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, LH_SILENT);
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 static SSL_CTX *ssl_ctx = NULL;
590 char *cookies = NULL;
592 /* Whether this connection will be kept alive after the HTTP request
596 /* Flags that detect the two ways of specifying HTTP keep-alive
598 int http_keep_alive_1, http_keep_alive_2;
600 /* Whether keep-alive should be inhibited. */
601 int inhibit_keep_alive;
603 /* Whether we need to print the host header with braces around host,
604 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
605 "Host: symbolic-name:1234". */
606 int squares_around_host = 0;
608 /* Headers sent when using POST. */
609 char *post_content_type, *post_content_length;
610 long post_data_size = 0;
613 /* initialize ssl_ctx on first run */
616 uerr_t err = init_ssl (&ssl_ctx);
621 case SSLERRCTXCREATE:
623 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
627 /* try without certfile */
628 logprintf (LOG_NOTQUIET,
629 _("Failed to load certificates from %s\n"),
632 logprintf (LOG_NOTQUIET,
633 _("Trying without the specified certificate\n"));
636 logprintf (LOG_NOTQUIET,
637 _("Failed to get certificate key from %s\n"),
640 logprintf (LOG_NOTQUIET,
641 _("Trying without the specified certificate\n"));
648 #endif /* HAVE_SSL */
650 if (!(*dt & HEAD_ONLY))
651 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
652 know the local filename so we can save to it. */
653 assert (*hs->local_file != NULL);
656 auth_tried_already = 0;
658 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
661 /* We need to come back here when the initial attempt to retrieve
662 without authorization header fails. (Expected to happen at least
663 for the Digest authorization scheme.) */
666 http_keep_alive_1 = http_keep_alive_2 = 0;
668 post_content_type = NULL;
669 post_content_length = NULL;
671 /* Initialize certain elements of struct http_stat. */
676 hs->remote_time = NULL;
679 /* If we're using a proxy, we will be connecting to the proxy
681 conn = proxy ? proxy : u;
683 /* First: establish the connection. */
684 if (inhibit_keep_alive
685 || !persistent_available_p (conn->host, conn->port,
687 u->scheme == SCHEME_HTTPS
693 sock = connect_to_host (conn->host, conn->port);
697 return CONNECT_ERROR (errno);
700 if (conn->scheme == SCHEME_HTTPS)
702 if (!connect_ssl (sock, ssl_ctx))
704 logputs (LOG_VERBOSE, "\n");
705 logprintf (LOG_NOTQUIET,
706 _("Unable to establish SSL connection.\n"));
712 #endif /* HAVE_SSL */
716 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
717 conn->host, conn->port);
718 /* #### pc_last_fd should be accessed through an accessor
721 using_ssl = pc_last_ssl_p;
722 DEBUGP (("Reusing fd %d.\n", sock));
727 else if (opt.post_file_name || opt.post_data)
735 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
736 sprintf (referer, "Referer: %s\r\n", hs->referer);
739 if (*dt & SEND_NOCACHE)
740 pragma_h = "Pragma: no-cache\r\n";
746 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
747 /* Gag me! Some servers (e.g. WebSitePro) have been known to
748 respond to the following `Range' format by generating a
749 multipart/x-byte-ranges MIME document! This MIME type was
750 present in an old draft of the byteranges specification.
751 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
752 only if multiple non-overlapping ranges are requested --
753 which Wget never does. */
754 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
759 STRDUP_ALLOCA (useragent, opt.useragent);
762 useragent = (char *)alloca (10 + strlen (version_string));
763 sprintf (useragent, "Wget/%s", version_string);
765 /* Construct the authentication, if userid is present. */
768 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
769 user = user ? user : opt.http_user;
770 passwd = passwd ? passwd : opt.http_passwd;
777 /* We have the username and the password, but haven't tried
778 any authorization yet. Let's see if the "Basic" method
779 works. If not, we'll come back here and construct a
780 proper authorization method with the right challenges.
782 If we didn't employ this kind of logic, every URL that
783 requires authorization would have to be processed twice,
784 which is very suboptimal and generates a bunch of false
785 "unauthorized" errors in the server log.
787 #### But this logic also has a serious problem when used
788 with stronger authentications: we *first* transmit the
789 username and the password in clear text, and *then*
790 attempt a stronger authentication scheme. That cannot be
791 right! We are only fortunate that almost everyone still
792 uses the `Basic' scheme anyway.
