2 Copyright (C) 1995, 1996, 1997, 1998, 2000 Free Software Foundation, Inc.
4 This file is part of Wget.
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
24 #include <sys/types.h>
36 #if TIME_WITH_SYS_TIME
37 # include <sys/time.h>
41 # include <sys/time.h>
50 # include <netdb.h> /* for h_errno */
67 # include "gen_sslfunc.h"
70 extern char *version_string;
82 #define TEXTHTML_S "text/html"
83 #define HTTP_ACCEPT "*/*"
85 /* Some status code validation macros: */
86 #define H_20X(x) (((x) >= 200) && ((x) < 300))
87 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
88 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
89 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
91 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
93 #define HTTP_STATUS_OK 200
94 #define HTTP_STATUS_CREATED 201
95 #define HTTP_STATUS_ACCEPTED 202
96 #define HTTP_STATUS_NO_CONTENT 204
97 #define HTTP_STATUS_PARTIAL_CONTENTS 206
99 /* Redirection 3xx. */
100 #define HTTP_STATUS_MULTIPLE_CHOICES 300
101 #define HTTP_STATUS_MOVED_PERMANENTLY 301
102 #define HTTP_STATUS_MOVED_TEMPORARILY 302
103 #define HTTP_STATUS_NOT_MODIFIED 304
105 /* Client error 4xx. */
106 #define HTTP_STATUS_BAD_REQUEST 400
107 #define HTTP_STATUS_UNAUTHORIZED 401
108 #define HTTP_STATUS_FORBIDDEN 403
109 #define HTTP_STATUS_NOT_FOUND 404
111 /* Server errors 5xx. */
112 #define HTTP_STATUS_INTERNAL 500
113 #define HTTP_STATUS_NOT_IMPLEMENTED 501
114 #define HTTP_STATUS_BAD_GATEWAY 502
115 #define HTTP_STATUS_UNAVAILABLE 503
118 /* Parse the HTTP status line, which is of format:
120 HTTP-Version SP Status-Code SP Reason-Phrase
122 The function returns the status-code, or -1 if the status line is
123 malformed. The pointer to reason-phrase is returned in RP. */
125 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
127 /* (the variables must not be named `major' and `minor', because
128 that breaks compilation with SunOS4 cc.) */
129 int mjr, mnr, statcode;
132 *reason_phrase_ptr = NULL;
134 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
135 major version, and Y is minor version. */
136 if (strncmp (line, "HTTP/", 5) != 0)
140 /* Calculate major HTTP version. */
142 for (mjr = 0; ISDIGIT (*line); line++)
143 mjr = 10 * mjr + (*line - '0');
144 if (*line != '.' || p == line)
148 /* Calculate minor HTTP version. */
150 for (mnr = 0; ISDIGIT (*line); line++)
151 mnr = 10 * mnr + (*line - '0');
152 if (*line != ' ' || p == line)
154 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
155 minor version can be safely ignored. */
160 /* Calculate status code. */
161 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
163 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
165 /* Set up the reason phrase pointer. */
167 /* RFC2068 requires SPC here, but we allow the string to finish
168 here, in case no reason-phrase is present. */
172 *reason_phrase_ptr = line;
177 *reason_phrase_ptr = line + 1;
182 /* Functions to be used as arguments to header_process(): */
184 struct http_process_range_closure {
190 /* Parse the `Content-Range' header and extract the information it
191 contains. Returns 1 if successful, -1 otherwise. */
193 http_process_range (const char *hdr, void *arg)
195 struct http_process_range_closure *closure
196 = (struct http_process_range_closure *)arg;
199 /* Certain versions of Nutscape proxy server send out
200 `Content-Length' without "bytes" specifier, which is a breach of
201 RFC2068 (as well as the HTTP/1.1 draft which was current at the
202 time). But hell, I must support it... */
203 if (!strncasecmp (hdr, "bytes", 5))
206 hdr += skip_lws (hdr);
212 for (num = 0; ISDIGIT (*hdr); hdr++)
213 num = 10 * num + (*hdr - '0');
214 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
216 closure->first_byte_pos = num;
218 for (num = 0; ISDIGIT (*hdr); hdr++)
219 num = 10 * num + (*hdr - '0');
220 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
222 closure->last_byte_pos = num;
224 for (num = 0; ISDIGIT (*hdr); hdr++)
225 num = 10 * num + (*hdr - '0');
226 closure->entity_length = num;
230 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
231 Used for `Accept-Ranges'. */
233 http_process_none (const char *hdr, void *arg)
235 int *where = (int *)arg;
237 if (strstr (hdr, "none"))
244 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
246 http_process_type (const char *hdr, void *arg)
248 char **result = (char **)arg;
249 /* Locate P on `;' or the terminating zero, whichever comes first. */
250 const char *p = strchr (hdr, ';');
252 p = hdr + strlen (hdr);
253 while (p > hdr && ISSPACE (*(p - 1)))
255 *result = strdupdelim (hdr, p);
259 /* Check whether the `Connection' header is set to "keep-alive". */
261 http_process_connection (const char *hdr, void *arg)
263 int *flag = (int *)arg;
264 if (!strcasecmp (hdr, "Keep-Alive"))
269 /* Persistent connections. Currently, we cache the most recently used
270 connection as persistent, provided that the HTTP server agrees to
271 make it such. The persistence data is stored in the variables
272 below. Ideally, it would be in a structure, and it should be
273 possible to cache an arbitrary fixed number of these connections.
275 I think the code is quite easy to extend in that direction. */
277 /* Whether a persistent connection is active. */
278 static int pc_active_p;
279 /* Host and port of currently active persistent connection. */
280 static unsigned char pc_last_host[4];
281 static unsigned short pc_last_port;
283 /* File descriptor of the currently active persistent connection. */
284 static int pc_last_fd;
287 /* Whether a ssl handshake has occoured on this connection */
288 static int pc_active_ssl;
289 /* SSL connection of the currently active persistent connection. */
290 static SSL *pc_last_ssl;
291 #endif /* HAVE_SSL */
293 /* Mark the persistent connection as invalid. This is used by the
294 CLOSE_* macros after they forcefully close a registered persistent
295 connection. This does not close the file descriptor -- it is left
296 to the caller to do that. (Maybe it should, though.) */
299 invalidate_persistent (void)
304 #endif /* HAVE_SSL */
305 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
308 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
309 persistent. This will enable someone to use the same connection
310 later. In the context of HTTP, this must be called only AFTER the
311 response has been received and the server has promised that the
312 connection will remain alive.
