2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001, 2002
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.
21 In addition, as a special exception, the Free Software Foundation
22 gives permission to link the code of its release of Wget with the
23 OpenSSL project's "OpenSSL" library (or with modified versions of it
24 that use the same license as the "OpenSSL" library), and distribute
25 the linked executables. You must obey the GNU General Public License
26 in all respects for all of the code used other than "OpenSSL". If you
27 modify this file, you may extend this exception to your version of the
28 file, but you are not obligated to do so. If you do not wish to do
29 so, delete this exception statement from your version. */
35 #include <sys/types.h>
46 #if TIME_WITH_SYS_TIME
47 # include <sys/time.h>
51 # include <sys/time.h>
70 # include "gen_sslfunc.h"
78 extern char *version_string;
79 extern LARGE_INT total_downloaded_bytes;
82 static int cookies_loaded_p;
83 struct cookie_jar *wget_cookie_jar;
85 #define TEXTHTML_S "text/html"
86 #define TEXTXHTML_S "application/xhtml+xml"
87 #define HTTP_ACCEPT "*/*"
89 /* Some status code validation macros: */
90 #define H_20X(x) (((x) >= 200) && ((x) < 300))
91 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
92 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
93 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
94 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
96 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
98 #define HTTP_STATUS_OK 200
99 #define HTTP_STATUS_CREATED 201
100 #define HTTP_STATUS_ACCEPTED 202
101 #define HTTP_STATUS_NO_CONTENT 204
102 #define HTTP_STATUS_PARTIAL_CONTENTS 206
104 /* Redirection 3xx. */
105 #define HTTP_STATUS_MULTIPLE_CHOICES 300
106 #define HTTP_STATUS_MOVED_PERMANENTLY 301
107 #define HTTP_STATUS_MOVED_TEMPORARILY 302
108 #define HTTP_STATUS_NOT_MODIFIED 304
109 #define HTTP_STATUS_TEMPORARY_REDIRECT 307
111 /* Client error 4xx. */
112 #define HTTP_STATUS_BAD_REQUEST 400
113 #define HTTP_STATUS_UNAUTHORIZED 401
114 #define HTTP_STATUS_FORBIDDEN 403
115 #define HTTP_STATUS_NOT_FOUND 404
117 /* Server errors 5xx. */
118 #define HTTP_STATUS_INTERNAL 500
119 #define HTTP_STATUS_NOT_IMPLEMENTED 501
120 #define HTTP_STATUS_BAD_GATEWAY 502
121 #define HTTP_STATUS_UNAVAILABLE 503
124 /* Parse the HTTP status line, which is of format:
126 HTTP-Version SP Status-Code SP Reason-Phrase
128 The function returns the status-code, or -1 if the status line is
129 malformed. The pointer to reason-phrase is returned in RP. */
131 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
133 /* (the variables must not be named `major' and `minor', because
134 that breaks compilation with SunOS4 cc.) */
135 int mjr, mnr, statcode;
138 *reason_phrase_ptr = NULL;
140 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
141 major version, and Y is minor version. */
142 if (strncmp (line, "HTTP/", 5) != 0)
146 /* Calculate major HTTP version. */
148 for (mjr = 0; ISDIGIT (*line); line++)
149 mjr = 10 * mjr + (*line - '0');
150 if (*line != '.' || p == line)
154 /* Calculate minor HTTP version. */
156 for (mnr = 0; ISDIGIT (*line); line++)
157 mnr = 10 * mnr + (*line - '0');
158 if (*line != ' ' || p == line)
160 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
161 minor version can be safely ignored. */
166 /* Calculate status code. */
167 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
169 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
171 /* Set up the reason phrase pointer. */
173 /* RFC2068 requires SPC here, but we allow the string to finish
174 here, in case no reason-phrase is present. */
178 *reason_phrase_ptr = line;
183 *reason_phrase_ptr = line + 1;
188 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
190 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
191 PROMISED_SIZE bytes are sent over the wire -- if the file is
192 longer, read only that much; if the file is shorter, report an error. */
195 post_file (int sock, const char *file_name, long promised_size)
197 static char chunk[8192];
202 DEBUGP (("[writing POST file %s ... ", file_name));
204 fp = fopen (file_name, "rb");
207 while (!feof (fp) && written < promised_size)
210 int length = fread (chunk, 1, sizeof (chunk), fp);
213 towrite = WMIN (promised_size - written, length);
214 write_error = xwrite (sock, chunk, towrite, -1);
224 /* If we've written less than was promised, report a (probably
225 nonsensical) error rather than break the promise. */
226 if (written < promised_size)
232 assert (written == promised_size);
233 DEBUGP (("done]\n"));
237 /* Functions to be used as arguments to header_process(): */
239 struct http_process_range_closure {
245 /* Parse the `Content-Range' header and extract the information it
246 contains. Returns 1 if successful, -1 otherwise. */
248 http_process_range (const char *hdr, void *arg)
250 struct http_process_range_closure *closure
251 = (struct http_process_range_closure *)arg;
254 /* Certain versions of Nutscape proxy server send out
255 `Content-Length' without "bytes" specifier, which is a breach of
256 RFC2068 (as well as the HTTP/1.1 draft which was current at the
257 time). But hell, I must support it... */
258 if (!strncasecmp (hdr, "bytes", 5))
261 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
265 hdr += skip_lws (hdr);
271 for (num = 0; ISDIGIT (*hdr); hdr++)
272 num = 10 * num + (*hdr - '0');
273 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
275 closure->first_byte_pos = num;
277 for (num = 0; ISDIGIT (*hdr); hdr++)
278 num = 10 * num + (*hdr - '0');
279 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
281 closure->last_byte_pos = num;
283 for (num = 0; ISDIGIT (*hdr); hdr++)
284 num = 10 * num + (*hdr - '0');
285 closure->entity_length = num;
289 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
290 Used for `Accept-Ranges'. */
292 http_process_none (const char *hdr, void *arg)
294 int *where = (int *)arg;
296 if (strstr (hdr, "none"))
303 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
305 http_process_type (const char *hdr, void *arg)
307 char **result = (char **)arg;
308 /* Locate P on `;' or the terminating zero, whichever comes first. */
309 const char *p = strchr (hdr, ';');
311 p = hdr + strlen (hdr);
312 while (p > hdr && ISSPACE (*(p - 1)))
314 *result = strdupdelim (hdr, p);
318 /* Check whether the `Connection' header is set to "keep-alive". */
320 http_process_connection (const char *hdr, void *arg)
322 int *flag = (int *)arg;
323 if (!strcasecmp (hdr, "Keep-Alive"))
328 /* Commit the cookie to the cookie jar. */
331 http_process_set_cookie (const char *hdr, void *arg)
333 struct url *u = (struct url *)arg;
335 /* The jar should have been created by now. */
336 assert (wget_cookie_jar != NULL);
338 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path, hdr);
343 /* Persistent connections. Currently, we cache the most recently used
344 connection as persistent, provided that the HTTP server agrees to
345 make it such. The persistence data is stored in the variables
346 below. Ideally, it would be in a structure, and it should be
347 possible to cache an arbitrary fixed number of these connections.