794 There should be an option to prevent this from happening,
795 for those who use strong authentication schemes and value
797 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
801 /* Use the full path, i.e. one that includes the leading
802 slash and the query string, but is independent of proxy
804 char *pth = url_full_path (u);
805 wwwauth = create_authorization_line (authenticate_h, user, passwd,
814 char *proxy_user, *proxy_passwd;
815 /* For normal username and password, URL components override
816 command-line/wgetrc parameters. With proxy authentication,
817 it's the reverse, because proxy URLs are normally the
818 "permanent" ones, so command-line args should take
820 if (opt.proxy_user && opt.proxy_passwd)
822 proxy_user = opt.proxy_user;
823 proxy_passwd = opt.proxy_passwd;
827 proxy_user = proxy->user;
828 proxy_passwd = proxy->passwd;
830 /* #### This does not appear right. Can't the proxy request,
831 say, `Digest' authentication? */
832 if (proxy_user && proxy_passwd)
833 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
834 "Proxy-Authorization");
837 /* String of the form :PORT. Used only for non-standard ports. */
839 if (u->port != scheme_default_port (u->scheme))
841 port_maybe = (char *)alloca (numdigit (u->port) + 2);
842 sprintf (port_maybe, ":%d", u->port);
845 if (!inhibit_keep_alive)
846 request_keep_alive = "Connection: Keep-Alive\r\n";
848 request_keep_alive = NULL;
851 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
853 u->scheme == SCHEME_HTTPS
859 if (opt.post_data || opt.post_file_name)
861 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
863 post_data_size = strlen (opt.post_data);
866 post_data_size = file_size (opt.post_file_name);
867 if (post_data_size == -1)
869 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
874 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
875 sprintf (post_content_length,
876 "Content-Length: %ld\r\n", post_data_size);
880 full_path = xstrdup (u->url);
882 /* Use the full path, i.e. one that includes the leading slash and
883 the query string. E.g. if u->path is "foo/bar" and u->query is
884 "param=value", full_path will be "/foo/bar?param=value". */
885 full_path = url_full_path (u);
887 if (strchr (u->host, ':'))
888 squares_around_host = 1;
890 /* Allocate the memory for the request. */
891 request = (char *)alloca (strlen (command)
895 + (port_maybe ? strlen (port_maybe) : 0)
896 + strlen (HTTP_ACCEPT)
897 + (request_keep_alive
898 ? strlen (request_keep_alive) : 0)
899 + (referer ? strlen (referer) : 0)
900 + (cookies ? strlen (cookies) : 0)
901 + (wwwauth ? strlen (wwwauth) : 0)
902 + (proxyauth ? strlen (proxyauth) : 0)
903 + (range ? strlen (range) : 0)
906 ? strlen (post_content_type) : 0)
907 + (post_content_length
908 ? strlen (post_content_length) : 0)
909 + (opt.user_header ? strlen (opt.user_header) : 0)
911 /* Construct the request. */
917 %s%s%s%s%s%s%s%s%s%s\r\n",
920 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
921 port_maybe ? port_maybe : "",
923 request_keep_alive ? request_keep_alive : "",
924 referer ? referer : "",
925 cookies ? cookies : "",
926 wwwauth ? wwwauth : "",
927 proxyauth ? proxyauth : "",
930 post_content_type ? post_content_type : "",
931 post_content_length ? post_content_length : "",
932 opt.user_header ? opt.user_header : "");
933 DEBUGP (("---request begin---\n%s", request));
935 /* Free the temporary memory. */
936 xfree_null (wwwauth);
937 xfree_null (proxyauth);
938 xfree_null (cookies);
941 /* Send the request to server. */
942 write_error = xwrite (sock, request, strlen (request), -1);
944 if (write_error >= 0)
948 DEBUGP (("[POST data: %s]\n", opt.post_data));
949 write_error = xwrite (sock, opt.post_data, post_data_size, -1);
951 else if (opt.post_file_name && post_data_size != 0)
952 write_error = post_file (sock, opt.post_file_name, post_data_size);
954 DEBUGP (("---request end---\n"));
958 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
960 CLOSE_INVALIDATE (sock);
963 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
964 proxy ? "Proxy" : "HTTP");
965 contlen = contrange = -1;
970 /* Before reading anything, initialize the rbuf. */
971 rbuf_initialize (&rbuf, sock);
974 /* Header-fetching loop. */
982 /* Get the header. */
983 status = header_get (&rbuf, &hdr,
984 /* Disallow continuations for status line. */
985 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
987 /* Check for errors. */
988 if (status == HG_EOF && *hdr)
990 /* This used to be an unconditional error, but that was
991 somewhat controversial, because of a large number of
992 broken CGI's that happily "forget" to send the second EOL
993 before closing the connection of a HEAD request.
995 So, the deal is to check whether the header is empty
996 (*hdr is zero if it is); if yes, it means that the
997 previous header was fully retrieved, and that -- most
998 probably -- the request is complete. "...be liberal in
999 what you accept." Oh boy. */
1000 logputs (LOG_VERBOSE, "\n");
1001 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1004 xfree_null (all_headers);
1005 CLOSE_INVALIDATE (sock);
1008 else if (status == HG_ERROR)
1010 logputs (LOG_VERBOSE, "\n");
1011 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1015 xfree_null (all_headers);
1016 CLOSE_INVALIDATE (sock);
1020 /* If the headers are to be saved to a file later, save them to
1022 if (opt.save_headers)
1024 int lh = strlen (hdr);
1025 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1026 memcpy (all_headers + all_length, hdr, lh);
1028 all_headers[all_length++] = '\n';
1029 all_headers[all_length] = '\0';
1032 /* Check for status line. */
1036 /* Parse the first line of server response. */
1037 statcode = parse_http_status_line (hdr, &error);