314 If a previous connection was persistent, it is closed. */
317 register_persistent (const char *host, unsigned short port, int fd
327 if (pc_last_fd == fd)
329 /* The connection FD is already registered. Nothing to
335 /* The old persistent connection is still active; let's
336 close it first. This situation arises whenever a
337 persistent connection exists, but we then connect to a
338 different host, and try to register a persistent
339 connection to that one. */
341 /* The ssl disconnect has to take place before the closing
344 shutdown_ssl(pc_last_ssl);
347 invalidate_persistent ();
351 /* This store_hostaddress may not fail, because it has the results
353 success = store_hostaddress (pc_last_host, host);
360 pc_active_ssl = ssl ? 1 : 0;
362 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
365 /* Return non-zero if a persistent connection is available for
366 connecting to HOST:PORT. */
369 persistent_available_p (const char *host, unsigned short port
375 unsigned char this_host[4];
376 /* First, check whether a persistent connection is active at all. */
379 /* Second, check if the active connection pertains to the correct
380 (HOST, PORT) ordered pair. */
381 if (port != pc_last_port)
384 /* Second, a): check if current connection is (not) ssl, too. This
385 test is unlikely to fail because HTTP and HTTPS typicaly use
386 different ports. Yet it is possible, or so I [Christian
387 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
389 if (ssl != pc_active_ssl)
391 #endif /* HAVE_SSL */
392 if (!store_hostaddress (this_host, host))
394 if (memcmp (pc_last_host, this_host, 4))
396 /* Third: check whether the connection is still open. This is
397 important because most server implement a liberal (short) timeout
398 on persistent connections. Wget can of course always reconnect
399 if the connection doesn't work out, but it's nicer to know in
400 advance. This test is a logical followup of the first test, but
401 is "expensive" and therefore placed at the end of the list. */
402 if (!test_socket_open (pc_last_fd))
404 /* Oops, the socket is no longer open. Now that we know that,
405 let's invalidate the persistent connection before returning
408 invalidate_persistent ();
415 # define SHUTDOWN_SSL(ssl) do { \
417 shutdown_ssl (ssl); \
420 # define SHUTDOWN_SSL(ssl)
423 /* The idea behind these two CLOSE macros is to distinguish between
424 two cases: one when the job we've been doing is finished, and we
425 want to close the connection and leave, and two when something is
426 seriously wrong and we're closing the connection as part of
429 In case of keep_alive, CLOSE_FINISH should leave the connection
430 open, while CLOSE_INVALIDATE should still close it.
432 Note that the semantics of the flag `keep_alive' is "this
433 connection *will* be reused (the server has promised not to close
434 the connection once we're done)", while the semantics of
435 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
436 active, registered connection". */
438 #define CLOSE_FINISH(fd) do { \
441 SHUTDOWN_SSL (ssl); \
443 if (pc_active_p && (fd) == pc_last_fd) \
444 invalidate_persistent (); \
448 #define CLOSE_INVALIDATE(fd) do { \
449 SHUTDOWN_SSL (ssl); \
451 if (pc_active_p && (fd) == pc_last_fd) \
452 invalidate_persistent (); \
457 long len; /* received length */
458 long contlen; /* expected length */
459 long restval; /* the restart value */
460 int res; /* the result of last read */
461 char *newloc; /* new location (redirection) */
462 char *remote_time; /* remote time-stamp string */
463 char *error; /* textual HTTP error */
464 int statcode; /* status code */
465 long dltime; /* time of the download */
468 /* Free the elements of hstat X. */
469 #define FREEHSTAT(x) do \
471 FREE_MAYBE ((x).newloc); \
472 FREE_MAYBE ((x).remote_time); \
473 FREE_MAYBE ((x).error); \
474 (x).newloc = (x).remote_time = (x).error = NULL; \
477 static char *create_authorization_line PARAMS ((const char *, const char *,
478 const char *, const char *,
480 static char *basic_authentication_encode PARAMS ((const char *, const char *,
482 static int known_authentication_scheme_p PARAMS ((const char *));
484 static time_t http_atotm PARAMS ((char *));
486 #define BEGINS_WITH(line, string_constant) \
487 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
488 && (ISSPACE (line[sizeof (string_constant) - 1]) \
489 || !line[sizeof (string_constant) - 1]))
491 /* Retrieve a document through HTTP protocol. It recognizes status
492 code, and correctly handles redirections. It closes the network
493 socket. If it receives an error from the functions below it, it
494 will print it if there is enough information to do so (almost
495 always), returning the error to the caller (i.e. http_loop).
