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>
35 #if TIME_WITH_SYS_TIME
36 # include <sys/time.h>
40 # include <sys/time.h>
49 # include <netdb.h> /* for h_errno */
66 # include "gen_sslfunc.h"
70 extern char *version_string;
81 static int cookies_loaded_p;
83 #define TEXTHTML_S "text/html"
84 #define HTTP_ACCEPT "*/*"
86 /* Some status code validation macros: */
87 #define H_20X(x) (((x) >= 200) && ((x) < 300))
88 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
89 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
90 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
92 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
94 #define HTTP_STATUS_OK 200
95 #define HTTP_STATUS_CREATED 201
96 #define HTTP_STATUS_ACCEPTED 202
97 #define HTTP_STATUS_NO_CONTENT 204
98 #define HTTP_STATUS_PARTIAL_CONTENTS 206
100 /* Redirection 3xx. */
101 #define HTTP_STATUS_MULTIPLE_CHOICES 300
102 #define HTTP_STATUS_MOVED_PERMANENTLY 301
103 #define HTTP_STATUS_MOVED_TEMPORARILY 302
104 #define HTTP_STATUS_NOT_MODIFIED 304
106 /* Client error 4xx. */
107 #define HTTP_STATUS_BAD_REQUEST 400
108 #define HTTP_STATUS_UNAUTHORIZED 401
109 #define HTTP_STATUS_FORBIDDEN 403
110 #define HTTP_STATUS_NOT_FOUND 404
112 /* Server errors 5xx. */
113 #define HTTP_STATUS_INTERNAL 500
114 #define HTTP_STATUS_NOT_IMPLEMENTED 501
115 #define HTTP_STATUS_BAD_GATEWAY 502
116 #define HTTP_STATUS_UNAVAILABLE 503
119 /* Parse the HTTP status line, which is of format:
121 HTTP-Version SP Status-Code SP Reason-Phrase
123 The function returns the status-code, or -1 if the status line is
124 malformed. The pointer to reason-phrase is returned in RP. */
126 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
128 /* (the variables must not be named `major' and `minor', because
129 that breaks compilation with SunOS4 cc.) */
130 int mjr, mnr, statcode;
133 *reason_phrase_ptr = NULL;
135 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
136 major version, and Y is minor version. */
137 if (strncmp (line, "HTTP/", 5) != 0)
141 /* Calculate major HTTP version. */
143 for (mjr = 0; ISDIGIT (*line); line++)
144 mjr = 10 * mjr + (*line - '0');
145 if (*line != '.' || p == line)
149 /* Calculate minor HTTP version. */
151 for (mnr = 0; ISDIGIT (*line); line++)
152 mnr = 10 * mnr + (*line - '0');
153 if (*line != ' ' || p == line)
155 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
156 minor version can be safely ignored. */
161 /* Calculate status code. */
162 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
164 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
166 /* Set up the reason phrase pointer. */
168 /* RFC2068 requires SPC here, but we allow the string to finish
169 here, in case no reason-phrase is present. */
173 *reason_phrase_ptr = line;
178 *reason_phrase_ptr = line + 1;
183 /* Functions to be used as arguments to header_process(): */
185 struct http_process_range_closure {
191 /* Parse the `Content-Range' header and extract the information it
192 contains. Returns 1 if successful, -1 otherwise. */
194 http_process_range (const char *hdr, void *arg)
196 struct http_process_range_closure *closure
197 = (struct http_process_range_closure *)arg;
200 /* Certain versions of Nutscape proxy server send out
201 `Content-Length' without "bytes" specifier, which is a breach of
202 RFC2068 (as well as the HTTP/1.1 draft which was current at the
203 time). But hell, I must support it... */
204 if (!strncasecmp (hdr, "bytes", 5))
207 hdr += skip_lws (hdr);
213 for (num = 0; ISDIGIT (*hdr); hdr++)
214 num = 10 * num + (*hdr - '0');
215 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
217 closure->first_byte_pos = num;
219 for (num = 0; ISDIGIT (*hdr); hdr++)
220 num = 10 * num + (*hdr - '0');
221 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
223 closure->last_byte_pos = num;
225 for (num = 0; ISDIGIT (*hdr); hdr++)
226 num = 10 * num + (*hdr - '0');
227 closure->entity_length = num;
231 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
232 Used for `Accept-Ranges'. */
234 http_process_none (const char *hdr, void *arg)
236 int *where = (int *)arg;
238 if (strstr (hdr, "none"))
245 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
247 http_process_type (const char *hdr, void *arg)
249 char **result = (char **)arg;
250 /* Locate P on `;' or the terminating zero, whichever comes first. */
251 const char *p = strchr (hdr, ';');
253 p = hdr + strlen (hdr);
254 while (p > hdr && ISSPACE (*(p - 1)))
256 *result = strdupdelim (hdr, p);
260 /* Check whether the `Connection' header is set to "keep-alive". */
262 http_process_connection (const char *hdr, void *arg)
264 int *flag = (int *)arg;
265 if (!strcasecmp (hdr, "Keep-Alive"))
270 /* Persistent connections. Currently, we cache the most recently used
271 connection as persistent, provided that the HTTP server agrees to
272 make it such. The persistence data is stored in the variables
273 below. Ideally, it would be in a structure, and it should be
274 possible to cache an arbitrary fixed number of these connections.
276 I think the code is quite easy to extend in that direction. */
278 /* Whether a persistent connection is active. */
279 static int pc_active_p;
280 /* Host and port of currently active persistent connection. */
281 static unsigned char pc_last_host[4];
282 static unsigned short pc_last_port;
284 /* File descriptor of the currently active persistent connection. */
285 static int pc_last_fd;
288 /* Whether a ssl handshake has occoured on this connection */
289 static int pc_active_ssl;
290 /* SSL connection of the currently active persistent connection. */
291 static SSL *pc_last_ssl;
292 #endif /* HAVE_SSL */
294 /* Mark the persistent connection as invalid. This is used by the
295 CLOSE_* macros after they forcefully close a registered persistent
296 connection. This does not close the file descriptor -- it is left
297 to the caller to do that. (Maybe it should, though.) */
300 invalidate_persistent (void)
305 #endif /* HAVE_SSL */
306 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
309 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
310 persistent. This will enable someone to use the same connection
311 later. In the context of HTTP, this must be called only AFTER the
312 response has been received and the server has promised that the
313 connection will remain alive.
