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>
68 # include "gen_sslfunc.h"
76 extern char *version_string;
77 extern LARGE_INT total_downloaded_bytes;
80 static int cookies_loaded_p;
81 struct cookie_jar *wget_cookie_jar;
83 #define TEXTHTML_S "text/html"
84 #define TEXTXHTML_S "application/xhtml+xml"
85 #define HTTP_ACCEPT "*/*"
87 /* Some status code validation macros: */
88 #define H_20X(x) (((x) >= 200) && ((x) < 300))
89 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
90 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
91 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
92 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
94 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
96 #define HTTP_STATUS_OK 200
97 #define HTTP_STATUS_CREATED 201
98 #define HTTP_STATUS_ACCEPTED 202
99 #define HTTP_STATUS_NO_CONTENT 204
100 #define HTTP_STATUS_PARTIAL_CONTENTS 206
102 /* Redirection 3xx. */
103 #define HTTP_STATUS_MULTIPLE_CHOICES 300
104 #define HTTP_STATUS_MOVED_PERMANENTLY 301
105 #define HTTP_STATUS_MOVED_TEMPORARILY 302
106 #define HTTP_STATUS_NOT_MODIFIED 304
107 #define HTTP_STATUS_TEMPORARY_REDIRECT 307
109 /* Client error 4xx. */
110 #define HTTP_STATUS_BAD_REQUEST 400
111 #define HTTP_STATUS_UNAUTHORIZED 401
112 #define HTTP_STATUS_FORBIDDEN 403
113 #define HTTP_STATUS_NOT_FOUND 404
115 /* Server errors 5xx. */
116 #define HTTP_STATUS_INTERNAL 500
117 #define HTTP_STATUS_NOT_IMPLEMENTED 501
118 #define HTTP_STATUS_BAD_GATEWAY 502
119 #define HTTP_STATUS_UNAVAILABLE 503
122 /* Parse the HTTP status line, which is of format:
124 HTTP-Version SP Status-Code SP Reason-Phrase
126 The function returns the status-code, or -1 if the status line is
127 malformed. The pointer to reason-phrase is returned in RP. */
129 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
131 /* (the variables must not be named `major' and `minor', because
132 that breaks compilation with SunOS4 cc.) */
133 int mjr, mnr, statcode;
136 *reason_phrase_ptr = NULL;
138 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
139 major version, and Y is minor version. */
140 if (strncmp (line, "HTTP/", 5) != 0)
144 /* Calculate major HTTP version. */
146 for (mjr = 0; ISDIGIT (*line); line++)
147 mjr = 10 * mjr + (*line - '0');
148 if (*line != '.' || p == line)
152 /* Calculate minor HTTP version. */
154 for (mnr = 0; ISDIGIT (*line); line++)
155 mnr = 10 * mnr + (*line - '0');
156 if (*line != ' ' || p == line)
158 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
159 minor version can be safely ignored. */
164 /* Calculate status code. */
165 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
167 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
169 /* Set up the reason phrase pointer. */
171 /* RFC2068 requires SPC here, but we allow the string to finish
172 here, in case no reason-phrase is present. */
176 *reason_phrase_ptr = line;
181 *reason_phrase_ptr = line + 1;
186 #define WMIN(x, y) ((x) > (y) ? (y) : (x))
188 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
189 PROMISED_SIZE bytes are sent over the wire -- if the file is
190 longer, read only that much; if the file is shorter, report an error. */
193 post_file (int sock, const char *file_name, long promised_size)
195 static char chunk[8192];
200 DEBUGP (("[writing POST file %s ... ", file_name));
202 fp = fopen (file_name, "rb");
205 while (!feof (fp) && written < promised_size)
208 int length = fread (chunk, 1, sizeof (chunk), fp);
211 towrite = WMIN (promised_size - written, length);
212 write_error = fd_write (sock, chunk, towrite, -1);
222 /* If we've written less than was promised, report a (probably
223 nonsensical) error rather than break the promise. */
224 if (written < promised_size)
230 assert (written == promised_size);
231 DEBUGP (("done]\n"));
236 next_header (const char *h)
238 const char *end = NULL;
242 p = strchr (p, '\n');
247 while (*p == ' ' || *p == '\t');
252 /* Skip LWS (linear white space), if present. Returns number of
253 characters to skip. */
255 skip_lws (const char *string)
257 const char *p = string;
259 while (*p == ' ' || *p == '\t' || *p == '\r' || *p == '\n')
264 /* Check whether HEADER begins with NAME and, if yes, skip the `:' and
265 the whitespace, and call PROCFUN with the arguments of HEADER's
266 contents (after the `:' and space) and ARG. Otherwise, return 0. */
268 header_process (const char *header, const char *name,
269 int (*procfun) (const char *, void *),
272 /* Check whether HEADER matches NAME. */
273 while (*name && (TOLOWER (*name) == TOLOWER (*header)))
275 if (*name || *header++ != ':')
278 header += skip_lws (header);
280 return ((*procfun) (header, arg));
283 /* Helper functions for use with header_process(). */
285 /* Extract a long integer from HEADER and store it to CLOSURE. If an
286 error is encountered, return 0, else 1. */
288 header_extract_number (const char *header, void *closure)
290 const char *p = header;
293 for (result = 0; ISDIGIT (*p); p++)
294 result = 10 * result + (*p - '0');
296 /* Failure if no number present. */
300 /* Skip trailing whitespace. */
303 /* Indicate failure if trailing garbage is present. */
307 *(long *)closure = result;
311 /* Strdup HEADER, and place the pointer to CLOSURE. */
313 header_strdup (const char *header, void *closure)
315 *(char **)closure = xstrdup (header);
319 /* Write the value 1 into the integer pointed to by CLOSURE. */
321 header_exists (const char *header, void *closure)
327 /* Functions to be used as arguments to header_process(): */
329 struct http_process_range_closure {
335 /* Parse the `Content-Range' header and extract the information it
336 contains. Returns 1 if successful, -1 otherwise. */
338 http_process_range (const char *hdr, void *arg)
340 struct http_process_range_closure *closure
341 = (struct http_process_range_closure *)arg;
344 /* Certain versions of Nutscape proxy server send out
345 `Content-Length' without "bytes" specifier, which is a breach of
346 RFC2068 (as well as the HTTP/1.1 draft which was current at the
347 time). But hell, I must support it... */
348 if (!strncasecmp (hdr, "bytes", 5))
351 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
355 hdr += skip_lws (hdr);
361 for (num = 0; ISDIGIT (*hdr); hdr++)
362 num = 10 * num + (*hdr - '0');
363 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
365 closure->first_byte_pos = num;
367 for (num = 0; ISDIGIT (*hdr); hdr++)
368 num = 10 * num + (*hdr - '0');
369 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
371 closure->last_byte_pos = num;
373 for (num = 0; ISDIGIT (*hdr); hdr++)
374 num = 10 * num + (*hdr - '0');
375 closure->entity_length = num;
379 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
380 Used for `Accept-Ranges'. */
382 http_process_none (const char *hdr, void *arg)
384 int *where = (int *)arg;
386 if (strstr (hdr, "none"))
393 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
395 http_process_type (const char *hdr, void *arg)
397 char **result = (char **)arg;
398 /* Locate P on `;' or the terminating zero, whichever comes first. */
399 const char *p = strchr (hdr, ';');
401 p = hdr + strlen (hdr);
402 while (p > hdr && ISSPACE (*(p - 1)))
404 *result = strdupdelim (hdr, p);
408 /* Check whether the `Connection' header is set to "keep-alive". */
410 http_process_connection (const char *hdr, void *arg)
412 int *flag = (int *)arg;
413 if (!strcasecmp (hdr, "Keep-Alive"))
418 /* Commit the cookie to the cookie jar. */
421 http_process_set_cookie (const char *hdr, void *arg)
423 struct url *u = (struct url *)arg;
425 /* The jar should have been created by now. */
426 assert (wget_cookie_jar != NULL);
428 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path, hdr);
433 /* Persistent connections. Currently, we cache the most recently used
434 connection as persistent, provided that the HTTP server agrees to
435 make it such. The persistence data is stored in the variables
436 below. Ideally, it should be possible to cache an arbitrary fixed
437 number of these connections. */
439 /* Whether a persistent connection is active. */
440 static int pconn_active;
443 /* The socket of the connection. */
446 /* Host and port of the currently active persistent connection. */
450 /* Whether a ssl handshake has occoured on this connection. */
454 /* Mark the persistent connection as invalid and free the resources it
455 uses. This is used by the CLOSE_* macros after they forcefully
456 close a registered persistent connection. */
459 invalidate_persistent (void)
461 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
463 fd_close (pconn.socket);
468 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
469 persistent. This will enable someone to use the same connection
470 later. In the context of HTTP, this must be called only AFTER the
471 response has been received and the server has promised that the
472 connection will remain alive.