497 Various HTTP parameters are stored to hs. Although it parses the
498 response code correctly, it is not used in a sane way. The caller
501 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
502 and u->proxy->url will be given to the proxy server (bad naming,
505 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
507 char *request, *type, *command, *path;
509 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
510 char *authenticate_h;
514 char *request_keep_alive;
515 int sock, hcount, num_written, all_length, remport, statcode;
516 long contlen, contrange;
520 int auth_tried_already;
523 static SSL_CTX *ssl_ctx = NULL;
525 #endif /* HAVE_SSL */
527 /* Whether this connection will be kept alive after the HTTP request
531 /* Flags that detect the two ways of specifying HTTP keep-alive
533 int http_keep_alive_1, http_keep_alive_2;
535 /* Whether keep-alive should be inhibited. */
536 int inhibit_keep_alive;
539 /* initialize ssl_ctx on first run */
542 #endif /* HAVE_SSL */
544 if (!(*dt & HEAD_ONLY))
545 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
546 know the local filename so we can save to it. */
547 assert (u->local != NULL);
550 auth_tried_already = 0;
552 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
555 /* We need to come back here when the initial attempt to retrieve
556 without authorization header fails. (Expected to happen at least
557 for the Digest authorization scheme.) */
560 http_keep_alive_1 = http_keep_alive_2 = 0;
562 /* Initialize certain elements of struct http_stat. */
567 hs->remote_time = NULL;
570 /* Which structure to use to retrieve the original URL data. */
576 /* First: establish the connection. */
577 if (inhibit_keep_alive
580 !persistent_available_p (u->host, u->port)
582 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
583 #endif /* HAVE_SSL */
586 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
587 err = make_connection (&sock, u->host, u->port);
591 logputs (LOG_VERBOSE, "\n");
592 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
596 logputs (LOG_VERBOSE, "\n");
597 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
601 logputs (LOG_VERBOSE, "\n");
602 logprintf (LOG_NOTQUIET,
603 _("Connection to %s:%hu refused.\n"), u->host, u->port);
607 logputs (LOG_VERBOSE, "\n");
608 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
613 /* Everything is fine! */
614 logputs (LOG_VERBOSE, _("connected!\n"));
621 if (u->proto == URLHTTPS)
622 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
624 logputs (LOG_VERBOSE, "\n");
625 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
629 #endif /* HAVE_SSL */
633 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
634 /* #### pc_last_fd should be accessed through an accessor
639 #endif /* HAVE_SSL */
640 DEBUGP (("Reusing fd %d.\n", sock));
644 path = u->proxy->url;
648 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
652 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
653 sprintf (referer, "Referer: %s\r\n", ou->referer);
655 if (*dt & SEND_NOCACHE)
656 pragma_h = "Pragma: no-cache\r\n";
661 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
662 /* Gag me! Some servers (e.g. WebSitePro) have been known to
663 respond to the following `Range' format by generating a
664 multipart/x-byte-ranges MIME document! This MIME type was
665 present in an old draft of the byteranges specification.
666 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
667 only if multiple non-overlapping ranges are requested --
668 which Wget never does. */
669 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
674 STRDUP_ALLOCA (useragent, opt.useragent);
677 useragent = (char *)alloca (10 + strlen (version_string));
678 sprintf (useragent, "Wget/%s", version_string);
680 /* Construct the authentication, if userid is present. */
683 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
684 user = user ? user : opt.http_user;
685 passwd = passwd ? passwd : opt.http_passwd;
692 /* We have the username and the password, but haven't tried
693 any authorization yet. Let's see if the "Basic" method
694 works. If not, we'll come back here and construct a
695 proper authorization method with the right challenges.
697 If we didn't employ this kind of logic, every URL that
698 requires authorization would have to be processed twice,
699 which is very suboptimal and generates a bunch of false
700 "unauthorized" errors in the server log.
702 #### But this logic also has a serious problem when used
703 with stronger authentications: we *first* transmit the
704 username and the password in clear text, and *then*
705 attempt a stronger authentication scheme. That cannot be
706 right! We are only fortunate that almost everyone still
707 uses the `Basic' scheme anyway.
709 There should be an option to prevent this from happening,
710 for those who use strong authentication schemes and value
712 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
716 wwwauth = create_authorization_line (authenticate_h, user, passwd,
724 char *proxy_user, *proxy_passwd;
725 /* For normal username and password, URL components override
726 command-line/wgetrc parameters. With proxy authentication,
727 it's the reverse, because proxy URLs are normally the
728 "permanent" ones, so command-line args should take
730 if (opt.proxy_user && opt.proxy_passwd)
732 proxy_user = opt.proxy_user;
733 proxy_passwd = opt.proxy_passwd;
737 proxy_user = u->user;
738 proxy_passwd = u->passwd;
740 /* #### This is junky. Can't the proxy request, say, `Digest'
742 if (proxy_user && proxy_passwd)
743 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
744 "Proxy-Authorization");
749 /* String of the form :PORT. Used only for non-standard ports. */
753 port_maybe = (char *)alloca (numdigit (remport) + 2);
754 sprintf (port_maybe, ":%d", remport);
757 if (!inhibit_keep_alive)
758 request_keep_alive = "Connection: Keep-Alive\r\n";
760 request_keep_alive = NULL;
762 /* Allocate the memory for the request. */
763 request = (char *)alloca (strlen (command) + strlen (path)
766 + (port_maybe ? strlen (port_maybe) : 0)
767 + strlen (HTTP_ACCEPT)
768 + (request_keep_alive
769 ? strlen (request_keep_alive) : 0)
770 + (referer ? strlen (referer) : 0)
771 + (wwwauth ? strlen (wwwauth) : 0)
772 + (proxyauth ? strlen (proxyauth) : 0)
773 + (range ? strlen (range) : 0)
775 + (opt.user_header ? strlen (opt.user_header) : 0)
777 /* Construct the request. */
784 command, path, useragent, remhost,
785 port_maybe ? port_maybe : "",
787 request_keep_alive ? request_keep_alive : "",
788 referer ? referer : "",
789 wwwauth ? wwwauth : "",
790 proxyauth ? proxyauth : "",
793 opt.user_header ? opt.user_header : "");
794 DEBUGP (("---request begin---\n%s---request end---\n", request));
795 /* Free the temporary memory. */
796 FREE_MAYBE (wwwauth);
797 FREE_MAYBE (proxyauth);
799 /* Send the request to server. */
801 if (u->proto == URLHTTPS)
802 num_written = ssl_iwrite (ssl, request, strlen (request));
804 #endif /* HAVE_SSL */
805 num_written = iwrite (sock, request, strlen (request));
809 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
811 CLOSE_INVALIDATE (sock);
814 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
815 u->proxy ? "Proxy" : "HTTP");
816 contlen = contrange = -1;
821 /* Before reading anything, initialize the rbuf. */
822 rbuf_initialize (&rbuf, sock);
824 if (u->proto == URLHTTPS)
828 #endif /* HAVE_SSL */
831 /* Header-fetching loop. */
839 /* Get the header. */
840 status = header_get (&rbuf, &hdr,
841 /* Disallow continuations for status line. */
842 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
844 /* Check for errors. */
845 if (status == HG_EOF && *hdr)
847 /* This used to be an unconditional error, but that was
848 somewhat controversial, because of a large number of
849 broken CGI's that happily "forget" to send the second EOL
850 before closing the connection of a HEAD request.