315 If a previous connection was persistent, it is closed. */
318 register_persistent (const char *host, unsigned short port, int fd
328 if (pc_last_fd == fd)
330 /* The connection FD is already registered. Nothing to
336 /* The old persistent connection is still active; let's
337 close it first. This situation arises whenever a
338 persistent connection exists, but we then connect to a
339 different host, and try to register a persistent
340 connection to that one. */
342 /* The ssl disconnect has to take place before the closing
345 shutdown_ssl(pc_last_ssl);
348 invalidate_persistent ();
352 /* This store_hostaddress may not fail, because it has the results
354 success = store_hostaddress (pc_last_host, host);
361 pc_active_ssl = ssl ? 1 : 0;
363 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
366 /* Return non-zero if a persistent connection is available for
367 connecting to HOST:PORT. */
370 persistent_available_p (const char *host, unsigned short port
376 unsigned char this_host[4];
377 /* First, check whether a persistent connection is active at all. */
380 /* Second, check if the active connection pertains to the correct
381 (HOST, PORT) ordered pair. */
382 if (port != pc_last_port)
385 /* Second, a): check if current connection is (not) ssl, too. This
386 test is unlikely to fail because HTTP and HTTPS typicaly use
387 different ports. Yet it is possible, or so I [Christian
388 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
390 if (ssl != pc_active_ssl)
392 #endif /* HAVE_SSL */
393 if (!store_hostaddress (this_host, host))
395 if (memcmp (pc_last_host, this_host, 4))
397 /* Third: check whether the connection is still open. This is
398 important because most server implement a liberal (short) timeout
399 on persistent connections. Wget can of course always reconnect
400 if the connection doesn't work out, but it's nicer to know in
401 advance. This test is a logical followup of the first test, but
402 is "expensive" and therefore placed at the end of the list. */
403 if (!test_socket_open (pc_last_fd))
405 /* Oops, the socket is no longer open. Now that we know that,
406 let's invalidate the persistent connection before returning
409 invalidate_persistent ();
416 # define SHUTDOWN_SSL(ssl) do { \
418 shutdown_ssl (ssl); \
421 # define SHUTDOWN_SSL(ssl)
424 /* The idea behind these two CLOSE macros is to distinguish between
425 two cases: one when the job we've been doing is finished, and we
426 want to close the connection and leave, and two when something is
427 seriously wrong and we're closing the connection as part of
430 In case of keep_alive, CLOSE_FINISH should leave the connection
431 open, while CLOSE_INVALIDATE should still close it.
433 Note that the semantics of the flag `keep_alive' is "this
434 connection *will* be reused (the server has promised not to close
435 the connection once we're done)", while the semantics of
436 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
437 active, registered connection". */
439 #define CLOSE_FINISH(fd) do { \
442 SHUTDOWN_SSL (ssl); \
444 if (pc_active_p && (fd) == pc_last_fd) \
445 invalidate_persistent (); \
449 #define CLOSE_INVALIDATE(fd) do { \
450 SHUTDOWN_SSL (ssl); \
452 if (pc_active_p && (fd) == pc_last_fd) \
453 invalidate_persistent (); \
458 long len; /* received length */
459 long contlen; /* expected length */
460 long restval; /* the restart value */
461 int res; /* the result of last read */
462 char *newloc; /* new location (redirection) */
463 char *remote_time; /* remote time-stamp string */
464 char *error; /* textual HTTP error */
465 int statcode; /* status code */
466 long dltime; /* time of the download */
467 int no_truncate; /* whether truncating the file is
471 /* Free the elements of hstat X. */
472 #define FREEHSTAT(x) do \
474 FREE_MAYBE ((x).newloc); \
475 FREE_MAYBE ((x).remote_time); \
476 FREE_MAYBE ((x).error); \
477 (x).newloc = (x).remote_time = (x).error = NULL; \
480 static char *create_authorization_line PARAMS ((const char *, const char *,
481 const char *, const char *,
483 static char *basic_authentication_encode PARAMS ((const char *, const char *,
485 static int known_authentication_scheme_p PARAMS ((const char *));
487 time_t http_atotm PARAMS ((char *));
489 #define BEGINS_WITH(line, string_constant) \
490 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
491 && (ISSPACE (line[sizeof (string_constant) - 1]) \
492 || !line[sizeof (string_constant) - 1]))
494 /* Retrieve a document through HTTP protocol. It recognizes status
495 code, and correctly handles redirections. It closes the network
496 socket. If it receives an error from the functions below it, it
497 will print it if there is enough information to do so (almost
498 always), returning the error to the caller (i.e. http_loop).
500 Various HTTP parameters are stored to hs. Although it parses the
501 response code correctly, it is not used in a sane way. The caller
504 If u->proxy is non-NULL, the URL u will be taken as a proxy URL,
505 and u->proxy->url will be given to the proxy server (bad naming,
508 gethttp (struct urlinfo *u, struct http_stat *hs, int *dt)
510 char *request, *type, *command, *path;
512 char *pragma_h, *referer, *useragent, *range, *wwwauth, *remhost;
513 char *authenticate_h;
517 char *request_keep_alive;
518 int sock, hcount, num_written, all_length, remport, statcode;
519 long contlen, contrange;
523 int auth_tried_already;
526 static SSL_CTX *ssl_ctx = NULL;
528 #endif /* HAVE_SSL */
529 char *cookies = NULL;
531 /* Whether this connection will be kept alive after the HTTP request
535 /* Flags that detect the two ways of specifying HTTP keep-alive
537 int http_keep_alive_1, http_keep_alive_2;
539 /* Whether keep-alive should be inhibited. */
540 int inhibit_keep_alive;
543 /* initialize ssl_ctx on first run */
546 err=init_ssl (&ssl_ctx);
551 case SSLERRCTXCREATE:
553 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
557 /* try without certfile */
558 logprintf (LOG_NOTQUIET,
559 _("Failed to load certificates from %s\n"),
562 logprintf (LOG_NOTQUIET,
563 _("Trying without the specified certificate\n"));
566 logprintf (LOG_NOTQUIET,
567 _("Failed to get certificate key from %s\n"),
570 logprintf (LOG_NOTQUIET,
571 _("Trying without the specified certificate\n"));
578 #endif /* HAVE_SSL */
580 if (!(*dt & HEAD_ONLY))
581 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
582 know the local filename so we can save to it. */
583 assert (u->local != NULL);
586 auth_tried_already = 0;
588 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
591 /* We need to come back here when the initial attempt to retrieve
592 without authorization header fails. (Expected to happen at least
593 for the Digest authorization scheme.) */
596 http_keep_alive_1 = http_keep_alive_2 = 0;
599 cookies = build_cookies_request (u->host, u->port, u->path,
600 u->proto == URLHTTPS);
602 /* Initialize certain elements of struct http_stat. */
607 hs->remote_time = NULL;
610 /* Which structure to use to retrieve the original URL data. */
616 /* First: establish the connection. */
617 if (inhibit_keep_alive
620 !persistent_available_p (u->host, u->port)
622 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
623 #endif /* HAVE_SSL */
626 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
627 err = make_connection (&sock, u->host, u->port);
631 logputs (LOG_VERBOSE, "\n");
632 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
636 logputs (LOG_VERBOSE, "\n");
637 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
641 logputs (LOG_VERBOSE, "\n");
642 logprintf (LOG_NOTQUIET,
643 _("Connection to %s:%hu refused.\n"), u->host, u->port);
647 logputs (LOG_VERBOSE, "\n");
648 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
653 /* Everything is fine! */
654 logputs (LOG_VERBOSE, _("connected!\n"));
661 if (u->proto == URLHTTPS)
662 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
664 logputs (LOG_VERBOSE, "\n");
665 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
669 #endif /* HAVE_SSL */
673 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
674 /* #### pc_last_fd should be accessed through an accessor
679 #endif /* HAVE_SSL */
680 DEBUGP (("Reusing fd %d.\n", sock));
684 path = u->proxy->url;
688 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
692 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
693 sprintf (referer, "Referer: %s\r\n", ou->referer);
695 if (*dt & SEND_NOCACHE)
696 pragma_h = "Pragma: no-cache\r\n";
701 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
702 /* Gag me! Some servers (e.g. WebSitePro) have been known to
703 respond to the following `Range' format by generating a
704 multipart/x-byte-ranges MIME document! This MIME type was
705 present in an old draft of the byteranges specification.
