local_dot_orig_file_exists = TRUE;
1504 local_filename = filename_plus_orig_suffix;
1508 if (!local_dot_orig_file_exists)
1509 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1510 if (stat (*hstat.local_file, &st) == 0)
1511 local_filename = *hstat.local_file;
1513 if (local_filename != NULL)
1514 /* There was a local file, so we'll check later to see if the version
1515 the server has is the same version we already have, allowing us to
1520 local_size = st.st_size;
1524 /* Reset the counter. */
1526 *dt = 0 | ACCEPTRANGES;
1530 /* Increment the pass counter. */
1532 sleep_between_retrievals (count);
1533 /* Get the current time string. */
1534 tms = time_str (NULL);
1535 /* Print fetch message, if opt.verbose. */
1538 char *hurl = url_string (u, 1);
1542 sprintf (tmp, _("(try:%2d)"), count);
1543 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1544 tms, hurl, tmp, locf);
1546 ws_changetitle (hurl, 1);
1551 /* Default document type is empty. However, if spider mode is
1552 on or time-stamping is employed, HEAD_ONLY commands is
1553 encoded within *dt. */
1554 if (opt.spider || (use_ts && !got_head))
1558 /* Assume no restarting. */
1560 /* Decide whether or not to restart. */
1561 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1562 /* #### this calls access() and then stat(); could be optimized. */
1563 && file_exists_p (locf))
1564 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1565 hstat.restval = st.st_size;
1567 /* In `-c' is used and the file is existing and non-empty,
1568 refuse to truncate it if the server doesn't support continued
1570 hstat.no_truncate = 0;
1571 if (opt.always_rest && hstat.restval)
1572 hstat.no_truncate = 1;
1574 /* Decide whether to send the no-cache directive. We send it in
1576 a) we're using a proxy, and we're past our first retrieval.
1577 Some proxies are notorious for caching incomplete data, so
1578 we require a fresh get.
1579 b) caching is explicitly inhibited. */
1580 if ((proxy && count > 1) /* a */
1581 || !opt.allow_cache /* b */
1583 *dt |= SEND_NOCACHE;
1585 *dt &= ~SEND_NOCACHE;
1587 /* Try fetching the document, or at least its head. */
1588 err = gethttp (u, &hstat, dt, proxy);
1590 /* It's unfortunate that wget determines the local filename before finding
1591 out the Content-Type of the file. Barring a major restructuring of the
1592 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1593 *hstat.local_file to tack on ".html". */
1594 if (!opt.output_document)
1595 locf = *hstat.local_file;
1597 locf = opt.output_document;
1600 tms = time_str (NULL);
1601 /* Get the new location (with or without the redirection). */
1603 *newloc = xstrdup (hstat.newloc);
1606 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1607 case CONERROR: case READERR: case WRITEFAILED:
1609 /* Non-fatal errors continue executing the loop, which will
1610 bring them to "while" statement at the end, to judge
1611 whether the number of tries was exceeded. */
1612 free_hstat (&hstat);
1613 printwhat (count, opt.ntry);
1616 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1617 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1618 /* Fatal errors just return from the function. */
1619 free_hstat (&hstat);
1623 case FWRITEERR: case FOPENERR:
1624 /* Another fatal error. */
1625 logputs (LOG_VERBOSE, "\n");
1626 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1627 *hstat.local_file, strerror (errno));
1628 free_hstat (&hstat);
1633 /* Another fatal error. */
1634 logputs (LOG_VERBOSE, "\n");
1635 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1636 free_hstat (&hstat);
1641 /* Return the new location to the caller. */
1644 logprintf (LOG_NOTQUIET,
1645 _("ERROR: Redirection (%d) without location.\n"),
1647 free_hstat (&hstat);
1651 free_hstat (&hstat);
1656 /* The file was already fully retrieved. */
1657 free_hstat (&hstat);
1662 /* Deal with you later. */
1665 /* All possibilities should have been exhausted. */
1668 if (!(*dt & RETROKF))
1672 /* #### Ugly ugly ugly! */
1673 char *hurl = url_string (u, 1);
1674 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1677 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1678 tms, hstat.statcode, hstat.error);
