2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
25 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
50 # include <netdb.h> /* for h_errno */
64 # include "gen_sslfunc.h"
68 extern char *version_string;
79 static int cookies_loaded_p;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 hdr += skip_lws (hdr);
211 for (num = 0; ISDIGIT (*hdr); hdr++)
212 num = 10 * num + (*hdr - '0');
213 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
215 closure->first_byte_pos = num;
217 for (num = 0; ISDIGIT (*hdr); hdr++)
218 num = 10 * num + (*hdr - '0');
219 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
221 closure->last_byte_pos = num;
223 for (num = 0; ISDIGIT (*hdr); hdr++)
224 num = 10 * num + (*hdr - '0');
225 closure->entity_length = num;
229 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
230 Used for `Accept-Ranges'. */
232 http_process_none (const char *hdr, void *arg)
234 int *where = (int *)arg;
236 if (strstr (hdr, "none"))
243 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
245 http_process_type (const char *hdr, void *arg)
247 char **result = (char **)arg;
248 /* Locate P on `;' or the terminating zero, whichever comes first. */
249 const char *p = strchr (hdr, ';');
251 p = hdr + strlen (hdr);
252 while (p > hdr && ISSPACE (*(p - 1)))
254 *result = strdupdelim (hdr, p);
258 /* Check whether the `Connection' header is set to "keep-alive". */
260 http_process_connection (const char *hdr, void *arg)
262 int *flag = (int *)arg;
263 if (!strcasecmp (hdr, "Keep-Alive"))
268 /* Persistent connections. Currently, we cache the most recently used
269 connection as persistent, provided that the HTTP server agrees to
270 make it such. The persistence data is stored in the variables
271 below. Ideally, it would be in a structure, and it should be
272 possible to cache an arbitrary fixed number of these connections.
274 I think the code is quite easy to extend in that direction. */
276 /* Whether a persistent connection is active. */
277 static int pc_active_p;
278 /* Host and port of currently active persistent connection. */
279 static unsigned char pc_last_host_ip[4];
280 static unsigned short pc_last_port;
282 /* File descriptor of the currently active persistent connection. */
283 static int pc_last_fd;
286 /* Whether a ssl handshake has occoured on this connection */
287 static int pc_active_ssl;
288 /* SSL connection of the currently active persistent connection. */
289 static SSL *pc_last_ssl;
290 #endif /* HAVE_SSL */
292 /* Mark the persistent connection as invalid. This is used by the
293 CLOSE_* macros after they forcefully close a registered persistent
294 connection. This does not close the file descriptor -- it is left
295 to the caller to do that. (Maybe it should, though.) */
298 invalidate_persistent (void)
303 #endif /* HAVE_SSL */
304 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
307 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
308 persistent. This will enable someone to use the same connection
309 later. In the context of HTTP, this must be called only AFTER the
310 response has been received and the server has promised that the
311 connection will remain alive.
313 If a previous connection was persistent, it is closed. */
316 register_persistent (const char *host, unsigned short port, int fd
326 if (pc_last_fd == fd)
328 /* The connection FD is already registered. Nothing to
334 /* The old persistent connection is still active; let's
335 close it first. This situation arises whenever a
336 persistent connection exists, but we then connect to a
337 different host, and try to register a persistent
338 connection to that one. */
340 /* The ssl disconnect has to take place before the closing
343 shutdown_ssl(pc_last_ssl);
346 invalidate_persistent ();
350 /* This lookup_host cannot fail, because it has the results in the
352 success = lookup_host (host, pc_last_host_ip);
359 pc_active_ssl = ssl ? 1 : 0;
361 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
364 /* Return non-zero if a persistent connection is available for
365 connecting to HOST:PORT. */
368 persistent_available_p (const char *host, unsigned short port
374 unsigned char this_host_ip[4];
375 /* First, check whether a persistent connection is active at all. */
378 /* Second, check if the active connection pertains to the correct
379 (HOST, PORT) ordered pair. */
380 if (port != pc_last_port)
383 /* Second, a): check if current connection is (not) ssl, too. This
384 test is unlikely to fail because HTTP and HTTPS typicaly use
385 different ports. Yet it is possible, or so I [Christian
386 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
388 if (ssl != pc_active_ssl)
390 #endif /* HAVE_SSL */
391 if (!lookup_host (host, this_host_ip))
393 if (memcmp (pc_last_host_ip, this_host_ip, 4))
395 /* Third: check whether the connection is still open. This is
396 important because most server implement a liberal (short) timeout
397 on persistent connections. Wget can of course always reconnect
398 if the connection doesn't work out, but it's nicer to know in
399 advance. This test is a logical followup of the first test, but
400 is "expensive" and therefore placed at the end of the list. */
401 if (!test_socket_open (pc_last_fd))
403 /* Oops, the socket is no longer open. Now that we know that,
404 let's invalidate the persistent connection before returning
407 invalidate_persistent ();
414 # define SHUTDOWN_SSL(ssl) do { \
416 shutdown_ssl (ssl); \
419 # define SHUTDOWN_SSL(ssl)
422 /* The idea behind these two CLOSE macros is to distinguish between
423 two cases: one when the job we've been doing is finished, and we
424 want to close the connection and leave, and two when something is
425 seriously wrong and we're closing the connection as part of
428 In case of keep_alive, CLOSE_FINISH should leave the connection
429 open, while CLOSE_INVALIDATE should still close it.