474 If a previous connection was persistent, it is closed. */
477 register_persistent (const char *host, int port, int fd, int ssl)
481 if (pconn.socket == fd)
483 /* The connection FD is already registered. */
488 /* The old persistent connection is still active; close it
489 first. This situation arises whenever a persistent
490 connection exists, but we then connect to a different
491 host, and try to register a persistent connection to that
493 invalidate_persistent ();
499 pconn.host = xstrdup (host);
503 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
506 /* Return non-zero if a persistent connection is available for
507 connecting to HOST:PORT. */
510 persistent_available_p (const char *host, int port, int ssl,
511 int *host_lookup_failed)
513 /* First, check whether a persistent connection is active at all. */
517 /* If we want SSL and the last connection wasn't or vice versa,
518 don't use it. Checking for host and port is not enough because
519 HTTP and HTTPS can apparently coexist on the same port. */
520 if (ssl != pconn.ssl)
523 /* If we're not connecting to the same port, we're not interested. */
524 if (port != pconn.port)
527 /* If the host is the same, we're in business. If not, there is
528 still hope -- read below. */
529 if (0 != strcasecmp (host, pconn.host))
531 /* If pconn.socket is already talking to HOST, we needn't
532 reconnect. This happens often when both sites are virtual
533 hosts distinguished only by name and served by the same
534 network interface, and hence the same web server (possibly
535 set up by the ISP and serving many different web sites).
536 This admittedly non-standard optimization does not contradict
537 HTTP and works well with popular server software. */
541 struct address_list *al;
544 /* Don't try to talk to two different SSL sites over the same
545 secure connection! (Besides, it's not clear if name-based
546 virtual hosting is even possible with SSL.) */
549 /* If pconn.socket's peer is one of the IP addresses HOST
550 resolves to, pconn.socket is for all intents and purposes
551 already talking to HOST. */
553 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
555 /* Can't get the peer's address -- something must be very
556 wrong with the connection. */
557 invalidate_persistent ();
560 al = lookup_host (host, 0);
563 *host_lookup_failed = 1;
567 found = address_list_contains (al, &ip);
568 address_list_release (al);
573 /* The persistent connection's peer address was found among the
574 addresses HOST resolved to; therefore, pconn.sock is in fact
575 already talking to HOST -- no need to reconnect. */
578 /* Finally, check whether the connection is still open. This is
579 important because most server implement a liberal (short) timeout
580 on persistent connections. Wget can of course always reconnect
581 if the connection doesn't work out, but it's nicer to know in
582 advance. This test is a logical followup of the first test, but
583 is "expensive" and therefore placed at the end of the list. */
585 if (!test_socket_open (pconn.socket))
587 /* Oops, the socket is no longer open. Now that we know that,
588 let's invalidate the persistent connection before returning
590 invalidate_persistent ();
597 /* The idea behind these two CLOSE macros is to distinguish between
598 two cases: one when the job we've been doing is finished, and we
599 want to close the connection and leave, and two when something is
600 seriously wrong and we're closing the connection as part of
603 In case of keep_alive, CLOSE_FINISH should leave the connection
604 open, while CLOSE_INVALIDATE should still close it.
606 Note that the semantics of the flag `keep_alive' is "this
607 connection *will* be reused (the server has promised not to close
608 the connection once we're done)", while the semantics of
609 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
610 active, registered connection". */
612 #define CLOSE_FINISH(fd) do { \
615 if (pconn_active && (fd) == pconn.socket) \
616 invalidate_persistent (); \
622 #define CLOSE_INVALIDATE(fd) do { \
623 if (pconn_active && (fd) == pconn.socket) \
624 invalidate_persistent (); \
631 long len; /* received length */
632 long contlen; /* expected length */
633 long restval; /* the restart value */
634 int res; /* the result of last read */
635 char *newloc; /* new location (redirection) */
636 char *remote_time; /* remote time-stamp string */
637 char *error; /* textual HTTP error */
638 int statcode; /* status code */
639 double dltime; /* time of the download in msecs */
640 int no_truncate; /* whether truncating the file is
642 const char *referer; /* value of the referer header. */
643 char **local_file; /* local file. */
647 free_hstat (struct http_stat *hs)
649 xfree_null (hs->newloc);
650 xfree_null (hs->remote_time);
651 xfree_null (hs->error);
653 /* Guard against being called twice. */
655 hs->remote_time = NULL;
659 static char *create_authorization_line PARAMS ((const char *, const char *,
660 const char *, const char *,
662 static char *basic_authentication_encode PARAMS ((const char *, const char *,
664 static int known_authentication_scheme_p PARAMS ((const char *));
666 time_t http_atotm PARAMS ((const char *));
668 #define BEGINS_WITH(line, string_constant) \
669 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
670 && (ISSPACE (line[sizeof (string_constant) - 1]) \
671 || !line[sizeof (string_constant) - 1]))
673 /* Retrieve a document through HTTP protocol. It recognizes status
674 code, and correctly handles redirections. It closes the network
675 socket. If it receives an error from the functions below it, it
676 will print it if there is enough information to do so (almost
677 always), returning the error to the caller (i.e. http_loop).