1038 hs->statcode = statcode;
1039 /* Store the descriptive response. */
1040 if (statcode == -1) /* malformed response */
1042 /* A common reason for "malformed response" error is the
1043 case when no data was actually received. Handle this
1046 hs->error = xstrdup (_("No data received"));
1048 hs->error = xstrdup (_("Malformed status line"));
1053 hs->error = xstrdup (_("(no description)"));
1055 hs->error = xstrdup (error);
1057 if ((statcode != -1)
1063 if (opt.server_response)
1064 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1066 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1072 /* Exit on empty header. */
1079 /* Print the header if requested. */
1080 if (opt.server_response && hcount != 1)
1081 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1083 /* Try getting content-length. */
1084 if (contlen == -1 && !opt.ignore_length)
1085 if (header_process (hdr, "Content-Length", header_extract_number,
1088 /* Try getting content-type. */
1090 if (header_process (hdr, "Content-Type", http_process_type, &type))
1092 /* Try getting location. */
1094 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1096 /* Try getting last-modified. */
1097 if (!hs->remote_time)
1098 if (header_process (hdr, "Last-Modified", header_strdup,
1101 /* Try getting cookies. */
1103 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1105 /* Try getting www-authentication. */
1106 if (!authenticate_h)
1107 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1110 /* Check for accept-ranges header. If it contains the word
1111 `none', disable the ranges. */
1112 if (*dt & ACCEPTRANGES)
1115 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1118 *dt &= ~ACCEPTRANGES;
1122 /* Try getting content-range. */
1123 if (contrange == -1)
1125 struct http_process_range_closure closure;
1126 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1128 contrange = closure.first_byte_pos;
1132 /* Check for keep-alive related responses. */
1133 if (!inhibit_keep_alive)
1135 /* Check for the `Keep-Alive' header. */
1136 if (!http_keep_alive_1)
1138 if (header_process (hdr, "Keep-Alive", header_exists,
1139 &http_keep_alive_1))
1142 /* Check for `Connection: Keep-Alive'. */
1143 if (!http_keep_alive_2)
1145 if (header_process (hdr, "Connection", http_process_connection,
1146 &http_keep_alive_2))
1154 logputs (LOG_VERBOSE, "\n");
1157 && (http_keep_alive_1 || http_keep_alive_2))
1159 assert (inhibit_keep_alive == 0);
1163 /* The server has promised that it will not close the connection
1164 when we're done. This means that we can register it. */
1165 register_persistent (conn->host, conn->port, sock, using_ssl);
1167 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1170 /* Authorization is required. */
1174 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1175 might be more bytes in the body. */
1176 if (auth_tried_already)
1178 /* If we have tried it already, then there is not point
1181 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1182 xfree (authenticate_h);
1185 else if (!known_authentication_scheme_p (authenticate_h))
1187 xfree (authenticate_h);
1188 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1191 else if (BEGINS_WITH (authenticate_h, "Basic"))
1193 /* The authentication scheme is basic, the one we try by
1194 default, and it failed. There's no sense in trying
1200 auth_tried_already = 1;
1204 /* We do not need this anymore. */
1207 xfree (authenticate_h);
1208 authenticate_h = NULL;
1211 /* 20x responses are counted among successful by default. */
1212 if (H_20X (statcode))
1215 /* Return if redirected. */
1216 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1218 /* RFC2068 says that in case of the 300 (multiple choices)
1219 response, the server can output a preferred URL through
1220 `Location' header; otherwise, the request should be treated
1221 like GET. So, if the location is set, it will be a
1222 redirection; otherwise, just proceed normally. */
1223 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1227 logprintf (LOG_VERBOSE,
1228 _("Location: %s%s\n"),
1229 hs->newloc ? hs->newloc : _("unspecified"),
1230 hs->newloc ? _(" [following]") : "");
1231 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1232 might be more bytes in the body. */
1234 xfree_null (all_headers);
1239 /* If content-type is not given, assume text/html. This is because
1240 of the multitude of broken CGI's that "forget" to generate the
1243 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1244 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1249 if (opt.html_extension && (*dt & TEXTHTML))
1250 /* -E / --html-extension / html_extension = on was specified, and this is a
1251 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1252 already the file's suffix, tack on ".html". */
1254 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1256 if (last_period_in_local_filename == NULL
1257 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1258 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1260 size_t local_filename_len = strlen(*hs->local_file);
1262 *hs->local_file = xrealloc(*hs->local_file,
1263 local_filename_len + sizeof(".html"));
1264 strcpy(*hs->local_file + local_filename_len, ".html");
1266 *dt |= ADDED_HTML_EXTENSION;
1270 if (contrange == -1)
1272 /* We did not get a content-range header. This means that the
1273 server did not honor our `Range' request. Normally, this
1274 means we should reset hs->restval and continue normally. */
1276 /* However, if `-c' is used, we need to be a bit more careful:
1278 1. If `-c' is specified and the file already existed when
1279 Wget was started, it would be a bad idea for us to start
1280 downloading it from scratch, effectively truncating it. I
1281 believe this cannot happen unless `-c' was specified.