431 Note that the semantics of the flag `keep_alive' is "this
432 connection *will* be reused (the server has promised not to close
433 the connection once we're done)", while the semantics of
434 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
435 active, registered connection". */
437 #define CLOSE_FINISH(fd) do { \
440 SHUTDOWN_SSL (ssl); \
442 if (pc_active_p && (fd) == pc_last_fd) \
443 invalidate_persistent (); \
447 #define CLOSE_INVALIDATE(fd) do { \
448 SHUTDOWN_SSL (ssl); \
450 if (pc_active_p && (fd) == pc_last_fd) \
451 invalidate_persistent (); \
456 long len; /* received length */
457 long contlen; /* expected length */
458 long restval; /* the restart value */
459 int res; /* the result of last read */
460 char *newloc; /* new location (redirection) */
461 char *remote_time; /* remote time-stamp string */
462 char *error; /* textual HTTP error */
463 int statcode; /* status code */
464 long dltime; /* time of the download */
465 int no_truncate; /* whether truncating the file is
467 const char *referer; /* value of the referer header. */
468 char **local_file; /* local file. */
472 free_hstat (struct http_stat *hs)
474 FREE_MAYBE (hs->newloc);
475 FREE_MAYBE (hs->remote_time);
476 FREE_MAYBE (hs->error);
478 /* Guard against being called twice. */
480 hs->remote_time = NULL;
484 static char *create_authorization_line PARAMS ((const char *, const char *,
485 const char *, const char *,
487 static char *basic_authentication_encode PARAMS ((const char *, const char *,
489 static int known_authentication_scheme_p PARAMS ((const char *));
491 time_t http_atotm PARAMS ((char *));
493 #define BEGINS_WITH(line, string_constant) \
494 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
495 && (ISSPACE (line[sizeof (string_constant) - 1]) \
496 || !line[sizeof (string_constant) - 1]))
498 /* Retrieve a document through HTTP protocol. It recognizes status
499 code, and correctly handles redirections. It closes the network
500 socket. If it receives an error from the functions below it, it
501 will print it if there is enough information to do so (almost
502 always), returning the error to the caller (i.e. http_loop).
504 Various HTTP parameters are stored to hs. Although it parses the
505 response code correctly, it is not used in a sane way. The caller
508 If PROXY is non-NULL, the connection will be made to the proxy
509 server, and u->url will be requested. */
511 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
513 char *request, *type, *command, *full_path;
515 char *pragma_h, *referer, *useragent, *range, *wwwauth;
516 char *authenticate_h;
520 char *request_keep_alive;
521 int sock, hcount, num_written, all_length, statcode;
522 long contlen, contrange;
526 int auth_tried_already;
529 static SSL_CTX *ssl_ctx = NULL;
531 #endif /* HAVE_SSL */
532 struct wget_timer *timer;
533 char *cookies = NULL;
535 /* Whether this connection will be kept alive after the HTTP request
539 /* Flags that detect the two ways of specifying HTTP keep-alive
541 int http_keep_alive_1, http_keep_alive_2;
543 /* Whether keep-alive should be inhibited. */
544 int inhibit_keep_alive;
547 /* initialize ssl_ctx on first run */
550 err = init_ssl (&ssl_ctx);
555 case SSLERRCTXCREATE:
557 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
561 /* try without certfile */
562 logprintf (LOG_NOTQUIET,
563 _("Failed to load certificates from %s\n"),
566 logprintf (LOG_NOTQUIET,
567 _("Trying without the specified certificate\n"));
570 logprintf (LOG_NOTQUIET,
571 _("Failed to get certificate key from %s\n"),
574 logprintf (LOG_NOTQUIET,
575 _("Trying without the specified certificate\n"));
582 #endif /* HAVE_SSL */
584 if (!(*dt & HEAD_ONLY))
585 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
586 know the local filename so we can save to it. */
587 assert (*hs->local_file != NULL);
590 auth_tried_already = 0;
592 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
595 /* We need to come back here when the initial attempt to retrieve
596 without authorization header fails. (Expected to happen at least
597 for the Digest authorization scheme.) */
600 http_keep_alive_1 = http_keep_alive_2 = 0;
602 /* Initialize certain elements of struct http_stat. */
607 hs->remote_time = NULL;
610 /* If we're using a proxy, we will be connecting to the proxy
612 conn = proxy ? proxy : u;
614 /* First: establish the connection. */
615 if (inhibit_keep_alive
618 !persistent_available_p (conn->host, conn->port)
620 !persistent_available_p (conn->host, conn->port,
621 u->scheme == SCHEME_HTTPS)
622 #endif /* HAVE_SSL */
625 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "),
626 conn->host, conn->port);
627 err = make_connection (&sock, conn->host, conn->port);
631 logputs (LOG_VERBOSE, "\n");
632 logprintf (LOG_NOTQUIET, "%s: %s.\n", conn->host, herrmsg (h_errno));
636 logputs (LOG_VERBOSE, "\n");
637 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
641 logputs (LOG_VERBOSE, "\n");
642 logprintf (LOG_NOTQUIET,
643 _("Connection to %s:%hu refused.\n"), conn->host,
648 logputs (LOG_VERBOSE, "\n");
649 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
654 /* Everything is fine! */
655 logputs (LOG_VERBOSE, _("connected!\n"));
662 if (conn->scheme == SCHEME_HTTPS)
663 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
665 logputs (LOG_VERBOSE, "\n");
666 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
671 #endif /* HAVE_SSL */
675 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
676 conn->host, conn->port);
677 /* #### pc_last_fd should be accessed through an accessor
682 #endif /* HAVE_SSL */
683 DEBUGP (("Reusing fd %d.\n", sock));
686 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
691 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
692 sprintf (referer, "Referer: %s\r\n", hs->referer);
695 if (*dt & SEND_NOCACHE)
696 pragma_h = "Pragma: no-cache\r\n";
702 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
703 /* Gag me! Some servers (e.g. WebSitePro) have been known to
704 respond to the following `Range' format by generating a
705 multipart/x-byte-ranges MIME document! This MIME type was
706 present in an old draft of the byteranges specification.
707 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
708 only if multiple non-overlapping ranges are requested --
709 which Wget never does. */
710 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
715 STRDUP_ALLOCA (useragent, opt.useragent);
718 useragent = (char *)alloca (10 + strlen (version_string));
719 sprintf (useragent, "Wget/%s", version_string);
721 /* Construct the authentication, if userid is present. */
724 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
725 user = user ? user : opt.http_user;
726 passwd = passwd ? passwd : opt.http_passwd;
733 /* We have the username and the password, but haven't tried
734 any authorization yet. Let's see if the "Basic" method
735 works. If not, we'll come back here and construct a
736 proper authorization method with the right challenges.
738 If we didn't employ this kind of logic, every URL that
739 requires authorization would have to be processed twice,
740 which is very suboptimal and generates a bunch of false
741 "unauthorized" errors in the server log.
743 #### But this logic also has a serious problem when used
744 with stronger authentications: we *first* transmit the
745 username and the password in clear text, and *then*
746 attempt a stronger authentication scheme. That cannot be
747 right! We are only fortunate that almost everyone still
748 uses the `Basic' scheme anyway.