852 So, the deal is to check whether the header is empty
853 (*hdr is zero if it is); if yes, it means that the
854 previous header was fully retrieved, and that -- most
855 probably -- the request is complete. "...be liberal in
856 what you accept." Oh boy. */
857 logputs (LOG_VERBOSE, "\n");
858 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
861 FREE_MAYBE (hs->newloc);
862 FREE_MAYBE (all_headers);
863 CLOSE_INVALIDATE (sock);
866 else if (status == HG_ERROR)
868 logputs (LOG_VERBOSE, "\n");
869 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
873 FREE_MAYBE (hs->newloc);
874 FREE_MAYBE (all_headers);
875 CLOSE_INVALIDATE (sock);
879 /* If the headers are to be saved to a file later, save them to
881 if (opt.save_headers)
883 int lh = strlen (hdr);
884 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
885 memcpy (all_headers + all_length, hdr, lh);
887 all_headers[all_length++] = '\n';
888 all_headers[all_length] = '\0';
891 /* Print the header if requested. */
892 if (opt.server_response && hcount != 1)
893 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
895 /* Check for status line. */
899 /* Parse the first line of server response. */
900 statcode = parse_http_status_line (hdr, &error);
901 hs->statcode = statcode;
902 /* Store the descriptive response. */
903 if (statcode == -1) /* malformed response */
905 /* A common reason for "malformed response" error is the
906 case when no data was actually received. Handle this
909 hs->error = xstrdup (_("No data received"));
911 hs->error = xstrdup (_("Malformed status line"));
916 hs->error = xstrdup (_("(no description)"));
918 hs->error = xstrdup (error);
925 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
930 /* Exit on empty header. */
937 /* Try getting content-length. */
938 if (contlen == -1 && !opt.ignore_length)
939 if (header_process (hdr, "Content-Length", header_extract_number,
942 /* Try getting content-type. */
944 if (header_process (hdr, "Content-Type", http_process_type, &type))
946 /* Try getting location. */
948 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
950 /* Try getting last-modified. */
951 if (!hs->remote_time)
952 if (header_process (hdr, "Last-Modified", header_strdup,
955 /* Try getting www-authentication. */
957 if (header_process (hdr, "WWW-Authenticate", header_strdup,
960 /* Check for accept-ranges header. If it contains the word
961 `none', disable the ranges. */
962 if (*dt & ACCEPTRANGES)
965 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
968 *dt &= ~ACCEPTRANGES;
972 /* Try getting content-range. */
975 struct http_process_range_closure closure;
976 if (header_process (hdr, "Content-Range", http_process_range, &closure))
978 contrange = closure.first_byte_pos;
982 /* Check for keep-alive related responses. */
983 if (!inhibit_keep_alive)
985 /* Check for the `Keep-Alive' header. */
986 if (!http_keep_alive_1)
988 if (header_process (hdr, "Keep-Alive", header_exists,
992 /* Check for `Connection: Keep-Alive'. */
993 if (!http_keep_alive_2)
995 if (header_process (hdr, "Connection", http_process_connection,
1004 logputs (LOG_VERBOSE, "\n");
1007 && (http_keep_alive_1 || http_keep_alive_2))
1009 assert (inhibit_keep_alive == 0);
1013 /* The server has promised that it will not close the connection
1014 when we're done. This means that we can register it. */
1016 register_persistent (u->host, u->port, sock);
1018 register_persistent (u->host, u->port, sock, ssl);
1019 #endif /* HAVE_SSL */
1021 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1024 /* Authorization is required. */
1028 CLOSE_FINISH (sock);
1029 if (auth_tried_already)
1031 /* If we have tried it already, then there is not point
1034 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1035 xfree (authenticate_h);
1038 else if (!known_authentication_scheme_p (authenticate_h))
1040 xfree (authenticate_h);
1041 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1044 else if (BEGINS_WITH (authenticate_h, "Basic"))
1046 /* The authentication scheme is basic, the one we try by
1047 default, and it failed. There's no sense in trying
1053 auth_tried_already = 1;
1057 /* We do not need this anymore. */
1060 xfree (authenticate_h);
1061 authenticate_h = NULL;
1064 /* 20x responses are counted among successful by default. */
1065 if (H_20X (statcode))
1068 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1071 /* We don't assume text/html by default. */
1074 if (opt.html_extension && (*dt & TEXTHTML))
1075 /* -E / --html-extension / html_extension = on was specified, and this is a
1076 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1077 already the file's suffix, tack on ".html". */
1079 char* last_period_in_local_filename = strrchr(u->local, '.');
1081 if (last_period_in_local_filename == NULL ||
1082 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1083 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1085 size_t local_filename_len = strlen(u->local);
1087 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1088 strcpy(u->local + local_filename_len, ".html");
1090 *dt |= ADDED_HTML_EXTENSION;
1094 if (contrange == -1)
1096 else if (contrange != hs->restval ||
1097 (H_PARTIAL (statcode) && contrange == -1))
1099 /* This means the whole request was somehow misunderstood by the
1100 server. Bail out. */
1102 FREE_MAYBE (hs->newloc);
1103 FREE_MAYBE (all_headers);
1104 CLOSE_INVALIDATE (sock);
1111 contlen += contrange;
1113 contrange = -1; /* If conent-length was not sent,