706 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
707 only if multiple non-overlapping ranges are requested --
708 which Wget never does. */
709 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
714 STRDUP_ALLOCA (useragent, opt.useragent);
717 useragent = (char *)alloca (10 + strlen (version_string));
718 sprintf (useragent, "Wget/%s", version_string);
720 /* Construct the authentication, if userid is present. */
723 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
724 user = user ? user : opt.http_user;
725 passwd = passwd ? passwd : opt.http_passwd;
732 /* We have the username and the password, but haven't tried
733 any authorization yet. Let's see if the "Basic" method
734 works. If not, we'll come back here and construct a
735 proper authorization method with the right challenges.
737 If we didn't employ this kind of logic, every URL that
738 requires authorization would have to be processed twice,
739 which is very suboptimal and generates a bunch of false
740 "unauthorized" errors in the server log.
742 #### But this logic also has a serious problem when used
743 with stronger authentications: we *first* transmit the
744 username and the password in clear text, and *then*
745 attempt a stronger authentication scheme. That cannot be
746 right! We are only fortunate that almost everyone still
747 uses the `Basic' scheme anyway.
749 There should be an option to prevent this from happening,
750 for those who use strong authentication schemes and value
752 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
756 wwwauth = create_authorization_line (authenticate_h, user, passwd,
764 char *proxy_user, *proxy_passwd;
765 /* For normal username and password, URL components override
766 command-line/wgetrc parameters. With proxy authentication,
767 it's the reverse, because proxy URLs are normally the
768 "permanent" ones, so command-line args should take
770 if (opt.proxy_user && opt.proxy_passwd)
772 proxy_user = opt.proxy_user;
773 proxy_passwd = opt.proxy_passwd;
777 proxy_user = u->user;
778 proxy_passwd = u->passwd;
780 /* #### This is junky. Can't the proxy request, say, `Digest'
782 if (proxy_user && proxy_passwd)
783 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
784 "Proxy-Authorization");
789 /* String of the form :PORT. Used only for non-standard ports. */
793 && remport != (u->proto == URLHTTPS
794 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
796 && remport != DEFAULT_HTTP_PORT
800 port_maybe = (char *)alloca (numdigit (remport) + 2);
801 sprintf (port_maybe, ":%d", remport);
804 if (!inhibit_keep_alive)
805 request_keep_alive = "Connection: Keep-Alive\r\n";
807 request_keep_alive = NULL;
809 /* Allocate the memory for the request. */
810 request = (char *)alloca (strlen (command) + strlen (path)
813 + (port_maybe ? strlen (port_maybe) : 0)
814 + strlen (HTTP_ACCEPT)
815 + (request_keep_alive
816 ? strlen (request_keep_alive) : 0)
817 + (referer ? strlen (referer) : 0)
818 + (cookies ? strlen (cookies) : 0)
819 + (wwwauth ? strlen (wwwauth) : 0)
820 + (proxyauth ? strlen (proxyauth) : 0)
821 + (range ? strlen (range) : 0)
823 + (opt.user_header ? strlen (opt.user_header) : 0)
825 /* Construct the request. */
831 %s%s%s%s%s%s%s%s\r\n",
832 command, path, useragent, remhost,
833 port_maybe ? port_maybe : "",
835 request_keep_alive ? request_keep_alive : "",
836 referer ? referer : "",
837 cookies ? cookies : "",
838 wwwauth ? wwwauth : "",
839 proxyauth ? proxyauth : "",
842 opt.user_header ? opt.user_header : "");
843 DEBUGP (("---request begin---\n%s---request end---\n", request));
844 /* Free the temporary memory. */
845 FREE_MAYBE (wwwauth);
846 FREE_MAYBE (proxyauth);
847 FREE_MAYBE (cookies);
849 /* Send the request to server. */
851 if (u->proto == URLHTTPS)
852 num_written = ssl_iwrite (ssl, request, strlen (request));
854 #endif /* HAVE_SSL */
855 num_written = iwrite (sock, request, strlen (request));
859 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
861 CLOSE_INVALIDATE (sock);
864 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
865 u->proxy ? "Proxy" : "HTTP");
866 contlen = contrange = -1;
871 /* Before reading anything, initialize the rbuf. */
872 rbuf_initialize (&rbuf, sock);
874 if (u->proto == URLHTTPS)
878 #endif /* HAVE_SSL */
881 /* Header-fetching loop. */
889 /* Get the header. */
890 status = header_get (&rbuf, &hdr,
891 /* Disallow continuations for status line. */
892 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
894 /* Check for errors. */
895 if (status == HG_EOF && *hdr)
897 /* This used to be an unconditional error, but that was
898 somewhat controversial, because of a large number of
899 broken CGI's that happily "forget" to send the second EOL
900 before closing the connection of a HEAD request.
