2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001
3 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
25 #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 */
64 # include "gen_sslfunc.h"
68 extern char *version_string;
79 static int cookies_loaded_p;
81 #define TEXTHTML_S "text/html"
82 #define HTTP_ACCEPT "*/*"
84 /* Some status code validation macros: */
85 #define H_20X(x) (((x) >= 200) && ((x) < 300))
86 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
87 #define H_REDIRECTED(x) (((x) == HTTP_STATUS_MOVED_PERMANENTLY) \
88 || ((x) == HTTP_STATUS_MOVED_TEMPORARILY))
90 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
92 #define HTTP_STATUS_OK 200
93 #define HTTP_STATUS_CREATED 201
94 #define HTTP_STATUS_ACCEPTED 202
95 #define HTTP_STATUS_NO_CONTENT 204
96 #define HTTP_STATUS_PARTIAL_CONTENTS 206
98 /* Redirection 3xx. */
99 #define HTTP_STATUS_MULTIPLE_CHOICES 300
100 #define HTTP_STATUS_MOVED_PERMANENTLY 301
101 #define HTTP_STATUS_MOVED_TEMPORARILY 302
102 #define HTTP_STATUS_NOT_MODIFIED 304
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
110 /* Server errors 5xx. */
111 #define HTTP_STATUS_INTERNAL 500
112 #define HTTP_STATUS_NOT_IMPLEMENTED 501
113 #define HTTP_STATUS_BAD_GATEWAY 502
114 #define HTTP_STATUS_UNAVAILABLE 503
117 /* Parse the HTTP status line, which is of format:
119 HTTP-Version SP Status-Code SP Reason-Phrase
121 The function returns the status-code, or -1 if the status line is
122 malformed. The pointer to reason-phrase is returned in RP. */
124 parse_http_status_line (const char *line, const char **reason_phrase_ptr)
126 /* (the variables must not be named `major' and `minor', because
127 that breaks compilation with SunOS4 cc.) */
128 int mjr, mnr, statcode;
131 *reason_phrase_ptr = NULL;
133 /* The standard format of HTTP-Version is: `HTTP/X.Y', where X is
134 major version, and Y is minor version. */
135 if (strncmp (line, "HTTP/", 5) != 0)
139 /* Calculate major HTTP version. */
141 for (mjr = 0; ISDIGIT (*line); line++)
142 mjr = 10 * mjr + (*line - '0');
143 if (*line != '.' || p == line)
147 /* Calculate minor HTTP version. */
149 for (mnr = 0; ISDIGIT (*line); line++)
150 mnr = 10 * mnr + (*line - '0');
151 if (*line != ' ' || p == line)
153 /* Wget will accept only 1.0 and higher HTTP-versions. The value of
154 minor version can be safely ignored. */
159 /* Calculate status code. */
160 if (!(ISDIGIT (*line) && ISDIGIT (line[1]) && ISDIGIT (line[2])))
162 statcode = 100 * (*line - '0') + 10 * (line[1] - '0') + (line[2] - '0');
164 /* Set up the reason phrase pointer. */
166 /* RFC2068 requires SPC here, but we allow the string to finish
167 here, in case no reason-phrase is present. */
171 *reason_phrase_ptr = line;
176 *reason_phrase_ptr = line + 1;
181 /* Functions to be used as arguments to header_process(): */
183 struct http_process_range_closure {
189 /* Parse the `Content-Range' header and extract the information it
190 contains. Returns 1 if successful, -1 otherwise. */
192 http_process_range (const char *hdr, void *arg)
194 struct http_process_range_closure *closure
195 = (struct http_process_range_closure *)arg;
198 /* Certain versions of Nutscape proxy server send out
199 `Content-Length' without "bytes" specifier, which is a breach of
200 RFC2068 (as well as the HTTP/1.1 draft which was current at the
201 time). But hell, I must support it... */
202 if (!strncasecmp (hdr, "bytes", 5))
205 hdr += skip_lws (hdr);
211 for (num = 0; ISDIGIT (*hdr); hdr++)
212 num = 10 * num + (*hdr - '0');
213 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
215 closure->first_byte_pos = num;
217 for (num = 0; ISDIGIT (*hdr); hdr++)
218 num = 10 * num + (*hdr - '0');
219 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
221 closure->last_byte_pos = num;
223 for (num = 0; ISDIGIT (*hdr); hdr++)
224 num = 10 * num + (*hdr - '0');
225 closure->entity_length = num;
229 /* Place 1 to ARG if the HDR contains the word "none", 0 otherwise.
230 Used for `Accept-Ranges'. */
232 http_process_none (const char *hdr, void *arg)
234 int *where = (int *)arg;
236 if (strstr (hdr, "none"))
243 /* Place the malloc-ed copy of HDR hdr, to the first `;' to ARG. */
245 http_process_type (const char *hdr, void *arg)
247 char **result = (char **)arg;
248 /* Locate P on `;' or the terminating zero, whichever comes first. */
249 const char *p = strchr (hdr, ';');
251 p = hdr + strlen (hdr);
252 while (p > hdr && ISSPACE (*(p - 1)))
254 *result = strdupdelim (hdr, p);
258 /* Check whether the `Connection' header is set to "keep-alive". */
260 http_process_connection (const char *hdr, void *arg)
262 int *flag = (int *)arg;
263 if (!strcasecmp (hdr, "Keep-Alive"))
268 /* Persistent connections. Currently, we cache the most recently used
269 connection as persistent, provided that the HTTP server agrees to
270 make it such. The persistence data is stored in the variables
271 below. Ideally, it would be in a structure, and it should be
272 possible to cache an arbitrary fixed number of these connections.
274 I think the code is quite easy to extend in that direction. */
276 /* Whether a persistent connection is active. */
277 static int pc_active_p;
278 /* Host and port of currently active persistent connection. */
279 static struct address_list *pc_last_host_ip;
280 static unsigned short pc_last_port;
282 /* File descriptor of the currently active persistent connection. */
283 static int pc_last_fd;
286 /* Whether a ssl handshake has occoured on this connection */
287 static int pc_active_ssl;
288 /* SSL connection of the currently active persistent connection. */
289 static SSL *pc_last_ssl;
290 #endif /* HAVE_SSL */
292 /* Mark the persistent connection as invalid. This is used by the
293 CLOSE_* macros after they forcefully close a registered persistent
294 connection. This does not close the file descriptor -- it is left
295 to the caller to do that. (Maybe it should, though.) */
298 invalidate_persistent (void)
303 #endif /* HAVE_SSL */
304 if (pc_last_host_ip != NULL)
306 address_list_release (pc_last_host_ip);
307 pc_last_host_ip = NULL;
309 DEBUGP (("Invalidating fd %d from further reuse.\n", pc_last_fd));
312 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
313 persistent. This will enable someone to use the same connection
314 later. In the context of HTTP, this must be called only AFTER the
315 response has been received and the server has promised that the
316 connection will remain alive.