1679 logputs (LOG_VERBOSE, "\n");
1680 free_hstat (&hstat);
1685 /* Did we get the time-stamp? */
1688 if (opt.timestamping && !hstat.remote_time)
1690 logputs (LOG_NOTQUIET, _("\
1691 Last-modified header missing -- time-stamps turned off.\n"));
1693 else if (hstat.remote_time)
1695 /* Convert the date-string into struct tm. */
1696 tmr = http_atotm (hstat.remote_time);
1697 if (tmr == (time_t) (-1))
1698 logputs (LOG_VERBOSE, _("\
1699 Last-modified header invalid -- time-stamp ignored.\n"));
1703 /* The time-stamping section. */
1708 use_ts = 0; /* no more time-stamping */
1709 count = 0; /* the retrieve count for HEAD is
1711 if (hstat.remote_time && tmr != (time_t) (-1))
1713 /* Now time-stamping can be used validly. Time-stamping
1714 means that if the sizes of the local and remote file
1715 match, and local file is newer than the remote file,
1716 it will not be retrieved. Otherwise, the normal
1717 download procedure is resumed. */
1719 (hstat.contlen == -1 || local_size == hstat.contlen))
1721 logprintf (LOG_VERBOSE, _("\
1722 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1724 free_hstat (&hstat);
1728 else if (tml >= tmr)
1729 logprintf (LOG_VERBOSE, _("\
1730 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1732 logputs (LOG_VERBOSE,
1733 _("Remote file is newer, retrieving.\n"));
1735 free_hstat (&hstat);
1738 if ((tmr != (time_t) (-1))
1740 && ((hstat.len == hstat.contlen) ||
1741 ((hstat.res == 0) &&
1742 ((hstat.contlen == -1) ||
1743 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1745 /* #### This code repeats in http.c and ftp.c. Move it to a
1747 const char *fl = NULL;
1748 if (opt.output_document)
1750 if (opt.od_known_regular)
1751 fl = opt.output_document;
1754 fl = *hstat.local_file;
1758 /* End of time-stamping section. */
1762 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1767 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1769 if (hstat.len == hstat.contlen)
1773 logprintf (LOG_VERBOSE,
1774 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1775 tms, tmrate, locf, hstat.len, hstat.contlen);
1776 logprintf (LOG_NONVERBOSE,
1777 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1778 tms, u->url, hstat.len, hstat.contlen, locf, count);
1781 downloaded_increase (hstat.len);
1783 /* Remember that we downloaded the file for later ".orig" code. */
1784 if (*dt & ADDED_HTML_EXTENSION)
1785 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1787 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1789 free_hstat (&hstat);
1793 else if (hstat.res == 0) /* No read error */
1795 if (hstat.contlen == -1) /* We don't know how much we were supposed
1796 to get, so assume we succeeded. */
1800 logprintf (LOG_VERBOSE,
1801 _("%s (%s) - `%s' saved [%ld]\n\n"),
1802 tms, tmrate, locf, hstat.len);
1803 logprintf (LOG_NONVERBOSE,
1804 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1805 tms, u->url, hstat.len, locf, count);
1808 downloaded_increase (hstat.len);
1810 /* Remember that we downloaded the file for later ".orig" code. */
1811 if (*dt & ADDED_HTML_EXTENSION)
1812 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1814 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1816 free_hstat (&hstat);
1820 else if (hstat.len < hstat.contlen) /* meaning we lost the
1821 connection too soon */
1823 logprintf (LOG_VERBOSE,
1824 _("%s (%s) - Connection closed at byte %ld. "),
1825 tms, tmrate, hstat.len);
1826 printwhat (count, opt.ntry);
1827 free_hstat (&hstat);
1830 else if (!opt.kill_longer) /* meaning we got more than expected */
1832 logprintf (LOG_VERBOSE,
1833 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1834 tms, tmrate, locf, hstat.len, hstat.contlen);
1835 logprintf (LOG_NONVERBOSE,
1836 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1837 tms, u->url, hstat.len, hstat.contlen, locf, count);
1839 downloaded_increase (hstat.len);
1841 /* Remember that we downloaded the file for later ".orig" code. */
1842 if (*dt & ADDED_HTML_EXTENSION)
1843 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1845 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1847 free_hstat (&hstat);
1851 else /* the same, but not accepted */