349 I think the code is quite easy to extend in that direction. */
351 /* Whether a persistent connection is active. */
352 static int pc_active_p;
354 /* Host and port of currently active persistent connection. */
355 static struct address_list *pc_last_host_ip;
356 static unsigned short pc_last_port;
358 /* File descriptor of the currently active persistent connection. */
359 static int pc_last_fd;
361 /* Whether a ssl handshake has occoured on this connection */
362 static int pc_last_ssl_p;
364 /* Mark the persistent connection as invalid. This is used by the
365 CLOSE_* macros after they forcefully close a registered persistent
366 connection. This does not close the file descriptor -- it is left
367 to the caller to do that. (Maybe it should, though.) */
370 invalidate_persistent (void)
374 if (pc_last_host_ip != NULL)
376 address_list_release (pc_last_host_ip);
377 pc_last_host_ip = NULL;
379 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
382 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
383 persistent. This will enable someone to use the same connection
384 later. In the context of HTTP, this must be called only AFTER the
385 response has been received and the server has promised that the
386 connection will remain alive.
388 If a previous connection was persistent, it is closed. */
391 register_persistent (const char *host, unsigned short port, int fd, int ssl)
395 if (pc_last_fd == fd)
397 /* The connection FD is already registered. Nothing to
403 /* The old persistent connection is still active; let's
404 close it first. This situation arises whenever a
405 persistent connection exists, but we then connect to a
406 different host, and try to register a persistent
407 connection to that one. */
409 invalidate_persistent ();
413 assert (pc_last_host_ip == NULL);
415 /* This lookup_host cannot fail, because it has the results in the
417 pc_last_host_ip = lookup_host (host, LH_SILENT);
418 assert (pc_last_host_ip != NULL);
424 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
427 /* Return non-zero if a persistent connection is available for
428 connecting to HOST:PORT. */
431 persistent_available_p (const char *host, unsigned short port, int ssl)
434 struct address_list *this_host_ip;
436 /* First, check whether a persistent connection is active at all. */
439 /* Second, check if the active connection pertains to the correct
440 (HOST, PORT) ordered pair. */
441 if (port != pc_last_port)
444 /* Second, a): check if current connection is (not) ssl, too. This
445 test is unlikely to fail because HTTP and HTTPS typicaly use
446 different ports. Yet it is possible, or so I [Christian
447 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
449 if (ssl != pc_last_ssl_p)
452 this_host_ip = lookup_host (host, 0);
456 /* To equate the two host names for the purposes of persistent
457 connections, they need to share all the IP addresses in the
459 success = address_list_match_all (pc_last_host_ip, this_host_ip);
460 address_list_release (this_host_ip);
464 /* Third: check whether the connection is still open. This is
465 important because most server implement a liberal (short) timeout
466 on persistent connections. Wget can of course always reconnect
467 if the connection doesn't work out, but it's nicer to know in
468 advance. This test is a logical followup of the first test, but
469 is "expensive" and therefore placed at the end of the list. */
470 if (!test_socket_open (pc_last_fd))
472 /* Oops, the socket is no longer open. Now that we know that,
473 let's invalidate the persistent connection before returning
476 invalidate_persistent ();
482 /* The idea behind these two CLOSE macros is to distinguish between
483 two cases: one when the job we've been doing is finished, and we
484 want to close the connection and leave, and two when something is
485 seriously wrong and we're closing the connection as part of
488 In case of keep_alive, CLOSE_FINISH should leave the connection
489 open, while CLOSE_INVALIDATE should still close it.
491 Note that the semantics of the flag `keep_alive' is "this
492 connection *will* be reused (the server has promised not to close
493 the connection once we're done)", while the semantics of
494 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
495 active, registered connection". */
497 #define CLOSE_FINISH(fd) do { \
501 if (pc_active_p && (fd) == pc_last_fd) \
502 invalidate_persistent (); \
506 #define CLOSE_INVALIDATE(fd) do { \
508 if (pc_active_p && (fd) == pc_last_fd) \
509 invalidate_persistent (); \
514 long len; /* received length */
515 long contlen; /* expected length */
516 long restval; /* the restart value */
517 int res; /* the result of last read */
518 char *newloc; /* new location (redirection) */
519 char *remote_time; /* remote time-stamp string */
520 char *error; /* textual HTTP error */
521 int statcode; /* status code */
522 double dltime; /* time of the download in msecs */
523 int no_truncate; /* whether truncating the file is
525 const char *referer; /* value of the referer header. */
526 char **local_file; /* local file. */
530 free_hstat (struct http_stat *hs)
532 xfree_null (hs->newloc);
533 xfree_null (hs->remote_time);
534 xfree_null (hs->error);
536 /* Guard against being called twice. */
538 hs->remote_time = NULL;
542 static char *create_authorization_line PARAMS ((const char *, const char *,
543 const char *, const char *,
545 static char *basic_authentication_encode PARAMS ((const char *, const char *,
547 static int known_authentication_scheme_p PARAMS ((const char *));
549 time_t http_atotm PARAMS ((const char *));
551 #define BEGINS_WITH(line, string_constant) \
552 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
553 && (ISSPACE (line[sizeof (string_constant) - 1]) \
554 || !line[sizeof (string_constant) - 1]))
556 /* Retrieve a document through HTTP protocol. It recognizes status
557 code, and correctly handles redirections. It closes the network
558 socket. If it receives an error from the functions below it, it
559 will print it if there is enough information to do so (almost
560 always), returning the error to the caller (i.e. http_loop).