679 Various HTTP parameters are stored to hs.
681 If PROXY is non-NULL, the connection will be made to the proxy
682 server, and u->url will be requested. */
684 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
686 char *request, *type, *command, *full_path;
688 char *pragma_h, *referer, *useragent, *range, *wwwauth;
689 char *authenticate_h;
692 char *request_keep_alive;
693 int sock, hcount, statcode;
695 long contlen, contrange;
698 int auth_tried_already;
700 char *cookies = NULL;
703 const char *hdr_beg, *hdr_end;
705 /* Whether this connection will be kept alive after the HTTP request
709 /* Flags that detect the two ways of specifying HTTP keep-alive
711 int http_keep_alive_1, http_keep_alive_2;
713 /* Whether keep-alive should be inhibited. */
714 int inhibit_keep_alive;
716 /* Whether we need to print the host header with braces around host,
717 e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the usual
718 "Host: symbolic-name:1234". */
719 int squares_around_host = 0;
721 /* Headers sent when using POST. */
722 char *post_content_type, *post_content_length;
723 long post_data_size = 0;
725 int host_lookup_failed;
728 /* Initialize the SSL context. After the first run, this is a
732 case SSLERRCTXCREATE:
734 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
735 return SSLERRCTXCREATE;
737 /* try without certfile */
738 logprintf (LOG_NOTQUIET,
739 _("Failed to load certificates from %s\n"),
741 logprintf (LOG_NOTQUIET,
742 _("Trying without the specified certificate\n"));
745 logprintf (LOG_NOTQUIET,
746 _("Failed to get certificate key from %s\n"),
748 logprintf (LOG_NOTQUIET,
749 _("Trying without the specified certificate\n"));
754 #endif /* HAVE_SSL */
756 if (!(*dt & HEAD_ONLY))
757 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
758 know the local filename so we can save to it. */
759 assert (*hs->local_file != NULL);
762 auth_tried_already = 0;
764 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
767 /* We need to come back here when the initial attempt to retrieve
768 without authorization header fails. (Expected to happen at least
769 for the Digest authorization scheme.) */
772 http_keep_alive_1 = http_keep_alive_2 = 0;
774 post_content_type = NULL;
775 post_content_length = NULL;
777 /* Initialize certain elements of struct http_stat. */
782 hs->remote_time = NULL;
785 /* If we're using a proxy, we will be connecting to the proxy
787 conn = proxy ? proxy : u;
789 host_lookup_failed = 0;
791 /* First: establish the connection. */
792 if (inhibit_keep_alive
793 || !persistent_available_p (conn->host, conn->port,
795 u->scheme == SCHEME_HTTPS
799 , &host_lookup_failed))
801 /* In its current implementation, persistent_available_p will
802 look up conn->host in some cases. If that lookup failed, we
803 don't need to bother with connect_to_host. */
804 if (host_lookup_failed)
807 sock = connect_to_host (conn->host, conn->port);
811 return (retryable_socket_connect_error (errno)
812 ? CONERROR : CONIMPOSSIBLE);
815 if (conn->scheme == SCHEME_HTTPS)
817 if (!ssl_connect (sock))
819 logputs (LOG_VERBOSE, "\n");
820 logprintf (LOG_NOTQUIET,
821 _("Unable to establish SSL connection.\n"));
827 #endif /* HAVE_SSL */
831 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
832 pconn.host, pconn.port);
834 using_ssl = pconn.ssl;
835 DEBUGP (("Reusing fd %d.\n", sock));
840 else if (opt.post_file_name || opt.post_data)
848 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
849 sprintf (referer, "Referer: %s\r\n", hs->referer);
852 if (*dt & SEND_NOCACHE)
853 pragma_h = "Pragma: no-cache\r\n";
859 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
860 /* Gag me! Some servers (e.g. WebSitePro) have been known to
861 respond to the following `Range' format by generating a
862 multipart/x-byte-ranges MIME document! This MIME type was
863 present in an old draft of the byteranges specification.
864 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
865 only if multiple non-overlapping ranges are requested --
866 which Wget never does. */
867 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
872 STRDUP_ALLOCA (useragent, opt.useragent);
875 useragent = (char *)alloca (10 + strlen (version_string));
876 sprintf (useragent, "Wget/%s", version_string);
878 /* Construct the authentication, if userid is present. */
881 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
882 user = user ? user : opt.http_user;
883 passwd = passwd ? passwd : opt.http_passwd;
890 /* We have the username and the password, but haven't tried
891 any authorization yet. Let's see if the "Basic" method
892 works. If not, we'll come back here and construct a
893 proper authorization method with the right challenges.
895 If we didn't employ this kind of logic, every URL that
896 requires authorization would have to be processed twice,
897 which is very suboptimal and generates a bunch of false
898 "unauthorized" errors in the server log.
900 #### But this logic also has a serious problem when used
901 with stronger authentications: we *first* transmit the
902 username and the password in clear text, and *then*
903 attempt a stronger authentication scheme. That cannot be
904 right! We are only fortunate that almost everyone still
905 uses the `Basic' scheme anyway.
907 There should be an option to prevent this from happening,
908 for those who use strong authentication schemes and value
910 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
914 /* Use the full path, i.e. one that includes the leading
915 slash and the query string, but is independent of proxy
917 char *pth = url_full_path (u);
918 wwwauth = create_authorization_line (authenticate_h, user, passwd,
927 char *proxy_user, *proxy_passwd;
928 /* For normal username and password, URL components override
929 command-line/wgetrc parameters. With proxy authentication,
930 it's the reverse, because proxy URLs are normally the
931 "permanent" ones, so command-line args should take
933 if (opt.proxy_user && opt.proxy_passwd)
935 proxy_user = opt.proxy_user;
936 proxy_passwd = opt.proxy_passwd;
940 proxy_user = proxy->user;
941 proxy_passwd = proxy->passwd;
943 /* #### This does not appear right. Can't the proxy request,
944 say, `Digest' authentication? */
945 if (proxy_user && proxy_passwd)
946 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
947 "Proxy-Authorization");
950 /* String of the form :PORT. Used only for non-standard ports. */
952 if (u->port != scheme_default_port (u->scheme))
954 port_maybe = (char *)alloca (numdigit (u->port) + 2);
955 sprintf (port_maybe, ":%d", u->port);
958 if (!inhibit_keep_alive)
959 request_keep_alive = "Connection: Keep-Alive\r\n";
961 request_keep_alive = NULL;
964 cookies = cookie_header (wget_cookie_jar, u->host, u->port, u->path,
966 u->scheme == SCHEME_HTTPS
972 if (opt.post_data || opt.post_file_name)
974 post_content_type = "Content-Type: application/x-www-form-urlencoded\r\n";
976 post_data_size = strlen (opt.post_data);
979 post_data_size = file_size (opt.post_file_name);
980 if (post_data_size == -1)
982 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
987 post_content_length = xmalloc (16 + numdigit (post_data_size) + 2 + 1);
988 sprintf (post_content_length,
989 "Content-Length: %ld\r\n", post_data_size);
993 full_path = xstrdup (u->url);
995 /* Use the full path, i.e. one that includes the leading slash and
996 the query string. E.g. if u->path is "foo/bar" and u->query is
997 "param=value", full_path will be "/foo/bar?param=value". */
998 full_path = url_full_path (u);
1000 if (strchr (u->host, ':'))
1001 squares_around_host = 1;
1003 /* Allocate the memory for the request. */
1004 request = (char *)alloca (strlen (command)
1005 + strlen (full_path)
1006 + strlen (useragent)
1008 + (port_maybe ? strlen (port_maybe) : 0)
1009 + strlen (HTTP_ACCEPT)
1010 + (request_keep_alive
1011 ? strlen (request_keep_alive) : 0)
1012 + (referer ? strlen (referer) : 0)
1013 + (cookies ? strlen (cookies) : 0)
1014 + (wwwauth ? strlen (wwwauth) : 0)
1015 + (proxyauth ? strlen (proxyauth) : 0)
