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 should be possible to cache an arbitrary fixed
347 number of these connections. */
349 /* Whether a persistent connection is active. */
350 static int pconn_active;
353 /* The socket of the connection. */
356 /* Host and port of the currently active persistent connection. */
360 /* Whether a ssl handshake has occoured on this connection. */
364 /* Mark the persistent connection as invalid and free the resources it
365 uses. This is used by the CLOSE_* macros after they forcefully
366 close a registered persistent connection. */
369 invalidate_persistent (void)
371 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
373 xclose (pconn.socket);
378 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
379 persistent. This will enable someone to use the same connection
380 later. In the context of HTTP, this must be called only AFTER the
381 response has been received and the server has promised that the
382 connection will remain alive.
384 If a previous connection was persistent, it is closed. */
387 register_persistent (const char *host, int port, int fd, int ssl)
391 if (pconn.socket == fd)
393 /* The connection FD is already registered. */
398 /* The old persistent connection is still active; close it
399 first. This situation arises whenever a persistent
400 connection exists, but we then connect to a different
401 host, and try to register a persistent connection to that
403 invalidate_persistent ();
409 pconn.host = xstrdup (host);
413 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
416 /* Return non-zero if a persistent connection is available for
417 connecting to HOST:PORT. */
420 persistent_available_p (const char *host, int port, int ssl,
421 int *host_lookup_failed)
423 /* First, check whether a persistent connection is active at all. */
427 /* If we want SSL and the last connection wasn't or vice versa,
428 don't use it. Checking for host and port is not enough because
429 HTTP and HTTPS can apparently coexist on the same port. */
430 if (ssl != pconn.ssl)
433 /* If we're not connecting to the same port, we're not interested. */
434 if (port != pconn.port)
437 /* If the host is the same, we're in business. If not, there is
438 still hope -- read below. */
439 if (0 != strcasecmp (host, pconn.host))
441 /* This is somewhat evil, but works in practice: if the address
442 that pconn.socket is connected to is one of the IP addresses
443 HOST resolves to, we don't need to reconnect. #### Is it
444 correct to do this by default? */
447 struct address_list *al;
449 if (!socket_ip_address (pconn.socket, &ip, 0))
451 /* Can't get the peer's address -- something must be wrong
452 with the connection. */
453 invalidate_persistent ();
456 al = lookup_host (host, 0);
459 *host_lookup_failed = 1;
463 found = address_list_find (al, &ip);
464 address_list_release (al);
469 /* HOST resolves to an address pconn.sock is connected to -- no
470 need to reconnect. */
473 /* Finally, check whether the connection is still open. This is
474 important because most server implement a liberal (short) timeout
475 on persistent connections. Wget can of course always reconnect
476 if the connection doesn't work out, but it's nicer to know in
477 advance. This test is a logical followup of the first test, but
478 is "expensive" and therefore placed at the end of the list. */
480 if (!test_socket_open (pconn.socket))
482 /* Oops, the socket is no longer open. Now that we know that,
483 let's invalidate the persistent connection before returning
485 invalidate_persistent ();
492 /* The idea behind these two CLOSE macros is to distinguish between
493 two cases: one when the job we've been doing is finished, and we
494 want to close the connection and leave, and two when something is
495 seriously wrong and we're closing the connection as part of
498 In case of keep_alive, CLOSE_FINISH should leave the connection
499 open, while CLOSE_INVALIDATE should still close it.
501 Note that the semantics of the flag `keep_alive' is "this
502 connection *will* be reused (the server has promised not to close
503 the connection once we're done)", while the semantics of
504 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
505 active, registered connection". */
507 #define CLOSE_FINISH(fd) do { \
510 if (pconn_active && (fd) == pconn.socket) \
511 invalidate_persistent (); \
517 #define CLOSE_INVALIDATE(fd) do { \
518 if (pconn_active && (fd) == pconn.socket) \
519 invalidate_persistent (); \
526 long len; /* received length */
527 long contlen; /* expected length */
528 long restval; /* the restart value */
529 int res; /* the result of last read */
530 char *newloc; /* new location (redirection) */
531 char *remote_time; /* remote time-stamp string */
532 char *error; /* textual HTTP error */
533 int statcode; /* status code */
534 double dltime; /* time of the download in msecs */
535 int no_truncate; /* whether truncating the file is
537 const char *referer; /* value of the referer header. */
538 char **local_file; /* local file. */
542 free_hstat (struct http_stat *hs)
544 xfree_null (hs->newloc);
545 xfree_null (hs->remote_time);
546 xfree_null (hs->error);
548 /* Guard against being called twice. */
550 hs->remote_time = NULL;
554 static char *create_authorization_line PARAMS ((const char *, const char *,
555 const char *, const char *,
557 static char *basic_authentication_encode PARAMS ((const char *, const char *,
559 static int known_authentication_scheme_p PARAMS ((const char *));
561 time_t http_atotm PARAMS ((const char *));
563 #define BEGINS_WITH(line, string_constant) \
564 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
565 && (ISSPACE (line[sizeof (string_constant) - 1]) \
566 || !line[sizeof (string_constant) - 1]))
568 /* Retrieve a document through HTTP protocol. It recognizes status
569 code, and correctly handles redirections. It closes the network
570 socket. If it receives an error from the functions below it, it
571 will print it if there is enough information to do so (almost
572 always), returning the error to the caller (i.e. http_loop).