562 Various HTTP parameters are stored to hs. Although it parses the
563 response code correctly, it is not used in a sane way. The caller
566 If PROXY is non-NULL, the connection will be made to the proxy
567 server, and u->url will be requested. */
569 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
571 char *request, *type, *command, *full_path;
573 char *pragma_h, *referer, *useragent, *range, *wwwauth;
574 char *authenticate_h;
578 char *request_keep_alive;
579 int sock, hcount, all_length, statcode;
581 long contlen, contrange;
584 int auth_tried_already;
587 char *cookies = NULL;
589 /* Whether this connection will be kept alive after the HTTP request
593 /* Flags that detect the two ways of specifying HTTP keep-alive
595 int http_keep_alive_1, http_keep_alive_2;
597 /* Whether keep-alive should be inhibited. */
598 int inhibit_keep_alive;
600 /* Whether we need to print the host header with braces around host,
601 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
602 "Host: symbolic-name:1234". */
603 int squares_around_host = 0;
605 /* Headers sent when using POST. */
606 char *post_content_type, *post_content_length;
607 long post_data_size = 0;
610 /* Initialize the SSL context. After the first run, this is a
614 case SSLERRCTXCREATE:
616 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
617 return SSLERRCTXCREATE;
619 /* try without certfile */
620 logprintf (LOG_NOTQUIET,
621 _("Failed to load certificates from %s\n"),
623 logprintf (LOG_NOTQUIET,
624 _("Trying without the specified certificate\n"));
627 logprintf (LOG_NOTQUIET,
628 _("Failed to get certificate key from %s\n"),
630 logprintf (LOG_NOTQUIET,
631 _("Trying without the specified certificate\n"));
636 #endif /* HAVE_SSL */
638 if (!(*dt & HEAD_ONLY))
639 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
640 know the local filename so we can save to it. */
641 assert (*hs->local_file != NULL);
644 auth_tried_already = 0;
646 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
649 /* We need to come back here when the initial attempt to retrieve
650 without authorization header fails. (Expected to happen at least
651 for the Digest authorization scheme.) */
654 http_keep_alive_1 = http_keep_alive_2 = 0;
656 post_content_type = NULL;
657 post_content_length = NULL;
659 /* Initialize certain elements of struct http_stat. */
664 hs->remote_time = NULL;
667 /* If we're using a proxy, we will be connecting to the proxy
669 conn = proxy ? proxy : u;
671 /* First: establish the connection. */
672 if (inhibit_keep_alive
673 || !persistent_available_p (conn->host, conn->port,
675 u->scheme == SCHEME_HTTPS
681 sock = connect_to_host (conn->host, conn->port);
685 return CONNECT_ERROR (errno);
688 if (conn->scheme == SCHEME_HTTPS)
690 if (!ssl_connect (sock))
692 logputs (LOG_VERBOSE, "\n");
693 logprintf (LOG_NOTQUIET,
694 _("Unable to establish SSL connection.\n"));
700 #endif /* HAVE_SSL */
704 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
705 conn->host, conn->port);
706 /* #### pc_last_fd should be accessed through an accessor
709 using_ssl = pc_last_ssl_p;
710 DEBUGP (("Reusing fd %d.\n", sock));
715 else if (opt.post_file_name || opt.post_data)
723 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
724 sprintf (referer, "Referer: %s\r\n", hs->referer);
727 if (*dt & SEND_NOCACHE)
728 pragma_h = "Pragma: no-cache\r\n";
734 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
735 /* Gag me! Some servers (e.g. WebSitePro) have been known to
736 respond to the following `Range' format by generating a
737 multipart/x-byte-ranges MIME document! This MIME type was
738 present in an old draft of the byteranges specification.
739 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
740 only if multiple non-overlapping ranges are requested --
741 which Wget never does. */
742 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
747 STRDUP_ALLOCA (useragent, opt.useragent);
750 useragent = (char *)alloca (10 + strlen (version_string));
751 sprintf (useragent, "Wget/%s", version_string);
753 /* Construct the authentication, if userid is present. */
756 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
757 user = user ? user : opt.http_user;
758 passwd = passwd ? passwd : opt.http_passwd;
765 /* We have the username and the password, but haven't tried
766 any authorization yet. Let's see if the "Basic" method
767 works. If not, we'll come back here and construct a
768 proper authorization method with the right challenges.
770 If we didn't employ this kind of logic, every URL that
771 requires authorization would have to be processed twice,
772 which is very suboptimal and generates a bunch of false
773 "unauthorized" errors in the server log.
775 #### But this logic also has a serious problem when used
776 with stronger authentications: we *first* transmit the
777 username and the password in clear text, and *then*
778 attempt a stronger authentication scheme. That cannot be
779 right! We are only fortunate that almost everyone still
780 uses the `Basic' scheme anyway.
782 There should be an option to prevent this from happening,
783 for those who use strong authentication schemes and value
785 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
789 /* Use the full path, i.e. one that includes the leading
790 slash and the query string, but is independent of proxy
792 char *pth = url_full_path (u);
793 wwwauth = create_authorization_line (authenticate_h, user, passwd,
802 char *proxy_user, *proxy_passwd;
803 /* For normal username and password, URL components override
804 command-line/wgetrc parameters. With proxy authentication,
805 it's the reverse, because proxy URLs are normally the
806 "permanent" ones, so command-line args should take
808 if (opt.proxy_user && opt.proxy_passwd)
810 proxy_user = opt.proxy_user;
811 proxy_passwd = opt.proxy_passwd;
815 proxy_user = proxy->user;
816 proxy_passwd = proxy->passwd;
818 /* #### This does not appear right. Can't the proxy request,
819 say, `Digest' authentication? */
820 if (proxy_user && proxy_passwd)
821 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
822 "Proxy-Authorization");
825 /* String of the form :PORT. Used only for non-standard ports. */
827 if (u->port != scheme_default_port (u->scheme))
829 port_maybe = (char *)alloca (numdigit (u->port) + 2);
830 sprintf (port_maybe, ":%d", u->port);
833 if (!inhibit_keep_alive)
834 request_keep_alive = "Connection: Keep-Alive\r\n";
836 request_keep_alive = NULL;
839 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
841 u->scheme == SCHEME_HTTPS
847 if (opt.post_data || opt.post_file_name)
849 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
851 post_data_size = strlen (opt.post_data);
854 post_data_size = file_size (opt.post_file_name);
855 if (post_data_size == -1)
857 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
862 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
863 sprintf (post_content_length,
864 "Content-Length: %ld\r\n", post_data_size);
868 full_path = xstrdup (u->url);
870 /* Use the full path, i.e. one that includes the leading slash and
871 the query string. E.g. if u->path is "foo/bar" and u->query is
872 "param=value", full_path will be "/foo/bar?param=value". */
873 full_path = url_full_path (u);
875 if (strchr (u->host, ':'))
876 squares_around_host = 1;
878 /* Allocate the memory for the request. */
879 request = (char *)alloca (strlen (command)
883 + (port_maybe ? strlen (port_maybe) : 0)
884 + strlen (HTTP_ACCEPT)
885 + (request_keep_alive
886 ? strlen (request_keep_alive) : 0)
887 + (referer ? strlen (referer) : 0)
888 + (cookies ? strlen (cookies) : 0)
889 + (wwwauth ? strlen (wwwauth) : 0)
890 + (proxyauth ? strlen (proxyauth) : 0)
891 + (range ? strlen (range) : 0)
894 ? strlen (post_content_type) : 0)
895 + (post_content_length
896 ? strlen (post_content_length) : 0)
897 + (opt.user_header ? strlen (opt.user_header) : 0)
899 /* Construct the request. */
905 %s%s%s%s%s%s%s%s%s%s\r\n",
908 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
909 port_maybe ? port_maybe : "",
911 request_keep_alive ? request_keep_alive : "",
912 referer ? referer : "",
913 cookies ? cookies : "",
914 wwwauth ? wwwauth : "",
915 proxyauth ? proxyauth : "",
918 post_content_type ? post_content_type : "",
919 post_content_length ? post_content_length : "",
920 opt.user_header ? opt.user_header : "");
921 DEBUGP (("---request begin---\n%s", request));
923 /* Free the temporary memory. */
924 xfree_null (wwwauth);
925 xfree_null (proxyauth);
926 xfree_null (cookies);
929 /* Send the request to server. */
930 write_error = xwrite (sock, request, strlen (request), -1);
932 if (write_error >= 0)
936 DEBUGP (("[POST data: %s]\n", opt.post_data));
937 write_error = xwrite (sock, opt.post_data, post_data_size, -1);
939 else if (opt.post_file_name && post_data_size != 0)
940 write_error = post_file (sock, opt.post_file_name, post_data_size);
942 DEBUGP (("---request end---\n"));
946 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
948 CLOSE_INVALIDATE (sock);
951 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
952 proxy ? "Proxy" : "HTTP");
953 contlen = contrange = -1;
958 /* Before reading anything, initialize the rbuf. */
959 rbuf_initialize (&rbuf, sock);
962 /* Header-fetching loop. */
970 /* Get the header. */
971 status = header_get (&rbuf, &hdr,
972 /* Disallow continuations for status line. */
973 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
975 /* Check for errors. */
976 if (status == HG_EOF && *hdr)
978 /* This used to be an unconditional error, but that was
979 somewhat controversial, because of a large number of
980 broken CGI's that happily "forget" to send the second EOL
981 before closing the connection of a HEAD request.