1016 + (range ? strlen (range) : 0)
1018 + (post_content_type
1019 ? strlen (post_content_type) : 0)
1020 + (post_content_length
1021 ? strlen (post_content_length) : 0)
1022 + (opt.user_header ? strlen (opt.user_header) : 0)
1024 /* Construct the request. */
1025 sprintf (request, "\
1030 %s%s%s%s%s%s%s%s%s%s\r\n",
1033 squares_around_host ? "[" : "", u->host, squares_around_host ? "]" : "",
1034 port_maybe ? port_maybe : "",
1036 request_keep_alive ? request_keep_alive : "",
1037 referer ? referer : "",
1038 cookies ? cookies : "",
1039 wwwauth ? wwwauth : "",
1040 proxyauth ? proxyauth : "",
1043 post_content_type ? post_content_type : "",
1044 post_content_length ? post_content_length : "",
1045 opt.user_header ? opt.user_header : "");
1046 DEBUGP (("\n---request begin---\n%s", request));
1048 /* Free the temporary memory. */
1049 xfree_null (wwwauth);
1050 xfree_null (proxyauth);
1051 xfree_null (cookies);
1054 /* Send the request to server. */
1055 write_error = fd_write (sock, request, strlen (request), -1);
1057 if (write_error >= 0)
1061 DEBUGP (("[POST data: %s]\n", opt.post_data));
1062 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1064 else if (opt.post_file_name && post_data_size != 0)
1065 write_error = post_file (sock, opt.post_file_name, post_data_size);
1067 DEBUGP (("---request end---\n"));
1069 if (write_error < 0)
1071 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1073 CLOSE_INVALIDATE (sock);
1076 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1077 proxy ? "Proxy" : "HTTP");
1078 contlen = contrange = -1;
1083 DEBUGP (("\n---response begin---\n"));
1085 head = fd_read_head (sock);
1088 logputs (LOG_VERBOSE, "\n");
1091 logputs (LOG_NOTQUIET, _("No data received.\n"));
1092 CLOSE_INVALIDATE (sock);
1097 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1099 CLOSE_INVALIDATE (sock);
1104 /* Loop through the headers and process them. */
1107 for (hdr_beg = head;
1108 (hdr_end = next_header (hdr_beg));
1111 char *hdr = strdupdelim (hdr_beg, hdr_end);
1113 char *tmp = hdr + strlen (hdr);
1114 if (tmp > hdr && tmp[-1] == '\n')
1116 if (tmp > hdr && tmp[-1] == '\r')
1121 /* Check for status line. */
1125 /* Parse the first line of server response. */
1126 statcode = parse_http_status_line (hdr, &error);
1127 hs->statcode = statcode;
1128 /* Store the descriptive response. */
1129 if (statcode == -1) /* malformed response */
1131 /* A common reason for "malformed response" error is the
1132 case when no data was actually received. Handle this
1135 hs->error = xstrdup (_("No data received"));
1137 hs->error = xstrdup (_("Malformed status line"));
1142 hs->error = xstrdup (_("(no description)"));
1144 hs->error = xstrdup (error);
1146 if ((statcode != -1)
1152 if (opt.server_response)
1153 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1155 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
1161 /* Exit on empty header. */
1168 /* Print the header if requested. */
1169 if (opt.server_response && hcount != 1)
1170 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
1172 /* Try getting content-length. */
1173 if (contlen == -1 && !opt.ignore_length)
1174 if (header_process (hdr, "Content-Length", header_extract_number,
1177 /* Try getting content-type. */
1179 if (header_process (hdr, "Content-Type", http_process_type, &type))
1181 /* Try getting location. */
1183 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1185 /* Try getting last-modified. */
1186 if (!hs->remote_time)
1187 if (header_process (hdr, "Last-Modified", header_strdup,
1190 /* Try getting cookies. */
1192 if (header_process (hdr, "Set-Cookie", http_process_set_cookie, u))
1194 /* Try getting www-authentication. */
1195 if (!authenticate_h)
1196 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1199 /* Check for accept-ranges header. If it contains the word
1200 `none', disable the ranges. */
1201 if (*dt & ACCEPTRANGES)
1204 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1207 *dt &= ~ACCEPTRANGES;
1211 /* Try getting content-range. */
1212 if (contrange == -1)
1214 struct http_process_range_closure closure;
1215 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1217 contrange = closure.first_byte_pos;
1221 /* Check for keep-alive related responses. */
1222 if (!inhibit_keep_alive)
1224 /* Check for the `Keep-Alive' header. */
1225 if (!http_keep_alive_1)
1227 if (header_process (hdr, "Keep-Alive", header_exists,
1228 &http_keep_alive_1))
1231 /* Check for `Connection: Keep-Alive'. */
1232 if (!http_keep_alive_2)
1234 if (header_process (hdr, "Connection", http_process_connection,
1235 &http_keep_alive_2))
1242 DEBUGP (("---response end---\n"));
1244 logputs (LOG_VERBOSE, "\n");
1247 && (http_keep_alive_1 || http_keep_alive_2))
1249 assert (inhibit_keep_alive == 0);
1253 /* The server has promised that it will not close the connection
1254 when we're done. This means that we can register it. */
1255 register_persistent (conn->host, conn->port, sock, using_ssl);
1257 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1260 /* Authorization is required. */
1264 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1265 might be more bytes in the body. */
1266 if (auth_tried_already)
1268 /* If we have tried it already, then there is not point
1271 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1272 xfree (authenticate_h);
1275 else if (!known_authentication_scheme_p (authenticate_h))
1277 xfree (authenticate_h);
1278 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1281 else if (BEGINS_WITH (authenticate_h, "Basic"))
1283 /* The authentication scheme is basic, the one we try by
1284 default, and it failed. There's no sense in trying
1290 auth_tried_already = 1;
1294 /* We do not need this anymore. */
1297 xfree (authenticate_h);
1298 authenticate_h = NULL;
1301 /* 20x responses are counted among successful by default. */
1302 if (H_20X (statcode))
1305 /* Return if redirected. */
1306 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1308 /* RFC2068 says that in case of the 300 (multiple choices)
1309 response, the server can output a preferred URL through
1310 `Location' header; otherwise, the request should be treated
1311 like GET. So, if the location is set, it will be a
1312 redirection; otherwise, just proceed normally. */
1313 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1317 logprintf (LOG_VERBOSE,
1318 _("Location: %s%s\n"),
1319 hs->newloc ? hs->newloc : _("unspecified"),
1320 hs->newloc ? _(" [following]") : "");
1321 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1322 might be more bytes in the body. */
1328 /* If content-type is not given, assume text/html. This is because
1329 of the multitude of broken CGI's that "forget" to generate the
1332 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1333 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1338 if (opt.html_extension && (*dt & TEXTHTML))
1339 /* -E / --html-extension / html_extension = on was specified, and this is a
1340 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1341 already the file's suffix, tack on ".html". */
1343 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1345 if (last_period_in_local_filename == NULL
1346 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1347 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1349 size_t local_filename_len = strlen(*hs->local_file);
1351 *hs->local_file = xrealloc(*hs->local_file,
1352 local_filename_len + sizeof(".html"));
1353 strcpy(*hs->local_file + local_filename_len, ".html");
1355 *dt |= ADDED_HTML_EXTENSION;
1359 if (contrange == -1)
1361 /* We did not get a content-range header. This means that the
1362 server did not honor our `Range' request. Normally, this
1363 means we should reset hs->restval and continue normally. */
1365 /* However, if `-c' is used, we need to be a bit more careful:
1367 1. If `-c' is specified and the file already existed when
1368 Wget was started, it would be a bad idea for us to start
1369 downloading it from scratch, effectively truncating it. I
1370 believe this cannot happen unless `-c' was specified.