574 Various HTTP parameters are stored to hs. Although it parses the
575 response code correctly, it is not used in a sane way. The caller
578 If PROXY is non-NULL, the connection will be made to the proxy
579 server, and u->url will be requested. */
581 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
583 char *request, *type, *command, *full_path;
585 char *pragma_h, *referer, *useragent, *range, *wwwauth;
586 char *authenticate_h;
590 char *request_keep_alive;
591 int sock, hcount, all_length, statcode;
593 long contlen, contrange;
596 int auth_tried_already;
599 char *cookies = NULL;
601 /* Whether this connection will be kept alive after the HTTP request
605 /* Flags that detect the two ways of specifying HTTP keep-alive
607 int http_keep_alive_1, http_keep_alive_2;
609 /* Whether keep-alive should be inhibited. */
610 int inhibit_keep_alive;
612 /* Whether we need to print the host header with braces around host,
613 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
614 "Host: symbolic-name:1234". */
615 int squares_around_host = 0;
617 /* Headers sent when using POST. */
618 char *post_content_type, *post_content_length;
619 long post_data_size = 0;
621 int host_lookup_failed;
624 /* Initialize the SSL context. After the first run, this is a
628 case SSLERRCTXCREATE:
630 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
631 return SSLERRCTXCREATE;
633 /* try without certfile */
634 logprintf (LOG_NOTQUIET,
635 _("Failed to load certificates from %s\n"),
637 logprintf (LOG_NOTQUIET,
638 _("Trying without the specified certificate\n"));
641 logprintf (LOG_NOTQUIET,
642 _("Failed to get certificate key from %s\n"),
644 logprintf (LOG_NOTQUIET,
645 _("Trying without the specified certificate\n"));
650 #endif /* HAVE_SSL */
652 if (!(*dt & HEAD_ONLY))
653 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
654 know the local filename so we can save to it. */
655 assert (*hs->local_file != NULL);
658 auth_tried_already = 0;
660 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
663 /* We need to come back here when the initial attempt to retrieve
664 without authorization header fails. (Expected to happen at least
665 for the Digest authorization scheme.) */
668 http_keep_alive_1 = http_keep_alive_2 = 0;
670 post_content_type = NULL;
671 post_content_length = NULL;
673 /* Initialize certain elements of struct http_stat. */
678 hs->remote_time = NULL;
681 /* If we're using a proxy, we will be connecting to the proxy
683 conn = proxy ? proxy : u;
685 host_lookup_failed = 0;
687 /* First: establish the connection. */
688 if (inhibit_keep_alive
689 || !persistent_available_p (conn->host, conn->port,
691 u->scheme == SCHEME_HTTPS
695 , &host_lookup_failed))
697 /* In its current implementation, persistent_available_p will
698 look up conn->host in some cases. If that lookup failed, we
699 don't need to bother with connect_to_host. */
700 if (host_lookup_failed)
703 sock = connect_to_host (conn->host, conn->port);
707 return (retryable_socket_connect_error (errno)
708 ? CONERROR : CONIMPOSSIBLE);
711 if (conn->scheme == SCHEME_HTTPS)
713 if (!ssl_connect (sock))
715 logputs (LOG_VERBOSE, "\n");
716 logprintf (LOG_NOTQUIET,
717 _("Unable to establish SSL connection.\n"));
723 #endif /* HAVE_SSL */
727 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
728 conn->host, conn->port);
729 /* #### pc_last_fd should be accessed through an accessor
732 using_ssl = pconn.ssl;
733 DEBUGP (("Reusing fd %d.\n", sock));
738 else if (opt.post_file_name || opt.post_data)
746 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
747 sprintf (referer, "Referer: %s\r\n", hs->referer);
750 if (*dt & SEND_NOCACHE)
751 pragma_h = "Pragma: no-cache\r\n";
757 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
758 /* Gag me! Some servers (e.g. WebSitePro) have been known to
759 respond to the following `Range' format by generating a
760 multipart/x-byte-ranges MIME document! This MIME type was
761 present in an old draft of the byteranges specification.
762 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
763 only if multiple non-overlapping ranges are requested --
764 which Wget never does. */
765 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
770 STRDUP_ALLOCA (useragent, opt.useragent);
773 useragent = (char *)alloca (10 + strlen (version_string));
774 sprintf (useragent, "Wget/%s", version_string);
776 /* Construct the authentication, if userid is present. */
779 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
780 user = user ? user : opt.http_user;
781 passwd = passwd ? passwd : opt.http_passwd;
788 /* We have the username and the password, but haven't tried
789 any authorization yet. Let's see if the "Basic" method
790 works. If not, we'll come back here and construct a
791 proper authorization method with the right challenges.
793 If we didn't employ this kind of logic, every URL that
794 requires authorization would have to be processed twice,
795 which is very suboptimal and generates a bunch of false
796 "unauthorized" errors in the server log.
798 #### But this logic also has a serious problem when used
799 with stronger authentications: we *first* transmit the
800 username and the password in clear text, and *then*
801 attempt a stronger authentication scheme. That cannot be
802 right! We are only fortunate that almost everyone still
803 uses the `Basic' scheme anyway.