983 So, the deal is to check whether the header is empty
984 (*hdr is zero if it is); if yes, it means that the
985 previous header was fully retrieved, and that -- most
986 probably -- the request is complete. "...be liberal in
987 what you accept." Oh boy. */
988 logputs (LOG_VERBOSE, "\n");
989 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
992 xfree_null (all_headers);
993 CLOSE_INVALIDATE (sock);
996 else if (status == HG_ERROR)
998 logputs (LOG_VERBOSE, "\n");
999 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1003 xfree_null (all_headers);
1004 CLOSE_INVALIDATE (sock);
1008 /* If the headers are to be saved to a file later, save them to
1010 if (opt.save_headers)
1012 int lh = strlen (hdr);
1013 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1014 memcpy (all_headers + all_length, hdr, lh);
1016 all_headers[all_length++] = '\n';
1017 all_headers[all_length] = '\0';
1020 /* Check for status line. */
1024 /* Parse the first line of server response. */
1025 statcode = parse_http_status_line (hdr, &error);
1026 hs->statcode = statcode;
1027 /* Store the descriptive response. */
1028 if (statcode == -1) /* malformed response */
1030 /* A common reason for "malformed response" error is the
1031 case when no data was actually received. Handle this
1034 hs->error = xstrdup (_("No data received"));
1036 hs->error = xstrdup (_("Malformed status line"));
1041 hs->error = xstrdup (_("(no description)"));
1043 hs->error = xstrdup (error);
1045 if ((statcode != -1)
1051 if (opt.server_response)
1052 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1054 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1060 /* Exit on empty header. */
1067 /* Print the header if requested. */
1068 if (opt.server_response && hcount != 1)
1069 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1071 /* Try getting content-length. */
1072 if (contlen == -1 && !opt.ignore_length)
1073 if (header_process (hdr, "Content-Length", header_extract_number,
1076 /* Try getting content-type. */
1078 if (header_process (hdr, "Content-Type", http_process_type, &type))
1080 /* Try getting location. */
1082 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1084 /* Try getting last-modified. */
1085 if (!hs->remote_time)
1086 if (header_process (hdr, "Last-Modified", header_strdup,
1089 /* Try getting cookies. */
1091 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1093 /* Try getting www-authentication. */
1094 if (!authenticate_h)
1095 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1098 /* Check for accept-ranges header. If it contains the word
1099 `none', disable the ranges. */
1100 if (*dt & ACCEPTRANGES)
1103 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1106 *dt &= ~ACCEPTRANGES;
1110 /* Try getting content-range. */
1111 if (contrange == -1)
1113 struct http_process_range_closure closure;
1114 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1116 contrange = closure.first_byte_pos;
1120 /* Check for keep-alive related responses. */
1121 if (!inhibit_keep_alive)
1123 /* Check for the `Keep-Alive' header. */
1124 if (!http_keep_alive_1)
1126 if (header_process (hdr, "Keep-Alive", header_exists,
1127 &http_keep_alive_1))
1130 /* Check for `Connection: Keep-Alive'. */
1131 if (!http_keep_alive_2)
1133 if (header_process (hdr, "Connection", http_process_connection,
1134 &http_keep_alive_2))
1142 logputs (LOG_VERBOSE, "\n");
1145 && (http_keep_alive_1 || http_keep_alive_2))
1147 assert (inhibit_keep_alive == 0);
1151 /* The server has promised that it will not close the connection
1152 when we're done. This means that we can register it. */
1153 register_persistent (conn->host, conn->port, sock, using_ssl);
1155 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1158 /* Authorization is required. */
1162 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1163 might be more bytes in the body. */
1164 if (auth_tried_already)
1166 /* If we have tried it already, then there is not point
1169 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1170 xfree (authenticate_h);
1173 else if (!known_authentication_scheme_p (authenticate_h))
1175 xfree (authenticate_h);
1176 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1179 else if (BEGINS_WITH (authenticate_h, "Basic"))
1181 /* The authentication scheme is basic, the one we try by
1182 default, and it failed. There's no sense in trying
1188 auth_tried_already = 1;
1192 /* We do not need this anymore. */
1195 xfree (authenticate_h);
1196 authenticate_h = NULL;
1199 /* 20x responses are counted among successful by default. */
1200 if (H_20X (statcode))
1203 /* Return if redirected. */
1204 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1206 /* RFC2068 says that in case of the 300 (multiple choices)
1207 response, the server can output a preferred URL through
1208 `Location' header; otherwise, the request should be treated
1209 like GET. So, if the location is set, it will be a
1210 redirection; otherwise, just proceed normally. */
1211 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1215 logprintf (LOG_VERBOSE,
1216 _("Location: %s%s\n"),
1217 hs->newloc ? hs->newloc : _("unspecified"),
1218 hs->newloc ? _(" [following]") : "");
1219 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1220 might be more bytes in the body. */
1222 xfree_null (all_headers);
1227 /* If content-type is not given, assume text/html. This is because
1228 of the multitude of broken CGI's that "forget" to generate the
1231 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1232 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1237 if (opt.html_extension && (*dt & TEXTHTML))
1238 /* -E / --html-extension / html_extension = on was specified, and this is a
1239 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1240 already the file's suffix, tack on ".html". */
1242 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1244 if (last_period_in_local_filename == NULL
1245 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1246 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1248 size_t local_filename_len = strlen(*hs->local_file);
1250 *hs->local_file = xrealloc(*hs->local_file,
1251 local_filename_len + sizeof(".html"));
1252 strcpy(*hs->local_file + local_filename_len, ".html");
1254 *dt |= ADDED_HTML_EXTENSION;
1258 if (contrange == -1)
1260 /* We did not get a content-range header. This means that the
1261 server did not honor our `Range' request. Normally, this
1262 means we should reset hs->restval and continue normally. */
1264 /* However, if `-c' is used, we need to be a bit more careful:
1266 1. If `-c' is specified and the file already existed when
1267 Wget was started, it would be a bad idea for us to start
1268 downloading it from scratch, effectively truncating it. I
1269 believe this cannot happen unless `-c' was specified.