1372 2. If `-c' is used on a file that is already fully
1373 downloaded, we're requesting bytes after the end of file,
1374 which can result in server not honoring `Range'. If this is
1375 the case, `Content-Length' will be equal to the length of the
1377 if (opt.always_rest)
1379 /* Check for condition #2. */
1380 if (hs->restval > 0 /* restart was requested. */
1381 && contlen != -1 /* we got content-length. */
1382 && hs->restval >= contlen /* file fully downloaded
1386 logputs (LOG_VERBOSE, _("\
1387 \n The file is already fully retrieved; nothing to do.\n\n"));
1388 /* In case the caller inspects. */
1391 /* Mark as successfully retrieved. */
1394 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1395 might be more bytes in the body. */
1396 return RETRUNNEEDED;
1399 /* Check for condition #1. */
1400 if (hs->no_truncate)
1402 logprintf (LOG_NOTQUIET,
1405 Continued download failed on this file, which conflicts with `-c'.\n\
1406 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1408 CLOSE_INVALIDATE (sock);
1409 return CONTNOTSUPPORTED;
1417 else if (contrange != hs->restval ||
1418 (H_PARTIAL (statcode) && contrange == -1))
1420 /* This means the whole request was somehow misunderstood by the
1421 server. Bail out. */
1423 CLOSE_INVALIDATE (sock);
1430 contlen += contrange;
1432 contrange = -1; /* If conent-length was not sent,
1433 content-range will be ignored. */
1435 hs->contlen = contlen;
1439 if ((*dt & RETROKF) && !opt.server_response)
1441 /* No need to print this output if the body won't be
1442 downloaded at all, or if the original server response is
1444 logputs (LOG_VERBOSE, _("Length: "));
1447 logputs (LOG_VERBOSE, legible (contlen));
1448 if (contrange != -1)
1449 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1450 legible (contlen - contrange));
1453 logputs (LOG_VERBOSE,
1454 opt.ignore_length ? _("ignored") : _("unspecified"));
1456 logprintf (LOG_VERBOSE, " [%s]\n", type);
1458 logputs (LOG_VERBOSE, "\n");
1462 type = NULL; /* We don't need it any more. */
1464 /* Return if we have no intention of further downloading. */
1465 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1467 /* In case the caller cares to look... */
1471 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1472 might be more bytes in the body. */
1473 return RETRFINISHED;
1476 /* Open the local file. */
1479 mkalldirs (*hs->local_file);
1481 rotate_backups (*hs->local_file);
1482 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1485 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1486 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1487 might be more bytes in the body. */
1493 extern int global_download_count;
1495 /* To ensure that repeated "from scratch" downloads work for -O
1496 files, we rewind the file pointer, unless restval is
1497 non-zero. (This works only when -O is used on regular files,
1498 but it's still a valuable feature.)
1500 However, this loses when more than one URL is specified on
1501 the command line the second rewinds eradicates the contents
1502 of the first download. Thus we disable the above trick for
1503 all the downloads except the very first one.
1505 #### A possible solution to this would be to remember the
1506 file position in the output document and to seek to that
1507 position, instead of rewinding.
1509 We don't truncate stdout, since that breaks
1510 "wget -O - [...] >> foo".
1512 if (!hs->restval && global_download_count == 0 && opt.dfp != stdout)
1514 /* This will silently fail for streams that don't correspond
1515 to regular files, but that's OK. */
1517 /* ftruncate is needed because opt.dfp is opened in append
1518 mode if opt.always_rest is set. */
1519 ftruncate (fileno (fp), 0);
1524 /* #### This confuses the code that checks for file size. There
1525 should be some overhead information. */
1526 if (opt.save_headers)
1527 fwrite (head, 1, strlen (head), fp);
1529 /* Get the contents of the document. */
1530 hs->res = fd_read_body (sock, fp, &hs->len, hs->restval,
1531 (contlen != -1 ? contlen : 0),
1532 keep_alive, &hs->dltime);
1535 CLOSE_FINISH (sock);
1537 CLOSE_INVALIDATE (sock);
1540 /* Close or flush the file. We have to be careful to check for
1541 error here. Checking the result of fwrite() is not enough --
1542 errors could go unnoticed! */
1545 flush_res = fclose (fp);
1547 flush_res = fflush (fp);
1548 if (flush_res == EOF)
1553 return RETRFINISHED;
1556 /* The genuine HTTP loop! This is the part where the retrieval is
1557 retried, and retried, and retried, and... */
1559 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1560 int *dt, struct url *proxy)
1563 int use_ts, got_head = 0; /* time-stamping info */
1564 char *filename_plus_orig_suffix;
1565 char *local_filename = NULL;
1566 char *tms, *locf, *tmrate;
1568 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1569 long local_size = 0; /* the size of the local file */
1570 size_t filename_len;
1571 struct http_stat hstat; /* HTTP status */
1575 /* This used to be done in main(), but it's a better idea to do it
1576 here so that we don't go through the hoops if we're just using
1580 if (!wget_cookie_jar)
1581 wget_cookie_jar = cookie_jar_new ();
1582 if (opt.cookies_input && !cookies_loaded_p)
1584 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1585 cookies_loaded_p = 1;
1591 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1592 has_wildcards_p because it would also warn on `?', and we know that
1593 shows up in CGI paths a *lot*. */
1594 if (strchr (u->url, '*'))
1595 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1597 /* Determine the local filename. */
1598 if (local_file && *local_file)
1599 hstat.local_file = local_file;
1600 else if (local_file)
1602 *local_file = url_file_name (u);
1603 hstat.local_file = local_file;
1607 dummy = url_file_name (u);
1608 hstat.local_file = &dummy;
1611 if (!opt.output_document)
1612 locf = *hstat.local_file;
1614 locf = opt.output_document;
1616 hstat.referer = referer;
1618 filename_len = strlen (*hstat.local_file);
1619 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1621 if (opt.noclobber && file_exists_p (*hstat.local_file))
1623 /* If opt.noclobber is turned on and file already exists, do not
1624 retrieve the file */
1625 logprintf (LOG_VERBOSE, _("\
1626 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1627 /* If the file is there, we suppose it's retrieved OK. */
1630 /* #### Bogusness alert. */
1631 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1632 if (has_html_suffix_p (*hstat.local_file))
1640 if (opt.timestamping)
1642 int local_dot_orig_file_exists = 0;
1644 if (opt.backup_converted)
1645 /* If -K is specified, we'll act on the assumption that it was specified
1646 last time these files were downloaded as well, and instead of just
1647 comparing local file X against server file X, we'll compare local
1648 file X.orig (if extant, else X) against server file X. If -K
1649 _wasn't_ specified last time, or the server contains files called
1650 *.orig, -N will be back to not operating correctly with -k. */
1652 /* Would a single s[n]printf() call be faster? --dan
1654 Definitely not. sprintf() is horribly slow. It's a
1655 different question whether the difference between the two
1656 affects a program. Usually I'd say "no", but at one
1657 point I profiled Wget, and found that a measurable and
1658 non-negligible amount of time was lost calling sprintf()
1659 in url.c. Replacing sprintf with inline calls to
1660 strcpy() and long_to_string() made a difference.