805 There should be an option to prevent this from happening,
806 for those who use strong authentication schemes and value
808 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
812 /* Use the full path, i.e. one that includes the leading
813 slash and the query string, but is independent of proxy
815 char *pth = url_full_path (u);
816 wwwauth = create_authorization_line (authenticate_h, user, passwd,
825 char *proxy_user, *proxy_passwd;
826 /* For normal username and password, URL components override
827 command-line/wgetrc parameters. With proxy authentication,
828 it's the reverse, because proxy URLs are normally the
829 "permanent" ones, so command-line args should take
831 if (opt.proxy_user && opt.proxy_passwd)
833 proxy_user = opt.proxy_user;
834 proxy_passwd = opt.proxy_passwd;
838 proxy_user = proxy->user;
839 proxy_passwd = proxy->passwd;
841 /* #### This does not appear right. Can't the proxy request,
842 say, `Digest' authentication? */
843 if (proxy_user && proxy_passwd)
844 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
845 "Proxy-Authorization");
848 /* String of the form :PORT. Used only for non-standard ports. */
850 if (u->port != scheme_default_port (u->scheme))
852 port_maybe = (char *)alloca (numdigit (u->port) + 2);
853 sprintf (port_maybe, ":%d", u->port);
856 if (!inhibit_keep_alive)
857 request_keep_alive = "Connection: Keep-Alive\r\n";
859 request_keep_alive = NULL;
862 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
864 u->scheme == SCHEME_HTTPS
870 if (opt.post_data || opt.post_file_name)
872 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
874 post_data_size = strlen (opt.post_data);
877 post_data_size = file_size (opt.post_file_name);
878 if (post_data_size == -1)
880 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
885 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
886 sprintf (post_content_length,
887 "Content-Length: %ld\r\n", post_data_size);
891 full_path = xstrdup (u->url);
893 /* Use the full path, i.e. one that includes the leading slash and
894 the query string. E.g. if u->path is "foo/bar" and u->query is
895 "param=value", full_path will be "/foo/bar?param=value". */
896 full_path = url_full_path (u);
898 if (strchr (u->host, ':'))
899 squares_around_host = 1;
901 /* Allocate the memory for the request. */
902 request = (char *)alloca (strlen (command)
906 + (port_maybe ? strlen (port_maybe) : 0)
907 + strlen (HTTP_ACCEPT)
908 + (request_keep_alive
909 ? strlen (request_keep_alive) : 0)
910 + (referer ? strlen (referer) : 0)
911 + (cookies ? strlen (cookies) : 0)
912 + (wwwauth ? strlen (wwwauth) : 0)
913 + (proxyauth ? strlen (proxyauth) : 0)
914 + (range ? strlen (range) : 0)
917 ? strlen (post_content_type) : 0)
918 + (post_content_length
919 ? strlen (post_content_length) : 0)
920 + (opt.user_header ? strlen (opt.user_header) : 0)
922 /* Construct the request. */
928 %s%s%s%s%s%s%s%s%s%s\r\n",
931 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
932 port_maybe ? port_maybe : "",
934 request_keep_alive ? request_keep_alive : "",
935 referer ? referer : "",
936 cookies ? cookies : "",
937 wwwauth ? wwwauth : "",
938 proxyauth ? proxyauth : "",
941 post_content_type ? post_content_type : "",
942 post_content_length ? post_content_length : "",
943 opt.user_header ? opt.user_header : "");
944 DEBUGP (("\n---request begin---\n%s", request));
946 /* Free the temporary memory. */
947 xfree_null (wwwauth);
948 xfree_null (proxyauth);
949 xfree_null (cookies);
952 /* Send the request to server. */
953 write_error = xwrite (sock, request, strlen (request), -1);
955 if (write_error >= 0)
959 DEBUGP (("[POST data: %s]\n", opt.post_data));
960 write_error = xwrite (sock, opt.post_data, post_data_size, -1);
962 else if (opt.post_file_name && post_data_size != 0)
963 write_error = post_file (sock, opt.post_file_name, post_data_size);
965 DEBUGP (("---request end---\n"));
969 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
971 CLOSE_INVALIDATE (sock);
974 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
975 proxy ? "Proxy" : "HTTP");
976 contlen = contrange = -1;
981 /* Before reading anything, initialize the rbuf. */
982 rbuf_initialize (&rbuf, sock);
986 DEBUGP (("\n---response begin---\n"));
988 /* Header-fetching loop. */
996 /* Get the header. */
997 status = header_get (&rbuf, &hdr,
998 /* Disallow continuations for status line. */
999 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
1001 /* Check for errors. */
1002 if (status == HG_EOF && *hdr)
1004 /* This used to be an unconditional error, but that was
1005 somewhat controversial, because of a large number of
1006 broken CGI's that happily "forget" to send the second EOL
1007 before closing the connection of a HEAD request.
1009 So, the deal is to check whether the header is empty
1010 (*hdr is zero if it is); if yes, it means that the
1011 previous header was fully retrieved, and that -- most
1012 probably -- the request is complete. "...be liberal in
1013 what you accept." Oh boy. */
1014 logputs (LOG_VERBOSE, "\n");
1015 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
1018 xfree_null (all_headers);
1019 CLOSE_INVALIDATE (sock);
1022 else if (status == HG_ERROR)
1024 logputs (LOG_VERBOSE, "\n");
1025 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1029 xfree_null (all_headers);
1030 CLOSE_INVALIDATE (sock);
1034 /* If the headers are to be saved to a file later, save them to
1036 if (opt.save_headers)
1038 int lh = strlen (hdr);
1039 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
1040 memcpy (all_headers + all_length, hdr, lh);
1042 all_headers[all_length++] = '\n';
1043 all_headers[all_length] = '\0';
1046 /* Check for status line. */
1050 /* Parse the first line of server response. */
1051 statcode = parse_http_status_line (hdr, &error);
1052 hs->statcode = statcode;
1053 /* Store the descriptive response. */
1054 if (statcode == -1) /* malformed response */
1056 /* A common reason for "malformed response" error is the
1057 case when no data was actually received. Handle this
1060 hs->error = xstrdup (_("No data received"));
1062 hs->error = xstrdup (_("Malformed status line"));
1067 hs->error = xstrdup (_("(no description)"));
1069 hs->error = xstrdup (error);
1071 if ((statcode != -1)
1077 if (opt.server_response)
1078 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1080 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1086 /* Exit on empty header. */
1093 /* Print the header if requested. */
1094 if (opt.server_response && hcount != 1)
1095 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1097 /* Try getting content-length. */
1098 if (contlen == -1 && !opt.ignore_length)
1099 if (header_process (hdr, "Content-Length", header_extract_number,
1102 /* Try getting content-type. */
1104 if (header_process (hdr, "Content-Type", http_process_type, &type))
1106 /* Try getting location. */
1108 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1110 /* Try getting last-modified. */
1111 if (!hs->remote_time)
1112 if (header_process (hdr, "Last-Modified", header_strdup,
1115 /* Try getting cookies. */
1117 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1119 /* Try getting www-authentication. */
1120 if (!authenticate_h)
1121 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1124 /* Check for accept-ranges header. If it contains the word
1125 `none', disable the ranges. */
1126 if (*dt & ACCEPTRANGES)
1129 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1132 *dt &= ~ACCEPTRANGES;