1283 2. If `-c' is used on a file that is already fully
1284 downloaded, we're requesting bytes after the end of file,
1285 which can result in server not honoring `Range'. If this is
1286 the case, `Content-Length' will be equal to the length of the
1288 if (opt.always_rest)
1290 /* Check for condition #2. */
1291 if (hs->restval > 0 /* restart was requested. */
1292 && contlen != -1 /* we got content-length. */
1293 && hs->restval >= contlen /* file fully downloaded
1297 logputs (LOG_VERBOSE, _("\
1298 \n The file is already fully retrieved; nothing to do.\n\n"));
1299 /* In case the caller inspects. */
1302 /* Mark as successfully retrieved. */
1305 xfree_null (all_headers);
1306 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1307 might be more bytes in the body. */
1308 return RETRUNNEEDED;
1311 /* Check for condition #1. */
1312 if (hs->no_truncate)
1314 logprintf (LOG_NOTQUIET,
1317 Continued download failed on this file, which conflicts with `-c'.\n\
1318 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1320 xfree_null (all_headers);
1321 CLOSE_INVALIDATE (sock);
1322 return CONTNOTSUPPORTED;
1330 else if (contrange != hs->restval ||
1331 (H_PARTIAL (statcode) && contrange == -1))
1333 /* This means the whole request was somehow misunderstood by the
1334 server. Bail out. */
1336 xfree_null (all_headers);
1337 CLOSE_INVALIDATE (sock);
1344 contlen += contrange;
1346 contrange = -1; /* If conent-length was not sent,
1347 content-range will be ignored. */
1349 hs->contlen = contlen;
1353 if ((*dt & RETROKF) && !opt.server_response)
1355 /* No need to print this output if the body won't be
1356 downloaded at all, or if the original server response is
1358 logputs (LOG_VERBOSE, _("Length: "));
1361 logputs (LOG_VERBOSE, legible (contlen));
1362 if (contrange != -1)
1363 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1364 legible (contlen - contrange));
1367 logputs (LOG_VERBOSE,
1368 opt.ignore_length ? _("ignored") : _("unspecified"));
1370 logprintf (LOG_VERBOSE, " [%s]\n", type);
1372 logputs (LOG_VERBOSE, "\n");
1376 type = NULL; /* We don't need it any more. */
1378 /* Return if we have no intention of further downloading. */
1379 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1381 /* In case the caller cares to look... */
1385 xfree_null (all_headers);
1386 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1387 might be more bytes in the body. */
1388 return RETRFINISHED;
1391 /* Open the local file. */
1394 mkalldirs (*hs->local_file);
1396 rotate_backups (*hs->local_file);
1397 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1400 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1401 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1402 might be more bytes in the body. */
1403 xfree_null (all_headers);
1409 extern int global_download_count;
1411 /* To ensure that repeated "from scratch" downloads work for -O
1412 files, we rewind the file pointer, unless restval is
1413 non-zero. (This works only when -O is used on regular files,
1414 but it's still a valuable feature.)
1416 However, this loses when more than one URL is specified on
1417 the command line the second rewinds eradicates the contents
1418 of the first download. Thus we disable the above trick for
1419 all the downloads except the very first one.
1421 #### A possible solution to this would be to remember the
1422 file position in the output document and to seek to that
1423 position, instead of rewinding.
1425 We don't truncate stdout, since that breaks
1426 "wget -O - [...] >> foo".
1428 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1430 /* This will silently fail for streams that don't correspond
1431 to regular files, but that's OK. */
1433 /* ftruncate is needed because opt.dfp is opened in append
1434 mode if opt.always_rest is set. */
1435 ftruncate (fileno (fp), 0);
1440 /* #### This confuses the code that checks for file size. There
1441 should be some overhead information. */
1442 if (opt.save_headers)
1443 fwrite (all_headers, 1, all_length, fp);
1445 /* Get the contents of the document. */
1446 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1447 (contlen != -1 ? contlen : 0),
1448 &rbuf, keep_alive, &hs->dltime);
1451 CLOSE_FINISH (sock);
1453 CLOSE_INVALIDATE (sock);
1456 /* Close or flush the file. We have to be careful to check for
1457 error here. Checking the result of fwrite() is not enough --
1458 errors could go unnoticed! */
1461 flush_res = fclose (fp);
1463 flush_res = fflush (fp);
1464 if (flush_res == EOF)
1467 xfree_null (all_headers);
1470 return RETRFINISHED;
1473 /* The genuine HTTP loop! This is the part where the retrieval is
1474 retried, and retried, and retried, and... */
1476 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1477 int *dt, struct url *proxy)
1480 int use_ts, got_head = 0; /* time-stamping info */
1481 char *filename_plus_orig_suffix;
1482 char *local_filename = NULL;
1483 char *tms, *locf, *tmrate;
1485 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1486 long local_size = 0; /* the size of the local file */
1487 size_t filename_len;
1488 struct http_stat hstat; /* HTTP status */
1492 /* This used to be done in main(), but it's a better idea to do it
1493 here so that we don't go through the hoops if we're just using
1497 if (!wget_cookie_jar)
1498 wget_cookie_jar = cookie_jar_new ();
1499 if (opt.cookies_input && !cookies_loaded_p)
1501 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1502 cookies_loaded_p = 1;
1508 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1509 has_wildcards_p because it would also warn on `?', and we know that
1510 shows up in CGI paths a *lot*. */
1511 if (strchr (u->url, '*'))
1512 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1514 /* Determine the local filename. */
1515 if (local_file && *local_file)
1516 hstat.local_file = local_file;
1517 else if (local_file)
1519 *local_file = url_file_name (u);
1520 hstat.local_file = local_file;
1524 dummy = url_file_name (u);
1525 hstat.local_file = &dummy;
1528 if (!opt.output_document)
1529 locf = *hstat.local_file;
1531 locf = opt.output_document;
1533 hstat.referer = referer;
1535 filename_len = strlen (*hstat.local_file);
1536 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1538 if (opt.noclobber && file_exists_p (*hstat.local_file))
1540 /* If opt.noclobber is turned on and file already exists, do not
1541 retrieve the file */
1542 logprintf (LOG_VERBOSE, _("\
1543 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1544 /* If the file is there, we suppose it's retrieved OK. */
1547 /* #### Bogusness alert. */
1548 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1549 if (has_html_suffix_p (*hstat.local_file))
1557 if (opt.timestamping)
1559 int local_dot_orig_file_exists = 0;
1561 if (opt.backup_converted)
1562 /* If -K is specified, we'll act on the assumption that it was specified
1563 last time these files were downloaded as well, and instead of just
1564 comparing local file X against server file X, we'll compare local
1565 file X.orig (if extant, else X) against server file X. If -K
1566 _wasn't_ specified last time, or the server contains files called
1567 *.orig, -N will be back to not operating correctly with -k. */
1569 /* Would a single s[n]printf() call be faster? --dan
1571 Definitely not. sprintf() is horribly slow. It's a
1572 different question whether the difference between the two
1573 affects a program. Usually I'd say "no", but at one
1574 point I profiled Wget, and found that a measurable and
1575 non-negligible amount of time was lost calling sprintf()
1576 in url.c. Replacing sprintf with inline calls to
1577 strcpy() and long_to_string() made a difference.