318 If a previous connection was persistent, it is closed. */
321 register_persistent (const char *host, unsigned short port, int fd
329 if (pc_last_fd == fd)
331 /* The connection FD is already registered. Nothing to
337 /* The old persistent connection is still active; let's
338 close it first. This situation arises whenever a
339 persistent connection exists, but we then connect to a
340 different host, and try to register a persistent
341 connection to that one. */
343 /* The ssl disconnect has to take place before the closing
346 shutdown_ssl(pc_last_ssl);
349 invalidate_persistent ();
353 assert (pc_last_host_ip == NULL);
355 /* This lookup_host cannot fail, because it has the results in the
357 pc_last_host_ip = lookup_host (host, 1);
358 assert (pc_last_host_ip != NULL);
365 pc_active_ssl = ssl ? 1 : 0;
367 DEBUGP (("Registered fd %d for persistent reuse.\n", fd));
370 /* Return non-zero if a persistent connection is available for
371 connecting to HOST:PORT. */
374 persistent_available_p (const char *host, unsigned short port
381 struct address_list *this_host_ip;
383 /* First, check whether a persistent connection is active at all. */
386 /* Second, check if the active connection pertains to the correct
387 (HOST, PORT) ordered pair. */
388 if (port != pc_last_port)
392 /* Second, a): check if current connection is (not) ssl, too. This
393 test is unlikely to fail because HTTP and HTTPS typicaly use
394 different ports. Yet it is possible, or so I [Christian
395 Fraenkel] have been told, to run HTTPS and HTTP simultaneus on
397 if (ssl != pc_active_ssl)
399 #endif /* HAVE_SSL */
401 this_host_ip = lookup_host (host, 1);
405 /* To equate the two host names for the purposes of persistent
406 connections, they need to share all the IP addresses in the
408 success = address_list_match_all (pc_last_host_ip, this_host_ip);
409 address_list_release (this_host_ip);
413 /* Third: check whether the connection is still open. This is
414 important because most server implement a liberal (short) timeout
415 on persistent connections. Wget can of course always reconnect
416 if the connection doesn't work out, but it's nicer to know in
417 advance. This test is a logical followup of the first test, but
418 is "expensive" and therefore placed at the end of the list. */
419 if (!test_socket_open (pc_last_fd))
421 /* Oops, the socket is no longer open. Now that we know that,
422 let's invalidate the persistent connection before returning
425 invalidate_persistent ();
432 # define SHUTDOWN_SSL(ssl) do { \
434 shutdown_ssl (ssl); \
437 # define SHUTDOWN_SSL(ssl)
440 /* The idea behind these two CLOSE macros is to distinguish between
441 two cases: one when the job we've been doing is finished, and we
442 want to close the connection and leave, and two when something is
443 seriously wrong and we're closing the connection as part of
446 In case of keep_alive, CLOSE_FINISH should leave the connection
447 open, while CLOSE_INVALIDATE should still close it.
449 Note that the semantics of the flag `keep_alive' is "this
450 connection *will* be reused (the server has promised not to close
451 the connection once we're done)", while the semantics of
452 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
453 active, registered connection". */
455 #define CLOSE_FINISH(fd) do { \
458 SHUTDOWN_SSL (ssl); \
460 if (pc_active_p && (fd) == pc_last_fd) \
461 invalidate_persistent (); \
465 #define CLOSE_INVALIDATE(fd) do { \
466 SHUTDOWN_SSL (ssl); \
468 if (pc_active_p && (fd) == pc_last_fd) \
469 invalidate_persistent (); \
474 long len; /* received length */
475 long contlen; /* expected length */
476 long restval; /* the restart value */
477 int res; /* the result of last read */
478 char *newloc; /* new location (redirection) */
479 char *remote_time; /* remote time-stamp string */
480 char *error; /* textual HTTP error */
481 int statcode; /* status code */
482 long dltime; /* time of the download */
483 int no_truncate; /* whether truncating the file is
485 const char *referer; /* value of the referer header. */
486 char **local_file; /* local file. */
490 free_hstat (struct http_stat *hs)
492 FREE_MAYBE (hs->newloc);
493 FREE_MAYBE (hs->remote_time);
494 FREE_MAYBE (hs->error);
496 /* Guard against being called twice. */
498 hs->remote_time = NULL;
502 static char *create_authorization_line PARAMS ((const char *, const char *,
503 const char *, const char *,
505 static char *basic_authentication_encode PARAMS ((const char *, const char *,
507 static int known_authentication_scheme_p PARAMS ((const char *));
509 time_t http_atotm PARAMS ((char *));
511 #define BEGINS_WITH(line, string_constant) \
512 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
513 && (ISSPACE (line[sizeof (string_constant) - 1]) \
514 || !line[sizeof (string_constant) - 1]))
516 /* Retrieve a document through HTTP protocol. It recognizes status
517 code, and correctly handles redirections. It closes the network
518 socket. If it receives an error from the functions below it, it
519 will print it if there is enough information to do so (almost
520 always), returning the error to the caller (i.e. http_loop).