1136 /* Try getting content-range. */
1137 if (contrange == -1)
1139 struct http_process_range_closure closure;
1140 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1142 contrange = closure.first_byte_pos;
1146 /* Check for keep-alive related responses. */
1147 if (!inhibit_keep_alive)
1149 /* Check for the `Keep-Alive' header. */
1150 if (!http_keep_alive_1)
1152 if (header_process (hdr, "Keep-Alive", header_exists,
1153 &http_keep_alive_1))
1156 /* Check for `Connection: Keep-Alive'. */
1157 if (!http_keep_alive_2)
1159 if (header_process (hdr, "Connection", http_process_connection,
1160 &http_keep_alive_2))
1167 DEBUGP (("---response end---\n"));
1169 logputs (LOG_VERBOSE, "\n");
1172 && (http_keep_alive_1 || http_keep_alive_2))
1174 assert (inhibit_keep_alive == 0);
1178 /* The server has promised that it will not close the connection
1179 when we're done. This means that we can register it. */
1180 register_persistent (conn->host, conn->port, sock, using_ssl);
1182 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1185 /* Authorization is required. */
1189 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1190 might be more bytes in the body. */
1191 if (auth_tried_already)
1193 /* If we have tried it already, then there is not point
1196 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1197 xfree (authenticate_h);
1200 else if (!known_authentication_scheme_p (authenticate_h))
1202 xfree (authenticate_h);
1203 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1206 else if (BEGINS_WITH (authenticate_h, "Basic"))
1208 /* The authentication scheme is basic, the one we try by
1209 default, and it failed. There's no sense in trying
1215 auth_tried_already = 1;
1219 /* We do not need this anymore. */
1222 xfree (authenticate_h);
1223 authenticate_h = NULL;
1226 /* 20x responses are counted among successful by default. */
1227 if (H_20X (statcode))
1230 /* Return if redirected. */
1231 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1233 /* RFC2068 says that in case of the 300 (multiple choices)
1234 response, the server can output a preferred URL through
1235 `Location' header; otherwise, the request should be treated
1236 like GET. So, if the location is set, it will be a
1237 redirection; otherwise, just proceed normally. */
1238 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1242 logprintf (LOG_VERBOSE,
1243 _("Location: %s%s\n"),
1244 hs->newloc ? hs->newloc : _("unspecified"),
1245 hs->newloc ? _(" [following]") : "");
1246 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1247 might be more bytes in the body. */
1249 xfree_null (all_headers);
1254 /* If content-type is not given, assume text/html. This is because
1255 of the multitude of broken CGI's that "forget" to generate the
1258 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1259 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1264 if (opt.html_extension && (*dt & TEXTHTML))
1265 /* -E / --html-extension / html_extension = on was specified, and this is a
1266 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1267 already the file's suffix, tack on ".html". */
1269 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1271 if (last_period_in_local_filename == NULL
1272 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1273 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1275 size_t local_filename_len = strlen(*hs->local_file);
1277 *hs->local_file = xrealloc(*hs->local_file,
1278 local_filename_len + sizeof(".html"));
1279 strcpy(*hs->local_file + local_filename_len, ".html");
1281 *dt |= ADDED_HTML_EXTENSION;
1285 if (contrange == -1)
1287 /* We did not get a content-range header. This means that the
1288 server did not honor our `Range' request. Normally, this
1289 means we should reset hs->restval and continue normally. */
1291 /* However, if `-c' is used, we need to be a bit more careful:
1293 1. If `-c' is specified and the file already existed when
1294 Wget was started, it would be a bad idea for us to start
1295 downloading it from scratch, effectively truncating it. I
1296 believe this cannot happen unless `-c' was specified.
1298 2. If `-c' is used on a file that is already fully
1299 downloaded, we're requesting bytes after the end of file,
1300 which can result in server not honoring `Range'. If this is
1301 the case, `Content-Length' will be equal to the length of the
1303 if (opt.always_rest)
1305 /* Check for condition #2. */
1306 if (hs->restval > 0 /* restart was requested. */
1307 && contlen != -1 /* we got content-length. */
1308 && hs->restval >= contlen /* file fully downloaded
1312 logputs (LOG_VERBOSE, _("\
1313 \n The file is already fully retrieved; nothing to do.\n\n"));
1314 /* In case the caller inspects. */
1317 /* Mark as successfully retrieved. */
1320 xfree_null (all_headers);
1321 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1322 might be more bytes in the body. */
1323 return RETRUNNEEDED;
1326 /* Check for condition #1. */
1327 if (hs->no_truncate)
1329 logprintf (LOG_NOTQUIET,
1332 Continued download failed on this file, which conflicts with `-c'.\n\
1333 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1335 xfree_null (all_headers);
1336 CLOSE_INVALIDATE (sock);
1337 return CONTNOTSUPPORTED;
1345 else if (contrange != hs->restval ||
1346 (H_PARTIAL (statcode) && contrange == -1))
1348 /* This means the whole request was somehow misunderstood by the
1349 server. Bail out. */
1351 xfree_null (all_headers);
1352 CLOSE_INVALIDATE (sock);
1359 contlen += contrange;
1361 contrange = -1; /* If conent-length was not sent,
1362 content-range will be ignored. */
1364 hs->contlen = contlen;
1368 if ((*dt & RETROKF) && !opt.server_response)
1370 /* No need to print this output if the body won't be
1371 downloaded at all, or if the original server response is
1373 logputs (LOG_VERBOSE, _("Length: "));
1376 logputs (LOG_VERBOSE, legible (contlen));
1377 if (contrange != -1)
1378 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1379 legible (contlen - contrange));
1382 logputs (LOG_VERBOSE,
1383 opt.ignore_length ? _("ignored") : _("unspecified"));
1385 logprintf (LOG_VERBOSE, " [%s]\n", type);
1387 logputs (LOG_VERBOSE, "\n");
1391 type = NULL; /* We don't need it any more. */
1393 /* Return if we have no intention of further downloading. */
1394 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1396 /* In case the caller cares to look... */
1400 xfree_null (all_headers);
1401 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1402 might be more bytes in the body. */
1403 return RETRFINISHED;
1406 /* Open the local file. */
1409 mkalldirs (*hs->local_file);
1411 rotate_backups (*hs->local_file);
1412 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1415 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1416 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1417 might be more bytes in the body. */
1418 xfree_null (all_headers);
1424 extern int global_download_count;
1426 /* To ensure that repeated "from scratch" downloads work for -O
1427 files, we rewind the file pointer, unless restval is
1428 non-zero. (This works only when -O is used on regular files,
1429 but it's still a valuable feature.)
1431 However, this loses when more than one URL is specified on
1432 the command line the second rewinds eradicates the contents
1433 of the first download. Thus we disable the above trick for
1434 all the downloads except the very first one.