902 So, the deal is to check whether the header is empty
903 (*hdr is zero if it is); if yes, it means that the
904 previous header was fully retrieved, and that -- most
905 probably -- the request is complete. "...be liberal in
906 what you accept." Oh boy. */
907 logputs (LOG_VERBOSE, "\n");
908 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
911 FREE_MAYBE (hs->newloc);
912 FREE_MAYBE (all_headers);
913 CLOSE_INVALIDATE (sock);
916 else if (status == HG_ERROR)
918 logputs (LOG_VERBOSE, "\n");
919 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
923 FREE_MAYBE (hs->newloc);
924 FREE_MAYBE (all_headers);
925 CLOSE_INVALIDATE (sock);
929 /* If the headers are to be saved to a file later, save them to
931 if (opt.save_headers)
933 int lh = strlen (hdr);
934 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
935 memcpy (all_headers + all_length, hdr, lh);
937 all_headers[all_length++] = '\n';
938 all_headers[all_length] = '\0';
941 /* Print the header if requested. */
942 if (opt.server_response && hcount != 1)
943 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
945 /* Check for status line. */
949 /* Parse the first line of server response. */
950 statcode = parse_http_status_line (hdr, &error);
951 hs->statcode = statcode;
952 /* Store the descriptive response. */
953 if (statcode == -1) /* malformed response */
955 /* A common reason for "malformed response" error is the
956 case when no data was actually received. Handle this
959 hs->error = xstrdup (_("No data received"));
961 hs->error = xstrdup (_("Malformed status line"));
966 hs->error = xstrdup (_("(no description)"));
968 hs->error = xstrdup (error);
975 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
980 /* Exit on empty header. */
987 /* Try getting content-length. */
988 if (contlen == -1 && !opt.ignore_length)
989 if (header_process (hdr, "Content-Length", header_extract_number,
992 /* Try getting content-type. */
994 if (header_process (hdr, "Content-Type", http_process_type, &type))
996 /* Try getting location. */
998 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1000 /* Try getting last-modified. */
1001 if (!hs->remote_time)
1002 if (header_process (hdr, "Last-Modified", header_strdup,
1005 /* Try getting cookies. */
1007 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1009 /* Try getting www-authentication. */
1010 if (!authenticate_h)
1011 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1014 /* Check for accept-ranges header. If it contains the word
1015 `none', disable the ranges. */
1016 if (*dt & ACCEPTRANGES)
1019 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1022 *dt &= ~ACCEPTRANGES;
1026 /* Try getting content-range. */
1027 if (contrange == -1)
1029 struct http_process_range_closure closure;
1030 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1032 contrange = closure.first_byte_pos;
1036 /* Check for keep-alive related responses. */
1037 if (!inhibit_keep_alive)
1039 /* Check for the `Keep-Alive' header. */
1040 if (!http_keep_alive_1)
1042 if (header_process (hdr, "Keep-Alive", header_exists,
1043 &http_keep_alive_1))
1046 /* Check for `Connection: Keep-Alive'. */
1047 if (!http_keep_alive_2)
1049 if (header_process (hdr, "Connection", http_process_connection,
1050 &http_keep_alive_2))
1058 logputs (LOG_VERBOSE, "\n");
1061 && (http_keep_alive_1 || http_keep_alive_2))
1063 assert (inhibit_keep_alive == 0);
1067 /* The server has promised that it will not close the connection
1068 when we're done. This means that we can register it. */
1070 register_persistent (u->host, u->port, sock);
1072 register_persistent (u->host, u->port, sock, ssl);
1073 #endif /* HAVE_SSL */
1075 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1078 /* Authorization is required. */
1082 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1083 might be more bytes in the body. */
1084 if (auth_tried_already)
1086 /* If we have tried it already, then there is not point
1089 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1090 xfree (authenticate_h);
1093 else if (!known_authentication_scheme_p (authenticate_h))
1095 xfree (authenticate_h);
1096 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1099 else if (BEGINS_WITH (authenticate_h, "Basic"))
1101 /* The authentication scheme is basic, the one we try by
1102 default, and it failed. There's no sense in trying
1108 auth_tried_already = 1;
1112 /* We do not need this anymore. */
1115 xfree (authenticate_h);
1116 authenticate_h = NULL;
1119 /* 20x responses are counted among successful by default. */
1120 if (H_20X (statcode))
1123 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1126 /* We don't assume text/html by default. */
1129 if (opt.html_extension && (*dt & TEXTHTML))
1130 /* -E / --html-extension / html_extension = on was specified, and this is a
1131 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1132 already the file's suffix, tack on ".html". */
1134 char* last_period_in_local_filename = strrchr(u->local, '.');
1136 if (last_period_in_local_filename == NULL ||
1137 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1138 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1140 size_t local_filename_len = strlen(u->local);
1142 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1143 strcpy(u->local + local_filename_len, ".html");
1145 *dt |= ADDED_HTML_EXTENSION;
1149 if (contrange == -1)
1151 /* We did not get a content-range header. This means that the
1152 server did not honor our `Range' request. Normally, this
1153 means we should reset hs->restval and continue normally. */
1155 /* However, if `-c' is used, we need to be a bit more careful:
1157 1. If `-c' is specified and the file already existed when
1158 Wget was started, it would be a bad idea for us to start
1159 downloading it from scratch, effectively truncating it. I
1160 believe this cannot happen unless `-c' was specified.