1579 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1580 memcpy (filename_plus_orig_suffix + filename_len,
1581 ".orig", sizeof (".orig"));
1583 /* Try to stat() the .orig file. */
1584 if (stat (filename_plus_orig_suffix, &st) == 0)
1586 local_dot_orig_file_exists = 1;
1587 local_filename = filename_plus_orig_suffix;
1591 if (!local_dot_orig_file_exists)
1592 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1593 if (stat (*hstat.local_file, &st) == 0)
1594 local_filename = *hstat.local_file;
1596 if (local_filename != NULL)
1597 /* There was a local file, so we'll check later to see if the version
1598 the server has is the same version we already have, allowing us to
1604 /* Modification time granularity is 2 seconds for Windows, so
1605 increase local time by 1 second for later comparison. */
1608 local_size = st.st_size;
1612 /* Reset the counter. */
1614 *dt = 0 | ACCEPTRANGES;
1618 /* Increment the pass counter. */
1620 sleep_between_retrievals (count);
1621 /* Get the current time string. */
1622 tms = time_str (NULL);
1623 /* Print fetch message, if opt.verbose. */
1626 char *hurl = url_string (u, 1);
1630 sprintf (tmp, _("(try:%2d)"), count);
1631 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1632 tms, hurl, tmp, locf);
1634 ws_changetitle (hurl, 1);
1639 /* Default document type is empty. However, if spider mode is
1640 on or time-stamping is employed, HEAD_ONLY commands is
1641 encoded within *dt. */
1642 if (opt.spider || (use_ts && !got_head))
1646 /* Assume no restarting. */
1648 /* Decide whether or not to restart. */
1649 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1650 /* #### this calls access() and then stat(); could be optimized. */
1651 && file_exists_p (locf))
1652 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1653 hstat.restval = st.st_size;
1655 /* In `-c' is used and the file is existing and non-empty,
1656 refuse to truncate it if the server doesn't support continued
1658 hstat.no_truncate = 0;
1659 if (opt.always_rest && hstat.restval)
1660 hstat.no_truncate = 1;
1662 /* Decide whether to send the no-cache directive. We send it in
1664 a) we're using a proxy, and we're past our first retrieval.
1665 Some proxies are notorious for caching incomplete data, so
1666 we require a fresh get.
1667 b) caching is explicitly inhibited. */
1668 if ((proxy && count > 1) /* a */
1669 || !opt.allow_cache /* b */
1671 *dt |= SEND_NOCACHE;
1673 *dt &= ~SEND_NOCACHE;
1675 /* Try fetching the document, or at least its head. */
1676 err = gethttp (u, &hstat, dt, proxy);
1678 /* It's unfortunate that wget determines the local filename before finding
1679 out the Content-Type of the file. Barring a major restructuring of the
1680 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1681 *hstat.local_file to tack on ".html". */
1682 if (!opt.output_document)
1683 locf = *hstat.local_file;
1685 locf = opt.output_document;
1688 tms = time_str (NULL);
1689 /* Get the new location (with or without the redirection). */
1691 *newloc = xstrdup (hstat.newloc);
1694 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1695 case CONERROR: case READERR: case WRITEFAILED:
1697 /* Non-fatal errors continue executing the loop, which will
1698 bring them to "while" statement at the end, to judge
1699 whether the number of tries was exceeded. */
1700 free_hstat (&hstat);
1701 printwhat (count, opt.ntry);
1704 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1705 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1706 /* Fatal errors just return from the function. */
1707 free_hstat (&hstat);
1711 case FWRITEERR: case FOPENERR:
1712 /* Another fatal error. */
1713 logputs (LOG_VERBOSE, "\n");
1714 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1715 *hstat.local_file, strerror (errno));
1716 free_hstat (&hstat);
1721 /* Another fatal error. */
1722 logputs (LOG_VERBOSE, "\n");
1723 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1724 free_hstat (&hstat);
1729 /* Return the new location to the caller. */
1732 logprintf (LOG_NOTQUIET,
1733 _("ERROR: Redirection (%d) without location.\n"),
1735 free_hstat (&hstat);
1739 free_hstat (&hstat);
1744 /* The file was already fully retrieved. */
1745 free_hstat (&hstat);
1750 /* Deal with you later. */
1753 /* All possibilities should have been exhausted. */
1756 if (!(*dt & RETROKF))
1760 /* #### Ugly ugly ugly! */
1761 char *hurl = url_string (u, 1);
1762 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1765 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1766 tms, hstat.statcode, hstat.error);
1767 logputs (LOG_VERBOSE, "\n");
1768 free_hstat (&hstat);
1773 /* Did we get the time-stamp? */
1776 if (opt.timestamping && !hstat.remote_time)
1778 logputs (LOG_NOTQUIET, _("\
1779 Last-modified header missing -- time-stamps turned off.\n"));
1781 else if (hstat.remote_time)
1783 /* Convert the date-string into struct tm. */