750 There should be an option to prevent this from happening,
751 for those who use strong authentication schemes and value
753 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
757 wwwauth = create_authorization_line (authenticate_h, user, passwd,
765 char *proxy_user, *proxy_passwd;
766 /* For normal username and password, URL components override
767 command-line/wgetrc parameters. With proxy authentication,
768 it's the reverse, because proxy URLs are normally the
769 "permanent" ones, so command-line args should take
771 if (opt.proxy_user && opt.proxy_passwd)
773 proxy_user = opt.proxy_user;
774 proxy_passwd = opt.proxy_passwd;
778 proxy_user = proxy->user;
779 proxy_passwd = proxy->passwd;
781 /* #### This does not appear right. Can't the proxy request,
782 say, `Digest' authentication? */
783 if (proxy_user && proxy_passwd)
784 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
785 "Proxy-Authorization");
788 /* String of the form :PORT. Used only for non-standard ports. */
790 if (u->port != scheme_default_port (u->scheme))
792 port_maybe = (char *)alloca (numdigit (u->port) + 2);
793 sprintf (port_maybe, ":%d", u->port);
796 if (!inhibit_keep_alive)
797 request_keep_alive = "Connection: Keep-Alive\r\n";
799 request_keep_alive = NULL;
802 cookies = build_cookies_request (u->host, u->port, u->path,
804 u->scheme == SCHEME_HTTPS
811 full_path = xstrdup (u->url);
813 full_path = url_full_path (u);
815 /* Allocate the memory for the request. */
816 request = (char *)alloca (strlen (command)
820 + (port_maybe ? strlen (port_maybe) : 0)
821 + strlen (HTTP_ACCEPT)
822 + (request_keep_alive
823 ? strlen (request_keep_alive) : 0)
824 + (referer ? strlen (referer) : 0)
825 + (cookies ? strlen (cookies) : 0)
826 + (wwwauth ? strlen (wwwauth) : 0)
827 + (proxyauth ? strlen (proxyauth) : 0)
828 + (range ? strlen (range) : 0)
830 + (opt.user_header ? strlen (opt.user_header) : 0)
832 /* Construct the request. */
838 %s%s%s%s%s%s%s%s\r\n",
841 port_maybe ? port_maybe : "",
843 request_keep_alive ? request_keep_alive : "",
844 referer ? referer : "",
845 cookies ? cookies : "",
846 wwwauth ? wwwauth : "",
847 proxyauth ? proxyauth : "",
850 opt.user_header ? opt.user_header : "");
851 DEBUGP (("---request begin---\n%s---request end---\n", request));
853 /* Free the temporary memory. */
854 FREE_MAYBE (wwwauth);
855 FREE_MAYBE (proxyauth);
856 FREE_MAYBE (cookies);
859 /* Send the request to server. */
861 if (u->scheme == SCHEME_HTTPS)
862 num_written = ssl_iwrite (ssl, request, strlen (request));
864 #endif /* HAVE_SSL */
865 num_written = iwrite (sock, request, strlen (request));
869 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
871 CLOSE_INVALIDATE (sock);
874 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
875 proxy ? "Proxy" : "HTTP");
876 contlen = contrange = -1;
881 /* Before reading anything, initialize the rbuf. */
882 rbuf_initialize (&rbuf, sock);
884 if (u->scheme == SCHEME_HTTPS)
888 #endif /* HAVE_SSL */
891 /* Header-fetching loop. */
899 /* Get the header. */
900 status = header_get (&rbuf, &hdr,
901 /* Disallow continuations for status line. */
902 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
904 /* Check for errors. */
905 if (status == HG_EOF && *hdr)
907 /* This used to be an unconditional error, but that was
908 somewhat controversial, because of a large number of
909 broken CGI's that happily "forget" to send the second EOL
910 before closing the connection of a HEAD request.
912 So, the deal is to check whether the header is empty
913 (*hdr is zero if it is); if yes, it means that the
914 previous header was fully retrieved, and that -- most
915 probably -- the request is complete. "...be liberal in
916 what you accept." Oh boy. */
917 logputs (LOG_VERBOSE, "\n");
918 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
921 FREE_MAYBE (all_headers);
922 CLOSE_INVALIDATE (sock);
925 else if (status == HG_ERROR)
927 logputs (LOG_VERBOSE, "\n");
928 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
932 FREE_MAYBE (all_headers);
933 CLOSE_INVALIDATE (sock);
937 /* If the headers are to be saved to a file later, save them to
939 if (opt.save_headers)
941 int lh = strlen (hdr);
942 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
943 memcpy (all_headers + all_length, hdr, lh);
945 all_headers[all_length++] = '\n';
946 all_headers[all_length] = '\0';
949 /* Check for status line. */
953 /* Parse the first line of server response. */
954 statcode = parse_http_status_line (hdr, &error);
955 hs->statcode = statcode;
956 /* Store the descriptive response. */
957 if (statcode == -1) /* malformed response */
959 /* A common reason for "malformed response" error is the
960 case when no data was actually received. Handle this
963 hs->error = xstrdup (_("No data received"));
965 hs->error = xstrdup (_("Malformed status line"));
970 hs->error = xstrdup (_("(no description)"));
972 hs->error = xstrdup (error);
980 if (opt.server_response)
981 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
983 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
989 /* Exit on empty header. */
996 /* Print the header if requested. */
997 if (opt.server_response && hcount != 1)
998 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1000 /* Try getting content-length. */
1001 if (contlen == -1 && !opt.ignore_length)
1002 if (header_process (hdr, "Content-Length", header_extract_number,
1005 /* Try getting content-type. */
1007 if (header_process (hdr, "Content-Type", http_process_type, &type))
1009 /* Try getting location. */
1011 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1013 /* Try getting last-modified. */
1014 if (!hs->remote_time)
1015 if (header_process (hdr, "Last-Modified", header_strdup,
1018 /* Try getting cookies. */
1020 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1022 /* Try getting www-authentication. */
1023 if (!authenticate_h)
1024 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1027 /* Check for accept-ranges header. If it contains the word
1028 `none', disable the ranges. */
1029 if (*dt & ACCEPTRANGES)
1032 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1035 *dt &= ~ACCEPTRANGES;
1039 /* Try getting content-range. */
1040 if (contrange == -1)
1042 struct http_process_range_closure closure;
1043 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1045 contrange = closure.first_byte_pos;
1049 /* Check for keep-alive related responses. */
1050 if (!inhibit_keep_alive)
1052 /* Check for the `Keep-Alive' header. */
1053 if (!http_keep_alive_1)
1055 if (header_process (hdr, "Keep-Alive", header_exists,