522 Various HTTP parameters are stored to hs. Although it parses the
523 response code correctly, it is not used in a sane way. The caller
526 If PROXY is non-NULL, the connection will be made to the proxy
527 server, and u->url will be requested. */
529 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
531 char *request, *type, *command, *full_path;
533 char *pragma_h, *referer, *useragent, *range, *wwwauth;
534 char *authenticate_h;
538 char *request_keep_alive;
539 int sock, hcount, num_written, all_length, statcode;
540 long contlen, contrange;
543 int auth_tried_already;
546 static SSL_CTX *ssl_ctx = NULL;
548 #endif /* HAVE_SSL */
549 char *cookies = NULL;
551 /* Whether this connection will be kept alive after the HTTP request
555 /* Flags that detect the two ways of specifying HTTP keep-alive
557 int http_keep_alive_1, http_keep_alive_2;
559 /* Whether keep-alive should be inhibited. */
560 int inhibit_keep_alive;
563 /* initialize ssl_ctx on first run */
566 err = init_ssl (&ssl_ctx);
571 case SSLERRCTXCREATE:
573 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
577 /* try without certfile */
578 logprintf (LOG_NOTQUIET,
579 _("Failed to load certificates from %s\n"),
582 logprintf (LOG_NOTQUIET,
583 _("Trying without the specified certificate\n"));
586 logprintf (LOG_NOTQUIET,
587 _("Failed to get certificate key from %s\n"),
590 logprintf (LOG_NOTQUIET,
591 _("Trying without the specified certificate\n"));
598 #endif /* HAVE_SSL */
600 if (!(*dt & HEAD_ONLY))
601 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
602 know the local filename so we can save to it. */
603 assert (*hs->local_file != NULL);
606 auth_tried_already = 0;
608 inhibit_keep_alive = !opt.http_keep_alive || proxy != NULL;
611 /* We need to come back here when the initial attempt to retrieve
612 without authorization header fails. (Expected to happen at least
613 for the Digest authorization scheme.) */
616 http_keep_alive_1 = http_keep_alive_2 = 0;
618 /* Initialize certain elements of struct http_stat. */
623 hs->remote_time = NULL;
626 /* If we're using a proxy, we will be connecting to the proxy
628 conn = proxy ? proxy : u;
630 /* First: establish the connection. */
631 if (inhibit_keep_alive
634 !persistent_available_p (conn->host, conn->port)
636 !persistent_available_p (conn->host, conn->port,
637 u->scheme == SCHEME_HTTPS)
638 #endif /* HAVE_SSL */
641 struct address_list *al = lookup_host (conn->host, 0);
644 set_connection_host_name (conn->host);
645 sock = connect_to_many (al, conn->port, 0);
646 set_connection_host_name (NULL);
647 address_list_release (al);
650 return errno == ECONNREFUSED ? CONREFUSED : CONERROR;
653 if (conn->scheme == SCHEME_HTTPS)
654 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
656 logputs (LOG_VERBOSE, "\n");
657 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
662 #endif /* HAVE_SSL */
666 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"),
667 conn->host, conn->port);
668 /* #### pc_last_fd should be accessed through an accessor
673 #endif /* HAVE_SSL */
674 DEBUGP (("Reusing fd %d.\n", sock));
677 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
682 referer = (char *)alloca (9 + strlen (hs->referer) + 3);
683 sprintf (referer, "Referer: %s\r\n", hs->referer);
686 if (*dt & SEND_NOCACHE)
687 pragma_h = "Pragma: no-cache\r\n";
693 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
694 /* Gag me! Some servers (e.g. WebSitePro) have been known to
695 respond to the following `Range' format by generating a
696 multipart/x-byte-ranges MIME document! This MIME type was
697 present in an old draft of the byteranges specification.
698 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
699 only if multiple non-overlapping ranges are requested --
700 which Wget never does. */
701 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
706 STRDUP_ALLOCA (useragent, opt.useragent);
709 useragent = (char *)alloca (10 + strlen (version_string));
710 sprintf (useragent, "Wget/%s", version_string);
712 /* Construct the authentication, if userid is present. */
715 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
716 user = user ? user : opt.http_user;
717 passwd = passwd ? passwd : opt.http_passwd;
724 /* We have the username and the password, but haven't tried
725 any authorization yet. Let's see if the "Basic" method
726 works. If not, we'll come back here and construct a
727 proper authorization method with the right challenges.
729 If we didn't employ this kind of logic, every URL that
730 requires authorization would have to be processed twice,
731 which is very suboptimal and generates a bunch of false
732 "unauthorized" errors in the server log.
734 #### But this logic also has a serious problem when used
735 with stronger authentications: we *first* transmit the
736 username and the password in clear text, and *then*
737 attempt a stronger authentication scheme. That cannot be
738 right! We are only fortunate that almost everyone still
739 uses the `Basic' scheme anyway.
741 There should be an option to prevent this from happening,
742 for those who use strong authentication schemes and value
744 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
748 wwwauth = create_authorization_line (authenticate_h, user, passwd,
756 char *proxy_user, *proxy_passwd;
757 /* For normal username and password, URL components override
758 command-line/wgetrc parameters. With proxy authentication,
759 it's the reverse, because proxy URLs are normally the
760 "permanent" ones, so command-line args should take
762 if (opt.proxy_user && opt.proxy_passwd)
764 proxy_user = opt.proxy_user;
765 proxy_passwd = opt.proxy_passwd;
769 proxy_user = proxy->user;
770 proxy_passwd = proxy->passwd;
772 /* #### This does not appear right. Can't the proxy request,
773 say, `Digest' authentication? */
774 if (proxy_user && proxy_passwd)
775 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
776 "Proxy-Authorization");
779 /* String of the form :PORT. Used only for non-standard ports. */
781 if (u->port != scheme_default_port (u->scheme))
783 port_maybe = (char *)alloca (numdigit (u->port) + 2);
784 sprintf (port_maybe, ":%d", u->port);
787 if (!inhibit_keep_alive)
788 request_keep_alive = "Connection: Keep-Alive\r\n";
790 request_keep_alive = NULL;
793 cookies = build_cookies_request (u->host, u->port, u->path,
795 u->scheme == SCHEME_HTTPS
802 full_path = xstrdup (u->url);
804 full_path = url_full_path (u);
806 /* Allocate the memory for the request. */
807 request = (char *)alloca (strlen (command)
811 + (port_maybe ? strlen (port_maybe) : 0)
812 + strlen (HTTP_ACCEPT)
813 + (request_keep_alive
814 ? strlen (request_keep_alive) : 0)
815 + (referer ? strlen (referer) : 0)
816 + (cookies ? strlen (cookies) : 0)
817 + (wwwauth ? strlen (wwwauth) : 0)
818 + (proxyauth ? strlen (proxyauth) : 0)
819 + (range ? strlen (range) : 0)
821 + (opt.user_header ? strlen (opt.user_header) : 0)
823 /* Construct the request. */
829 %s%s%s%s%s%s%s%s\r\n",
832 port_maybe ? port_maybe : "",
834 request_keep_alive ? request_keep_alive : "",
835 referer ? referer : "",
836 cookies ? cookies : "",
837 wwwauth ? wwwauth : "",
838 proxyauth ? proxyauth : "",
841 opt.user_header ? opt.user_header : "");
842 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);
850 /* Send the request to server. */
852 if (u->scheme == SCHEME_HTTPS)
853 num_written = ssl_iwrite (ssl, request, strlen (request));
855 #endif /* HAVE_SSL */
856 num_written = iwrite (sock, request, strlen (request));
860 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
862 CLOSE_INVALIDATE (sock);
865 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
866 proxy ? "Proxy" : "HTTP");
867 contlen = contrange = -1;
872 /* Before reading anything, initialize the rbuf. */
873 rbuf_initialize (&rbuf, sock);
875 if (u->scheme == SCHEME_HTTPS)
879 #endif /* HAVE_SSL */
882 /* Header-fetching loop. */
890 /* Get the header. */
891 status = header_get (&rbuf, &hdr,
892 /* Disallow continuations for status line. */
893 (hcount == 1 ? HG_NO_CONTINUATIONS : HG_NONE));
895 /* Check for errors. */
896 if (status == HG_EOF && *hdr)
898 /* This used to be an unconditional error, but that was
899 somewhat controversial, because of a large number of
900 broken CGI's that happily "forget" to send the second EOL
901 before closing the connection of a HEAD request.