1784 tmr = http_atotm (hstat.remote_time);
1785 if (tmr == (time_t) (-1))
1786 logputs (LOG_VERBOSE, _("\
1787 Last-modified header invalid -- time-stamp ignored.\n"));
1791 /* The time-stamping section. */
1796 use_ts = 0; /* no more time-stamping */
1797 count = 0; /* the retrieve count for HEAD is
1799 if (hstat.remote_time && tmr != (time_t) (-1))
1801 /* Now time-stamping can be used validly. Time-stamping
1802 means that if the sizes of the local and remote file
1803 match, and local file is newer than the remote file,
1804 it will not be retrieved. Otherwise, the normal
1805 download procedure is resumed. */
1807 (hstat.contlen == -1 || local_size == hstat.contlen))
1809 logprintf (LOG_VERBOSE, _("\
1810 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1812 free_hstat (&hstat);
1816 else if (tml >= tmr)
1817 logprintf (LOG_VERBOSE, _("\
1818 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1820 logputs (LOG_VERBOSE,
1821 _("Remote file is newer, retrieving.\n"));
1823 free_hstat (&hstat);
1826 if ((tmr != (time_t) (-1))
1828 && ((hstat.len == hstat.contlen) ||
1829 ((hstat.res == 0) &&
1830 ((hstat.contlen == -1) ||
1831 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1833 /* #### This code repeats in http.c and ftp.c. Move it to a
1835 const char *fl = NULL;
1836 if (opt.output_document)
1838 if (opt.od_known_regular)
1839 fl = opt.output_document;
1842 fl = *hstat.local_file;
1846 /* End of time-stamping section. */
1850 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1855 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1857 if (hstat.len == hstat.contlen)
1861 logprintf (LOG_VERBOSE,
1862 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1863 tms, tmrate, locf, hstat.len, hstat.contlen);
1864 logprintf (LOG_NONVERBOSE,
1865 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1866 tms, u->url, hstat.len, hstat.contlen, locf, count);
1869 total_downloaded_bytes += hstat.len;
1871 /* Remember that we downloaded the file for later ".orig" code. */
1872 if (*dt & ADDED_HTML_EXTENSION)
1873 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1875 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1877 free_hstat (&hstat);
1881 else if (hstat.res == 0) /* No read error */
1883 if (hstat.contlen == -1) /* We don't know how much we were supposed
1884 to get, so assume we succeeded. */
1888 logprintf (LOG_VERBOSE,
1889 _("%s (%s) - `%s' saved [%ld]\n\n"),
1890 tms, tmrate, locf, hstat.len);
1891 logprintf (LOG_NONVERBOSE,
1892 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1893 tms, u->url, hstat.len, locf, count);
1896 total_downloaded_bytes += hstat.len;
1898 /* Remember that we downloaded the file for later ".orig" code. */
1899 if (*dt & ADDED_HTML_EXTENSION)
1900 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1902 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1904 free_hstat (&hstat);
1908 else if (hstat.len < hstat.contlen) /* meaning we lost the
1909 connection too soon */
1911 logprintf (LOG_VERBOSE,
1912 _("%s (%s) - Connection closed at byte %ld. "),
1913 tms, tmrate, hstat.len);
1914 printwhat (count, opt.ntry);
1915 free_hstat (&hstat);
1918 else if (!opt.kill_longer) /* meaning we got more than expected */
1920 logprintf (LOG_VERBOSE,
1921 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1922 tms, tmrate, locf, hstat.len, hstat.contlen);
1923 logprintf (LOG_NONVERBOSE,
1924 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1925 tms, u->url, hstat.len, hstat.contlen, locf, count);
1927 total_downloaded_bytes += hstat.len;
1929 /* Remember that we downloaded the file for later ".orig" code. */
1930 if (*dt & ADDED_HTML_EXTENSION)
1931 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1933 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1935 free_hstat (&hstat);
1939 else /* the same, but not accepted */
1941 logprintf (LOG_VERBOSE,
1942 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1943 tms, tmrate, hstat.len, hstat.contlen);
1944 printwhat (count, opt.ntry);
1945 free_hstat (&hstat);
1949 else /* now hstat.res can only be -1 */
1951 if (hstat.contlen == -1)
1953 logprintf (LOG_VERBOSE,
1954 _("%s (%s) - Read error at byte %ld (%s)."),
1955 tms, tmrate, hstat.len, strerror (errno));
1956 printwhat (count, opt.ntry);
1957 free_hstat (&hstat);
1960 else /* hstat.res == -1 and contlen is given */
1962 logprintf (LOG_VERBOSE,
1963 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1964 tms, tmrate, hstat.len, hstat.contlen,
1966 printwhat (count, opt.ntry);
1967 free_hstat (&hstat);
1974 while (!opt.ntry || (count < opt.ntry));
1978 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1979 than local timezone.
1981 mktime is similar but assumes struct tm, also known as the
1982 "broken-down" form of time, is in local time zone. mktime_from_utc
1983 uses mktime to make the conversion understanding that an offset
1984 will be introduced by the local time assumption.