1853 logprintf (LOG_VERBOSE,
1854 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1855 tms, tmrate, hstat.len, hstat.contlen);
1856 printwhat (count, opt.ntry);
1857 free_hstat (&hstat);
1861 else /* now hstat.res can only be -1 */
1863 if (hstat.contlen == -1)
1865 logprintf (LOG_VERBOSE,
1866 _("%s (%s) - Read error at byte %ld (%s)."),
1867 tms, tmrate, hstat.len, strerror (errno));
1868 printwhat (count, opt.ntry);
1869 free_hstat (&hstat);
1872 else /* hstat.res == -1 and contlen is given */
1874 logprintf (LOG_VERBOSE,
1875 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1876 tms, tmrate, hstat.len, hstat.contlen,
1878 printwhat (count, opt.ntry);
1879 free_hstat (&hstat);
1886 while (!opt.ntry || (count < opt.ntry));
1890 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1891 than local timezone.
1893 mktime is similar but assumes struct tm, also known as the
1894 "broken-down" form of time, is in local time zone. mktime_from_utc
1895 uses mktime to make the conversion understanding that an offset
1896 will be introduced by the local time assumption.
1898 mktime_from_utc then measures the introduced offset by applying
1899 gmtime to the initial result and applying mktime to the resulting
1900 "broken-down" form. The difference between the two mktime results
1901 is the measured offset which is then subtracted from the initial
1902 mktime result to yield a calendar time which is the value returned.
1904 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1905 consistent offset (the non DST offset) since tm and tm+o might be
1906 on opposite sides of a DST change.
1908 Some implementations of mktime return -1 for the nonexistent
1909 localtime hour at the beginning of DST. In this event, use
1910 mktime(tm - 1hr) + 3600.
1914 gmtime(t+o) --> tm+o
1915 mktime(tm+o) --> t+2o
1916 t+o - (t+2o - t+o) = t
1918 Note that glibc contains a function of the same purpose named
1919 `timegm' (reverse of gmtime). But obviously, it is not universally
1920 available, and unfortunately it is not straightforwardly
1921 extractable for use here. Perhaps configure should detect timegm
1922 and use it where available.
1924 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1925 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1926 Further improved by Roger with assistance from Edward J. Sabol
1927 based on input by Jamie Zawinski. */
1930 mktime_from_utc (struct tm *t)
1941 return -1; /* can't deal with output from strptime */
1952 return -1; /* can't deal with output from gmtime */
1955 return (tl - (tb - tl));
1958 /* Check whether the result of strptime() indicates success.
1959 strptime() returns the pointer to how far it got to in the string.
1960 The processing has been successful if the string is at `GMT' or
1961 `+X', or at the end of the string.
1963 In extended regexp parlance, the function returns 1 if P matches
1964 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1965 can return) is considered a failure and 0 is returned. */
1967 check_end (const char *p)
1971 while (ISSPACE (*p))
1974 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1975 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1981 /* Convert the textual specification of time in TIME_STRING to the
1982 number of seconds since the Epoch.
1984 TIME_STRING can be in any of the three formats RFC2068 allows the
1985 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1986 Timezones are ignored, and should be GMT.
1988 Return the computed time_t representation, or -1 if the conversion
1991 This function uses strptime with various string formats for parsing
1992 TIME_STRING. This results in a parser that is not as lenient in
1993 interpreting TIME_STRING as I would like it to be. Being based on
1994 strptime, it always allows shortened months, one-digit days, etc.,
1995 but due to the multitude of formats in which time can be
1996 represented, an ideal HTTP time parser would be even more
1997 forgiving. It should completely ignore things like week days and
1998 concentrate only on the various forms of representing years,
1999 months, days, hours, minutes, and seconds. For example, it would
2000 be nice if it accepted ISO 8601 out of the box.