1162 2. If `-c' is used on a file that is already fully
1163 downloaded, we're requesting bytes after the end of file,
1164 which can result in server not honoring `Range'. If this is
1165 the case, `Content-Length' will be equal to the length of the
1167 if (opt.always_rest)
1169 /* Check for condition #2. */
1170 if (hs->restval == contlen)
1172 logputs (LOG_VERBOSE, _("\
1173 \n The file is already fully retrieved; nothing to do.\n\n"));
1174 /* In case the caller inspects. */
1178 FREE_MAYBE (hs->newloc);
1179 FREE_MAYBE (all_headers);
1180 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1181 might be more bytes in the body. */
1182 return RETRFINISHED;
1185 /* Check for condition #1. */
1186 if (hs->no_truncate)
1188 logprintf (LOG_NOTQUIET,
1191 The server does not support continued download;\n\
1192 refusing to truncate `%s'.\n\n"), u->local);
1193 return CONTNOTSUPPORTED;
1202 else if (contrange != hs->restval ||
1203 (H_PARTIAL (statcode) && contrange == -1))
1205 /* This means the whole request was somehow misunderstood by the
1206 server. Bail out. */
1208 FREE_MAYBE (hs->newloc);
1209 FREE_MAYBE (all_headers);
1210 CLOSE_INVALIDATE (sock);
1217 contlen += contrange;
1219 contrange = -1; /* If conent-length was not sent,
1220 content-range will be ignored. */
1222 hs->contlen = contlen;
1224 /* Return if redirected. */
1225 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1227 /* RFC2068 says that in case of the 300 (multiple choices)
1228 response, the server can output a preferred URL through
1229 `Location' header; otherwise, the request should be treated
1230 like GET. So, if the location is set, it will be a
1231 redirection; otherwise, just proceed normally. */
1232 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1236 logprintf (LOG_VERBOSE,
1237 _("Location: %s%s\n"),
1238 hs->newloc ? hs->newloc : _("unspecified"),
1239 hs->newloc ? _(" [following]") : "");
1240 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1241 might be more bytes in the body. */
1243 FREE_MAYBE (all_headers);
1249 if ((*dt & RETROKF) && !opt.server_response)
1251 /* No need to print this output if the body won't be
1252 downloaded at all, or if the original server response is
1254 logputs (LOG_VERBOSE, _("Length: "));
1257 logputs (LOG_VERBOSE, legible (contlen));
1258 if (contrange != -1)
1259 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1260 legible (contlen - contrange));
1263 logputs (LOG_VERBOSE,
1264 opt.ignore_length ? _("ignored") : _("unspecified"));
1266 logprintf (LOG_VERBOSE, " [%s]\n", type);
1268 logputs (LOG_VERBOSE, "\n");
1272 type = NULL; /* We don't need it any more. */
1274 /* Return if we have no intention of further downloading. */
1275 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1277 /* In case the caller cares to look... */
1281 FREE_MAYBE (all_headers);
1282 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1283 might be more bytes in the body. */
1284 return RETRFINISHED;
1287 /* Open the local file. */
1290 mkalldirs (u->local);
1292 rotate_backups (u->local);
1293 fp = fopen (u->local, hs->restval ? "ab" : "wb");
1296 logprintf (LOG_NOTQUIET, "%s: %s\n", u->local, strerror (errno));
1297 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1298 might be more bytes in the body. */
1299 FREE_MAYBE (all_headers);
1305 extern int global_download_count;
1307 /* To ensure that repeated "from scratch" downloads work for -O
1308 files, we rewind the file pointer, unless restval is
1309 non-zero. (This works only when -O is used on regular files,
1310 but it's still a valuable feature.)
1312 However, this loses when more than one URL is specified on
1313 the command line the second rewinds eradicates the contents
1314 of the first download. Thus we disable the above trick for
1315 all the downloads except the very first one.
1317 #### A possible solution to this would be to remember the
1318 file position in the output document and to seek to that
1319 position, instead of rewinding. */
1320 if (!hs->restval && global_download_count == 0)
1322 /* This will silently fail for streams that don't correspond
1323 to regular files, but that's OK. */
1325 /* ftruncate is needed because opt.dfp is opened in append
1326 mode if opt.always_rest is set. */
1327 ftruncate (fileno (fp), 0);
1332 /* #### This confuses the code that checks for file size. There
1333 should be some overhead information. */
1334 if (opt.save_headers)
1335 fwrite (all_headers, 1, all_length, fp);
1337 /* Get the contents of the document. */
1338 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1339 (contlen != -1 ? contlen : 0),
1341 hs->dltime = elapsed_time ();
1343 /* Close or flush the file. We have to be careful to check for
1344 error here. Checking the result of fwrite() is not enough --
1345 errors could go unnoticed! */
1348 flush_res = fclose (fp);
1350 flush_res = fflush (fp);
1351 if (flush_res == EOF)
1354 FREE_MAYBE (all_headers);
1355 CLOSE_FINISH (sock);
1358 return RETRFINISHED;
1361 /* The genuine HTTP loop! This is the part where the retrieval is
1362 retried, and retried, and retried, and... */
1364 http_loop (struct urlinfo *u, char **newloc, int *dt)
1367 int use_ts, got_head = 0; /* time-stamping info */
1368 char *filename_plus_orig_suffix;
1369 char *local_filename = NULL;
1370 char *tms, *suf, *locf, *tmrate;
1372 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1373 long local_size = 0; /* the size of the local file */
1374 size_t filename_len;
1375 struct http_stat hstat; /* HTTP status */
1378 /* This used to be done in main(), but it's a better idea to do it
1379 here so that we don't go through the hoops if we're just using
1381 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1382 load_cookies (opt.cookies_input);
1386 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1387 has_wildcards_p because it would also warn on `?', and we know that
1388 shows up in CGI paths a *lot*. */
1389 if (strchr (u->url, '*'))
1390 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1392 /* Determine the local filename. */
1394 u->local = url_filename (u->proxy ? u->proxy : u);
1396 if (!opt.output_document)
1399 locf = opt.output_document;
1401 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1402 xmalloc() here before EACH return. This is one reason it's better to set
1403 flags that influence flow control and then return once at the end. */
1404 filename_len = strlen(u->local);
1405 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1407 if (opt.noclobber && file_exists_p (u->local))
1409 /* If opt.noclobber is turned on and file already exists, do not
1410 retrieve the file */
1411 logprintf (LOG_VERBOSE, _("\
1412 File `%s' already there, will not retrieve.\n"), u->local);
1413 /* If the file is there, we suppose it's retrieved OK. */
1416 /* #### Bogusness alert. */
1417 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1418 text/html, a harmless lie. */
1419 if (((suf = suffix (u->local)) != NULL)
1420 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1423 xfree (filename_plus_orig_suffix); /* must precede every return! */
1424 /* Another harmless lie: */
1429 if (opt.timestamping)
1431 boolean local_dot_orig_file_exists = FALSE;
1433 if (opt.backup_converted)
1434 /* If -K is specified, we'll act on the assumption that it was specified
1435 last time these files were downloaded as well, and instead of just
1436 comparing local file X against server file X, we'll compare local
1437 file X.orig (if extant, else X) against server file X. If -K
1438 _wasn't_ specified last time, or the server contains files called
1439 *.orig, -N will be back to not operating correctly with -k. */
1441 /* Would a single s[n]printf() call be faster? --dan
1443 Definitely not. sprintf() is horribly slow. It's a
1444 different question whether the difference between the two
1445 affects a program. Usually I'd say "no", but at one
1446 point I profiled Wget, and found that a measurable and
1447 non-negligible amount of time was lost calling sprintf()
1448 in url.c. Replacing sprintf with inline calls to
1449 strcpy() and long_to_string() made a difference.