903 So, the deal is to check whether the header is empty
904 (*hdr is zero if it is); if yes, it means that the
905 previous header was fully retrieved, and that -- most
906 probably -- the request is complete. "...be liberal in
907 what you accept." Oh boy. */
908 logputs (LOG_VERBOSE, "\n");
909 logputs (LOG_NOTQUIET, _("End of file while parsing headers.\n"));
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 (all_headers);
924 CLOSE_INVALIDATE (sock);
928 /* If the headers are to be saved to a file later, save them to
930 if (opt.save_headers)
932 int lh = strlen (hdr);
933 all_headers = (char *)xrealloc (all_headers, all_length + lh + 2);
934 memcpy (all_headers + all_length, hdr, lh);
936 all_headers[all_length++] = '\n';
937 all_headers[all_length] = '\0';
940 /* Check for status line. */
944 /* Parse the first line of server response. */
945 statcode = parse_http_status_line (hdr, &error);
946 hs->statcode = statcode;
947 /* Store the descriptive response. */
948 if (statcode == -1) /* malformed response */
950 /* A common reason for "malformed response" error is the
951 case when no data was actually received. Handle this
954 hs->error = xstrdup (_("No data received"));
956 hs->error = xstrdup (_("Malformed status line"));
961 hs->error = xstrdup (_("(no description)"));
963 hs->error = xstrdup (error);
971 if (opt.server_response)
972 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
974 logprintf (LOG_VERBOSE, "%2d %s", statcode, error);
980 /* Exit on empty header. */
987 /* Print the header if requested. */
988 if (opt.server_response && hcount != 1)
989 logprintf (LOG_VERBOSE, "\n%2d %s", hcount, hdr);
991 /* Try getting content-length. */
992 if (contlen == -1 && !opt.ignore_length)
993 if (header_process (hdr, "Content-Length", header_extract_number,
996 /* Try getting content-type. */
998 if (header_process (hdr, "Content-Type", http_process_type, &type))
1000 /* Try getting location. */
1002 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
1004 /* Try getting last-modified. */
1005 if (!hs->remote_time)
1006 if (header_process (hdr, "Last-Modified", header_strdup,
1009 /* Try getting cookies. */
1011 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1013 /* Try getting www-authentication. */
1014 if (!authenticate_h)
1015 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1018 /* Check for accept-ranges header. If it contains the word
1019 `none', disable the ranges. */
1020 if (*dt & ACCEPTRANGES)
1023 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1026 *dt &= ~ACCEPTRANGES;
1030 /* Try getting content-range. */
1031 if (contrange == -1)
1033 struct http_process_range_closure closure;
1034 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1036 contrange = closure.first_byte_pos;
1040 /* Check for keep-alive related responses. */
1041 if (!inhibit_keep_alive)
1043 /* Check for the `Keep-Alive' header. */
1044 if (!http_keep_alive_1)
1046 if (header_process (hdr, "Keep-Alive", header_exists,
1047 &http_keep_alive_1))
1050 /* Check for `Connection: Keep-Alive'. */
1051 if (!http_keep_alive_2)
1053 if (header_process (hdr, "Connection", http_process_connection,
1054 &http_keep_alive_2))
1062 logputs (LOG_VERBOSE, "\n");
1065 && (http_keep_alive_1 || http_keep_alive_2))
1067 assert (inhibit_keep_alive == 0);
1071 /* The server has promised that it will not close the connection
1072 when we're done. This means that we can register it. */
1074 register_persistent (conn->host, conn->port, sock);
1076 register_persistent (conn->host, conn->port, sock, ssl);
1077 #endif /* HAVE_SSL */
1079 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1082 /* Authorization is required. */
1086 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1087 might be more bytes in the body. */
1088 if (auth_tried_already)
1090 /* If we have tried it already, then there is not point
1093 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1094 xfree (authenticate_h);
1097 else if (!known_authentication_scheme_p (authenticate_h))
1099 xfree (authenticate_h);
1100 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1103 else if (BEGINS_WITH (authenticate_h, "Basic"))
1105 /* The authentication scheme is basic, the one we try by
1106 default, and it failed. There's no sense in trying
1112 auth_tried_already = 1;
1116 /* We do not need this anymore. */
1119 xfree (authenticate_h);
1120 authenticate_h = NULL;
1123 /* 20x responses are counted among successful by default. */
1124 if (H_20X (statcode))
1127 /* Return if redirected. */
1128 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1130 /* RFC2068 says that in case of the 300 (multiple choices)
1131 response, the server can output a preferred URL through
1132 `Location' header; otherwise, the request should be treated
1133 like GET. So, if the location is set, it will be a
1134 redirection; otherwise, just proceed normally. */
1135 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1139 logprintf (LOG_VERBOSE,
1140 _("Location: %s%s\n"),
1141 hs->newloc ? hs->newloc : _("unspecified"),
1142 hs->newloc ? _(" [following]") : "");
1143 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1144 might be more bytes in the body. */
1146 FREE_MAYBE (all_headers);
1151 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1154 /* We don't assume text/html by default. */
1157 if (opt.html_extension && (*dt & TEXTHTML))
1158 /* -E / --html-extension / html_extension = on was specified, and this is a
1159 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1160 already the file's suffix, tack on ".html". */
1162 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1164 if (last_period_in_local_filename == NULL ||
1165 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1166 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1168 size_t local_filename_len = strlen(*hs->local_file);
1170 *hs->local_file = xrealloc(*hs->local_file,
1171 local_filename_len + sizeof(".html"));
1172 strcpy(*hs->local_file + local_filename_len, ".html");
1174 *dt |= ADDED_HTML_EXTENSION;
1178 if (contrange == -1)
1180 /* We did not get a content-range header. This means that the
1181 server did not honor our `Range' request. Normally, this
1182 means we should reset hs->restval and continue normally. */
1184 /* However, if `-c' is used, we need to be a bit more careful:
1186 1. If `-c' is specified and the file already existed when
1187 Wget was started, it would be a bad idea for us to start
1188 downloading it from scratch, effectively truncating it. I
1189 believe this cannot happen unless `-c' was specified.