1662 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1663 memcpy (filename_plus_orig_suffix + filename_len,
1664 ".orig", sizeof (".orig"));
1666 /* Try to stat() the .orig file. */
1667 if (stat (filename_plus_orig_suffix, &st) == 0)
1669 local_dot_orig_file_exists = 1;
1670 local_filename = filename_plus_orig_suffix;
1674 if (!local_dot_orig_file_exists)
1675 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1676 if (stat (*hstat.local_file, &st) == 0)
1677 local_filename = *hstat.local_file;
1679 if (local_filename != NULL)
1680 /* There was a local file, so we'll check later to see if the version
1681 the server has is the same version we already have, allowing us to
1687 /* Modification time granularity is 2 seconds for Windows, so
1688 increase local time by 1 second for later comparison. */
1691 local_size = st.st_size;
1695 /* Reset the counter. */
1697 *dt = 0 | ACCEPTRANGES;
1701 /* Increment the pass counter. */
1703 sleep_between_retrievals (count);
1704 /* Get the current time string. */
1705 tms = time_str (NULL);
1706 /* Print fetch message, if opt.verbose. */
1709 char *hurl = url_string (u, 1);
1713 sprintf (tmp, _("(try:%2d)"), count);
1714 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1715 tms, hurl, tmp, locf);
1717 ws_changetitle (hurl, 1);
1722 /* Default document type is empty. However, if spider mode is
1723 on or time-stamping is employed, HEAD_ONLY commands is
1724 encoded within *dt. */
1725 if (opt.spider || (use_ts && !got_head))
1729 /* Assume no restarting. */
1731 /* Decide whether or not to restart. */
1732 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1733 /* #### this calls access() and then stat(); could be optimized. */
1734 && file_exists_p (locf))
1735 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1736 hstat.restval = st.st_size;
1738 /* In `-c' is used and the file is existing and non-empty,
1739 refuse to truncate it if the server doesn't support continued
1741 hstat.no_truncate = 0;
1742 if (opt.always_rest && hstat.restval)
1743 hstat.no_truncate = 1;
1745 /* Decide whether to send the no-cache directive. We send it in
1747 a) we're using a proxy, and we're past our first retrieval.
1748 Some proxies are notorious for caching incomplete data, so
1749 we require a fresh get.
1750 b) caching is explicitly inhibited. */
1751 if ((proxy && count > 1) /* a */
1752 || !opt.allow_cache /* b */
1754 *dt |= SEND_NOCACHE;
1756 *dt &= ~SEND_NOCACHE;
1758 /* Try fetching the document, or at least its head. */
1759 err = gethttp (u, &hstat, dt, proxy);
1761 /* It's unfortunate that wget determines the local filename before finding
1762 out the Content-Type of the file. Barring a major restructuring of the
1763 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1764 *hstat.local_file to tack on ".html". */
1765 if (!opt.output_document)
1766 locf = *hstat.local_file;
1768 locf = opt.output_document;
1771 tms = time_str (NULL);
1772 /* Get the new location (with or without the redirection). */
1774 *newloc = xstrdup (hstat.newloc);
1777 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1778 case CONERROR: case READERR: case WRITEFAILED:
1780 /* Non-fatal errors continue executing the loop, which will
1781 bring them to "while" statement at the end, to judge
1782 whether the number of tries was exceeded. */
1783 free_hstat (&hstat);
1784 printwhat (count, opt.ntry);
1787 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1788 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1789 /* Fatal errors just return from the function. */
1790 free_hstat (&hstat);
1794 case FWRITEERR: case FOPENERR:
1795 /* Another fatal error. */
1796 logputs (LOG_VERBOSE, "\n");
1797 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1798 *hstat.local_file, strerror (errno));
1799 free_hstat (&hstat);
1804 /* Another fatal error. */
1805 logputs (LOG_VERBOSE, "\n");
1806 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1807 free_hstat (&hstat);
1812 /* Return the new location to the caller. */
1815 logprintf (LOG_NOTQUIET,
1816 _("ERROR: Redirection (%d) without location.\n"),
1818 free_hstat (&hstat);
1822 free_hstat (&hstat);
1827 /* The file was already fully retrieved. */
1828 free_hstat (&hstat);
1833 /* Deal with you later. */
1836 /* All possibilities should have been exhausted. */
1839 if (!(*dt & RETROKF))
1843 /* #### Ugly ugly ugly! */
1844 char *hurl = url_string (u, 1);
1845 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1848 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1849 tms, hstat.statcode, hstat.error);
1850 logputs (LOG_VERBOSE, "\n");
1851 free_hstat (&hstat);
1856 /* Did we get the time-stamp? */
1859 if (opt.timestamping && !hstat.remote_time)
1861 logputs (LOG_NOTQUIET, _("\
1862 Last-modified header missing -- time-stamps turned off.\n"));
1864 else if (hstat.remote_time)
1866 /* Convert the date-string into struct tm. */
1867 tmr = http_atotm (hstat.remote_time);
1868 if (tmr == (time_t) (-1))
1869 logputs (LOG_VERBOSE, _("\
1870 Last-modified header invalid -- time-stamp ignored.\n"));
1874 /* The time-stamping section. */
1879 use_ts = 0; /* no more time-stamping */
1880 count = 0; /* the retrieve count for HEAD is
1882 if (hstat.remote_time && tmr != (time_t) (-1))
1884 /* Now time-stamping can be used validly. Time-stamping
1885 means that if the sizes of the local and remote file
1886 match, and local file is newer than the remote file,
1887 it will not be retrieved. Otherwise, the normal
1888 download procedure is resumed. */
1890 (hstat.contlen == -1 || local_size == hstat.contlen))
1892 logprintf (LOG_VERBOSE, _("\
1893 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1895 free_hstat (&hstat);
1899 else if (tml >= tmr)
1900 logprintf (LOG_VERBOSE, _("\
1901 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1903 logputs (LOG_VERBOSE,
1904 _("Remote file is newer, retrieving.\n"));
1906 free_hstat (&hstat);
1909 if ((tmr != (time_t) (-1))
1911 && ((hstat.len == hstat.contlen) ||
1912 ((hstat.res == 0) &&
1913 ((hstat.contlen == -1) ||
1914 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1916 /* #### This code repeats in http.c and ftp.c. Move it to a
1918 const char *fl = NULL;
1919 if (opt.output_document)
1921 if (opt.od_known_regular)
1922 fl = opt.output_document;
1925 fl = *hstat.local_file;
1929 /* End of time-stamping section. */
1933 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1938 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1940 if (hstat.len == hstat.contlen)
1944 logprintf (LOG_VERBOSE,
1945 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1946 tms, tmrate, locf, hstat.len, hstat.contlen);
1947 logprintf (LOG_NONVERBOSE,
1948 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1949 tms, u->url, hstat.len, hstat.contlen, locf, count);
1952 total_downloaded_bytes += hstat.len;
1954 /* Remember that we downloaded the file for later ".orig" code. */
1955 if (*dt & ADDED_HTML_EXTENSION)
1956 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1958 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1960 free_hstat (&hstat);