1451 strcpy(filename_plus_orig_suffix, u->local);
1452 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1454 /* Try to stat() the .orig file. */
1455 if (stat(filename_plus_orig_suffix, &st) == 0)
1457 local_dot_orig_file_exists = TRUE;
1458 local_filename = filename_plus_orig_suffix;
1462 if (!local_dot_orig_file_exists)
1463 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1464 if (stat (u->local, &st) == 0)
1465 local_filename = u->local;
1467 if (local_filename != NULL)
1468 /* There was a local file, so we'll check later to see if the version
1469 the server has is the same version we already have, allowing us to
1474 local_size = st.st_size;
1478 /* Reset the counter. */
1480 *dt = 0 | ACCEPTRANGES;
1484 /* Increment the pass counter. */
1486 sleep_between_retrievals (count);
1487 /* Get the current time string. */
1488 tms = time_str (NULL);
1489 /* Print fetch message, if opt.verbose. */
1492 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1496 sprintf (tmp, _("(try:%2d)"), count);
1497 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1498 tms, hurl, tmp, locf);
1500 ws_changetitle (hurl, 1);
1505 /* Default document type is empty. However, if spider mode is
1506 on or time-stamping is employed, HEAD_ONLY commands is
1507 encoded within *dt. */
1508 if (opt.spider || (use_ts && !got_head))
1512 /* Assume no restarting. */
1514 /* Decide whether or not to restart. */
1515 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1516 && file_exists_p (locf))
1517 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1518 hstat.restval = st.st_size;
1519 /* Decide whether to send the no-cache directive. */
1520 if (u->proxy && (count > 1 || (opt.proxy_cache == 0)))
1521 *dt |= SEND_NOCACHE;
1523 *dt &= ~SEND_NOCACHE;
1525 /* Try fetching the document, or at least its head. :-) */
1526 err = gethttp (u, &hstat, dt);
1528 /* It's unfortunate that wget determines the local filename before finding
1529 out the Content-Type of the file. Barring a major restructuring of the
1530 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1531 u->local to tack on ".html". */
1532 if (!opt.output_document)
1535 locf = opt.output_document;
1537 /* In `-c' is used, check whether the file we're writing to
1538 exists before we've done anything. If so, we'll refuse to
1539 truncate it if the server doesn't support continued
1541 if (opt.always_rest)
1542 hstat.no_truncate = file_exists_p (locf);
1545 tms = time_str (NULL);
1546 /* Get the new location (with or without the redirection). */
1548 *newloc = xstrdup (hstat.newloc);
1551 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1552 case CONERROR: case READERR: case WRITEFAILED:
1554 /* Non-fatal errors continue executing the loop, which will
1555 bring them to "while" statement at the end, to judge
1556 whether the number of tries was exceeded. */
1558 printwhat (count, opt.ntry);
1561 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1562 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1563 /* Fatal errors just return from the function. */
1565 xfree (filename_plus_orig_suffix); /* must precede every return! */
1568 case FWRITEERR: case FOPENERR:
1569 /* Another fatal error. */
1570 logputs (LOG_VERBOSE, "\n");
1571 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1572 u->local, strerror (errno));
1577 /* Another fatal error. */
1578 logputs (LOG_VERBOSE, "\n");
1579 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1581 xfree (filename_plus_orig_suffix); /* must precede every return! */
1585 /* Return the new location to the caller. */
1588 logprintf (LOG_NOTQUIET,
1589 _("ERROR: Redirection (%d) without location.\n"),
1591 xfree (filename_plus_orig_suffix); /* must precede every return! */
1595 xfree (filename_plus_orig_suffix); /* must precede every return! */
1599 /* Deal with you later. */
1602 /* All possibilities should have been exhausted. */
1605 if (!(*dt & RETROKF))
1609 /* #### Ugly ugly ugly! */
1610 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1611 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1614 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1615 tms, hstat.statcode, hstat.error);
1616 logputs (LOG_VERBOSE, "\n");
1618 xfree (filename_plus_orig_suffix); /* must precede every return! */
1622 /* Did we get the time-stamp? */
1625 if (opt.timestamping && !hstat.remote_time)
1627 logputs (LOG_NOTQUIET, _("\
1628 Last-modified header missing -- time-stamps turned off.\n"));
1630 else if (hstat.remote_time)
1632 /* Convert the date-string into struct tm. */
1633 tmr = http_atotm (hstat.remote_time);
1634 if (tmr == (time_t) (-1))
1635 logputs (LOG_VERBOSE, _("\
1636 Last-modified header invalid -- time-stamp ignored.\n"));
1640 /* The time-stamping section. */
1645 use_ts = 0; /* no more time-stamping */
1646 count = 0; /* the retrieve count for HEAD is
1648 if (hstat.remote_time && tmr != (time_t) (-1))
1650 /* Now time-stamping can be used validly. Time-stamping
1651 means that if the sizes of the local and remote file
1652 match, and local file is newer than the remote file,
1653 it will not be retrieved. Otherwise, the normal
1654 download procedure is resumed. */
1656 (hstat.contlen == -1 || local_size == hstat.contlen))
1658 logprintf (LOG_VERBOSE, _("\
1659 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1662 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1665 else if (tml >= tmr)
1666 logprintf (LOG_VERBOSE, _("\
1667 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1669 logputs (LOG_VERBOSE,
1670 _("Remote file is newer, retrieving.\n"));
1675 if ((tmr != (time_t) (-1))
1677 && ((hstat.len == hstat.contlen) ||
1678 ((hstat.res == 0) &&
1679 ((hstat.contlen == -1) ||
1680 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1682 /* #### This code repeats in http.c and ftp.c. Move it to a
1684 const char *fl = NULL;
1685 if (opt.output_document)
1687 if (opt.od_known_regular)
1688 fl = opt.output_document;
1695 /* End of time-stamping section. */
1699 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1700 xfree (filename_plus_orig_suffix); /* must precede every return! */
1704 /* It is now safe to free the remainder of hstat, since the
1705 strings within it will no longer be used. */
1708 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1710 if (hstat.len == hstat.contlen)
1714 logprintf (LOG_VERBOSE,
1715 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1716 tms, tmrate, locf, hstat.len, hstat.contlen);
1717 logprintf (LOG_NONVERBOSE,
1718 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1719 tms, u->url, hstat.len, hstat.contlen, locf, count);
1722 downloaded_increase (hstat.len);
1724 /* Remember that we downloaded the file for later ".orig" code. */
1725 if (*dt & ADDED_HTML_EXTENSION)
1726 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1728 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1730 xfree(filename_plus_orig_suffix); /* must precede every return! */
1733 else if (hstat.res == 0) /* No read error */
1735 if (hstat.contlen == -1) /* We don't know how much we were supposed
1736 to get, so assume we succeeded. */
1740 logprintf (LOG_VERBOSE,
1741 _("%s (%s) - `%s' saved [%ld]\n\n"),
1742 tms, tmrate, locf, hstat.len);
1743 logprintf (LOG_NONVERBOSE,
1744 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1745 tms, u->url, hstat.len, locf, count);
1748 downloaded_increase (hstat.len);
1750 /* Remember that we downloaded the file for later ".orig" code. */
1751 if (*dt & ADDED_HTML_EXTENSION)
1752 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1754 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1756 xfree (filename_plus_orig_suffix); /* must precede every return! */