1114 content-range will be ignored. */
1116 hs->contlen = contlen;
1118 /* Return if redirected. */
1119 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1121 /* RFC2068 says that in case of the 300 (multiple choices)
1122 response, the server can output a preferred URL through
1123 `Location' header; otherwise, the request should be treated
1124 like GET. So, if the location is set, it will be a
1125 redirection; otherwise, just proceed normally. */
1126 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1130 logprintf (LOG_VERBOSE,
1131 _("Location: %s%s\n"),
1132 hs->newloc ? hs->newloc : _("unspecified"),
1133 hs->newloc ? _(" [following]") : "");
1134 CLOSE_FINISH (sock);
1136 FREE_MAYBE (all_headers);
1142 if ((*dt & RETROKF) && !opt.server_response)
1144 /* No need to print this output if the body won't be
1145 downloaded at all, or if the original server response is
1147 logputs (LOG_VERBOSE, _("Length: "));
1150 logputs (LOG_VERBOSE, legible (contlen));
1151 if (contrange != -1)
1152 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1153 legible (contlen - contrange));
1156 logputs (LOG_VERBOSE,
1157 opt.ignore_length ? _("ignored") : _("unspecified"));
1159 logprintf (LOG_VERBOSE, " [%s]\n", type);
1161 logputs (LOG_VERBOSE, "\n");
1165 type = NULL; /* We don't need it any more. */
1167 /* Return if we have no intention of further downloading. */
1168 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1170 /* In case someone cares to look... */
1174 FREE_MAYBE (all_headers);
1175 CLOSE_FINISH (sock);
1176 return RETRFINISHED;
1179 /* Open the local file. */
1182 mkalldirs (u->local);
1184 rotate_backups (u->local);
1185 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1188 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1189 CLOSE_FINISH (sock);
1190 FREE_MAYBE (all_headers);
1199 /* This will silently fail for streams that don't correspond
1200 to regular files, but that's OK. */
1206 /* #### This confuses the code that checks for file size. There
1207 should be some overhead information. */
1208 if (opt.save_headers)
1209 fwrite (all_headers, 1, all_length, fp);
1211 /* Get the contents of the document. */
1212 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1213 (contlen != -1 ? contlen : 0),
1215 hs->dltime = elapsed_time ();
1217 /* Close or flush the file. We have to be careful to check for
1218 error here. Checking the result of fwrite() is not enough --
1219 errors could go unnoticed! */
1222 flush_res = fclose (fp);
1224 flush_res = fflush (fp);
1225 if (flush_res == EOF)
1228 FREE_MAYBE (all_headers);
1229 CLOSE_FINISH (sock);
1232 return RETRFINISHED;
1235 /* The genuine HTTP loop! This is the part where the retrieval is
1236 retried, and retried, and retried, and... */
1238 http_loop (struct urlinfo *u, char **newloc, int *dt)
1241 int use_ts, got_head = 0; /* time-stamping info */
1242 char *filename_plus_orig_suffix;
1243 char *local_filename = NULL;
1244 char *tms, *suf, *locf, *tmrate;
1246 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1247 long local_size = 0; /* the size of the local file */
1248 size_t filename_len;
1249 struct http_stat hstat; /* HTTP status */
1254 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1255 has_wildcards_p because it would also warn on `?', and we know that
1256 shows up in CGI paths a *lot*. */
1257 if (strchr (u->url, '*'))
1258 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1260 /* Determine the local filename. */
1262 u->local = url_filename (u->proxy ? u->proxy : u);
1264 if (!opt.output_document)
1267 locf = opt.output_document;
1269 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1270 xmalloc() here before EACH return. This is one reason it's better to set
1271 flags that influence flow control and then return once at the end. */
1272 filename_len = strlen(u->local);
1273 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1275 if (opt.noclobber && file_exists_p (u->local))
1277 /* If opt.noclobber is turned on and file already exists, do not
1278 retrieve the file */
1279 logprintf (LOG_VERBOSE, _("\
1280 File `%s' already there, will not retrieve.\n"), u->local);
1281 /* If the file is there, we suppose it's retrieved OK. */
1284 /* #### Bogusness alert. */
1285 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1286 text/html, a harmless lie. */
1287 if (((suf = suffix (u->local)) != NULL)
1288 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1291 xfree (filename_plus_orig_suffix); /* must precede every return! */
1292 /* Another harmless lie: */
1297 if (opt.timestamping)
1299 boolean local_dot_orig_file_exists = FALSE;
1301 if (opt.backup_converted)
1302 /* If -K is specified, we'll act on the assumption that it was specified
1303 last time these files were downloaded as well, and instead of just
1304 comparing local file X against server file X, we'll compare local
1305 file X.orig (if extant, else X) against server file X. If -K
1306 _wasn't_ specified last time, or the server contains files called
1307 *.orig, -N will be back to not operating correctly with -k. */
1309 /* Would a single s[n]printf() call be faster? --dan
1311 It wouldn't. sprintf() is horribly slow. At one point I
1312 profiled Wget, and found that a measurable and
1313 non-negligible amount of time was lost calling sprintf()
1314 in url.c. Replacing sprintf with inline calls to
1315 strcpy() and long_to_string() made a difference.