1191 2. If `-c' is used on a file that is already fully
1192 downloaded, we're requesting bytes after the end of file,
1193 which can result in server not honoring `Range'. If this is
1194 the case, `Content-Length' will be equal to the length of the
1196 if (opt.always_rest)
1198 /* Check for condition #2. */
1199 if (hs->restval > 0 /* restart was requested. */
1200 && contlen != -1 /* we got content-length. */
1201 && hs->restval >= contlen /* file fully downloaded
1205 logputs (LOG_VERBOSE, _("\
1206 \n The file is already fully retrieved; nothing to do.\n\n"));
1207 /* In case the caller inspects. */
1211 FREE_MAYBE (all_headers);
1212 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1213 might be more bytes in the body. */
1214 return RETRUNNEEDED;
1217 /* Check for condition #1. */
1218 if (hs->no_truncate)
1220 logprintf (LOG_NOTQUIET,
1223 Continued download failed on this file, which conflicts with `-c'.\n\
1224 Refusing to truncate existing file `%s'.\n\n"), *hs->local_file);
1226 FREE_MAYBE (all_headers);
1227 CLOSE_INVALIDATE (sock);
1228 return CONTNOTSUPPORTED;
1236 else if (contrange != hs->restval ||
1237 (H_PARTIAL (statcode) && contrange == -1))
1239 /* This means the whole request was somehow misunderstood by the
1240 server. Bail out. */
1242 FREE_MAYBE (all_headers);
1243 CLOSE_INVALIDATE (sock);
1250 contlen += contrange;
1252 contrange = -1; /* If conent-length was not sent,
1253 content-range will be ignored. */
1255 hs->contlen = contlen;
1259 if ((*dt & RETROKF) && !opt.server_response)
1261 /* No need to print this output if the body won't be
1262 downloaded at all, or if the original server response is
1264 logputs (LOG_VERBOSE, _("Length: "));
1267 logputs (LOG_VERBOSE, legible (contlen));
1268 if (contrange != -1)
1269 logprintf (LOG_VERBOSE, _(" (%s to go)"),
1270 legible (contlen - contrange));
1273 logputs (LOG_VERBOSE,
1274 opt.ignore_length ? _("ignored") : _("unspecified"));
1276 logprintf (LOG_VERBOSE, " [%s]\n", type);
1278 logputs (LOG_VERBOSE, "\n");
1282 type = NULL; /* We don't need it any more. */
1284 /* Return if we have no intention of further downloading. */
1285 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1287 /* In case the caller cares to look... */
1291 FREE_MAYBE (all_headers);
1292 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1293 might be more bytes in the body. */
1294 return RETRFINISHED;
1297 /* Open the local file. */
1300 mkalldirs (*hs->local_file);
1302 rotate_backups (*hs->local_file);
1303 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1306 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1307 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1308 might be more bytes in the body. */
1309 FREE_MAYBE (all_headers);
1315 extern int global_download_count;
1317 /* To ensure that repeated "from scratch" downloads work for -O
1318 files, we rewind the file pointer, unless restval is
1319 non-zero. (This works only when -O is used on regular files,
1320 but it's still a valuable feature.)
1322 However, this loses when more than one URL is specified on
1323 the command line the second rewinds eradicates the contents
1324 of the first download. Thus we disable the above trick for
1325 all the downloads except the very first one.
1327 #### A possible solution to this would be to remember the
1328 file position in the output document and to seek to that
1329 position, instead of rewinding. */
1330 if (!hs->restval && global_download_count == 0)
1332 /* This will silently fail for streams that don't correspond
1333 to regular files, but that's OK. */
1335 /* ftruncate is needed because opt.dfp is opened in append
1336 mode if opt.always_rest is set. */
1337 ftruncate (fileno (fp), 0);
1342 /* #### This confuses the code that checks for file size. There
1343 should be some overhead information. */
1344 if (opt.save_headers)
1345 fwrite (all_headers, 1, all_length, fp);
1347 /* Get the contents of the document. */
1348 hs->res = get_contents (sock, fp, &hs->len, hs->restval,
1349 (contlen != -1 ? contlen : 0),
1350 &rbuf, keep_alive, &hs->dltime);
1353 /* Close or flush the file. We have to be careful to check for
1354 error here. Checking the result of fwrite() is not enough --
1355 errors could go unnoticed! */
1358 flush_res = fclose (fp);
1360 flush_res = fflush (fp);
1361 if (flush_res == EOF)
1364 FREE_MAYBE (all_headers);
1365 CLOSE_FINISH (sock);
1368 return RETRFINISHED;
1371 /* The genuine HTTP loop! This is the part where the retrieval is
1372 retried, and retried, and retried, and... */
1374 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1375 int *dt, struct url *proxy)
1378 int use_ts, got_head = 0; /* time-stamping info */
1379 char *filename_plus_orig_suffix;
1380 char *local_filename = NULL;
1381 char *tms, *suf, *locf, *tmrate;
1383 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1384 long local_size = 0; /* the size of the local file */
1385 size_t filename_len;
1386 struct http_stat hstat; /* HTTP status */
1390 /* This used to be done in main(), but it's a better idea to do it
1391 here so that we don't go through the hoops if we're just using
1393 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1395 load_cookies (opt.cookies_input);
1396 cookies_loaded_p = 1;
1401 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1402 has_wildcards_p because it would also warn on `?', and we know that
1403 shows up in CGI paths a *lot*. */
1404 if (strchr (u->url, '*'))
1405 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1407 /* Determine the local filename. */
1408 if (local_file && *local_file)
1409 hstat.local_file = local_file;
1410 else if (local_file)
1412 *local_file = url_filename (u);
1413 hstat.local_file = local_file;
1417 dummy = url_filename (u);
1418 hstat.local_file = &dummy;
1421 if (!opt.output_document)
1422 locf = *hstat.local_file;
1424 locf = opt.output_document;
1426 hstat.referer = referer;
1428 filename_len = strlen (*hstat.local_file);
1429 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1431 if (opt.noclobber && file_exists_p (*hstat.local_file))
1433 /* If opt.noclobber is turned on and file already exists, do not
1434 retrieve the file */
1435 logprintf (LOG_VERBOSE, _("\
1436 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1437 /* If the file is there, we suppose it's retrieved OK. */
1440 /* #### Bogusness alert. */
1441 /* If its suffix is "html" or "htm", assume text/html. */
1442 if (((suf = suffix (*hstat.local_file)) != NULL)
1443 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1451 if (opt.timestamping)
1453 boolean local_dot_orig_file_exists = FALSE;
1455 if (opt.backup_converted)
1456 /* If -K is specified, we'll act on the assumption that it was specified
1457 last time these files were downloaded as well, and instead of just
1458 comparing local file X against server file X, we'll compare local
1459 file X.orig (if extant, else X) against server file X. If -K
1460 _wasn't_ specified last time, or the server contains files called
1461 *.orig, -N will be back to not operating correctly with -k. */
1463 /* Would a single s[n]printf() call be faster? --dan
1465 Definitely not. sprintf() is horribly slow. It's a
1466 different question whether the difference between the two
1467 affects a program. Usually I'd say "no", but at one
1468 point I profiled Wget, and found that a measurable and
1469 non-negligible amount of time was lost calling sprintf()
1470 in url.c. Replacing sprintf with inline calls to
1471 strcpy() and long_to_string() made a difference.