1759 else if (hstat.len < hstat.contlen) /* meaning we lost the
1760 connection too soon */
1762 logprintf (LOG_VERBOSE,
1763 _("%s (%s) - Connection closed at byte %ld. "),
1764 tms, tmrate, hstat.len);
1765 printwhat (count, opt.ntry);
1768 else if (!opt.kill_longer) /* meaning we got more than expected */
1770 logprintf (LOG_VERBOSE,
1771 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1772 tms, tmrate, locf, hstat.len, hstat.contlen);
1773 logprintf (LOG_NONVERBOSE,
1774 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1775 tms, u->url, hstat.len, hstat.contlen, locf, count);
1777 downloaded_increase (hstat.len);
1779 /* Remember that we downloaded the file for later ".orig" code. */
1780 if (*dt & ADDED_HTML_EXTENSION)
1781 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1783 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1785 xfree (filename_plus_orig_suffix); /* must precede every return! */
1788 else /* the same, but not accepted */
1790 logprintf (LOG_VERBOSE,
1791 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1792 tms, tmrate, hstat.len, hstat.contlen);
1793 printwhat (count, opt.ntry);
1797 else /* now hstat.res can only be -1 */
1799 if (hstat.contlen == -1)
1801 logprintf (LOG_VERBOSE,
1802 _("%s (%s) - Read error at byte %ld (%s)."),
1803 tms, tmrate, hstat.len, strerror (errno));
1804 printwhat (count, opt.ntry);
1807 else /* hstat.res == -1 and contlen is given */
1809 logprintf (LOG_VERBOSE,
1810 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1811 tms, tmrate, hstat.len, hstat.contlen,
1813 printwhat (count, opt.ntry);
1820 while (!opt.ntry || (count < opt.ntry));
1821 xfree (filename_plus_orig_suffix); /* must precede every return! */
1825 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1826 than local timezone (mktime assumes the latter).
1828 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1829 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO. */
1831 mktime_from_utc (struct tm *t)
1838 tb = mktime (gmtime (&tl));
1839 return (tl <= tb ? (tl + (tl - tb)) : (tl - (tb - tl)));
1842 /* Check whether the result of strptime() indicates success.
1843 strptime() returns the pointer to how far it got to in the string.
1844 The processing has been successful if the string is at `GMT' or
1845 `+X', or at the end of the string.
1847 In extended regexp parlance, the function returns 1 if P matches
1848 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (a valid result of
1849 strptime()) is considered a failure and 0 is returned. */
1851 check_end (const char *p)
1855 while (ISSPACE (*p))
1858 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1859 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1865 /* Convert TIME_STRING time to time_t. TIME_STRING can be in any of
1866 the three formats RFC2068 allows the HTTP servers to emit --
1867 RFC1123-date, RFC850-date or asctime-date. Timezones are ignored,
1870 We use strptime() to recognize various dates, which makes it a
1871 little bit slacker than the RFC1123/RFC850/asctime (e.g. it always
1872 allows shortened dates and months, one-digit days, etc.). It also
1873 allows more than one space anywhere where the specs require one SP.
1874 The routine should probably be even more forgiving (as recommended
1875 by RFC2068), but I do not have the time to write one.
1877 Return the computed time_t representation, or -1 if all the
1880 Needless to say, what we *really* need here is something like
1881 Marcus Hennecke's atotm(), which is forgiving, fast, to-the-point,
1882 and does not use strptime(). atotm() is to be found in the sources
1883 of `phttpd', a little-known HTTP server written by Peter Erikson. */
1885 http_atotm (char *time_string)
1889 /* Roger Beeman says: "This function dynamically allocates struct tm
1890 t, but does no initialization. The only field that actually
1891 needs initialization is tm_isdst, since the others will be set by
1892 strptime. Since strptime does not set tm_isdst, it will return
1893 the data structure with whatever data was in tm_isdst to begin
1894 with. For those of us in timezones where DST can occur, there
1895 can be a one hour shift depending on the previous contents of the
1896 data area where the data structure is allocated." */
1899 /* Note that under foreign locales Solaris strptime() fails to
1900 recognize English dates, which renders this function useless. I
1901 assume that other non-GNU strptime's are plagued by the same
1902 disease. We solve this by setting only LC_MESSAGES in
1903 i18n_initialize(), instead of LC_ALL.
1905 Another solution could be to temporarily set locale to C, invoke
1906 strptime(), and restore it back. This is slow and dirty,
1907 however, and locale support other than LC_MESSAGES can mess other
1908 things, so I rather chose to stick with just setting LC_MESSAGES.
1910 Also note that none of this is necessary under GNU strptime(),
1911 because it recognizes both international and local dates. */
1913 /* NOTE: We don't use `%n' for white space, as OSF's strptime uses
1914 it to eat all white space up to (and including) a newline, and
1915 the function fails if there is no newline (!).
1917 Let's hope all strptime() implementations use ` ' to skip *all*
1918 whitespace instead of just one (it works that way on all the
1919 systems I've tested it on). */
1921 /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1922 if (check_end (strptime (time_string, "%a, %d %b %Y %T", &t)))
1923 return mktime_from_utc (&t);
1924 /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1925 if (check_end (strptime (time_string, "%A, %d-%b-%y %T", &t)))
1926 return mktime_from_utc (&t);
1927 /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1928 (google.com uses this for their cookies.)*/
1929 if (check_end (strptime (time_string, "%a, %d-%b-%Y %T", &t)))
1930 return mktime_from_utc (&t);
1931 /* asctime: Thu Jan 29 22:12:57 1998 */
1932 if (check_end (strptime (time_string, "%a %b %d %T %Y", &t)))
1933 return mktime_from_utc (&t);
1938 /* Authorization support: We support two authorization schemes:
1940 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
1942 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
1943 consisting of answering to the server's challenge with the proper
1946 /* How many bytes it will take to store LEN bytes in base64. */
1947 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
1949 /* Encode the string S of length LENGTH to base64 format and place it
1950 to STORE. STORE will be 0-terminated, and must point to a writable
1951 buffer of at least 1+BASE64_LENGTH(length) bytes. */
1953 base64_encode (const char *s, char *store, int length)
1955 /* Conversion table. */
1956 static char tbl[64] = {
1957 'A','B','C','D','E','F','G','H',
1958 'I','J','K','L','M','N','O','P',
1959 'Q','R','S','T','U','V','W','X',
1960 'Y','Z','a','b','c','d','e','f',
1961 'g','h','i','j','k','l','m','n',
1962 'o','p','q','r','s','t','u','v',
1963 'w','x','y','z','0','1','2','3',
1964 '4','5','6','7','8','9','+','/'
1967 unsigned char *p = (unsigned char *)store;
1969 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
1970 for (i = 0; i < length; i += 3)
1972 *p++ = tbl[s[0] >> 2];
1973 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
1974 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
1975 *p++ = tbl[s[2] & 0x3f];
1978 /* Pad the result if necessary... */
1979 if (i == length + 1)
1981 else if (i == length + 2)
1982 *(p - 1) = *(p - 2) = '=';
1983 /* ...and zero-terminate it. */
1987 /* Create the authentication header contents for the `Basic' scheme.