1317 strcpy(filename_plus_orig_suffix, u->local);
1318 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1320 /* Try to stat() the .orig file. */
1321 if (stat(filename_plus_orig_suffix, &st) == 0)
1323 local_dot_orig_file_exists = TRUE;
1324 local_filename = filename_plus_orig_suffix;
1328 if (!local_dot_orig_file_exists)
1329 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1330 if (stat (u->local, &st) == 0)
1331 local_filename = u->local;
1333 if (local_filename != NULL)
1334 /* There was a local file, so we'll check later to see if the version
1335 the server has is the same version we already have, allowing us to
1340 local_size = st.st_size;
1344 /* Reset the counter. */
1346 *dt = 0 | ACCEPTRANGES;
1350 /* Increment the pass counter. */
1352 sleep_between_retrievals (count);
1353 /* Get the current time string. */
1354 tms = time_str (NULL);
1355 /* Print fetch message, if opt.verbose. */
1358 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1362 sprintf (tmp, _("(try:%2d)"), count);
1363 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1364 tms, hurl, tmp, locf);
1366 ws_changetitle (hurl, 1);
1371 /* Default document type is empty. However, if spider mode is
1372 on or time-stamping is employed, HEAD_ONLY commands is
1373 encoded within *dt. */
1374 if (opt.spider || (use_ts && !got_head))
1378 /* Assume no restarting. */
1380 /* Decide whether or not to restart. */
1381 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1382 && file_exists_p (u->local))
1383 if (stat (u->local, &st) == 0)
1384 hstat.restval = st.st_size;
1385 /* Decide whether to send the no-cache directive. */
1386 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1387 *dt |= SEND_NOCACHE;
1389 *dt &= ~SEND_NOCACHE;
1391 /* Try fetching the document, or at least its head. :-) */
1392 err = gethttp (u, &hstat, dt);
1394 /* It's unfortunate that wget determines the local filename before finding
1395 out the Content-Type of the file. Barring a major restructuring of the
1396 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1397 u->local to tack on ".html". */
1398 if (!opt.output_document)
1401 locf = opt.output_document;
1404 tms = time_str (NULL);
1405 /* Get the new location (with or without the redirection). */
1407 *newloc = xstrdup (hstat.newloc);
1410 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1411 case CONERROR: case READERR: case WRITEFAILED:
1413 /* Non-fatal errors continue executing the loop, which will
1414 bring them to "while" statement at the end, to judge
1415 whether the number of tries was exceeded. */
1417 printwhat (count, opt.ntry);
1420 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1421 /* Fatal errors just return from the function. */
1423 xfree (filename_plus_orig_suffix); /* must precede every return! */
1426 case FWRITEERR: case FOPENERR:
1427 /* Another fatal error. */
1428 logputs (LOG_VERBOSE, "\n");
1429 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1430 u->local, strerror (errno));
1435 /* Another fatal error. */
1436 logputs (LOG_VERBOSE, "\n");
1437 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1439 xfree (filename_plus_orig_suffix); /* must precede every return! */
1443 /* Return the new location to the caller. */
1446 logprintf (LOG_NOTQUIET,
1447 _("ERROR: Redirection (%d) without location.\n"),
1449 xfree (filename_plus_orig_suffix); /* must precede every return! */
1453 xfree (filename_plus_orig_suffix); /* must precede every return! */
1457 /* Deal with you later. */
1460 /* All possibilities should have been exhausted. */
1463 if (!(*dt & RETROKF))
1467 /* #### Ugly ugly ugly! */
1468 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1469 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1472 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1473 tms, hstat.statcode, hstat.error);
1474 logputs (LOG_VERBOSE, "\n");
1476 xfree (filename_plus_orig_suffix); /* must precede every return! */
1480 /* Did we get the time-stamp? */
1483 if (opt.timestamping && !hstat.remote_time)
1485 logputs (LOG_NOTQUIET, _("\
1486 Last-modified header missing -- time-stamps turned off.\n"));
1488 else if (hstat.remote_time)
1490 /* Convert the date-string into struct tm. */
1491 tmr = http_atotm (hstat.remote_time);
1492 if (tmr == (time_t) (-1))
1493 logputs (LOG_VERBOSE, _("\
1494 Last-modified header invalid -- time-stamp ignored.\n"));
1498 /* The time-stamping section. */
1503 use_ts = 0; /* no more time-stamping */
1504 count = 0; /* the retrieve count for HEAD is
1506 if (hstat.remote_time && tmr != (time_t) (-1))
1508 /* Now time-stamping can be used validly. Time-stamping
1509 means that if the sizes of the local and remote file
1510 match, and local file is newer than the remote file,
1511 it will not be retrieved. Otherwise, the normal
1512 download procedure is resumed. */
1514 (hstat.contlen == -1 || local_size == hstat.contlen))
1516 logprintf (LOG_VERBOSE, _("\
1517 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1520 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1523 else if (tml >= tmr)
1524 logprintf (LOG_VERBOSE, _("\
1525 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1527 logputs (LOG_VERBOSE,
1528 _("Remote file is newer, retrieving.\n"));
1533 if ((tmr != (time_t) (-1))
1535 && ((hstat.len == hstat.contlen) ||
1536 ((hstat.res == 0) &&
1537 ((hstat.contlen == -1) ||
1538 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1540 /* #### This code repeats in http.c and ftp.c. Move it to a
1542 const char *fl = NULL;
1543 if (opt.output_document)
1545 if (opt.od_known_regular)
1546 fl = opt.output_document;
1553 /* End of time-stamping section. */
1557 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1558 xfree (filename_plus_orig_suffix); /* must precede every return! */
1562 /* It is now safe to free the remainder of hstat, since the
1563 strings within it will no longer be used. */
1566 tmrate = rate (hstat.len - hstat.restval, hstat.dltime);
1568 if (hstat.len == hstat.contlen)
1572 logprintf (LOG_VERBOSE,
1573 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1574 tms, tmrate, locf, hstat.len, hstat.contlen);
1575 logprintf (LOG_NONVERBOSE,
1576 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1577 tms, u->url, hstat.len, hstat.contlen, locf, count);
1580 downloaded_increase (hstat.len);
1582 /* Remember that we downloaded the file for later ".orig" code. */
1583 if (*dt & ADDED_HTML_EXTENSION)
1584 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1586 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1588 xfree(filename_plus_orig_suffix); /* must precede every return! */
1591 else if (hstat.res == 0) /* No read error */
1593 if (hstat.contlen == -1) /* We don't know how much we were supposed
1594 to get, so assume we succeeded. */
1598 logprintf (LOG_VERBOSE,
1599 _("%s (%s) - `%s' saved [%ld]\n\n"),
1600 tms, tmrate, locf, hstat.len);
1601 logprintf (LOG_NONVERBOSE,
1602 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1603 tms, u->url, hstat.len, locf, count);
1606 downloaded_increase (hstat.len);
1608 /* Remember that we downloaded the file for later ".orig" code. */
1609 if (*dt & ADDED_HTML_EXTENSION)
1610 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1612 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1614 xfree (filename_plus_orig_suffix); /* must precede every return! */
1617 else if (hstat.len < hstat.contlen) /* meaning we lost the
1618 connection too soon */
1620 logprintf (LOG_VERBOSE,
1621 _("%s (%s) - Connection closed at byte %ld. "),
1622 tms, tmrate, hstat.len);
1623 printwhat (count, opt.ntry);
1626 else if (!opt.kill_longer) /* meaning we got more than expected */
1628 logprintf (LOG_VERBOSE,
1629 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1630 tms, tmrate, locf, hstat.len, hstat.contlen);
1631 logprintf (LOG_NONVERBOSE,
1632 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1633 tms, u->url, hstat.len, hstat.contlen, locf, count);
1635 downloaded_increase (hstat.len);
1637 /* Remember that we downloaded the file for later ".orig" code. */
1638 if (*dt & ADDED_HTML_EXTENSION)
1639 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1641 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1643 xfree (filename_plus_orig_suffix); /* must precede every return! */
1646 else /* the same, but not accepted */
1648 logprintf (LOG_VERBOSE,
1649 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1650 tms, tmrate, hstat.len, hstat.contlen);
1651 printwhat (count, opt.ntry);
1655 else /* now hstat.res can only be -1 */
1657 if (hstat.contlen == -1)
1659 logprintf (LOG_VERBOSE,
1660 _("%s (%s) - Read error at byte %ld (%s)."),
1661 tms, tmrate, hstat.len, strerror (errno));
1662 printwhat (count, opt.ntry);
1665 else /* hstat.res == -1 and contlen is given */
1667 logprintf (LOG_VERBOSE,
1668 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1669 tms, tmrate, hstat.len, hstat.contlen,
1671 printwhat (count, opt.ntry);
1678 while (!opt.ntry || (count < opt.ntry));
1679 xfree (filename_plus_orig_suffix); /* must precede every return! */
1683 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1684 than local timezone (mktime assumes the latter).