1436 #### A possible solution to this would be to remember the
1437 file position in the output document and to seek to that
1438 position, instead of rewinding.
1440 We don't truncate stdout, since that breaks
1441 "wget -O - [...] >> foo".
1443 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1445 /* This will silently fail for streams that don't correspond
1446 to regular files, but that's OK. */
1448 /* ftruncate is needed because opt.dfp is opened in append
1449 mode if opt.always_rest is set. */
1450 ftruncate (fileno (fp), 0);
1455 /* #### This confuses the code that checks for file size. There
1456 should be some overhead information. */
1457 if (opt.save_headers)
1458 fwrite (all_headers, 1, all_length, fp);
1460 /* Get the contents of the document. */
1461 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1462 (contlen != -1 ? contlen : 0),
1463 &rbuf, keep_alive, &hs->dltime);
1466 CLOSE_FINISH (sock);
1468 CLOSE_INVALIDATE (sock);
1471 /* Close or flush the file. We have to be careful to check for
1472 error here. Checking the result of fwrite() is not enough --
1473 errors could go unnoticed! */
1476 flush_res = fclose (fp);
1478 flush_res = fflush (fp);
1479 if (flush_res == EOF)
1482 xfree_null (all_headers);
1485 return RETRFINISHED;
1488 /* The genuine HTTP loop! This is the part where the retrieval is
1489 retried, and retried, and retried, and... */
1491 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1492 int *dt, struct url *proxy)
1495 int use_ts, got_head = 0; /* time-stamping info */
1496 char *filename_plus_orig_suffix;
1497 char *local_filename = NULL;
1498 char *tms, *locf, *tmrate;
1500 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1501 long local_size = 0; /* the size of the local file */
1502 size_t filename_len;
1503 struct http_stat hstat; /* HTTP status */
1507 /* This used to be done in main(), but it's a better idea to do it
1508 here so that we don't go through the hoops if we're just using
1512 if (!wget_cookie_jar)
1513 wget_cookie_jar = cookie_jar_new ();
1514 if (opt.cookies_input && !cookies_loaded_p)
1516 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1517 cookies_loaded_p = 1;
1523 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1524 has_wildcards_p because it would also warn on `?', and we know that
1525 shows up in CGI paths a *lot*. */
1526 if (strchr (u->url, '*'))
1527 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1529 /* Determine the local filename. */
1530 if (local_file && *local_file)
1531 hstat.local_file = local_file;
1532 else if (local_file)
1534 *local_file = url_file_name (u);
1535 hstat.local_file = local_file;
1539 dummy = url_file_name (u);
1540 hstat.local_file = &dummy;
1543 if (!opt.output_document)
1544 locf = *hstat.local_file;
1546 locf = opt.output_document;
1548 hstat.referer = referer;
1550 filename_len = strlen (*hstat.local_file);
1551 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1553 if (opt.noclobber && file_exists_p (*hstat.local_file))
1555 /* If opt.noclobber is turned on and file already exists, do not
1556 retrieve the file */
1557 logprintf (LOG_VERBOSE, _("\
1558 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1559 /* If the file is there, we suppose it's retrieved OK. */
1562 /* #### Bogusness alert. */
1563 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1564 if (has_html_suffix_p (*hstat.local_file))
1572 if (opt.timestamping)
1574 int local_dot_orig_file_exists = 0;
1576 if (opt.backup_converted)
1577 /* If -K is specified, we'll act on the assumption that it was specified
1578 last time these files were downloaded as well, and instead of just
1579 comparing local file X against server file X, we'll compare local
1580 file X.orig (if extant, else X) against server file X. If -K
1581 _wasn't_ specified last time, or the server contains files called
1582 *.orig, -N will be back to not operating correctly with -k. */
1584 /* Would a single s[n]printf() call be faster? --dan
1586 Definitely not. sprintf() is horribly slow. It's a
1587 different question whether the difference between the two
1588 affects a program. Usually I'd say "no", but at one
1589 point I profiled Wget, and found that a measurable and
1590 non-negligible amount of time was lost calling sprintf()
1591 in url.c. Replacing sprintf with inline calls to
1592 strcpy() and long_to_string() made a difference.
1594 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1595 memcpy (filename_plus_orig_suffix + filename_len,
1596 ".orig", sizeof (".orig"));
1598 /* Try to stat() the .orig file. */
1599 if (stat (filename_plus_orig_suffix, &st) == 0)
1601 local_dot_orig_file_exists = 1;
1602 local_filename = filename_plus_orig_suffix;
1606 if (!local_dot_orig_file_exists)
1607 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1608 if (stat (*hstat.local_file, &st) == 0)
1609 local_filename = *hstat.local_file;
1611 if (local_filename != NULL)
1612 /* There was a local file, so we'll check later to see if the version
1613 the server has is the same version we already have, allowing us to
1619 /* Modification time granularity is 2 seconds for Windows, so
1620 increase local time by 1 second for later comparison. */
1623 local_size = st.st_size;
1627 /* Reset the counter. */
1629 *dt = 0 | ACCEPTRANGES;
1633 /* Increment the pass counter. */
1635 sleep_between_retrievals (count);
1636 /* Get the current time string. */
1637 tms = time_str (NULL);
1638 /* Print fetch message, if opt.verbose. */
1641 char *hurl = url_string (u, 1);
1645 sprintf (tmp, _("(try:%2d)"), count);
1646 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1647 tms, hurl, tmp, locf);
1649 ws_changetitle (hurl, 1);
1654 /* Default document type is empty. However, if spider mode is
1655 on or time-stamping is employed, HEAD_ONLY commands is
1656 encoded within *dt. */
1657 if (opt.spider || (use_ts && !got_head))
1661 /* Assume no restarting. */
1663 /* Decide whether or not to restart. */
1664 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1665 /* #### this calls access() and then stat(); could be optimized. */
1666 && file_exists_p (locf))
1667 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1668 hstat.restval = st.st_size;
1670 /* In `-c' is used and the file is existing and non-empty,
1671 refuse to truncate it if the server doesn't support continued
1673 hstat.no_truncate = 0;
1674 if (opt.always_rest && hstat.restval)
1675 hstat.no_truncate = 1;
1677 /* Decide whether to send the no-cache directive. We send it in
1679 a) we're using a proxy, and we're past our first retrieval.