2002 I've investigated free and PD code for this purpose, but none was
2003 usable. getdate was big and unwieldy, and had potential copyright
2004 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2005 distributed with phttpd, is excellent, but we cannot use it because
2006 it is not assigned to the FSF. So I stuck it with strptime. */
2009 http_atotm (char *time_string)
2011 /* NOTE: Solaris strptime man page claims that %n and %t match white
2012 space, but that's not universally available. Instead, we simply
2013 use ` ' to mean "skip all WS", which works under all strptime
2014 implementations I've tested. */
2016 static const char *time_formats[] = {
2017 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2018 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2019 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2020 (google.com uses this for their cookies.) */
2021 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2027 /* According to Roger Beeman, we need to initialize tm_isdst, since
2028 strptime won't do it. */
2031 /* Note that under foreign locales Solaris strptime() fails to
2032 recognize English dates, which renders this function useless. We
2033 solve this by being careful not to affect LC_TIME when
2034 initializing locale.
2036 Another solution would be to temporarily set locale to C, invoke
2037 strptime(), and restore it back. This is slow and dirty,
2038 however, and locale support other than LC_MESSAGES can mess other
2039 things, so I rather chose to stick with just setting LC_MESSAGES.
2041 GNU strptime does not have this problem because it recognizes
2042 both international and local dates. */
2044 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2045 if (check_end (strptime (time_string, time_formats[i], &t)))
2046 return mktime_from_utc (&t);
2048 /* All formats have failed. */
2052 /* Authorization support: We support two authorization schemes:
2054 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2056 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2057 consisting of answering to the server's challenge with the proper
2060 /* How many bytes it will take to store LEN bytes in base64. */
2061 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2063 /* Encode the string S of length LENGTH to base64 format and place it
2064 to STORE. STORE will be 0-terminated, and must point to a writable
2065 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2067 base64_encode (const char *s, char *store, int length)
2069 /* Conversion table. */
2070 static char tbl[64] = {
2071 'A','B','C','D','E','F','G','H',
2072 'I','J','K','L','M','N','O','P',
2073 'Q','R','S','T','U','V','W','X',
2074 'Y','Z','a','b','c','d','e','f',
2075 'g','h','i','j','k','l','m','n',
2076 'o','p','q','r','s','t','u','v',
2077 'w','x','y','z','0','1','2','3',
2078 '4','5','6','7','8','9','+','/'
2081 unsigned char *p = (unsigned char *)store;
2083 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2084 for (i = 0; i < length; i += 3)
2086 *p++ = tbl[s[0] >> 2];
2087 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2088 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2089 *p++ = tbl[s[2] & 0x3f];
2092 /* Pad the result if necessary... */
2093 if (i == length + 1)
2095 else if (i == length + 2)
2096 *(p - 1) = *(p - 2) = '=';
2097 /* ...and zero-terminate it. */
2101 /* Create the authentication header contents for the `Basic' scheme.