1964 else if (hstat.res == 0) /* No read error */
1966 if (hstat.contlen == -1) /* We don't know how much we were supposed
1967 to get, so assume we succeeded. */
1971 logprintf (LOG_VERBOSE,
1972 _("%s (%s) - `%s' saved [%ld]\n\n"),
1973 tms, tmrate, locf, hstat.len);
1974 logprintf (LOG_NONVERBOSE,
1975 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1976 tms, u->url, hstat.len, locf, count);
1979 total_downloaded_bytes += hstat.len;
1981 /* Remember that we downloaded the file for later ".orig" code. */
1982 if (*dt & ADDED_HTML_EXTENSION)
1983 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1985 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1987 free_hstat (&hstat);
1991 else if (hstat.len < hstat.contlen) /* meaning we lost the
1992 connection too soon */
1994 logprintf (LOG_VERBOSE,
1995 _("%s (%s) - Connection closed at byte %ld. "),
1996 tms, tmrate, hstat.len);
1997 printwhat (count, opt.ntry);
1998 free_hstat (&hstat);
2001 else if (!opt.kill_longer) /* meaning we got more than expected */
2003 logprintf (LOG_VERBOSE,
2004 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2005 tms, tmrate, locf, hstat.len, hstat.contlen);
2006 logprintf (LOG_NONVERBOSE,
2007 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2008 tms, u->url, hstat.len, hstat.contlen, locf, count);
2010 total_downloaded_bytes += hstat.len;
2012 /* Remember that we downloaded the file for later ".orig" code. */
2013 if (*dt & ADDED_HTML_EXTENSION)
2014 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2016 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2018 free_hstat (&hstat);
2022 else /* the same, but not accepted */
2024 logprintf (LOG_VERBOSE,
2025 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2026 tms, tmrate, hstat.len, hstat.contlen);
2027 printwhat (count, opt.ntry);
2028 free_hstat (&hstat);
2032 else /* now hstat.res can only be -1 */
2034 if (hstat.contlen == -1)
2036 logprintf (LOG_VERBOSE,
2037 _("%s (%s) - Read error at byte %ld (%s)."),
2038 tms, tmrate, hstat.len, strerror (errno));
2039 printwhat (count, opt.ntry);
2040 free_hstat (&hstat);
2043 else /* hstat.res == -1 and contlen is given */
2045 logprintf (LOG_VERBOSE,
2046 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2047 tms, tmrate, hstat.len, hstat.contlen,
2049 printwhat (count, opt.ntry);
2050 free_hstat (&hstat);
2057 while (!opt.ntry || (count < opt.ntry));
2061 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2062 than local timezone.
2064 mktime is similar but assumes struct tm, also known as the
2065 "broken-down" form of time, is in local time zone. mktime_from_utc
2066 uses mktime to make the conversion understanding that an offset
2067 will be introduced by the local time assumption.
2069 mktime_from_utc then measures the introduced offset by applying
2070 gmtime to the initial result and applying mktime to the resulting
2071 "broken-down" form. The difference between the two mktime results
2072 is the measured offset which is then subtracted from the initial
2073 mktime result to yield a calendar time which is the value returned.
2075 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2076 consistent offset (the non DST offset) since tm and tm+o might be
2077 on opposite sides of a DST change.
2079 Some implementations of mktime return -1 for the nonexistent
2080 localtime hour at the beginning of DST. In this event, use
2081 mktime(tm - 1hr) + 3600.
2085 gmtime(t+o) --> tm+o
2086 mktime(tm+o) --> t+2o
2087 t+o - (t+2o - t+o) = t
2089 Note that glibc contains a function of the same purpose named
2090 `timegm' (reverse of gmtime). But obviously, it is not universally
2091 available, and unfortunately it is not straightforwardly
2092 extractable for use here. Perhaps configure should detect timegm
2093 and use it where available.
2095 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2096 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2097 Further improved by Roger with assistance from Edward J. Sabol
2098 based on input by Jamie Zawinski. */
2101 mktime_from_utc (struct tm *t)
2112 return -1; /* can't deal with output from strptime */
2123 return -1; /* can't deal with output from gmtime */
2126 return (tl - (tb - tl));
2129 /* Check whether the result of strptime() indicates success.
2130 strptime() returns the pointer to how far it got to in the string.
2131 The processing has been successful if the string is at `GMT' or
2132 `+X', or at the end of the string.
2134 In extended regexp parlance, the function returns 1 if P matches
2135 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2136 can return) is considered a failure and 0 is returned. */
2138 check_end (const char *p)
2142 while (ISSPACE (*p))
2145 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2146 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2152 /* Convert the textual specification of time in TIME_STRING to the
2153 number of seconds since the Epoch.
2155 TIME_STRING can be in any of the three formats RFC2068 allows the
2156 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2157 Timezones are ignored, and should be GMT.
2159 Return the computed time_t representation, or -1 if the conversion
2162 This function uses strptime with various string formats for parsing
2163 TIME_STRING. This results in a parser that is not as lenient in
2164 interpreting TIME_STRING as I would like it to be. Being based on
2165 strptime, it always allows shortened months, one-digit days, etc.,
2166 but due to the multitude of formats in which time can be
2167 represented, an ideal HTTP time parser would be even more
2168 forgiving. It should completely ignore things like week days and
2169 concentrate only on the various forms of representing years,
2170 months, days, hours, minutes, and seconds. For example, it would
2171 be nice if it accepted ISO 8601 out of the box.
2173 I've investigated free and PD code for this purpose, but none was
2174 usable. getdate was big and unwieldy, and had potential copyright
2175 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2176 distributed with phttpd, is excellent, but we cannot use it because
2177 it is not assigned to the FSF. So I stuck it with strptime. */
2180 http_atotm (const char *time_string)
2182 /* NOTE: Solaris strptime man page claims that %n and %t match white
2183 space, but that's not universally available. Instead, we simply
2184 use ` ' to mean "skip all WS", which works under all strptime
2185 implementations I've tested. */
2187 static const char *time_formats[] = {
2188 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2189 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2190 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2191 (google.com uses this for their cookies.) */
2192 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2198 /* According to Roger Beeman, we need to initialize tm_isdst, since
2199 strptime won't do it. */
2202 /* Note that under foreign locales Solaris strptime() fails to
2203 recognize English dates, which renders this function useless. We
2204 solve this by being careful not to affect LC_TIME when
2205 initializing locale.