1680 Some proxies are notorious for caching incomplete data, so
1681 we require a fresh get.
1682 b) caching is explicitly inhibited. */
1683 if ((proxy && count > 1) /* a */
1684 || !opt.allow_cache /* b */
1686 *dt |= SEND_NOCACHE;
1688 *dt &= ~SEND_NOCACHE;
1690 /* Try fetching the document, or at least its head. */
1691 err = gethttp (u, &hstat, dt, proxy);
1693 /* It's unfortunate that wget determines the local filename before finding
1694 out the Content-Type of the file. Barring a major restructuring of the
1695 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1696 *hstat.local_file to tack on ".html". */
1697 if (!opt.output_document)
1698 locf = *hstat.local_file;
1700 locf = opt.output_document;
1703 tms = time_str (NULL);
1704 /* Get the new location (with or without the redirection). */
1706 *newloc = xstrdup (hstat.newloc);
1709 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1710 case CONERROR: case READERR: case WRITEFAILED:
1712 /* Non-fatal errors continue executing the loop, which will
1713 bring them to "while" statement at the end, to judge
1714 whether the number of tries was exceeded. */
1715 free_hstat (&hstat);
1716 printwhat (count, opt.ntry);
1719 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1720 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1721 /* Fatal errors just return from the function. */
1722 free_hstat (&hstat);
1726 case FWRITEERR: case FOPENERR:
1727 /* Another fatal error. */
1728 logputs (LOG_VERBOSE, "\n");
1729 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1730 *hstat.local_file, strerror (errno));
1731 free_hstat (&hstat);
1736 /* Another fatal error. */
1737 logputs (LOG_VERBOSE, "\n");
1738 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1739 free_hstat (&hstat);
1744 /* Return the new location to the caller. */
1747 logprintf (LOG_NOTQUIET,
1748 _("ERROR: Redirection (%d) without location.\n"),
1750 free_hstat (&hstat);
1754 free_hstat (&hstat);
1759 /* The file was already fully retrieved. */
1760 free_hstat (&hstat);
1765 /* Deal with you later. */
1768 /* All possibilities should have been exhausted. */
1771 if (!(*dt & RETROKF))
1775 /* #### Ugly ugly ugly! */
1776 char *hurl = url_string (u, 1);
1777 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1780 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1781 tms, hstat.statcode, hstat.error);
1782 logputs (LOG_VERBOSE, "\n");
1783 free_hstat (&hstat);
1788 /* Did we get the time-stamp? */
1791 if (opt.timestamping && !hstat.remote_time)
1793 logputs (LOG_NOTQUIET, _("\
1794 Last-modified header missing -- time-stamps turned off.\n"));
1796 else if (hstat.remote_time)
1798 /* Convert the date-string into struct tm. */
1799 tmr = http_atotm (hstat.remote_time);
1800 if (tmr == (time_t) (-1))
1801 logputs (LOG_VERBOSE, _("\
1802 Last-modified header invalid -- time-stamp ignored.\n"));
1806 /* The time-stamping section. */
1811 use_ts = 0; /* no more time-stamping */
1812 count = 0; /* the retrieve count for HEAD is
1814 if (hstat.remote_time && tmr != (time_t) (-1))
1816 /* Now time-stamping can be used validly. Time-stamping
1817 means that if the sizes of the local and remote file
1818 match, and local file is newer than the remote file,
1819 it will not be retrieved. Otherwise, the normal
1820 download procedure is resumed. */
1822 (hstat.contlen == -1 || local_size == hstat.contlen))
1824 logprintf (LOG_VERBOSE, _("\
1825 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1827 free_hstat (&hstat);
1831 else if (tml >= tmr)
1832 logprintf (LOG_VERBOSE, _("\
1833 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1835 logputs (LOG_VERBOSE,
1836 _("Remote file is newer, retrieving.\n"));
1838 free_hstat (&hstat);
1841 if ((tmr != (time_t) (-1))
1843 && ((hstat.len == hstat.contlen) ||
1844 ((hstat.res == 0) &&
1845 ((hstat.contlen == -1) ||
1846 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1848 /* #### This code repeats in http.c and ftp.c. Move it to a
1850 const char *fl = NULL;
1851 if (opt.output_document)
1853 if (opt.od_known_regular)
1854 fl = opt.output_document;
1857 fl = *hstat.local_file;
1861 /* End of time-stamping section. */
1865 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1870 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1872 if (hstat.len == hstat.contlen)
1876 logprintf (LOG_VERBOSE,
1877 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1878 tms, tmrate, locf, hstat.len, hstat.contlen);
1879 logprintf (LOG_NONVERBOSE,
1880 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1881 tms, u->url, hstat.len, hstat.contlen, 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.res == 0) /* No read error */
1898 if (hstat.contlen == -1) /* We don't know how much we were supposed
1899 to get, so assume we succeeded. */
1903 logprintf (LOG_VERBOSE,
1904 _("%s (%s) - `%s' saved [%ld]\n\n"),
1905 tms, tmrate, locf, hstat.len);
1906 logprintf (LOG_NONVERBOSE,
1907 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1908 tms, u->url, hstat.len, locf, count);
1911 total_downloaded_bytes += hstat.len;
1913 /* Remember that we downloaded the file for later ".orig" code. */
1914 if (*dt & ADDED_HTML_EXTENSION)
1915 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1917 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1919 free_hstat (&hstat);
1923 else if (hstat.len < hstat.contlen) /* meaning we lost the
1924 connection too soon */
1926 logprintf (LOG_VERBOSE,
1927 _("%s (%s) - Connection closed at byte %ld. "),
1928 tms, tmrate, hstat.len);
1929 printwhat (count, opt.ntry);
1930 free_hstat (&hstat);
1933 else if (!opt.kill_longer) /* meaning we got more than expected */
1935 logprintf (LOG_VERBOSE,
1936 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1937 tms, tmrate, locf, hstat.len, hstat.contlen);
1938 logprintf (LOG_NONVERBOSE,
1939 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1940 tms, u->url, hstat.len, hstat.contlen, locf, count);
1942 total_downloaded_bytes += hstat.len;
1944 /* Remember that we downloaded the file for later ".orig" code. */
1945 if (*dt & ADDED_HTML_EXTENSION)
1946 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1948 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1950 free_hstat (&hstat);
1954 else /* the same, but not accepted */
1956 logprintf (LOG_VERBOSE,
1957 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1958 tms, tmrate, hstat.len, hstat.contlen);
1959 printwhat (count, opt.ntry);
1960 free_hstat (&hstat);
1964 else /* now hstat.res can only be -1 */
1966 if (hstat.contlen == -1)
1968 logprintf (LOG_VERBOSE,
1969 _("%s (%s) - Read error at byte %ld (%s)."),
1970 tms, tmrate, hstat.len, strerror (errno));
1971 printwhat (count, opt.ntry);
1972 free_hstat (&hstat);
1975 else /* hstat.res == -1 and contlen is given */
1977 logprintf (LOG_VERBOSE,
1978 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1979 tms, tmrate, hstat.len, hstat.contlen,
1981 printwhat (count, opt.ntry);
1982 free_hstat (&hstat);
1989 while (!opt.ntry || (count < opt.ntry));
1993 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1994 than local timezone.