1473 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1474 memcpy (filename_plus_orig_suffix + filename_len,
1475 ".orig", sizeof (".orig"));
1477 /* Try to stat() the .orig file. */
1478 if (stat (filename_plus_orig_suffix, &st) == 0)
1480 local_dot_orig_file_exists = TRUE;
1481 local_filename = filename_plus_orig_suffix;
1485 if (!local_dot_orig_file_exists)
1486 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1487 if (stat (*hstat.local_file, &st) == 0)
1488 local_filename = *hstat.local_file;
1490 if (local_filename != NULL)
1491 /* There was a local file, so we'll check later to see if the version
1492 the server has is the same version we already have, allowing us to
1497 local_size = st.st_size;
1501 /* Reset the counter. */
1503 *dt = 0 | ACCEPTRANGES;
1507 /* Increment the pass counter. */
1509 sleep_between_retrievals (count);
1510 /* Get the current time string. */
1511 tms = time_str (NULL);
1512 /* Print fetch message, if opt.verbose. */
1515 char *hurl = url_string (u, 1);
1519 sprintf (tmp, _("(try:%2d)"), count);
1520 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1521 tms, hurl, tmp, locf);
1523 ws_changetitle (hurl, 1);
1528 /* Default document type is empty. However, if spider mode is
1529 on or time-stamping is employed, HEAD_ONLY commands is
1530 encoded within *dt. */
1531 if (opt.spider || (use_ts && !got_head))
1535 /* Assume no restarting. */
1537 /* Decide whether or not to restart. */
1538 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1539 /* #### this calls access() and then stat(); could be optimized. */
1540 && file_exists_p (locf))
1541 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1542 hstat.restval = st.st_size;
1544 /* In `-c' is used and the file is existing and non-empty,
1545 refuse to truncate it if the server doesn't support continued
1547 hstat.no_truncate = 0;
1548 if (opt.always_rest && hstat.restval)
1549 hstat.no_truncate = 1;
1551 /* Decide whether to send the no-cache directive. We send it in
1553 a) we're using a proxy, and we're past our first retrieval.
1554 Some proxies are notorious for caching incomplete data, so
1555 we require a fresh get.
1556 b) caching is explicitly inhibited. */
1557 if ((proxy && count > 1) /* a */
1558 || !opt.allow_cache /* b */
1560 *dt |= SEND_NOCACHE;
1562 *dt &= ~SEND_NOCACHE;
1564 /* Try fetching the document, or at least its head. */
1565 err = gethttp (u, &hstat, dt, proxy);
1567 /* It's unfortunate that wget determines the local filename before finding
1568 out the Content-Type of the file. Barring a major restructuring of the
1569 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1570 *hstat.local_file to tack on ".html". */
1571 if (!opt.output_document)
1572 locf = *hstat.local_file;
1574 locf = opt.output_document;
1577 tms = time_str (NULL);
1578 /* Get the new location (with or without the redirection). */
1580 *newloc = xstrdup (hstat.newloc);
1583 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1584 case CONERROR: case READERR: case WRITEFAILED:
1586 /* Non-fatal errors continue executing the loop, which will
1587 bring them to "while" statement at the end, to judge
1588 whether the number of tries was exceeded. */
1589 free_hstat (&hstat);
1590 printwhat (count, opt.ntry);
1593 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1594 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1595 /* Fatal errors just return from the function. */
1596 free_hstat (&hstat);
1600 case FWRITEERR: case FOPENERR:
1601 /* Another fatal error. */
1602 logputs (LOG_VERBOSE, "\n");
1603 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1604 *hstat.local_file, strerror (errno));
1605 free_hstat (&hstat);
1610 /* Another fatal error. */
1611 logputs (LOG_VERBOSE, "\n");
1612 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1613 free_hstat (&hstat);
1618 /* Return the new location to the caller. */
1621 logprintf (LOG_NOTQUIET,
1622 _("ERROR: Redirection (%d) without location.\n"),
1624 free_hstat (&hstat);
1628 free_hstat (&hstat);
1633 /* The file was already fully retrieved. */
1634 free_hstat (&hstat);
1639 /* Deal with you later. */
1642 /* All possibilities should have been exhausted. */
1645 if (!(*dt & RETROKF))
1649 /* #### Ugly ugly ugly! */
1650 char *hurl = url_string (u, 1);
1651 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1654 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1655 tms, hstat.statcode, hstat.error);
1656 logputs (LOG_VERBOSE, "\n");
1657 free_hstat (&hstat);
1662 /* Did we get the time-stamp? */
1665 if (opt.timestamping && !hstat.remote_time)
1667 logputs (LOG_NOTQUIET, _("\
1668 Last-modified header missing -- time-stamps turned off.\n"));
1670 else if (hstat.remote_time)
1672 /* Convert the date-string into struct tm. */
1673 tmr = http_atotm (hstat.remote_time);
1674 if (tmr == (time_t) (-1))
1675 logputs (LOG_VERBOSE, _("\
1676 Last-modified header invalid -- time-stamp ignored.\n"));
1680 /* The time-stamping section. */
1685 use_ts = 0; /* no more time-stamping */
1686 count = 0; /* the retrieve count for HEAD is
1688 if (hstat.remote_time && tmr != (time_t) (-1))
1690 /* Now time-stamping can be used validly. Time-stamping
1691 means that if the sizes of the local and remote file
1692 match, and local file is newer than the remote file,
1693 it will not be retrieved. Otherwise, the normal
1694 download procedure is resumed. */
1696 (hstat.contlen == -1 || local_size == hstat.contlen))
1698 logprintf (LOG_VERBOSE, _("\
1699 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1701 free_hstat (&hstat);
1705 else if (tml >= tmr)
1706 logprintf (LOG_VERBOSE, _("\
1707 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1709 logputs (LOG_VERBOSE,
1710 _("Remote file is newer, retrieving.\n"));
1712 free_hstat (&hstat);
1715 if ((tmr != (time_t) (-1))
1717 && ((hstat.len == hstat.contlen) ||
1718 ((hstat.res == 0) &&
1719 ((hstat.contlen == -1) ||
1720 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1722 /* #### This code repeats in http.c and ftp.c. Move it to a
1724 const char *fl = NULL;
1725 if (opt.output_document)
1727 if (opt.od_known_regular)
1728 fl = opt.output_document;
1731 fl = *hstat.local_file;
1735 /* End of time-stamping section. */
1739 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1744 tmrate = retr_rate (hstat.len - hstat.restval, hstat.dltime, 0);
1746 if (hstat.len == hstat.contlen)
1750 logprintf (LOG_VERBOSE,
1751 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1752 tms, tmrate, locf, hstat.len, hstat.contlen);
1753 logprintf (LOG_NONVERBOSE,
1754 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1755 tms, u->url, hstat.len, hstat.contlen, locf, count);
1758 downloaded_increase (hstat.len);
1760 /* Remember that we downloaded the file for later ".orig" code. */
1761 if (*dt & ADDED_HTML_EXTENSION)
1762 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1764 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1766 free_hstat (&hstat);
1770 else if (hstat.res == 0) /* No read error */
1772 if (hstat.contlen == -1) /* We don't know how much we were supposed
1773 to get, so assume we succeeded. */
1777 logprintf (LOG_VERBOSE,