1056 &http_keep_alive_1))
1059 /* Check for `Connection: Keep-Alive'. */
1060 if (!http_keep_alive_2)
1062 if (header_process (hdr, "Connection", http_process_connection,
1063 &http_keep_alive_2))
1071 logputs (LOG_VERBOSE, "\n");
1074 && (http_keep_alive_1 || http_keep_alive_2))
1076 assert (inhibit_keep_alive == 0);
1080 /* The server has promised that it will not close the connection
1081 when we're done. This means that we can register it. */
1083 register_persistent (conn->host, conn->port, sock);
1085 register_persistent (conn->host, conn->port, sock, ssl);
1086 #endif /* HAVE_SSL */
1088 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1091 /* Authorization is required. */
1095 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1096 might be more bytes in the body. */
1097 if (auth_tried_already)
1099 /* If we have tried it already, then there is not point
1102 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1103 xfree (authenticate_h);
1106 else if (!known_authentication_scheme_p (authenticate_h))
1108 xfree (authenticate_h);
1109 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1112 else if (BEGINS_WITH (authenticate_h, "Basic"))
1114 /* The authentication scheme is basic, the one we try by
1115 default, and it failed. There's no sense in trying
1121 auth_tried_already = 1;
1125 /* We do not need this anymore. */
1128 xfree (authenticate_h);
1129 authenticate_h = NULL;
1132 /* 20x responses are counted among successful by default. */
1133 if (H_20X (statcode))
1136 /* Return if redirected. */
1137 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1139 /* RFC2068 says that in case of the 300 (multiple choices)
1140 response, the server can output a preferred URL through
1141 `Location' header; otherwise, the request should be treated
1142 like GET. So, if the location is set, it will be a
1143 redirection; otherwise, just proceed normally. */
1144 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1148 logprintf (LOG_VERBOSE,
1149 _("Location: %s%s\n"),
1150 hs->newloc ? hs->newloc : _("unspecified"),
1151 hs->newloc ? _(" [following]") : "");
1152 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1153 might be more bytes in the body. */
1155 FREE_MAYBE (all_headers);
1160 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1163 /* We don't assume text/html by default. */
1166 if (opt.html_extension && (*dt & TEXTHTML))
1167 /* -E / --html-extension / html_extension = on was specified, and this is a
1168 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1169 already the file's suffix, tack on ".html". */
1171 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1173 if (last_period_in_local_filename == NULL ||
1174 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1175 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1177 size_t local_filename_len = strlen(*hs->local_file);
1179 *hs->local_file = xrealloc(*hs->local_file,
1180 local_filename_len + sizeof(".html"));
1181 strcpy(*hs->local_file + local_filename_len, ".html");
1183 *dt |= ADDED_HTML_EXTENSION;
1187 if (contrange == -1)
1189 /* We did not get a content-range header. This means that the
1190 server did not honor our `Range' request. Normally, this
1191 means we should reset hs->restval and continue normally. */
1193 /* However, if `-c' is used, we need to be a bit more careful:
1195 1. If `-c' is specified and the file already existed when
1196 Wget was started, it would be a bad idea for us to start
1197 downloading it from scratch, effectively truncating it. I
1198 believe this cannot happen unless `-c' was specified.
1200 2. If `-c' is used on a file that is already fully
1201 downloaded, we're requesting bytes after the end of file,
1202 which can result in server not honoring `Range'. If this is
1203 the case, `Content-Length' will be equal to the length of the
1205 if (opt.always_rest)
1207 /* Check for condition #2. */
1208 if (hs->restval > 0 /* restart was requested. */
1209 && contlen != -1 /* we got content-length. */
1210 && hs->restval >= contlen /* file fully downloaded
1214 logputs (LOG_VERBOSE, _("\
1215 \n The file is already fully retrieved; nothing to do.\n\n"));
1216 /* In case the caller inspects. */
1220 FREE_MAYBE (all_headers);
1221 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1222 might be more bytes in the body. */
1223 return RETRUNNEEDED;
1226 /* Check for condition #1. */
1227 if (hs->no_truncate)
1229 logprintf (LOG_NOTQUIET,
1232 Continued download failed on this file, which conflicts with `-c'.\n\
1233 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1235 FREE_MAYBE (all_headers);
1236 CLOSE_INVALIDATE (sock);
1237 return CONTNOTSUPPORTED;
1245 else if (contrange != hs->restval ||
1246 (H_PARTIAL (statcode) && contrange == -1))
1248 /* This means the whole request was somehow misunderstood by the
1249 server. Bail out. */
1251 FREE_MAYBE (all_headers);
1252 CLOSE_INVALIDATE (sock);
1259 contlen += contrange;
1261 contrange = -1; /* If conent-length was not sent,
1262 content-range will be ignored. */
1264 hs->contlen = contlen;
1268 if ((*dt & RETROKF) && !opt.server_response)
1270 /* No need to print this output if the body won't be
1271 downloaded at all, or if the original server response is
1273 logputs (LOG_VERBOSE, _("Length: "));
1276 logputs (LOG_VERBOSE, legible (contlen));
1277 if (contrange != -1)
1278 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1279 legible (contlen - contrange));
1282 logputs (LOG_VERBOSE,
1283 opt.ignore_length ? _("ignored") : _("unspecified"));
1285 logprintf (LOG_VERBOSE, " [%s]\n", type);
1287 logputs (LOG_VERBOSE, "\n");
1291 type = NULL; /* We don't need it any more. */
1293 /* Return if we have no intention of further downloading. */
1294 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1296 /* In case the caller cares to look... */
1300 FREE_MAYBE (all_headers);
1301 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1302 might be more bytes in the body. */
1303 return RETRFINISHED;
1306 /* Open the local file. */
1309 mkalldirs (*hs->local_file);
1311 rotate_backups (*hs->local_file);
1312 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1315 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1316 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1317 might be more bytes in the body. */
1318 FREE_MAYBE (all_headers);
1324 extern int global_download_count;
1326 /* To ensure that repeated "from scratch" downloads work for -O
1327 files, we rewind the file pointer, unless restval is
1328 non-zero. (This works only when -O is used on regular files,
1329 but it's still a valuable feature.)