1996 mktime is similar but assumes struct tm, also known as the
1997 "broken-down" form of time, is in local time zone. mktime_from_utc
1998 uses mktime to make the conversion understanding that an offset
1999 will be introduced by the local time assumption.
2001 mktime_from_utc then measures the introduced offset by applying
2002 gmtime to the initial result and applying mktime to the resulting
2003 "broken-down" form. The difference between the two mktime results
2004 is the measured offset which is then subtracted from the initial
2005 mktime result to yield a calendar time which is the value returned.
2007 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2008 consistent offset (the non DST offset) since tm and tm+o might be
2009 on opposite sides of a DST change.
2011 Some implementations of mktime return -1 for the nonexistent
2012 localtime hour at the beginning of DST. In this event, use
2013 mktime(tm - 1hr) + 3600.
2017 gmtime(t+o) --> tm+o
2018 mktime(tm+o) --> t+2o
2019 t+o - (t+2o - t+o) = t
2021 Note that glibc contains a function of the same purpose named
2022 `timegm' (reverse of gmtime). But obviously, it is not universally
2023 available, and unfortunately it is not straightforwardly
2024 extractable for use here. Perhaps configure should detect timegm
2025 and use it where available.
2027 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2028 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2029 Further improved by Roger with assistance from Edward J. Sabol
2030 based on input by Jamie Zawinski. */
2033 mktime_from_utc (struct tm *t)
2044 return -1; /* can't deal with output from strptime */
2055 return -1; /* can't deal with output from gmtime */
2058 return (tl - (tb - tl));
2061 /* Check whether the result of strptime() indicates success.
2062 strptime() returns the pointer to how far it got to in the string.
2063 The processing has been successful if the string is at `GMT' or
2064 `+X', or at the end of the string.
2066 In extended regexp parlance, the function returns 1 if P matches
2067 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2068 can return) is considered a failure and 0 is returned. */
2070 check_end (const char *p)
2074 while (ISSPACE (*p))
2077 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2078 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2084 /* Convert the textual specification of time in TIME_STRING to the
2085 number of seconds since the Epoch.
2087 TIME_STRING can be in any of the three formats RFC2068 allows the
2088 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2089 Timezones are ignored, and should be GMT.
2091 Return the computed time_t representation, or -1 if the conversion
2094 This function uses strptime with various string formats for parsing
2095 TIME_STRING. This results in a parser that is not as lenient in
2096 interpreting TIME_STRING as I would like it to be. Being based on
2097 strptime, it always allows shortened months, one-digit days, etc.,
2098 but due to the multitude of formats in which time can be
2099 represented, an ideal HTTP time parser would be even more
2100 forgiving. It should completely ignore things like week days and
2101 concentrate only on the various forms of representing years,
2102 months, days, hours, minutes, and seconds. For example, it would
2103 be nice if it accepted ISO 8601 out of the box.
2105 I've investigated free and PD code for this purpose, but none was
2106 usable. getdate was big and unwieldy, and had potential copyright
2107 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2108 distributed with phttpd, is excellent, but we cannot use it because
2109 it is not assigned to the FSF. So I stuck it with strptime. */
2112 http_atotm (const char *time_string)
2114 /* NOTE: Solaris strptime man page claims that %n and %t match white
2115 space, but that's not universally available. Instead, we simply
2116 use ` ' to mean "skip all WS", which works under all strptime
2117 implementations I've tested. */
2119 static const char *time_formats[] = {
2120 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2121 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2122 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2123 (google.com uses this for their cookies.) */
2124 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2130 /* According to Roger Beeman, we need to initialize tm_isdst, since
2131 strptime won't do it. */
2134 /* Note that under foreign locales Solaris strptime() fails to
2135 recognize English dates, which renders this function useless. We
2136 solve this by being careful not to affect LC_TIME when
2137 initializing locale.
2139 Another solution would be to temporarily set locale to C, invoke
2140 strptime(), and restore it back. This is slow and dirty,
2141 however, and locale support other than LC_MESSAGES can mess other
2142 things, so I rather chose to stick with just setting LC_MESSAGES.