1986 mktime_from_utc then measures the introduced offset by applying
1987 gmtime to the initial result and applying mktime to the resulting
1988 "broken-down" form. The difference between the two mktime results
1989 is the measured offset which is then subtracted from the initial
1990 mktime result to yield a calendar time which is the value returned.
1992 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1993 consistent offset (the non DST offset) since tm and tm+o might be
1994 on opposite sides of a DST change.
1996 Some implementations of mktime return -1 for the nonexistent
1997 localtime hour at the beginning of DST. In this event, use
1998 mktime(tm - 1hr) + 3600.
2002 gmtime(t+o) --> tm+o
2003 mktime(tm+o) --> t+2o
2004 t+o - (t+2o - t+o) = t
2006 Note that glibc contains a function of the same purpose named
2007 `timegm' (reverse of gmtime). But obviously, it is not universally
2008 available, and unfortunately it is not straightforwardly
2009 extractable for use here. Perhaps configure should detect timegm
2010 and use it where available.
2012 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2013 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2014 Further improved by Roger with assistance from Edward J. Sabol
2015 based on input by Jamie Zawinski. */
2018 mktime_from_utc (struct tm *t)
2029 return -1; /* can't deal with output from strptime */
2040 return -1; /* can't deal with output from gmtime */
2043 return (tl - (tb - tl));
2046 /* Check whether the result of strptime() indicates success.
2047 strptime() returns the pointer to how far it got to in the string.
2048 The processing has been successful if the string is at `GMT' or
2049 `+X', or at the end of the string.
2051 In extended regexp parlance, the function returns 1 if P matches
2052 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2053 can return) is considered a failure and 0 is returned. */
2055 check_end (const char *p)
2059 while (ISSPACE (*p))
2062 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2063 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2069 /* Convert the textual specification of time in TIME_STRING to the
2070 number of seconds since the Epoch.
2072 TIME_STRING can be in any of the three formats RFC2068 allows the
2073 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2074 Timezones are ignored, and should be GMT.
2076 Return the computed time_t representation, or -1 if the conversion
2079 This function uses strptime with various string formats for parsing
2080 TIME_STRING. This results in a parser that is not as lenient in
2081 interpreting TIME_STRING as I would like it to be. Being based on
2082 strptime, it always allows shortened months, one-digit days, etc.,
2083 but due to the multitude of formats in which time can be
2084 represented, an ideal HTTP time parser would be even more
2085 forgiving. It should completely ignore things like week days and
2086 concentrate only on the various forms of representing years,
2087 months, days, hours, minutes, and seconds. For example, it would
2088 be nice if it accepted ISO 8601 out of the box.
2090 I've investigated free and PD code for this purpose, but none was
2091 usable. getdate was big and unwieldy, and had potential copyright
2092 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2093 distributed with phttpd, is excellent, but we cannot use it because
2094 it is not assigned to the FSF. So I stuck it with strptime. */
2097 http_atotm (const char *time_string)
2099 /* NOTE: Solaris strptime man page claims that %n and %t match white
2100 space, but that's not universally available. Instead, we simply
2101 use ` ' to mean "skip all WS", which works under all strptime
2102 implementations I've tested. */
2104 static const char *time_formats[] = {
2105 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2106 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2107 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2108 (google.com uses this for their cookies.) */
2109 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2115 /* According to Roger Beeman, we need to initialize tm_isdst, since
2116 strptime won't do it. */
2119 /* Note that under foreign locales Solaris strptime() fails to
2120 recognize English dates, which renders this function useless. We
2121 solve this by being careful not to affect LC_TIME when
2122 initializing locale.
2124 Another solution would be to temporarily set locale to C, invoke
2125 strptime(), and restore it back. This is slow and dirty,
2126 however, and locale support other than LC_MESSAGES can mess other
2127 things, so I rather chose to stick with just setting LC_MESSAGES.
2129 GNU strptime does not have this problem because it recognizes
2130 both international and local dates. */
2132 for (i = 0; i < countof (time_formats); i++)
2133 if (check_end (strptime (time_string, time_formats[i], &t)))
2134 return mktime_from_utc (&t);
2136 /* All formats have failed. */
2140 /* Authorization support: We support two authorization schemes:
2142 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2144 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2145 consisting of answering to the server's challenge with the proper
2148 /* How many bytes it will take to store LEN bytes in base64. */
2149 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2151 /* Encode the string S of length LENGTH to base64 format and place it
2152 to STORE. STORE will be 0-terminated, and must point to a writable
2153 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2155 base64_encode (const char *s, char *store, int length)
2157 /* Conversion table. */
2158 static char tbl[64] = {
2159 'A','B','C','D','E','F','G','H',
2160 'I','J','K','L','M','N','O','P',
2161 'Q','R','S','T','U','V','W','X',
2162 'Y','Z','a','b','c','d','e','f',
2163 'g','h','i','j','k','l','m','n',
2164 'o','p','q','r','s','t','u','v',
2165 'w','x','y','z','0','1','2','3',
2166 '4','5','6','7','8','9','+','/'
2169 unsigned char *p = (unsigned char *)store;
2171 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2172 for (i = 0; i < length; i += 3)
2174 *p++ = tbl[s[0] >> 2];
2175 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2176 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2177 *p++ = tbl[s[2] & 0x3f];
2180 /* Pad the result if necessary... */
2181 if (i == length + 1)
2183 else if (i == length + 2)
2184 *(p - 1) = *(p - 2) = '=';
2185 /* ...and zero-terminate it. */
2189 /* Create the authentication header contents for the `Basic' scheme.