1331 However, this loses when more than one URL is specified on
1332 the command line the second rewinds eradicates the contents
1333 of the first download. Thus we disable the above trick for
1334 all the downloads except the very first one.
1336 #### A possible solution to this would be to remember the
1337 file position in the output document and to seek to that
1338 position, instead of rewinding. */
1339 if (!hs->restval && global_download_count == 0)
1341 /* This will silently fail for streams that don't correspond
1342 to regular files, but that's OK. */
1344 /* ftruncate is needed because opt.dfp is opened in append
1345 mode if opt.always_rest is set. */
1346 ftruncate (fileno (fp), 0);
1351 /* #### This confuses the code that checks for file size. There
1352 should be some overhead information. */
1353 if (opt.save_headers)
1354 fwrite (all_headers, 1, all_length, fp);
1355 timer = wtimer_new ();
1356 /* Get the contents of the document. */
1357 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1358 (contlen != -1 ? contlen : 0),
1360 hs->dltime = wtimer_elapsed (timer);
1361 wtimer_delete (timer);
1363 /* Close or flush the file. We have to be careful to check for
1364 error here. Checking the result of fwrite() is not enough --
1365 errors could go unnoticed! */
1368 flush_res = fclose (fp);
1370 flush_res = fflush (fp);
1371 if (flush_res == EOF)
1374 FREE_MAYBE (all_headers);
1375 CLOSE_FINISH (sock);
1378 return RETRFINISHED;
1381 /* The genuine HTTP loop! This is the part where the retrieval is
1382 retried, and retried, and retried, and... */
1384 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1385 int *dt, struct url *proxy)
1388 int use_ts, got_head = 0; /* time-stamping info */
1389 char *filename_plus_orig_suffix;
1390 char *local_filename = NULL;
1391 char *tms, *suf, *locf, *tmrate;
1393 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1394 long local_size = 0; /* the size of the local file */
1395 size_t filename_len;
1396 struct http_stat hstat; /* HTTP status */
1400 /* This used to be done in main(), but it's a better idea to do it
1401 here so that we don't go through the hoops if we're just using
1403 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1405 load_cookies (opt.cookies_input);
1406 cookies_loaded_p = 1;
1411 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1412 has_wildcards_p because it would also warn on `?', and we know that
1413 shows up in CGI paths a *lot*. */
1414 if (strchr (u->url, '*'))
1415 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1417 /* Determine the local filename. */
1418 if (local_file && *local_file)
1419 hstat.local_file = local_file;
1420 else if (local_file)
1422 *local_file = url_filename (u);
1423 hstat.local_file = local_file;
1427 dummy = url_filename (u);
1428 hstat.local_file = &dummy;
1431 if (!opt.output_document)
1432 locf = *hstat.local_file;
1434 locf = opt.output_document;
1436 hstat.referer = referer;
1438 filename_len = strlen (*hstat.local_file);
1439 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1441 if (opt.noclobber && file_exists_p (*hstat.local_file))
1443 /* If opt.noclobber is turned on and file already exists, do not
1444 retrieve the file */
1445 logprintf (LOG_VERBOSE, _("\
1446 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1447 /* If the file is there, we suppose it's retrieved OK. */
1450 /* #### Bogusness alert. */
1451 /* If its suffix is "html" or "htm", assume text/html. */
1452 if (((suf = suffix (*hstat.local_file)) != NULL)
1453 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1461 if (opt.timestamping)
1463 boolean local_dot_orig_file_exists = FALSE;
1465 if (opt.backup_converted)
1466 /* If -K is specified, we'll act on the assumption that it was specified
1467 last time these files were downloaded as well, and instead of just
1468 comparing local file X against server file X, we'll compare local
1469 file X.orig (if extant, else X) against server file X. If -K
1470 _wasn't_ specified last time, or the server contains files called
1471 *.orig, -N will be back to not operating correctly with -k. */
1473 /* Would a single s[n]printf() call be faster? --dan
1475 Definitely not. sprintf() is horribly slow. It's a
1476 different question whether the difference between the two
1477 affects a program. Usually I'd say "no", but at one
1478 point I profiled Wget, and found that a measurable and
1479 non-negligible amount of time was lost calling sprintf()
1480 in url.c. Replacing sprintf with inline calls to
1481 strcpy() and long_to_string() made a difference.
1483 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1484 memcpy (filename_plus_orig_suffix + filename_len,
1485 ".orig", sizeof (".orig"));
1487 /* Try to stat() the .orig file. */
1488 if (stat (filename_plus_orig_suffix, &st) == 0)
1490 local_dot_orig_file_exists = TRUE;
1491 local_filename = filename_plus_orig_suffix;
1495 if (!local_dot_orig_file_exists)
1496 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1497 if (stat (*hstat.local_file, &st) == 0)
1498 local_filename = *hstat.local_file;
1500 if (local_filename != NULL)
1501 /* There was a local file, so we'll check later to see if the version
1502 the server has is the same version we already have, allowing us to
1507 local_size = st.st_size;
1511 /* Reset the counter. */
1513 *dt = 0 | ACCEPTRANGES;
1517 /* Increment the pass counter. */
1519 sleep_between_retrievals (count);
1520 /* Get the current time string. */
1521 tms = time_str (NULL);
1522 /* Print fetch message, if opt.verbose. */
1525 char *hurl = url_string (u, 1);
1529 sprintf (tmp, _("(try:%2d)"), count);
1530 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1531 tms, hurl, tmp, locf);
1533 ws_changetitle (hurl, 1);
1538 /* Default document type is empty. However, if spider mode is
1539 on or time-stamping is employed, HEAD_ONLY commands is
1540 encoded within *dt. */
1541 if (opt.spider || (use_ts && !got_head))
1545 /* Assume no restarting. */
1547 /* Decide whether or not to restart. */
1548 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1549 /* #### this calls access() and then stat(); could be optimized. */
1550 && file_exists_p (locf))
1551 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1552 hstat.restval = st.st_size;
1554 /* In `-c' is used and the file is existing and non-empty,
1555 refuse to truncate it if the server doesn't support continued
1557 hstat.no_truncate = 0;
1558 if (opt.always_rest && hstat.restval)
1559 hstat.no_truncate = 1;
1561 /* Decide whether to send the no-cache directive. We send it in
1563 a) we're using a proxy, and we're past our first retrieval.