2144 GNU strptime does not have this problem because it recognizes
2145 both international and local dates. */
2147 for (i = 0; i < countof (time_formats); i++)
2148 if (check_end (strptime (time_string, time_formats[i], &t)))
2149 return mktime_from_utc (&t);
2151 /* All formats have failed. */
2155 /* Authorization support: We support two authorization schemes:
2157 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2159 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2160 consisting of answering to the server's challenge with the proper
2163 /* How many bytes it will take to store LEN bytes in base64. */
2164 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2166 /* Encode the string S of length LENGTH to base64 format and place it
2167 to STORE. STORE will be 0-terminated, and must point to a writable
2168 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2170 base64_encode (const char *s, char *store, int length)
2172 /* Conversion table. */
2173 static char tbl[64] = {
2174 'A','B','C','D','E','F','G','H',
2175 'I','J','K','L','M','N','O','P',
2176 'Q','R','S','T','U','V','W','X',
2177 'Y','Z','a','b','c','d','e','f',
2178 'g','h','i','j','k','l','m','n',
2179 'o','p','q','r','s','t','u','v',
2180 'w','x','y','z','0','1','2','3',
2181 '4','5','6','7','8','9','+','/'
2184 unsigned char *p = (unsigned char *)store;
2186 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2187 for (i = 0; i < length; i += 3)
2189 *p++ = tbl[s[0] >> 2];
2190 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2191 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2192 *p++ = tbl[s[2] & 0x3f];
2195 /* Pad the result if necessary... */
2196 if (i == length + 1)
2198 else if (i == length + 2)
2199 *(p - 1) = *(p - 2) = '=';
2200 /* ...and zero-terminate it. */
2204 /* Create the authentication header contents for the `Basic' scheme.
2205 This is done by encoding the string `USER:PASS' in base64 and
2206 prepending `HEADER: Basic ' to it. */
2208 basic_authentication_encode (const char *user, const char *passwd,
2211 char *t1, *t2, *res;
2212 int len1 = strlen (user) + 1 + strlen (passwd);
2213 int len2 = BASE64_LENGTH (len1);
2215 t1 = (char *)alloca (len1 + 1);
2216 sprintf (t1, "%s:%s", user, passwd);
2217 t2 = (char *)alloca (1 + len2);
2218 base64_encode (t1, t2, len1);
2219 res = (char *)xmalloc (len2 + 11 + strlen (header));
2220 sprintf (res, "%s: Basic %s\r\n", header, t2);
2226 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2227 of a field in such a header. If the field is the one specified by
2228 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2229 digest authorization code), extract its value in the (char*)
2230 variable pointed by RET. Returns negative on a malformed header,
2231 or number of bytes that have been parsed by this call. */
2233 extract_header_attr (const char *au, const char *attr_name, char **ret)
2235 const char *cp, *ep;
2239 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2241 cp += strlen (attr_name);
2244 cp += skip_lws (cp);
2249 cp += skip_lws (cp);
2254 for (ep = cp; *ep && *ep != '\"'; ep++)
2259 *ret = strdupdelim (cp, ep);
2266 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2267 an array of 16 bytes containing the hash keys, and BUF should be a
2268 buffer of 33 writable characters (32 for hex digits plus one for
2269 zero termination). */
2271 dump_hash (unsigned char *buf, const unsigned char *hash)
2275 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2277 *buf++ = XNUM_TO_digit (*hash >> 4);
2278 *buf++ = XNUM_TO_digit (*hash & 0xf);
2283 /* Take the line apart to find the challenge, and compose a digest
2284 authorization header. See RFC2069 section 2.1.2. */
2286 digest_authentication_encode (const char *au, const char *user,
2287 const char *passwd, const char *method,
2290 static char *realm, *opaque, *nonce;
2295 { "realm", &realm },
2296 { "opaque", &opaque },
2301 realm = opaque = nonce = NULL;
2303 au += 6; /* skip over `Digest' */
2308 au += skip_lws (au);
2309 for (i = 0; i < countof (options); i++)
2311 int skip = extract_header_attr (au, options[i].name,
2312 options[i].variable);
2316 xfree_null (opaque);
2326 if (i == countof (options))
2328 while (*au && *au != '=')
2332 au += skip_lws (au);
2336 while (*au && *au != '\"')
2343 while (*au && *au != ',')
2348 if (!realm || !nonce || !user || !passwd || !path || !method)
2351 xfree_null (opaque);
2356 /* Calculate the digest value. */
2358 ALLOCA_MD5_CONTEXT (ctx);
2359 unsigned char hash[MD5_HASHLEN];
2360 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2361 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2363 /* A1BUF = H(user ":" realm ":" password) */
2365 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2366 gen_md5_update ((unsigned char *)":", 1, ctx);
2367 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2368 gen_md5_update ((unsigned char *)":", 1, ctx);
2369 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2370 gen_md5_finish (ctx, hash);
2371 dump_hash (a1buf, hash);
2373 /* A2BUF = H(method ":" path) */
2375 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2376 gen_md5_update ((unsigned char *)":", 1, ctx);
2377 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2378 gen_md5_finish (ctx, hash);
2379 dump_hash (a2buf, hash);
2381 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2383 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2384 gen_md5_update ((unsigned char *)":", 1, ctx);
2385 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2386 gen_md5_update ((unsigned char *)":", 1, ctx);
2387 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2388 gen_md5_finish (ctx, hash);
2389 dump_hash (response_digest, hash);
2391 res = (char*) xmalloc (strlen (user)
2396 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2397 + (opaque ? strlen (opaque) : 0)
2399 sprintf (res, "Authorization: Digest \
2400 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2401 user, realm, nonce, path, response_digest);
2404 char *p = res + strlen (res);
2405 strcat (p, ", opaque=\"");
2409 strcat (res, "\r\n");
2413 #endif /* USE_DIGEST */
2416 #define BEGINS_WITH(line, string_constant) \
2417 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2418 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2419 || !line[sizeof (string_constant) - 1]))
2422 known_authentication_scheme_p (const char *au)
2424 return BEGINS_WITH (au, "Basic")
2425 || BEGINS_WITH (au, "Digest")
2426 || BEGINS_WITH (au, "NTLM");
2431 /* Create the HTTP authorization request header. When the
2432 `WWW-Authenticate' response header is seen, according to the
2433 authorization scheme specified in that header (`Basic' and `Digest'
2434 are supported by the current implementation), produce an
2435 appropriate HTTP authorization request header. */
2437 create_authorization_line (const char *au, const char *user,
2438 const char *passwd, const char *method,
2441 char *wwwauth = NULL;
2443 if (!strncasecmp (au, "Basic", 5))
2444 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2445 if (!strncasecmp (au, "NTLM", 4))
2446 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2448 else if (!strncasecmp (au, "Digest", 6))
2449 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2450 #endif /* USE_DIGEST */