1988 This is done by encoding the string `USER:PASS' in base64 and
1989 prepending `HEADER: Basic ' to it. */
1991 basic_authentication_encode (const char *user, const char *passwd,
1994 char *t1, *t2, *res;
1995 int len1 = strlen (user) + 1 + strlen (passwd);
1996 int len2 = BASE64_LENGTH (len1);
1998 t1 = (char *)alloca (len1 + 1);
1999 sprintf (t1, "%s:%s", user, passwd);
2000 t2 = (char *)alloca (1 + len2);
2001 base64_encode (t1, t2, len1);
2002 res = (char *)xmalloc (len2 + 11 + strlen (header));
2003 sprintf (res, "%s: Basic %s\r\n", header, t2);
2009 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2010 of a field in such a header. If the field is the one specified by
2011 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2012 digest authorization code), extract its value in the (char*)
2013 variable pointed by RET. Returns negative on a malformed header,
2014 or number of bytes that have been parsed by this call. */
2016 extract_header_attr (const char *au, const char *attr_name, char **ret)
2018 const char *cp, *ep;
2022 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2024 cp += strlen (attr_name);
2027 cp += skip_lws (cp);
2032 cp += skip_lws (cp);
2037 for (ep = cp; *ep && *ep != '\"'; ep++)
2042 *ret = strdupdelim (cp, ep);
2049 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2050 an array of 16 bytes containing the hash keys, and BUF should be a
2051 buffer of 33 writable characters (32 for hex digits plus one for
2052 zero termination). */
2054 dump_hash (unsigned char *buf, const unsigned char *hash)
2058 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2060 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2061 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2066 /* Take the line apart to find the challenge, and compose a digest
2067 authorization header. See RFC2069 section 2.1.2. */
2069 digest_authentication_encode (const char *au, const char *user,
2070 const char *passwd, const char *method,
2073 static char *realm, *opaque, *nonce;
2078 { "realm", &realm },
2079 { "opaque", &opaque },
2084 realm = opaque = nonce = NULL;
2086 au += 6; /* skip over `Digest' */
2091 au += skip_lws (au);
2092 for (i = 0; i < ARRAY_SIZE (options); i++)
2094 int skip = extract_header_attr (au, options[i].name,
2095 options[i].variable);
2099 FREE_MAYBE (opaque);
2109 if (i == ARRAY_SIZE (options))
2111 while (*au && *au != '=')
2115 au += skip_lws (au);
2119 while (*au && *au != '\"')
2126 while (*au && *au != ',')
2131 if (!realm || !nonce || !user || !passwd || !path || !method)
2134 FREE_MAYBE (opaque);
2139 /* Calculate the digest value. */
2142 unsigned char hash[MD5_HASHLEN];
2143 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2144 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2146 /* A1BUF = H(user ":" realm ":" password) */
2147 md5_init_ctx (&ctx);
2148 md5_process_bytes (user, strlen (user), &ctx);
2149 md5_process_bytes (":", 1, &ctx);
2150 md5_process_bytes (realm, strlen (realm), &ctx);
2151 md5_process_bytes (":", 1, &ctx);
2152 md5_process_bytes (passwd, strlen (passwd), &ctx);
2153 md5_finish_ctx (&ctx, hash);
2154 dump_hash (a1buf, hash);
2156 /* A2BUF = H(method ":" path) */
2157 md5_init_ctx (&ctx);
2158 md5_process_bytes (method, strlen (method), &ctx);
2159 md5_process_bytes (":", 1, &ctx);
2160 md5_process_bytes (path, strlen (path), &ctx);
2161 md5_finish_ctx (&ctx, hash);
2162 dump_hash (a2buf, hash);
2164 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2165 md5_init_ctx (&ctx);
2166 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2167 md5_process_bytes (":", 1, &ctx);
2168 md5_process_bytes (nonce, strlen (nonce), &ctx);
2169 md5_process_bytes (":", 1, &ctx);
2170 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2171 md5_finish_ctx (&ctx, hash);
2172 dump_hash (response_digest, hash);
2174 res = (char*) xmalloc (strlen (user)
2179 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2180 + (opaque ? strlen (opaque) : 0)
2182 sprintf (res, "Authorization: Digest \
2183 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2184 user, realm, nonce, path, response_digest);
2187 char *p = res + strlen (res);
2188 strcat (p, ", opaque=\"");
2192 strcat (res, "\r\n");
2196 #endif /* USE_DIGEST */
2199 #define BEGINS_WITH(line, string_constant) \
2200 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2201 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2202 || !line[sizeof (string_constant) - 1]))
2205 known_authentication_scheme_p (const char *au)
2207 return BEGINS_WITH (au, "Basic")
2208 || BEGINS_WITH (au, "Digest")
2209 || BEGINS_WITH (au, "NTLM");
2214 /* Create the HTTP authorization request header. When the
2215 `WWW-Authenticate' response header is seen, according to the
2216 authorization scheme specified in that header (`Basic' and `Digest'
2217 are supported by the current implementation), produce an
2218 appropriate HTTP authorization request header. */
2220 create_authorization_line (const char *au, const char *user,
2221 const char *passwd, const char *method,
2224 char *wwwauth = NULL;
2226 if (!strncasecmp (au, "Basic", 5))
2227 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2228 if (!strncasecmp (au, "NTLM", 4))
2229 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2231 else if (!strncasecmp (au, "Digest", 6))
2232 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2233 #endif /* USE_DIGEST */