1271 2. If `-c' is used on a file that is already fully
1272 downloaded, we're requesting bytes after the end of file,
1273 which can result in server not honoring `Range'. If this is
1274 the case, `Content-Length' will be equal to the length of the
1276 if (opt.always_rest)
1278 /* Check for condition #2. */
1279 if (hs->restval > 0 /* restart was requested. */
1280 && contlen != -1 /* we got content-length. */
1281 && hs->restval >= contlen /* file fully downloaded
1285 logputs (LOG_VERBOSE, _("\
1286 \n The file is already fully retrieved; nothing to do.\n\n"));
1287 /* In case the caller inspects. */
1290 /* Mark as successfully retrieved. */
1293 xfree_null (all_headers);
1294 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1295 might be more bytes in the body. */
1296 return RETRUNNEEDED;
1299 /* Check for condition #1. */
1300 if (hs->no_truncate)
1302 logprintf (LOG_NOTQUIET,
1305 Continued download failed on this file, which conflicts with `-c'.\n\
1306 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1308 xfree_null (all_headers);
1309 CLOSE_INVALIDATE (sock);
1310 return CONTNOTSUPPORTED;
1318 else if (contrange != hs->restval ||
1319 (H_PARTIAL (statcode) && contrange == -1))
1321 /* This means the whole request was somehow misunderstood by the
1322 server. Bail out. */
1324 xfree_null (all_headers);
1325 CLOSE_INVALIDATE (sock);
1332 contlen += contrange;
1334 contrange = -1; /* If conent-length was not sent,
1335 content-range will be ignored. */
1337 hs->contlen = contlen;
1341 if ((*dt & RETROKF) && !opt.server_response)
1343 /* No need to print this output if the body won't be
1344 downloaded at all, or if the original server response is
1346 logputs (LOG_VERBOSE, _("Length: "));
1349 logputs (LOG_VERBOSE, legible (contlen));
1350 if (contrange != -1)
1351 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1352 legible (contlen - contrange));
1355 logputs (LOG_VERBOSE,
1356 opt.ignore_length ? _("ignored") : _("unspecified"));
1358 logprintf (LOG_VERBOSE, " [%s]\n", type);
1360 logputs (LOG_VERBOSE, "\n");
1364 type = NULL; /* We don't need it any more. */
1366 /* Return if we have no intention of further downloading. */
1367 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1369 /* In case the caller cares to look... */
1373 xfree_null (all_headers);
1374 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1375 might be more bytes in the body. */
1376 return RETRFINISHED;
1379 /* Open the local file. */
1382 mkalldirs (*hs->local_file);
1384 rotate_backups (*hs->local_file);
1385 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1388 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1389 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1390 might be more bytes in the body. */
1391 xfree_null (all_headers);
1397 extern int global_download_count;
1399 /* To ensure that repeated "from scratch" downloads work for -O
1400 files, we rewind the file pointer, unless restval is
1401 non-zero. (This works only when -O is used on regular files,
1402 but it's still a valuable feature.)
1404 However, this loses when more than one URL is specified on
1405 the command line the second rewinds eradicates the contents
1406 of the first download. Thus we disable the above trick for
1407 all the downloads except the very first one.
1409 #### A possible solution to this would be to remember the
1410 file position in the output document and to seek to that
1411 position, instead of rewinding.
1413 We don't truncate stdout, since that breaks
1414 "wget -O - [...] >> foo".
1416 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1418 /* This will silently fail for streams that don't correspond
1419 to regular files, but that's OK. */
1421 /* ftruncate is needed because opt.dfp is opened in append
1422 mode if opt.always_rest is set. */
1423 ftruncate (fileno (fp), 0);
1428 /* #### This confuses the code that checks for file size. There
1429 should be some overhead information. */
1430 if (opt.save_headers)
1431 fwrite (all_headers, 1, all_length, fp);
1433 /* Get the contents of the document. */
1434 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1435 (contlen != -1 ? contlen : 0),
1436 &rbuf, keep_alive, &hs->dltime);
1439 CLOSE_FINISH (sock);
1441 CLOSE_INVALIDATE (sock);
1444 /* Close or flush the file. We have to be careful to check for
1445 error here. Checking the result of fwrite() is not enough --
1446 errors could go unnoticed! */
1449 flush_res = fclose (fp);
1451 flush_res = fflush (fp);
1452 if (flush_res == EOF)
1455 xfree_null (all_headers);
1458 return RETRFINISHED;
1461 /* The genuine HTTP loop! This is the part where the retrieval is
1462 retried, and retried, and retried, and... */
1464 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1465 int *dt, struct url *proxy)
1468 int use_ts, got_head = 0; /* time-stamping info */
1469 char *filename_plus_orig_suffix;
1470 char *local_filename = NULL;
1471 char *tms, *locf, *tmrate;
1473 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1474 long local_size = 0; /* the size of the local file */
1475 size_t filename_len;
1476 struct http_stat hstat; /* HTTP status */
1480 /* This used to be done in main(), but it's a better idea to do it
1481 here so that we don't go through the hoops if we're just using
1485 if (!wget_cookie_jar)
1486 wget_cookie_jar = cookie_jar_new ();
1487 if (opt.cookies_input && !cookies_loaded_p)
1489 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1490 cookies_loaded_p = 1;
1496 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1497 has_wildcards_p because it would also warn on `?', and we know that
1498 shows up in CGI paths a *lot*. */
1499 if (strchr (u->url, '*'))
1500 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1502 /* Determine the local filename. */
1503 if (local_file && *local_file)
1504 hstat.local_file = local_file;
1505 else if (local_file)
1507 *local_file = url_file_name (u);
1508 hstat.local_file = local_file;
1512 dummy = url_file_name (u);
1513 hstat.local_file = &dummy;
1516 if (!opt.output_document)
1517 locf = *hstat.local_file;
1519 locf = opt.output_document;
1521 hstat.referer = referer;
1523 filename_len = strlen (*hstat.local_file);
1524 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1526 if (opt.noclobber && file_exists_p (*hstat.local_file))
1528 /* If opt.noclobber is turned on and file already exists, do not
1529 retrieve the file */
1530 logprintf (LOG_VERBOSE, _("\
1531 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1532 /* If the file is there, we suppose it's retrieved OK. */
1535 /* #### Bogusness alert. */
1536 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1537 if (has_html_suffix_p (*hstat.local_file))
1545 if (opt.timestamping)
1547 int local_dot_orig_file_exists = 0;
1549 if (opt.backup_converted)
1550 /* If -K is specified, we'll act on the assumption that it was specified
1551 last time these files were downloaded as well, and instead of just
1552 comparing local file X against server file X, we'll compare local
1553 file X.orig (if extant, else X) against server file X. If -K
1554 _wasn't_ specified last time, or the server contains files called
1555 *.orig, -N will be back to not operating correctly with -k. */
1557 /* Would a single s[n]printf() call be faster? --dan
1559 Definitely not. sprintf() is horribly slow. It's a
1560 different question whether the difference between the two
1561 affects a program. Usually I'd say "no", but at one
1562 point I profiled Wget, and found that a measurable and
1563 non-negligible amount of time was lost calling sprintf()
1564 in url.c. Replacing sprintf with inline calls to
1565 strcpy() and long_to_string() made a difference.