2190 This is done by encoding the string `USER:PASS' in base64 and
2191 prepending `HEADER: Basic ' to it. */
2193 basic_authentication_encode (const char *user, const char *passwd,
2196 char *t1, *t2, *res;
2197 int len1 = strlen (user) + 1 + strlen (passwd);
2198 int len2 = BASE64_LENGTH (len1);
2200 t1 = (char *)alloca (len1 + 1);
2201 sprintf (t1, "%s:%s", user, passwd);
2202 t2 = (char *)alloca (1 + len2);
2203 base64_encode (t1, t2, len1);
2204 res = (char *)xmalloc (len2 + 11 + strlen (header));
2205 sprintf (res, "%s: Basic %s\r\n", header, t2);
2211 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2212 of a field in such a header. If the field is the one specified by
2213 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2214 digest authorization code), extract its value in the (char*)
2215 variable pointed by RET. Returns negative on a malformed header,
2216 or number of bytes that have been parsed by this call. */
2218 extract_header_attr (const char *au, const char *attr_name, char **ret)
2220 const char *cp, *ep;
2224 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2226 cp += strlen (attr_name);
2229 cp += skip_lws (cp);
2234 cp += skip_lws (cp);
2239 for (ep = cp; *ep && *ep != '\"'; ep++)
2244 *ret = strdupdelim (cp, ep);
2251 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2252 an array of 16 bytes containing the hash keys, and BUF should be a
2253 buffer of 33 writable characters (32 for hex digits plus one for
2254 zero termination). */
2256 dump_hash (unsigned char *buf, const unsigned char *hash)
2260 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2262 *buf++ = XNUM_TO_digit (*hash >> 4);
2263 *buf++ = XNUM_TO_digit (*hash & 0xf);
2268 /* Take the line apart to find the challenge, and compose a digest
2269 authorization header. See RFC2069 section 2.1.2. */
2271 digest_authentication_encode (const char *au, const char *user,
2272 const char *passwd, const char *method,
2275 static char *realm, *opaque, *nonce;
2280 { "realm", &realm },
2281 { "opaque", &opaque },
2286 realm = opaque = nonce = NULL;
2288 au += 6; /* skip over `Digest' */
2293 au += skip_lws (au);
2294 for (i = 0; i < countof (options); i++)
2296 int skip = extract_header_attr (au, options[i].name,
2297 options[i].variable);
2301 xfree_null (opaque);
2311 if (i == countof (options))
2313 while (*au && *au != '=')
2317 au += skip_lws (au);
2321 while (*au && *au != '\"')
2328 while (*au && *au != ',')
2333 if (!realm || !nonce || !user || !passwd || !path || !method)
2336 xfree_null (opaque);
2341 /* Calculate the digest value. */
2343 ALLOCA_MD5_CONTEXT (ctx);
2344 unsigned char hash[MD5_HASHLEN];
2345 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2346 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2348 /* A1BUF = H(user ":" realm ":" password) */
2350 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2351 gen_md5_update ((unsigned char *)":", 1, ctx);
2352 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2353 gen_md5_update ((unsigned char *)":", 1, ctx);
2354 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2355 gen_md5_finish (ctx, hash);
2356 dump_hash (a1buf, hash);
2358 /* A2BUF = H(method ":" path) */
2360 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2361 gen_md5_update ((unsigned char *)":", 1, ctx);
2362 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2363 gen_md5_finish (ctx, hash);
2364 dump_hash (a2buf, hash);
2366 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2368 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2369 gen_md5_update ((unsigned char *)":", 1, ctx);
2370 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2371 gen_md5_update ((unsigned char *)":", 1, ctx);
2372 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2373 gen_md5_finish (ctx, hash);
2374 dump_hash (response_digest, hash);
2376 res = (char*) xmalloc (strlen (user)
2381 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2382 + (opaque ? strlen (opaque) : 0)
2384 sprintf (res, "Authorization: Digest \
2385 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2386 user, realm, nonce, path, response_digest);
2389 char *p = res + strlen (res);
2390 strcat (p, ", opaque=\"");
2394 strcat (res, "\r\n");
2398 #endif /* USE_DIGEST */
2401 #define BEGINS_WITH(line, string_constant) \
2402 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2403 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2404 || !line[sizeof (string_constant) - 1]))
2407 known_authentication_scheme_p (const char *au)
2409 return BEGINS_WITH (au, "Basic")
2410 || BEGINS_WITH (au, "Digest")
2411 || BEGINS_WITH (au, "NTLM");
2416 /* Create the HTTP authorization request header. When the
2417 `WWW-Authenticate' response header is seen, according to the
2418 authorization scheme specified in that header (`Basic' and `Digest'
2419 are supported by the current implementation), produce an
2420 appropriate HTTP authorization request header. */
2422 create_authorization_line (const char *au, const char *user,
2423 const char *passwd, const char *method,
2426 char *wwwauth = NULL;
2428 if (!strncasecmp (au, "Basic", 5))
2429 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2430 if (!strncasecmp (au, "NTLM", 4))
2431 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2433 else if (!strncasecmp (au, "Digest", 6))
2434 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2435 #endif /* USE_DIGEST */
2442 if (pc_last_host_ip)
2443 address_list_release (pc_last_host_ip);