2207 Another solution would be to temporarily set locale to C, invoke
2208 strptime(), and restore it back. This is slow and dirty,
2209 however, and locale support other than LC_MESSAGES can mess other
2210 things, so I rather chose to stick with just setting LC_MESSAGES.
2212 GNU strptime does not have this problem because it recognizes
2213 both international and local dates. */
2215 for (i = 0; i < countof (time_formats); i++)
2216 if (check_end (strptime (time_string, time_formats[i], &t)))
2217 return mktime_from_utc (&t);
2219 /* All formats have failed. */
2223 /* Authorization support: We support two authorization schemes:
2225 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2227 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2228 consisting of answering to the server's challenge with the proper
2231 /* How many bytes it will take to store LEN bytes in base64. */
2232 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2234 /* Encode the string S of length LENGTH to base64 format and place it
2235 to STORE. STORE will be 0-terminated, and must point to a writable
2236 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2238 base64_encode (const char *s, char *store, int length)
2240 /* Conversion table. */
2241 static char tbl[64] = {
2242 'A','B','C','D','E','F','G','H',
2243 'I','J','K','L','M','N','O','P',
2244 'Q','R','S','T','U','V','W','X',
2245 'Y','Z','a','b','c','d','e','f',
2246 'g','h','i','j','k','l','m','n',
2247 'o','p','q','r','s','t','u','v',
2248 'w','x','y','z','0','1','2','3',
2249 '4','5','6','7','8','9','+','/'
2252 unsigned char *p = (unsigned char *)store;
2254 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2255 for (i = 0; i < length; i += 3)
2257 *p++ = tbl[s[0] >> 2];
2258 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2259 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2260 *p++ = tbl[s[2] & 0x3f];
2263 /* Pad the result if necessary... */
2264 if (i == length + 1)
2266 else if (i == length + 2)
2267 *(p - 1) = *(p - 2) = '=';
2268 /* ...and zero-terminate it. */
2272 /* Create the authentication header contents for the `Basic' scheme.
2273 This is done by encoding the string `USER:PASS' in base64 and
2274 prepending `HEADER: Basic ' to it. */
2276 basic_authentication_encode (const char *user, const char *passwd,
2279 char *t1, *t2, *res;
2280 int len1 = strlen (user) + 1 + strlen (passwd);
2281 int len2 = BASE64_LENGTH (len1);
2283 t1 = (char *)alloca (len1 + 1);
2284 sprintf (t1, "%s:%s", user, passwd);
2285 t2 = (char *)alloca (1 + len2);
2286 base64_encode (t1, t2, len1);
2287 res = (char *)xmalloc (len2 + 11 + strlen (header));
2288 sprintf (res, "%s: Basic %s\r\n", header, t2);
2294 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2295 of a field in such a header. If the field is the one specified by
2296 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2297 digest authorization code), extract its value in the (char*)
2298 variable pointed by RET. Returns negative on a malformed header,
2299 or number of bytes that have been parsed by this call. */
2301 extract_header_attr (const char *au, const char *attr_name, char **ret)
2303 const char *cp, *ep;
2307 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2309 cp += strlen (attr_name);
2312 cp += skip_lws (cp);
2317 cp += skip_lws (cp);
2322 for (ep = cp; *ep && *ep != '\"'; ep++)
2327 *ret = strdupdelim (cp, ep);
2334 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2335 an array of 16 bytes containing the hash keys, and BUF should be a
2336 buffer of 33 writable characters (32 for hex digits plus one for
2337 zero termination). */
2339 dump_hash (unsigned char *buf, const unsigned char *hash)
2343 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2345 *buf++ = XNUM_TO_digit (*hash >> 4);
2346 *buf++ = XNUM_TO_digit (*hash & 0xf);
2351 /* Take the line apart to find the challenge, and compose a digest
2352 authorization header. See RFC2069 section 2.1.2. */
2354 digest_authentication_encode (const char *au, const char *user,
2355 const char *passwd, const char *method,
2358 static char *realm, *opaque, *nonce;
2363 { "realm", &realm },
2364 { "opaque", &opaque },
2369 realm = opaque = nonce = NULL;
2371 au += 6; /* skip over `Digest' */
2376 au += skip_lws (au);
2377 for (i = 0; i < countof (options); i++)
2379 int skip = extract_header_attr (au, options[i].name,
2380 options[i].variable);
2384 xfree_null (opaque);
2394 if (i == countof (options))
2396 while (*au && *au != '=')
2400 au += skip_lws (au);
2404 while (*au && *au != '\"')
2411 while (*au && *au != ',')
2416 if (!realm || !nonce || !user || !passwd || !path || !method)
2419 xfree_null (opaque);
2424 /* Calculate the digest value. */
2426 ALLOCA_MD5_CONTEXT (ctx);
2427 unsigned char hash[MD5_HASHLEN];
2428 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2429 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2431 /* A1BUF = H(user ":" realm ":" password) */
2433 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2434 gen_md5_update ((unsigned char *)":", 1, ctx);
2435 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2436 gen_md5_update ((unsigned char *)":", 1, ctx);
2437 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2438 gen_md5_finish (ctx, hash);
2439 dump_hash (a1buf, hash);
2441 /* A2BUF = H(method ":" path) */
2443 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2444 gen_md5_update ((unsigned char *)":", 1, ctx);
2445 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2446 gen_md5_finish (ctx, hash);
2447 dump_hash (a2buf, hash);
2449 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2451 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2452 gen_md5_update ((unsigned char *)":", 1, ctx);
2453 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2454 gen_md5_update ((unsigned char *)":", 1, ctx);
2455 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2456 gen_md5_finish (ctx, hash);
2457 dump_hash (response_digest, hash);
2459 res = (char*) xmalloc (strlen (user)
2464 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2465 + (opaque ? strlen (opaque) : 0)
2467 sprintf (res, "Authorization: Digest \
2468 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2469 user, realm, nonce, path, response_digest);
2472 char *p = res + strlen (res);
2473 strcat (p, ", opaque=\"");
2477 strcat (res, "\r\n");
2481 #endif /* USE_DIGEST */
2484 #define BEGINS_WITH(line, string_constant) \
2485 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2486 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2487 || !line[sizeof (string_constant) - 1]))
2490 known_authentication_scheme_p (const char *au)
2492 return BEGINS_WITH (au, "Basic")
2493 || BEGINS_WITH (au, "Digest")
2494 || BEGINS_WITH (au, "NTLM");
2499 /* Create the HTTP authorization request header. When the
2500 `WWW-Authenticate' response header is seen, according to the
2501 authorization scheme specified in that header (`Basic' and `Digest'
2502 are supported by the current implementation), produce an
2503 appropriate HTTP authorization request header. */
2505 create_authorization_line (const char *au, const char *user,
2506 const char *passwd, const char *method,
2509 char *wwwauth = NULL;
2511 if (!strncasecmp (au, "Basic", 5))
2512 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2513 if (!strncasecmp (au, "NTLM", 4))
2514 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2516 else if (!strncasecmp (au, "Digest", 6))
2517 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2518 #endif /* USE_DIGEST */