1567 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1568 memcpy (filename_plus_orig_suffix + filename_len,
1569 ".orig", sizeof (".orig"));
1571 /* Try to stat() the .orig file. */
1572 if (stat (filename_plus_orig_suffix, &st) == 0)
1574 local_dot_orig_file_exists = 1;
1575 local_filename = filename_plus_orig_suffix;
1579 if (!local_dot_orig_file_exists)
1580 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1581 if (stat (*hstat.local_file, &st) == 0)
1582 local_filename = *hstat.local_file;
1584 if (local_filename != NULL)
1585 /* There was a local file, so we'll check later to see if the version
1586 the server has is the same version we already have, allowing us to
1592 /* Modification time granularity is 2 seconds for Windows, so
1593 increase local time by 1 second for later comparison. */
1596 local_size = st.st_size;
1600 /* Reset the counter. */
1602 *dt = 0 | ACCEPTRANGES;
1606 /* Increment the pass counter. */
1608 sleep_between_retrievals (count);
1609 /* Get the current time string. */
1610 tms = time_str (NULL);
1611 /* Print fetch message, if opt.verbose. */
1614 char *hurl = url_string (u, 1);
1618 sprintf (tmp, _("(try:%2d)"), count);
1619 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1620 tms, hurl, tmp, locf);
1622 ws_changetitle (hurl, 1);
1627 /* Default document type is empty. However, if spider mode is
1628 on or time-stamping is employed, HEAD_ONLY commands is
1629 encoded within *dt. */
1630 if (opt.spider || (use_ts && !got_head))
1634 /* Assume no restarting. */
1636 /* Decide whether or not to restart. */
1637 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1638 /* #### this calls access() and then stat(); could be optimized. */
1639 && file_exists_p (locf))
1640 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1641 hstat.restval = st.st_size;
1643 /* In `-c' is used and the file is existing and non-empty,
1644 refuse to truncate it if the server doesn't support continued
1646 hstat.no_truncate = 0;
1647 if (opt.always_rest && hstat.restval)
1648 hstat.no_truncate = 1;
1650 /* Decide whether to send the no-cache directive. We send it in
1652 a) we're using a proxy, and we're past our first retrieval.
1653 Some proxies are notorious for caching incomplete data, so
1654 we require a fresh get.
1655 b) caching is explicitly inhibited. */
1656 if ((proxy && count > 1) /* a */
1657 || !opt.allow_cache /* b */
1659 *dt |= SEND_NOCACHE;
1661 *dt &= ~SEND_NOCACHE;
1663 /* Try fetching the document, or at least its head. */
1664 err = gethttp (u, &hstat, dt, proxy);
1666 /* It's unfortunate that wget determines the local filename before finding
1667 out the Content-Type of the file. Barring a major restructuring of the
1668 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1669 *hstat.local_file to tack on ".html". */
1670 if (!opt.output_document)
1671 locf = *hstat.local_file;
1673 locf = opt.output_document;
1676 tms = time_str (NULL);
1677 /* Get the new location (with or without the redirection). */
1679 *newloc = xstrdup (hstat.newloc);
1682 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1683 case CONERROR: case READERR: case WRITEFAILED:
1685 /* Non-fatal errors continue executing the loop, which will
1686 bring them to "while" statement at the end, to judge
1687 whether the number of tries was exceeded. */
1688 free_hstat (&hstat);
1689 printwhat (count, opt.ntry);
1692 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1693 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1694 /* Fatal errors just return from the function. */
1695 free_hstat (&hstat);
1699 case FWRITEERR: case FOPENERR:
1700 /* Another fatal error. */
1701 logputs (LOG_VERBOSE, "\n");
1702 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1703 *hstat.local_file, strerror (errno));
1704 free_hstat (&hstat);
1709 /* Another fatal error. */
1710 logputs (LOG_VERBOSE, "\n");
1711 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1712 free_hstat (&hstat);
1717 /* Return the new location to the caller. */
1720 logprintf (LOG_NOTQUIET,
1721 _("ERROR: Redirection (%d) without location.\n"),
1723 free_hstat (&hstat);
1727 free_hstat (&hstat);
1732 /* The file was already fully retrieved. */
1733 free_hstat (&hstat);
1738 /* Deal with you later. */
1741 /* All possibilities should have been exhausted. */
1744 if (!(*dt & RETROKF))
1748 /* #### Ugly ugly ugly! */
1749 char *hurl = url_string (u, 1);
1750 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1753 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1754 tms, hstat.statcode, hstat.error);
1755 logputs (LOG_VERBOSE, "\n");
1756 free_hstat (&hstat);
1761 /* Did we get the time-stamp? */
1764 if (opt.timestamping && !hstat.remote_time)
1766 logputs (LOG_NOTQUIET, _("\
1767 Last-modified header missing -- time-stamps turned off.\n"));
1769 else if (hstat.remote_time)
1771 /* Convert the date-string into struct tm. */
1772 tmr = http_atotm (hstat.remote_time);
1773 if (tmr == (time_t) (-1))
1774 logputs (LOG_VERBOSE, _("\
1775 Last-modified header invalid -- time-stamp ignored.\n"));
1779 /* The time-stamping section. */
1784 use_ts = 0; /* no more time-stamping */
1785 count = 0; /* the retrieve count for HEAD is
1787 if (hstat.remote_time && tmr != (time_t) (-1))
1789 /* Now time-stamping can be used validly. Time-stamping
1790 means that if the sizes of the local and remote file
1791 match, and local file is newer than the remote file,
1792 it will not be retrieved. Otherwise, the normal
1793 download procedure is resumed. */
1795 (hstat.contlen == -1 || local_size == hstat.contlen))
1797 logprintf (LOG_VERBOSE, _("\
1798 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1800 free_hstat (&hstat);
1804 else if (tml >= tmr)
1805 logprintf (LOG_VERBOSE, _("\
1806 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1808 logputs (LOG_VERBOSE,
1809 _("Remote file is newer, retrieving.\n"));
1811 free_hstat (&hstat);
1814 if ((tmr != (time_t) (-1))
1816 && ((hstat.len == hstat.contlen) ||
1817 ((hstat.res == 0) &&
1818 ((hstat.contlen == -1) ||
1819 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1821 /* #### This code repeats in http.c and ftp.c. Move it to a
1823 const char *fl = NULL;
1824 if (opt.output_document)
1826 if (opt.od_known_regular)
1827 fl = opt.output_document;
1830 fl = *hstat.local_file;
1834 /* End of time-stamping section. */
1838 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1843 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1845 if (hstat.len == hstat.contlen)
1849 logprintf (LOG_VERBOSE,
1850 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1851 tms, tmrate, locf, hstat.len, hstat.contlen);
1852 logprintf (LOG_NONVERBOSE,
1853 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1854 tms, u->url, hstat.len, hstat.contlen, locf, count);
1857 total_downloaded_bytes += hstat.len;
1859 /* Remember that we downloaded the file for later ".orig" code. */
1860 if (*dt & ADDED_HTML_EXTENSION)
1861 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1863 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1865 free_hstat (&hstat);
1869 else if (hstat.res == 0) /* No read error */
1871 if (hstat.contlen == -1) /* We don't know how much we were supposed
1872 to get, so assume we succeeded. */