1778 _("%s (%s) - `%s' saved [%ld]\n\n"),
1779 tms, tmrate, locf, hstat.len);
1780 logprintf (LOG_NONVERBOSE,
1781 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1782 tms, u->url, hstat.len, locf, count);
1785 downloaded_increase (hstat.len);
1787 /* Remember that we downloaded the file for later ".orig" code. */
1788 if (*dt & ADDED_HTML_EXTENSION)
1789 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1791 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1793 free_hstat (&hstat);
1797 else if (hstat.len < hstat.contlen) /* meaning we lost the
1798 connection too soon */
1800 logprintf (LOG_VERBOSE,
1801 _("%s (%s) - Connection closed at byte %ld. "),
1802 tms, tmrate, hstat.len);
1803 printwhat (count, opt.ntry);
1804 free_hstat (&hstat);
1807 else if (!opt.kill_longer) /* meaning we got more than expected */
1809 logprintf (LOG_VERBOSE,
1810 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1811 tms, tmrate, locf, hstat.len, hstat.contlen);
1812 logprintf (LOG_NONVERBOSE,
1813 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1814 tms, u->url, hstat.len, hstat.contlen, locf, count);
1816 downloaded_increase (hstat.len);
1818 /* Remember that we downloaded the file for later ".orig" code. */
1819 if (*dt & ADDED_HTML_EXTENSION)
1820 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1822 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1824 free_hstat (&hstat);
1828 else /* the same, but not accepted */
1830 logprintf (LOG_VERBOSE,
1831 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1832 tms, tmrate, hstat.len, hstat.contlen);
1833 printwhat (count, opt.ntry);
1834 free_hstat (&hstat);
1838 else /* now hstat.res can only be -1 */
1840 if (hstat.contlen == -1)
1842 logprintf (LOG_VERBOSE,
1843 _("%s (%s) - Read error at byte %ld (%s)."),
1844 tms, tmrate, hstat.len, strerror (errno));
1845 printwhat (count, opt.ntry);
1846 free_hstat (&hstat);
1849 else /* hstat.res == -1 and contlen is given */
1851 logprintf (LOG_VERBOSE,
1852 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1853 tms, tmrate, hstat.len, hstat.contlen,
1855 printwhat (count, opt.ntry);
1856 free_hstat (&hstat);
1863 while (!opt.ntry || (count < opt.ntry));
1867 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1868 than local timezone.
1870 mktime is similar but assumes struct tm, also known as the
1871 "broken-down" form of time, is in local time zone. mktime_from_utc
1872 uses mktime to make the conversion understanding that an offset
1873 will be introduced by the local time assumption.
1875 mktime_from_utc then measures the introduced offset by applying
1876 gmtime to the initial result and applying mktime to the resulting
1877 "broken-down" form. The difference between the two mktime results
1878 is the measured offset which is then subtracted from the initial
1879 mktime result to yield a calendar time which is the value returned.
1881 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1882 consistent offset (the non DST offset) since tm and tm+o might be
1883 on opposite sides of a DST change.
1885 Some implementations of mktime return -1 for the nonexistent
1886 localtime hour at the beginning of DST. In this event, use
1887 mktime(tm - 1hr) + 3600.
1891 gmtime(t+o) --> tm+o
1892 mktime(tm+o) --> t+2o
1893 t+o - (t+2o - t+o) = t
1895 Note that glibc contains a function of the same purpose named
1896 `timegm' (reverse of gmtime). But obviously, it is not universally
1897 available, and unfortunately it is not straightforwardly
1898 extractable for use here. Perhaps configure should detect timegm
1899 and use it where available.
1901 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1902 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1903 Further improved by Roger with assistance from Edward J. Sabol
1904 based on input by Jamie Zawinski. */
1907 mktime_from_utc (struct tm *t)
1918 return -1; /* can't deal with output from strptime */
1929 return -1; /* can't deal with output from gmtime */
1932 return (tl - (tb - tl));
1935 /* Check whether the result of strptime() indicates success.
1936 strptime() returns the pointer to how far it got to in the string.
1937 The processing has been successful if the string is at `GMT' or
1938 `+X', or at the end of the string.
1940 In extended regexp parlance, the function returns 1 if P matches
1941 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1942 can return) is considered a failure and 0 is returned. */
1944 check_end (const char *p)
1948 while (ISSPACE (*p))
1951 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1952 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1958 /* Convert the textual specification of time in TIME_STRING to the
1959 number of seconds since the Epoch.
1961 TIME_STRING can be in any of the three formats RFC2068 allows the
1962 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1963 Timezones are ignored, and should be GMT.
1965 Return the computed time_t representation, or -1 if the conversion
1968 This function uses strptime with various string formats for parsing
1969 TIME_STRING. This results in a parser that is not as lenient in
1970 interpreting TIME_STRING as I would like it to be. Being based on
1971 strptime, it always allows shortened months, one-digit days, etc.,
1972 but due to the multitude of formats in which time can be
1973 represented, an ideal HTTP time parser would be even more
1974 forgiving. It should completely ignore things like week days and
1975 concentrate only on the various forms of representing years,
1976 months, days, hours, minutes, and seconds. For example, it would
1977 be nice if it accepted ISO 8601 out of the box.
1979 I've investigated free and PD code for this purpose, but none was
1980 usable. getdate was big and unwieldy, and had potential copyright
1981 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1982 distributed with phttpd, is excellent, but we cannot use it because
1983 it is not assigned to the FSF. So I stuck it with strptime. */
1986 http_atotm (char *time_string)
1988 /* NOTE: Solaris strptime man page claims that %n and %t match white
1989 space, but that's not universally available. Instead, we simply
1990 use ` ' to mean "skip all WS", which works under all strptime
1991 implementations I've tested. */
1993 static const char *time_formats[] = {
1994 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1995 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1996 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1997 (google.com uses this for their cookies.) */
1998 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2004 /* According to Roger Beeman, we need to initialize tm_isdst, since
2005 strptime won't do it. */
2008 /* Note that under foreign locales Solaris strptime() fails to
2009 recognize English dates, which renders this function useless. We
2010 solve this by being careful not to affect LC_TIME when
2011 initializing locale.
2013 Another solution would be to temporarily set locale to C, invoke
2014 strptime(), and restore it back. This is slow and dirty,
2015 however, and locale support other than LC_MESSAGES can mess other
2016 things, so I rather chose to stick with just setting LC_MESSAGES.