1564 Some proxies are notorious for caching incomplete data, so
1565 we require a fresh get.
1566 b) caching is explicitly inhibited. */
1567 if ((proxy && count > 1) /* a */
1568 || !opt.allow_cache /* b */
1570 *dt |= SEND_NOCACHE;
1572 *dt &= ~SEND_NOCACHE;
1574 /* Try fetching the document, or at least its head. */
1575 err = gethttp (u, &hstat, dt, proxy);
1577 /* It's unfortunate that wget determines the local filename before finding
1578 out the Content-Type of the file. Barring a major restructuring of the
1579 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1580 *hstat.local_file to tack on ".html". */
1581 if (!opt.output_document)
1582 locf = *hstat.local_file;
1584 locf = opt.output_document;
1587 tms = time_str (NULL);
1588 /* Get the new location (with or without the redirection). */
1590 *newloc = xstrdup (hstat.newloc);
1593 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1594 case CONERROR: case READERR: case WRITEFAILED:
1596 /* Non-fatal errors continue executing the loop, which will
1597 bring them to "while" statement at the end, to judge
1598 whether the number of tries was exceeded. */
1599 free_hstat (&hstat);
1600 printwhat (count, opt.ntry);
1603 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1604 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1605 /* Fatal errors just return from the function. */
1606 free_hstat (&hstat);
1610 case FWRITEERR: case FOPENERR:
1611 /* Another fatal error. */
1612 logputs (LOG_VERBOSE, "\n");
1613 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1614 *hstat.local_file, strerror (errno));
1615 free_hstat (&hstat);
1620 /* Another fatal error. */
1621 logputs (LOG_VERBOSE, "\n");
1622 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1623 free_hstat (&hstat);
1628 /* Return the new location to the caller. */
1631 logprintf (LOG_NOTQUIET,
1632 _("ERROR: Redirection (%d) without location.\n"),
1634 free_hstat (&hstat);
1638 free_hstat (&hstat);
1643 /* The file was already fully retrieved. */
1644 free_hstat (&hstat);
1649 /* Deal with you later. */
1652 /* All possibilities should have been exhausted. */
1655 if (!(*dt & RETROKF))
1659 /* #### Ugly ugly ugly! */
1660 char *hurl = url_string (u, 1);
1661 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1664 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1665 tms, hstat.statcode, hstat.error);
1666 logputs (LOG_VERBOSE, "\n");
1667 free_hstat (&hstat);
1672 /* Did we get the time-stamp? */
1675 if (opt.timestamping && !hstat.remote_time)
1677 logputs (LOG_NOTQUIET, _("\
1678 Last-modified header missing -- time-stamps turned off.\n"));
1680 else if (hstat.remote_time)
1682 /* Convert the date-string into struct tm. */
1683 tmr = http_atotm (hstat.remote_time);
1684 if (tmr == (time_t) (-1))
1685 logputs (LOG_VERBOSE, _("\
1686 Last-modified header invalid -- time-stamp ignored.\n"));
1690 /* The time-stamping section. */
1695 use_ts = 0; /* no more time-stamping */
1696 count = 0; /* the retrieve count for HEAD is
1698 if (hstat.remote_time && tmr != (time_t) (-1))
1700 /* Now time-stamping can be used validly. Time-stamping
1701 means that if the sizes of the local and remote file
1702 match, and local file is newer than the remote file,
1703 it will not be retrieved. Otherwise, the normal
1704 download procedure is resumed. */
1706 (hstat.contlen == -1 || local_size == hstat.contlen))
1708 logprintf (LOG_VERBOSE, _("\
1709 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1711 free_hstat (&hstat);
1715 else if (tml >= tmr)
1716 logprintf (LOG_VERBOSE, _("\
1717 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1719 logputs (LOG_VERBOSE,
1720 _("Remote file is newer, retrieving.\n"));
1722 free_hstat (&hstat);
1725 if ((tmr != (time_t) (-1))
1727 && ((hstat.len == hstat.contlen) ||
1728 ((hstat.res == 0) &&
1729 ((hstat.contlen == -1) ||
1730 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1732 /* #### This code repeats in http.c and ftp.c. Move it to a
1734 const char *fl = NULL;
1735 if (opt.output_document)
1737 if (opt.od_known_regular)
1738 fl = opt.output_document;
1741 fl = *hstat.local_file;
1745 /* End of time-stamping section. */
1749 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1754 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1756 if (hstat.len == hstat.contlen)
1760 logprintf (LOG_VERBOSE,
1761 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1762 tms, tmrate, locf, hstat.len, hstat.contlen);
1763 logprintf (LOG_NONVERBOSE,
1764 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1765 tms, u->url, hstat.len, hstat.contlen, locf, count);
1768 downloaded_increase (hstat.len);
1770 /* Remember that we downloaded the file for later ".orig" code. */
1771 if (*dt & ADDED_HTML_EXTENSION)
1772 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1774 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1776 free_hstat (&hstat);
1780 else if (hstat.res == 0) /* No read error */
1782 if (hstat.contlen == -1) /* We don't know how much we were supposed
1783 to get, so assume we succeeded. */
1787 logprintf (LOG_VERBOSE,
1788 _("%s (%s) - `%s' saved [%ld]\n\n"),
1789 tms, tmrate, locf, hstat.len);
1790 logprintf (LOG_NONVERBOSE,
1791 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1792 tms, u->url, hstat.len, locf, count);
1795 downloaded_increase (hstat.len);
1797 /* Remember that we downloaded the file for later ".orig" code. */
1798 if (*dt & ADDED_HTML_EXTENSION)
1799 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1801 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1803 free_hstat (&hstat);
1807 else if (hstat.len < hstat.contlen) /* meaning we lost the
1808 connection too soon */
1810 logprintf (LOG_VERBOSE,
1811 _("%s (%s) - Connection closed at byte %ld. "),
1812 tms, tmrate, hstat.len);
1813 printwhat (count, opt.ntry);
1814 free_hstat (&hstat);
1817 else if (!opt.kill_longer) /* meaning we got more than expected */
1819 logprintf (LOG_VERBOSE,
1820 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1821 tms, tmrate, locf, hstat.len, hstat.contlen);
1822 logprintf (LOG_NONVERBOSE,
1823 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1824 tms, u->url, hstat.len, hstat.contlen, locf, count);
1826 downloaded_increase (hstat.len);
1828 /* Remember that we downloaded the file for later ".orig" code. */
1829 if (*dt & ADDED_HTML_EXTENSION)
1830 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1832 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1834 free_hstat (&hstat);
1838 else /* the same, but not accepted */
1840 logprintf (LOG_VERBOSE,
1841 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1842 tms, tmrate, hstat.len, hstat.contlen);
1843 printwhat (count, opt.ntry);
1844 free_hstat (&hstat);
1848 else /* now hstat.res can only be -1 */
1850 if (hstat.contlen == -1)
1852 logprintf (LOG_VERBOSE,
1853 _("%s (%s) - Read error at byte %ld (%s)."),
1854 tms, tmrate, hstat.len, strerror (errno));
1855 printwhat (count, opt.ntry);
1856 free_hstat (&hstat);
1859 else /* hstat.res == -1 and contlen is given */
1861 logprintf (LOG_VERBOSE,
1862 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1863 tms, tmrate, hstat.len, hstat.contlen,
1865 printwhat (count, opt.ntry);
1866 free_hstat (&hstat);
1873 while (!opt.ntry || (count < opt.ntry));
1877 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1878 than local timezone.