1876 logprintf (LOG_VERBOSE,
1877 _("%s (%s) - `%s' saved [%ld]\n\n"),
1878 tms, tmrate, locf, hstat.len);
1879 logprintf (LOG_NONVERBOSE,
1880 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1881 tms, u->url, hstat.len, locf, count);
1884 total_downloaded_bytes += hstat.len;
1886 /* Remember that we downloaded the file for later ".orig" code. */
1887 if (*dt & ADDED_HTML_EXTENSION)
1888 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1890 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1892 free_hstat (&hstat);
1896 else if (hstat.len < hstat.contlen) /* meaning we lost the
1897 connection too soon */
1899 logprintf (LOG_VERBOSE,
1900 _("%s (%s) - Connection closed at byte %ld. "),
1901 tms, tmrate, hstat.len);
1902 printwhat (count, opt.ntry);
1903 free_hstat (&hstat);
1906 else if (!opt.kill_longer) /* meaning we got more than expected */
1908 logprintf (LOG_VERBOSE,
1909 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1910 tms, tmrate, locf, hstat.len, hstat.contlen);
1911 logprintf (LOG_NONVERBOSE,
1912 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1913 tms, u->url, hstat.len, hstat.contlen, locf, count);
1915 total_downloaded_bytes += hstat.len;
1917 /* Remember that we downloaded the file for later ".orig" code. */
1918 if (*dt & ADDED_HTML_EXTENSION)
1919 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1921 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1923 free_hstat (&hstat);
1927 else /* the same, but not accepted */
1929 logprintf (LOG_VERBOSE,
1930 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1931 tms, tmrate, hstat.len, hstat.contlen);
1932 printwhat (count, opt.ntry);
1933 free_hstat (&hstat);
1937 else /* now hstat.res can only be -1 */
1939 if (hstat.contlen == -1)
1941 logprintf (LOG_VERBOSE,
1942 _("%s (%s) - Read error at byte %ld (%s)."),
1943 tms, tmrate, hstat.len, strerror (errno));
1944 printwhat (count, opt.ntry);
1945 free_hstat (&hstat);
1948 else /* hstat.res == -1 and contlen is given */
1950 logprintf (LOG_VERBOSE,
1951 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1952 tms, tmrate, hstat.len, hstat.contlen,
1954 printwhat (count, opt.ntry);
1955 free_hstat (&hstat);
1962 while (!opt.ntry || (count < opt.ntry));
1966 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1967 than local timezone.
1969 mktime is similar but assumes struct tm, also known as the
1970 "broken-down" form of time, is in local time zone. mktime_from_utc
1971 uses mktime to make the conversion understanding that an offset
1972 will be introduced by the local time assumption.
1974 mktime_from_utc then measures the introduced offset by applying
1975 gmtime to the initial result and applying mktime to the resulting
1976 "broken-down" form. The difference between the two mktime results
1977 is the measured offset which is then subtracted from the initial
1978 mktime result to yield a calendar time which is the value returned.
1980 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1981 consistent offset (the non DST offset) since tm and tm+o might be
1982 on opposite sides of a DST change.
1984 Some implementations of mktime return -1 for the nonexistent
1985 localtime hour at the beginning of DST. In this event, use
1986 mktime(tm - 1hr) + 3600.
1990 gmtime(t+o) --> tm+o
1991 mktime(tm+o) --> t+2o
1992 t+o - (t+2o - t+o) = t
1994 Note that glibc contains a function of the same purpose named
1995 `timegm' (reverse of gmtime). But obviously, it is not universally
1996 available, and unfortunately it is not straightforwardly
1997 extractable for use here. Perhaps configure should detect timegm
1998 and use it where available.
2000 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2001 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2002 Further improved by Roger with assistance from Edward J. Sabol
2003 based on input by Jamie Zawinski. */
2006 mktime_from_utc (struct tm *t)
2017 return -1; /* can't deal with output from strptime */
2028 return -1; /* can't deal with output from gmtime */
2031 return (tl - (tb - tl));
2034 /* Check whether the result of strptime() indicates success.
2035 strptime() returns the pointer to how far it got to in the string.
2036 The processing has been successful if the string is at `GMT' or
2037 `+X', or at the end of the string.
2039 In extended regexp parlance, the function returns 1 if P matches
2040 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2041 can return) is considered a failure and 0 is returned. */
2043 check_end (const char *p)
2047 while (ISSPACE (*p))
2050 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2051 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2057 /* Convert the textual specification of time in TIME_STRING to the
2058 number of seconds since the Epoch.
2060 TIME_STRING can be in any of the three formats RFC2068 allows the
2061 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2062 Timezones are ignored, and should be GMT.
2064 Return the computed time_t representation, or -1 if the conversion
2067 This function uses strptime with various string formats for parsing
2068 TIME_STRING. This results in a parser that is not as lenient in
2069 interpreting TIME_STRING as I would like it to be. Being based on
2070 strptime, it always allows shortened months, one-digit days, etc.,
2071 but due to the multitude of formats in which time can be
2072 represented, an ideal HTTP time parser would be even more
2073 forgiving. It should completely ignore things like week days and
2074 concentrate only on the various forms of representing years,
2075 months, days, hours, minutes, and seconds. For example, it would
2076 be nice if it accepted ISO 8601 out of the box.
2078 I've investigated free and PD code for this purpose, but none was
2079 usable. getdate was big and unwieldy, and had potential copyright
2080 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2081 distributed with phttpd, is excellent, but we cannot use it because
2082 it is not assigned to the FSF. So I stuck it with strptime. */
2085 http_atotm (const char *time_string)
2087 /* NOTE: Solaris strptime man page claims that %n and %t match white
2088 space, but that's not universally available. Instead, we simply
2089 use ` ' to mean "skip all WS", which works under all strptime
2090 implementations I've tested. */
2092 static const char *time_formats[] = {
2093 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2094 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2095 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2096 (google.com uses this for their cookies.) */
2097 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2103 /* According to Roger Beeman, we need to initialize tm_isdst, since
2104 strptime won't do it. */
2107 /* Note that under foreign locales Solaris strptime() fails to
2108 recognize English dates, which renders this function useless. We
2109 solve this by being careful not to affect LC_TIME when
2110 initializing locale.
2112 Another solution would be to temporarily set locale to C, invoke
2113 strptime(), and restore it back. This is slow and dirty,
2114 however, and locale support other than LC_MESSAGES can mess other
2115 things, so I rather chose to stick with just setting LC_MESSAGES.