2018 GNU strptime does not have this problem because it recognizes
2019 both international and local dates. */
2021 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
2022 if (check_end (strptime (time_string, time_formats[i], &t)))
2023 return mktime_from_utc (&t);
2025 /* All formats have failed. */
2029 /* Authorization support: We support two authorization schemes:
2031 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2033 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2034 consisting of answering to the server's challenge with the proper
2037 /* How many bytes it will take to store LEN bytes in base64. */
2038 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2040 /* Encode the string S of length LENGTH to base64 format and place it
2041 to STORE. STORE will be 0-terminated, and must point to a writable
2042 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2044 base64_encode (const char *s, char *store, int length)
2046 /* Conversion table. */
2047 static char tbl[64] = {
2048 'A','B','C','D','E','F','G','H',
2049 'I','J','K','L','M','N','O','P',
2050 'Q','R','S','T','U','V','W','X',
2051 'Y','Z','a','b','c','d','e','f',
2052 'g','h','i','j','k','l','m','n',
2053 'o','p','q','r','s','t','u','v',
2054 'w','x','y','z','0','1','2','3',
2055 '4','5','6','7','8','9','+','/'
2058 unsigned char *p = (unsigned char *)store;
2060 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2061 for (i = 0; i < length; i += 3)
2063 *p++ = tbl[s[0] >> 2];
2064 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2065 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2066 *p++ = tbl[s[2] & 0x3f];
2069 /* Pad the result if necessary... */
2070 if (i == length + 1)
2072 else if (i == length + 2)
2073 *(p - 1) = *(p - 2) = '=';
2074 /* ...and zero-terminate it. */
2078 /* Create the authentication header contents for the `Basic' scheme.
2079 This is done by encoding the string `USER:PASS' in base64 and
2080 prepending `HEADER: Basic ' to it. */
2082 basic_authentication_encode (const char *user, const char *passwd,
2085 char *t1, *t2, *res;
2086 int len1 = strlen (user) + 1 + strlen (passwd);
2087 int len2 = BASE64_LENGTH (len1);
2089 t1 = (char *)alloca (len1 + 1);
2090 sprintf (t1, "%s:%s", user, passwd);
2091 t2 = (char *)alloca (1 + len2);
2092 base64_encode (t1, t2, len1);
2093 res = (char *)xmalloc (len2 + 11 + strlen (header));
2094 sprintf (res, "%s: Basic %s\r\n", header, t2);
2100 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2101 of a field in such a header. If the field is the one specified by
2102 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2103 digest authorization code), extract its value in the (char*)
2104 variable pointed by RET. Returns negative on a malformed header,
2105 or number of bytes that have been parsed by this call. */
2107 extract_header_attr (const char *au, const char *attr_name, char **ret)
2109 const char *cp, *ep;
2113 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2115 cp += strlen (attr_name);
2118 cp += skip_lws (cp);
2123 cp += skip_lws (cp);
2128 for (ep = cp; *ep && *ep != '\"'; ep++)
2133 *ret = strdupdelim (cp, ep);
2140 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2141 an array of 16 bytes containing the hash keys, and BUF should be a
2142 buffer of 33 writable characters (32 for hex digits plus one for
2143 zero termination). */
2145 dump_hash (unsigned char *buf, const unsigned char *hash)
2149 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2151 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2152 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2157 /* Take the line apart to find the challenge, and compose a digest
2158 authorization header. See RFC2069 section 2.1.2. */
2160 digest_authentication_encode (const char *au, const char *user,
2161 const char *passwd, const char *method,
2164 static char *realm, *opaque, *nonce;
2169 { "realm", &realm },
2170 { "opaque", &opaque },
2175 realm = opaque = nonce = NULL;
2177 au += 6; /* skip over `Digest' */
2182 au += skip_lws (au);
2183 for (i = 0; i < ARRAY_SIZE (options); i++)
2185 int skip = extract_header_attr (au, options[i].name,
2186 options[i].variable);
2190 FREE_MAYBE (opaque);
2200 if (i == ARRAY_SIZE (options))
2202 while (*au && *au != '=')
2206 au += skip_lws (au);
2210 while (*au && *au != '\"')
2217 while (*au && *au != ',')
2222 if (!realm || !nonce || !user || !passwd || !path || !method)
2225 FREE_MAYBE (opaque);
2230 /* Calculate the digest value. */
2232 MD5_CONTEXT_TYPE ctx;
2233 unsigned char hash[MD5_HASHLEN];
2234 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2235 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2237 /* A1BUF = H(user ":" realm ":" password) */
2239 MD5_UPDATE (user, strlen (user), &ctx);
2240 MD5_UPDATE (":", 1, &ctx);
2241 MD5_UPDATE (realm, strlen (realm), &ctx);
2242 MD5_UPDATE (":", 1, &ctx);
2243 MD5_UPDATE (passwd, strlen (passwd), &ctx);
2244 MD5_FINISH (&ctx, hash);
2245 dump_hash (a1buf, hash);
2247 /* A2BUF = H(method ":" path) */
2249 MD5_UPDATE (method, strlen (method), &ctx);
2250 MD5_UPDATE (":", 1, &ctx);
2251 MD5_UPDATE (path, strlen (path), &ctx);
2252 MD5_FINISH (&ctx, hash);
2253 dump_hash (a2buf, hash);
2255 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2257 MD5_UPDATE (a1buf, MD5_HASHLEN * 2, &ctx);
2258 MD5_UPDATE (":", 1, &ctx);
2259 MD5_UPDATE (nonce, strlen (nonce), &ctx);
2260 MD5_UPDATE (":", 1, &ctx);
2261 MD5_UPDATE (a2buf, MD5_HASHLEN * 2, &ctx);
2262 MD5_FINISH (&ctx, hash);
2263 dump_hash (response_digest, hash);
2265 res = (char*) xmalloc (strlen (user)
2270 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2271 + (opaque ? strlen (opaque) : 0)
2273 sprintf (res, "Authorization: Digest \
2274 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2275 user, realm, nonce, path, response_digest);
2278 char *p = res + strlen (res);
2279 strcat (p, ", opaque=\"");
2283 strcat (res, "\r\n");
2287 #endif /* USE_DIGEST */
2290 #define BEGINS_WITH(line, string_constant) \
2291 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2292 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2293 || !line[sizeof (string_constant) - 1]))
2296 known_authentication_scheme_p (const char *au)
2298 return BEGINS_WITH (au, "Basic")
2299 || BEGINS_WITH (au, "Digest")
2300 || BEGINS_WITH (au, "NTLM");
2305 /* Create the HTTP authorization request header. When the
2306 `WWW-Authenticate' response header is seen, according to the
2307 authorization scheme specified in that header (`Basic' and `Digest'
2308 are supported by the current implementation), produce an
2309 appropriate HTTP authorization request header. */
2311 create_authorization_line (const char *au, const char *user,
2312 const char *passwd, const char *method,
2315 char *wwwauth = NULL;
2317 if (!strncasecmp (au, "Basic", 5))
2318 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2319 if (!strncasecmp (au, "NTLM", 4))
2320 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2322 else if (!strncasecmp (au, "Digest", 6))
2323 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2324 #endif /* USE_DIGEST */
2331 if (pc_last_host_ip)
2332 address_list_release (pc_last_host_ip);