1686 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1687 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1689 mktime_from_utc (struct tm *t)
1696 tb = mktime (gmtime (&tl));
1697 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1700 /* Check whether the result of strptime() indicates success.
1701 strptime() returns the pointer to how far it got to in the string.
1702 The processing has been successful if the string is at `GMT' or
1703 `+X', or at the end of the string.
1705 In extended regexp parlance, the function returns 1 if P matches
1706 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1707 strptime()) is considered a failure and 0 is returned. */
1709 check_end (const char *p)
1713 while (ISSPACE (*p))
1716 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1717 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1723 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1724 the three formats RFC2068 allows the HTTP servers to emit --
1725 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1728 We use strptime() to recognize various dates, which makes it a
1729 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1730 allows shortened dates and months, one-digit days, etc.). It also
1731 allows more than one space anywhere where the specs require one SP.
1732 The routine should probably be even more forgiving (as recommended
1733 by RFC2068), but I do not have the time to write one.
1735 Return the computed time_t representation, or -1 if all the
1738 Needless to say, what we *really* need here is something like
1739 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1740 and does not use strptime(). atotm() is to be found in the sources
1741 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1743 http_atotm (char *time_string)
1747 /* Roger Beeman says: "This function dynamically allocates struct tm
1748 t, but does no initialization. The only field that actually
1749 needs initialization is tm_isdst, since the others will be set by
1750 strptime. Since strptime does not set tm_isdst, it will return
1751 the data structure with whatever data was in tm_isdst to begin
1752 with. For those of us in timezones where DST can occur, there
1753 can be a one hour shift depending on the previous contents of the
1754 data area where the data structure is allocated." */
1757 /* Note that under foreign locales Solaris strptime() fails to
1758 recognize English dates, which renders this function useless. I
1759 assume that other non-GNU strptime's are plagued by the same
1760 disease. We solve this by setting only LC_MESSAGES in
1761 i18n_initialize(), instead of LC_ALL.
1763 Another solution could be to temporarily set locale to C, invoke
1764 strptime(), and restore it back. This is slow and dirty,
1765 however, and locale support other than LC_MESSAGES can mess other
1766 things, so I rather chose to stick with just setting LC_MESSAGES.
1768 Also note that none of this is necessary under GNU strptime(),
1769 because it recognizes both international and local dates. */
1771 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1772 it to eat all white space up to (and including) a newline, and
1773 the function fails if there is no newline (!).
1775 Let's hope all strptime() implementations use ` ' to skip *all*
1776 whitespace instead of just one (it works that way on all the
1777 systems I've tested it on). */
1779 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1780 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1781 return mktime_from_utc (&t);
1782 /* RFC850: Thu, 29-Jan-98 22:12:57 */
1783 if (check_end (strptime (time_string, "%a, %d-%b-%y %T", &t)))
1784 return mktime_from_utc (&t);
1785 /* asctime: Thu Jan 29 22:12:57 1998 */
1786 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1787 return mktime_from_utc (&t);
1792 /* Authorization support: We support two authorization schemes:
1794 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1796 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1797 consisting of answering to the server's challenge with the proper
1800 /* How many bytes it will take to store LEN bytes in base64. */
1801 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1803 /* Encode the string S of length LENGTH to base64 format and place it
1804 to STORE. STORE will be 0-terminated, and must point to a writable
1805 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1807 base64_encode (const char *s, char *store, int length)
1809 /* Conversion table. */
1810 static char tbl[64] = {
1811 'A','B','C','D','E','F','G','H',
1812 'I','J','K','L','M','N','O','P',
1813 'Q','R','S','T','U','V','W','X',
1814 'Y','Z','a','b','c','d','e','f',
1815 'g','h','i','j','k','l','m','n',
1816 'o','p','q','r','s','t','u','v',
1817 'w','x','y','z','0','1','2','3',
1818 '4','5','6','7','8','9','+','/'
1821 unsigned char *p = (unsigned char *)store;
1823 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1824 for (i = 0; i < length; i += 3)
1826 *p++ = tbl[s[0] >> 2];
1827 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1828 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1829 *p++ = tbl[s[2] & 0x3f];
1832 /* Pad the result if necessary... */
1833 if (i == length + 1)
1835 else if (i == length + 2)
1836 *(p - 1) = *(p - 2) = '=';
1837 /* ...and zero-terminate it. */
1841 /* Create the authentication header contents for the `Basic' scheme.