2117 GNU strptime does not have this problem because it recognizes
2118 both international and local dates. */
2120 for (i = 0; i < countof (time_formats); i++)
2121 if (check_end (strptime (time_string, time_formats[i], &t)))
2122 return mktime_from_utc (&t);
2124 /* All formats have failed. */
2128 /* Authorization support: We support two authorization schemes:
2130 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2132 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2133 consisting of answering to the server's challenge with the proper
2136 /* How many bytes it will take to store LEN bytes in base64. */
2137 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2139 /* Encode the string S of length LENGTH to base64 format and place it
2140 to STORE. STORE will be 0-terminated, and must point to a writable
2141 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2143 base64_encode (const char *s, char *store, int length)
2145 /* Conversion table. */
2146 static char tbl[64] = {
2147 'A','B','C','D','E','F','G','H',
2148 'I','J','K','L','M','N','O','P',
2149 'Q','R','S','T','U','V','W','X',
2150 'Y','Z','a','b','c','d','e','f',
2151 'g','h','i','j','k','l','m','n',
2152 'o','p','q','r','s','t','u','v',
2153 'w','x','y','z','0','1','2','3',
2154 '4','5','6','7','8','9','+','/'
2157 unsigned char *p = (unsigned char *)store;
2159 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2160 for (i = 0; i < length; i += 3)
2162 *p++ = tbl[s[0] >> 2];
2163 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2164 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2165 *p++ = tbl[s[2] & 0x3f];
2168 /* Pad the result if necessary... */
2169 if (i == length + 1)
2171 else if (i == length + 2)
2172 *(p - 1) = *(p - 2) = '=';
2173 /* ...and zero-terminate it. */
2177 /* Create the authentication header contents for the `Basic' scheme.
2178 This is done by encoding the string `USER:PASS' in base64 and
2179 prepending `HEADER: Basic ' to it. */
2181 basic_authentication_encode (const char *user, const char *passwd,
2184 char *t1, *t2, *res;
2185 int len1 = strlen (user) + 1 + strlen (passwd);
2186 int len2 = BASE64_LENGTH (len1);
2188 t1 = (char *)alloca (len1 + 1);
2189 sprintf (t1, "%s:%s", user, passwd);
2190 t2 = (char *)alloca (1 + len2);
2191 base64_encode (t1, t2, len1);
2192 res = (char *)xmalloc (len2 + 11 + strlen (header));
2193 sprintf (res, "%s: Basic %s\r\n", header, t2);
2199 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2200 of a field in such a header. If the field is the one specified by
2201 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2202 digest authorization code), extract its value in the (char*)
2203 variable pointed by RET. Returns negative on a malformed header,
2204 or number of bytes that have been parsed by this call. */
2206 extract_header_attr (const char *au, const char *attr_name, char **ret)
2208 const char *cp, *ep;
2212 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2214 cp += strlen (attr_name);
2217 cp += skip_lws (cp);
2222 cp += skip_lws (cp);
2227 for (ep = cp; *ep && *ep != '\"'; ep++)
2232 *ret = strdupdelim (cp, ep);
2239 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2240 an array of 16 bytes containing the hash keys, and BUF should be a
2241 buffer of 33 writable characters (32 for hex digits plus one for
2242 zero termination). */
2244 dump_hash (unsigned char *buf, const unsigned char *hash)
2248 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2250 *buf++ = XNUM_TO_digit (*hash >> 4);
2251 *buf++ = XNUM_TO_digit (*hash & 0xf);
2256 /* Take the line apart to find the challenge, and compose a digest
2257 authorization header. See RFC2069 section 2.1.2. */
2259 digest_authentication_encode (const char *au, const char *user,
2260 const char *passwd, const char *method,
2263 static char *realm, *opaque, *nonce;
2268 { "realm", &realm },
2269 { "opaque", &opaque },
2274 realm = opaque = nonce = NULL;
2276 au += 6; /* skip over `Digest' */
2281 au += skip_lws (au);
2282 for (i = 0; i < countof (options); i++)
2284 int skip = extract_header_attr (au, options[i].name,
2285 options[i].variable);
2289 xfree_null (opaque);
2299 if (i == countof (options))
2301 while (*au && *au != '=')
2305 au += skip_lws (au);
2309 while (*au && *au != '\"')
2316 while (*au && *au != ',')
2321 if (!realm || !nonce || !user || !passwd || !path || !method)
2324 xfree_null (opaque);
2329 /* Calculate the digest value. */
2331 ALLOCA_MD5_CONTEXT (ctx);
2332 unsigned char hash[MD5_HASHLEN];
2333 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2334 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2336 /* A1BUF = H(user ":" realm ":" password) */
2338 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2339 gen_md5_update ((unsigned char *)":", 1, ctx);
2340 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2341 gen_md5_update ((unsigned char *)":", 1, ctx);
2342 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2343 gen_md5_finish (ctx, hash);
2344 dump_hash (a1buf, hash);
2346 /* A2BUF = H(method ":" path) */
2348 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2349 gen_md5_update ((unsigned char *)":", 1, ctx);
2350 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2351 gen_md5_finish (ctx, hash);
2352 dump_hash (a2buf, hash);
2354 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2356 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2357 gen_md5_update ((unsigned char *)":", 1, ctx);
2358 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2359 gen_md5_update ((unsigned char *)":", 1, ctx);
2360 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2361 gen_md5_finish (ctx, hash);
2362 dump_hash (response_digest, hash);
2364 res = (char*) xmalloc (strlen (user)
2369 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2370 + (opaque ? strlen (opaque) : 0)
2372 sprintf (res, "Authorization: Digest \
2373 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2374 user, realm, nonce, path, response_digest);
2377 char *p = res + strlen (res);
2378 strcat (p, ", opaque=\"");
2382 strcat (res, "\r\n");
2386 #endif /* USE_DIGEST */
2389 #define BEGINS_WITH(line, string_constant) \
2390 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2391 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2392 || !line[sizeof (string_constant) - 1]))
2395 known_authentication_scheme_p (const char *au)
2397 return BEGINS_WITH (au, "Basic")
2398 || BEGINS_WITH (au, "Digest")
2399 || BEGINS_WITH (au, "NTLM");
2404 /* Create the HTTP authorization request header. When the
2405 `WWW-Authenticate' response header is seen, according to the
2406 authorization scheme specified in that header (`Basic' and `Digest'
2407 are supported by the current implementation), produce an
2408 appropriate HTTP authorization request header. */
2410 create_authorization_line (const char *au, const char *user,
2411 const char *passwd, const char *method,
2414 char *wwwauth = NULL;
2416 if (!strncasecmp (au, "Basic", 5))
2417 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2418 if (!strncasecmp (au, "NTLM", 4))
2419 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2421 else if (!strncasecmp (au, "Digest", 6))
2422 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2423 #endif /* USE_DIGEST */
2430 if (pc_last_host_ip)
2431 address_list_release (pc_last_host_ip);