1880 mktime is similar but assumes struct tm, also known as the
1881 "broken-down" form of time, is in local time zone. mktime_from_utc
1882 uses mktime to make the conversion understanding that an offset
1883 will be introduced by the local time assumption.
1885 mktime_from_utc then measures the introduced offset by applying
1886 gmtime to the initial result and applying mktime to the resulting
1887 "broken-down" form. The difference between the two mktime results
1888 is the measured offset which is then subtracted from the initial
1889 mktime result to yield a calendar time which is the value returned.
1891 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1892 consistent offset (the non DST offset) since tm and tm+o might be
1893 on opposite sides of a DST change.
1895 Some implementations of mktime return -1 for the nonexistent
1896 localtime hour at the beginning of DST. In this event, use
1897 mktime(tm - 1hr) + 3600.
1901 gmtime(t+o) --> tm+o
1902 mktime(tm+o) --> t+2o
1903 t+o - (t+2o - t+o) = t
1905 Note that glibc contains a function of the same purpose named
1906 `timegm' (reverse of gmtime). But obviously, it is not universally
1907 available, and unfortunately it is not straightforwardly
1908 extractable for use here. Perhaps configure should detect timegm
1909 and use it where available.
1911 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1912 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1913 Further improved by Roger with assistance from Edward J. Sabol
1914 based on input by Jamie Zawinski. */
1917 mktime_from_utc (struct tm *t)
1928 return -1; /* can't deal with output from strptime */
1939 return -1; /* can't deal with output from gmtime */
1942 return (tl - (tb - tl));
1945 /* Check whether the result of strptime() indicates success.
1946 strptime() returns the pointer to how far it got to in the string.
1947 The processing has been successful if the string is at `GMT' or
1948 `+X', or at the end of the string.
1950 In extended regexp parlance, the function returns 1 if P matches
1951 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1952 can return) is considered a failure and 0 is returned. */
1954 check_end (const char *p)
1958 while (ISSPACE (*p))
1961 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1962 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1968 /* Convert the textual specification of time in TIME_STRING to the
1969 number of seconds since the Epoch.
1971 TIME_STRING can be in any of the three formats RFC2068 allows the
1972 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1973 Timezones are ignored, and should be GMT.
1975 Return the computed time_t representation, or -1 if the conversion
1978 This function uses strptime with various string formats for parsing
1979 TIME_STRING. This results in a parser that is not as lenient in
1980 interpreting TIME_STRING as I would like it to be. Being based on
1981 strptime, it always allows shortened months, one-digit days, etc.,
1982 but due to the multitude of formats in which time can be
1983 represented, an ideal HTTP time parser would be even more
1984 forgiving. It should completely ignore things like week days and
1985 concentrate only on the various forms of representing years,
1986 months, days, hours, minutes, and seconds. For example, it would
1987 be nice if it accepted ISO 8601 out of the box.
1989 I've investigated free and PD code for this purpose, but none was
1990 usable. getdate was big and unwieldy, and had potential copyright
1991 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1992 distributed with phttpd, is excellent, but we cannot use it because
1993 it is not assigned to the FSF. So I stuck it with strptime. */
1996 http_atotm (char *time_string)
1998 /* NOTE: Solaris strptime man page claims that %n and %t match white
1999 space, but that's not universally available. Instead, we simply
2000 use ` ' to mean "skip all WS", which works under all strptime
2001 implementations I've tested. */
2003 static const char *time_formats[] = {
2004 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2005 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2006 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2007 (google.com uses this for their cookies.) */
2008 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2014 /* According to Roger Beeman, we need to initialize tm_isdst, since
2015 strptime won't do it. */
2018 /* Note that under foreign locales Solaris strptime() fails to
2019 recognize English dates, which renders this function useless. We
2020 solve this by being careful not to affect LC_TIME when
2021 initializing locale.
2023 Another solution would be to temporarily set locale to C, invoke
2024 strptime(), and restore it back. This is slow and dirty,
2025 however, and locale support other than LC_MESSAGES can mess other
2026 things, so I rather chose to stick with just setting LC_MESSAGES.