2102 This is done by encoding the string `USER:PASS' in base64 and
2103 prepending `HEADER: Basic ' to it. */
2105 basic_authentication_encode (const char *user, const char *passwd,
2108 char *t1, *t2, *res;
2109 int len1 = strlen (user) + 1 + strlen (passwd);
2110 int len2 = BASE64_LENGTH (len1);
2112 t1 = (char *)alloca (len1 + 1);
2113 sprintf (t1, "%s:%s", user, passwd);
2114 t2 = (char *)alloca (1 + len2);
2115 base64_encode (t1, t2, len1);
2116 res = (char *)xmalloc (len2 + 11 + strlen (header));
2117 sprintf (res, "%s: Basic %s\r\n", header, t2);
2123 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2124 of a field in such a header. If the field is the one specified by
2125 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2126 digest authorization code), extract its value in the (char*)
2127 variable pointed by RET. Returns negative on a malformed header,
2128 or number of bytes that have been parsed by this call. */
2130 extract_header_attr (const char *au, const char *attr_name, char **ret)
2132 const char *cp, *ep;
2136 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2138 cp += strlen (attr_name);
2141 cp += skip_lws (cp);
2146 cp += skip_lws (cp);
2151 for (ep = cp; *ep && *ep != '\"'; ep++)
2156 *ret = strdupdelim (cp, ep);
2163 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2164 an array of 16 bytes containing the hash keys, and BUF should be a
2165 buffer of 33 writable characters (32 for hex digits plus one for
2166 zero termination). */
2168 dump_hash (unsigned char *buf, const unsigned char *hash)
2172 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2174 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2175 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2180 /* Take the line apart to find the challenge, and compose a digest
2181 authorization header. See RFC2069 section 2.1.2. */
2183 digest_authentication_encode (const char *au, const char *user,
2184 const char *passwd, const char *method,
2187 static char *realm, *opaque, *nonce;
2192 { "realm", &realm },
2193 { "opaque", &opaque },
2198 realm = opaque = nonce = NULL;
2200 au += 6; /* skip over `Digest' */
2205 au += skip_lws (au);
2206 for (i = 0; i < ARRAY_SIZE (options); i++)
2208 int skip = extract_header_attr (au, options[i].name,
2209 options[i].variable);
2213 FREE_MAYBE (opaque);
2223 if (i == ARRAY_SIZE (options))
2225 while (*au && *au != '=')
2229 au += skip_lws (au);
2233 while (*au && *au != '\"')
2240 while (*au && *au != ',')
2245 if (!realm || !nonce || !user || !passwd || !path || !method)
2248 FREE_MAYBE (opaque);
2253 /* Calculate the digest value. */
2255 ALLOCA_MD5_CONTEXT (ctx);
2256 unsigned char hash[MD5_HASHLEN];
2257 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2258 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2260 /* A1BUF = H(user ":" realm ":" password) */
2262 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2263 gen_md5_update ((unsigned char *)":", 1, ctx);
2264 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2265 gen_md5_update ((unsigned char *)":", 1, ctx);
2266 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2267 gen_md5_finish (ctx, hash);
2268 dump_hash (a1buf, hash);
2270 /* A2BUF = H(method ":" path) */
2272 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2273 gen_md5_update ((unsigned char *)":", 1, ctx);
2274 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2275 gen_md5_finish (ctx, hash);
2276 dump_hash (a2buf, hash);
2278 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2280 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2281 gen_md5_update ((unsigned char *)":", 1, ctx);
2282 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2283 gen_md5_update ((unsigned char *)":", 1, ctx);
2284 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2285 gen_md5_finish (ctx, hash);
2286 dump_hash (response_digest, hash);
2288 res = (char*) xmalloc (strlen (user)
2293 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2294 + (opaque ? strlen (opaque) : 0)
2296 sprintf (res, "Authorization: Digest \
2297 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2298 user, realm, nonce, path, response_digest);
2301 char *p = res + strlen (res);
2302 strcat (p, ", opaque=\"");
2306 strcat (res, "\r\n");
2310 #endif /* USE_DIGEST */
2313 #define BEGINS_WITH(line, string_constant) \
2314 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2315 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2316 || !line[sizeof (string_constant) - 1]))
2319 known_authentication_scheme_p (const char *au)
2321 return BEGINS_WITH (au, "Basic")
2322 || BEGINS_WITH (au, "Digest")
2323 || BEGINS_WITH (au, "NTLM");
2328 /* Create the HTTP authorization request header. When the
2329 `WWW-Authenticate' response header is seen, according to the
2330 authorization scheme specified in that header (`Basic' and `Digest'
2331 are supported by the current implementation), produce an
2332 appropriate HTTP authorization request header. */
2334 create_authorization_line (const char *au, const char *user,
2335 const char *passwd, const char *method,
2338 char *wwwauth = NULL;
2340 if (!strncasecmp (au, "Basic", 5))
2341 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2342 if (!strncasecmp (au, "NTLM", 4))
2343 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2345 else if (!strncasecmp (au, "Digest", 6))
2346 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2347 #endif /* USE_DIGEST */
2354 if (pc_last_host_ip)
2355 address_list_release (pc_last_host_ip);