1842 This is done by encoding the string `USER:PASS' in base64 and
1843 prepending `HEADER: Basic ' to it. */
1845 basic_authentication_encode (const char *user, const char *passwd,
1848 char *t1, *t2, *res;
1849 int len1 = strlen (user) + 1 + strlen (passwd);
1850 int len2 = BASE64_LENGTH (len1);
1852 t1 = (char *)alloca (len1 + 1);
1853 sprintf (t1, "%s:%s", user, passwd);
1854 t2 = (char *)alloca (1 + len2);
1855 base64_encode (t1, t2, len1);
1856 res = (char *)xmalloc (len2 + 11 + strlen (header));
1857 sprintf (res, "%s: Basic %s\r\n", header, t2);
1863 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
1864 of a field in such a header. If the field is the one specified by
1865 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
1866 digest authorization code), extract its value in the (char*)
1867 variable pointed by RET. Returns negative on a malformed header,
1868 or number of bytes that have been parsed by this call. */
1870 extract_header_attr (const char *au, const char *attr_name, char **ret)
1872 const char *cp, *ep;
1876 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
1878 cp += strlen (attr_name);
1881 cp += skip_lws (cp);
1886 cp += skip_lws (cp);
1891 for (ep = cp; *ep && *ep != '\"'; ep++)
1896 *ret = strdupdelim (cp, ep);
1903 /* Response value needs to be in lowercase, so we cannot use HEXD2ASC
1904 from url.h. See RFC 2069 2.1.2 for the syntax of response-digest. */
1905 #define HEXD2asc(x) (((x) < 10) ? ((x) + '0') : ((x) - 10 + 'a'))
1907 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
1908 an array of 16 bytes containing the hash keys, and BUF should be a
1909 buffer of 33 writable characters (32 for hex digits plus one for
1910 zero termination). */
1912 dump_hash (unsigned char *buf, const unsigned char *hash)
1916 for (i = 0; i < MD5_HASHLEN; i++, hash++)
1918 *buf++ = HEXD2asc (*hash >> 4);
1919 *buf++ = HEXD2asc (*hash & 0xf);
1924 /* Take the line apart to find the challenge, and compose a digest
1925 authorization header. See RFC2069 section 2.1.2. */
1927 digest_authentication_encode (const char *au, const char *user,
1928 const char *passwd, const char *method,
1931 static char *realm, *opaque, *nonce;
1936 { "realm", &realm },
1937 { "opaque", &opaque },
1942 realm = opaque = nonce = NULL;
1944 au += 6; /* skip over `Digest' */
1949 au += skip_lws (au);
1950 for (i = 0; i < ARRAY_SIZE (options); i++)
1952 int skip = extract_header_attr (au, options[i].name,
1953 options[i].variable);
1957 FREE_MAYBE (opaque);
1967 if (i == ARRAY_SIZE (options))
1969 while (*au && *au != '=')
1973 au += skip_lws (au);
1977 while (*au && *au != '\"')
1984 while (*au && *au != ',')
1989 if (!realm || !nonce || !user || !passwd || !path || !method)
1992 FREE_MAYBE (opaque);
1997 /* Calculate the digest value. */
2000 unsigned char hash[MD5_HASHLEN];
2001 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2002 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2004 /* A1BUF = H(user ":" realm ":" password) */
2005 md5_init_ctx (&ctx);
2006 md5_process_bytes (user, strlen (user), &ctx);
2007 md5_process_bytes (":", 1, &ctx);
2008 md5_process_bytes (realm, strlen (realm), &ctx);
2009 md5_process_bytes (":", 1, &ctx);
2010 md5_process_bytes (passwd, strlen (passwd), &ctx);
2011 md5_finish_ctx (&ctx, hash);
2012 dump_hash (a1buf, hash);
2014 /* A2BUF = H(method ":" path) */
2015 md5_init_ctx (&ctx);
2016 md5_process_bytes (method, strlen (method), &ctx);
2017 md5_process_bytes (":", 1, &ctx);
2018 md5_process_bytes (path, strlen (path), &ctx);
2019 md5_finish_ctx (&ctx, hash);
2020 dump_hash (a2buf, hash);
2022 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2023 md5_init_ctx (&ctx);
2024 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2025 md5_process_bytes (":", 1, &ctx);
2026 md5_process_bytes (nonce, strlen (nonce), &ctx);
2027 md5_process_bytes (":", 1, &ctx);
2028 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2029 md5_finish_ctx (&ctx, hash);
2030 dump_hash (response_digest, hash);
2032 res = (char*) xmalloc (strlen (user)
2037 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2038 + (opaque ? strlen (opaque) : 0)
2040 sprintf (res, "Authorization: Digest \
2041 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2042 user, realm, nonce, path, response_digest);
2045 char *p = res + strlen (res);
2046 strcat (p, ", opaque=\"");
2050 strcat (res, "\r\n");
2054 #endif /* USE_DIGEST */
2057 #define BEGINS_WITH(line, string_constant) \
2058 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2059 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2060 || !line[sizeof (string_constant) - 1]))
2063 known_authentication_scheme_p (const char *au)
2065 return BEGINS_WITH (au, "Basic")
2066 || BEGINS_WITH (au, "Digest")
2067 || BEGINS_WITH (au, "NTLM");
2072 /* Create the HTTP authorization request header. When the
2073 `WWW-Authenticate' response header is seen, according to the
2074 authorization scheme specified in that header (`Basic' and `Digest'
2075 are supported by the current implementation), produce an
2076 appropriate HTTP authorization request header. */
2078 create_authorization_line (const char *au, const char *user,
2079 const char *passwd, const char *method,
2082 char *wwwauth = NULL;
2084 if (!strncasecmp (au, "Basic", 5))
2085 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2086 if (!strncasecmp (au, "NTLM", 4))
2087 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2089 else if (!strncasecmp (au, "Digest", 6))
2090 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2091 #endif /* USE_DIGEST */