2028 GNU strptime does not have this problem because it recognizes
2029 both international and local dates. */
2031 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2032 if (check_end (strptime (time_string, time_formats[i], &t)))
2033 return mktime_from_utc (&t);
2035 /* All formats have failed. */
2039 /* Authorization support: We support two authorization schemes:
2041 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2043 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2044 consisting of answering to the server's challenge with the proper
2047 /* How many bytes it will take to store LEN bytes in base64. */
2048 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2050 /* Encode the string S of length LENGTH to base64 format and place it
2051 to STORE. STORE will be 0-terminated, and must point to a writable
2052 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2054 base64_encode (const char *s, char *store, int length)
2056 /* Conversion table. */
2057 static char tbl[64] = {
2058 'A','B','C','D','E','F','G','H',
2059 'I','J','K','L','M','N','O','P',
2060 'Q','R','S','T','U','V','W','X',
2061 'Y','Z','a','b','c','d','e','f',
2062 'g','h','i','j','k','l','m','n',
2063 'o','p','q','r','s','t','u','v',
2064 'w','x','y','z','0','1','2','3',
2065 '4','5','6','7','8','9','+','/'
2068 unsigned char *p = (unsigned char *)store;
2070 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2071 for (i = 0; i < length; i += 3)
2073 *p++ = tbl[s[0] >> 2];
2074 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2075 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2076 *p++ = tbl[s[2] & 0x3f];
2079 /* Pad the result if necessary... */
2080 if (i == length + 1)
2082 else if (i == length + 2)
2083 *(p - 1) = *(p - 2) = '=';
2084 /* ...and zero-terminate it. */
2088 /* Create the authentication header contents for the `Basic' scheme.
2089 This is done by encoding the string `USER:PASS' in base64 and
2090 prepending `HEADER: Basic ' to it. */
2092 basic_authentication_encode (const char *user, const char *passwd,
2095 char *t1, *t2, *res;
2096 int len1 = strlen (user) + 1 + strlen (passwd);
2097 int len2 = BASE64_LENGTH (len1);
2099 t1 = (char *)alloca (len1 + 1);
2100 sprintf (t1, "%s:%s", user, passwd);
2101 t2 = (char *)alloca (1 + len2);
2102 base64_encode (t1, t2, len1);
2103 res = (char *)xmalloc (len2 + 11 + strlen (header));
2104 sprintf (res, "%s: Basic %s\r\n", header, t2);
2110 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2111 of a field in such a header. If the field is the one specified by
2112 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2113 digest authorization code), extract its value in the (char*)
2114 variable pointed by RET. Returns negative on a malformed header,
2115 or number of bytes that have been parsed by this call. */
2117 extract_header_attr (const char *au, const char *attr_name, char **ret)
2119 const char *cp, *ep;
2123 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2125 cp += strlen (attr_name);
2128 cp += skip_lws (cp);
2133 cp += skip_lws (cp);
2138 for (ep = cp; *ep && *ep != '\"'; ep++)
2143 *ret = strdupdelim (cp, ep);
2150 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2151 an array of 16 bytes containing the hash keys, and BUF should be a
2152 buffer of 33 writable characters (32 for hex digits plus one for
2153 zero termination). */
2155 dump_hash (unsigned char *buf, const unsigned char *hash)
2159 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2161 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2162 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2167 /* Take the line apart to find the challenge, and compose a digest
2168 authorization header. See RFC2069 section 2.1.2. */
2170 digest_authentication_encode (const char *au, const char *user,
2171 const char *passwd, const char *method,
2174 static char *realm, *opaque, *nonce;
2179 { "realm", &realm },
2180 { "opaque", &opaque },
2185 realm = opaque = nonce = NULL;
2187 au += 6; /* skip over `Digest' */
2192 au += skip_lws (au);
2193 for (i = 0; i < ARRAY_SIZE (options); i++)
2195 int skip = extract_header_attr (au, options[i].name,
2196 options[i].variable);
2200 FREE_MAYBE (opaque);
2210 if (i == ARRAY_SIZE (options))
2212 while (*au && *au != '=')
2216 au += skip_lws (au);
2220 while (*au && *au != '\"')
2227 while (*au && *au != ',')
2232 if (!realm || !nonce || !user || !passwd || !path || !method)
2235 FREE_MAYBE (opaque);
2240 /* Calculate the digest value. */
2242 MD5_CONTEXT_TYPE ctx;
2243 unsigned char hash[MD5_HASHLEN];
2244 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2245 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2247 /* A1BUF = H(user ":" realm ":" password) */
2249 MD5_UPDATE (user, strlen (user), &ctx);
2250 MD5_UPDATE (":", 1, &ctx);
2251 MD5_UPDATE (realm, strlen (realm), &ctx);
2252 MD5_UPDATE (":", 1, &ctx);
2253 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2254 MD5_FINISH (&ctx, hash);
2255 dump_hash (a1buf, hash);
2257 /* A2BUF = H(method ":" path) */
2259 MD5_UPDATE (method, strlen (method), &ctx);
2260 MD5_UPDATE (":", 1, &ctx);
2261 MD5_UPDATE (path, strlen (path), &ctx);
2262 MD5_FINISH (&ctx, hash);
2263 dump_hash (a2buf, hash);
2265 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2267 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2268 MD5_UPDATE (":", 1, &ctx);
2269 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2270 MD5_UPDATE (":", 1, &ctx);
2271 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2272 MD5_FINISH (&ctx, hash);
2273 dump_hash (response_digest, hash);
2275 res = (char*) xmalloc (strlen (user)
2280 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2281 + (opaque ? strlen (opaque) : 0)
2283 sprintf (res, "Authorization: Digest \
2284 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2285 user, realm, nonce, path, response_digest);
2288 char *p = res + strlen (res);
2289 strcat (p, ", opaque=\"");
2293 strcat (res, "\r\n");
2297 #endif /* USE_DIGEST */
2300 #define BEGINS_WITH(line, string_constant) \
2301 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2302 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2303 || !line[sizeof (string_constant) - 1]))
2306 known_authentication_scheme_p (const char *au)
2308 return BEGINS_WITH (au, "Basic")
2309 || BEGINS_WITH (au, "Digest")
2310 || BEGINS_WITH (au, "NTLM");
2315 /* Create the HTTP authorization request header. When the
2316 `WWW-Authenticate' response header is seen, according to the
2317 authorization scheme specified in that header (`Basic' and `Digest'
2318 are supported by the current implementation), produce an
2319 appropriate HTTP authorization request header. */
2321 create_authorization_line (const char *au, const char *user,
2322 const char *passwd, const char *method,
2325 char *wwwauth = NULL;
2327 if (!strncasecmp (au, "Basic", 5))
2328 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2329 if (!strncasecmp (au, "NTLM", 4))
2330 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2332 else if (!strncasecmp (au, "Digest", 6))
2333 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2334 #endif /* USE_DIGEST */