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 struct wget_timer *timer;
530 char *cookies = NULL;
532 /* Whether this connection will be kept alive after the HTTP request
536 /* Flags that detect the two ways of specifying HTTP keep-alive
538 int http_keep_alive_1, http_keep_alive_2;
540 /* Whether keep-alive should be inhibited. */
541 int inhibit_keep_alive;
544 /* initialize ssl_ctx on first run */
547 err=init_ssl (&ssl_ctx);
552 case SSLERRCTXCREATE:
554 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
558 /* try without certfile */
559 logprintf (LOG_NOTQUIET,
560 _("Failed to load certificates from %s\n"),
563 logprintf (LOG_NOTQUIET,
564 _("Trying without the specified certificate\n"));
567 logprintf (LOG_NOTQUIET,
568 _("Failed to get certificate key from %s\n"),
571 logprintf (LOG_NOTQUIET,
572 _("Trying without the specified certificate\n"));
579 #endif /* HAVE_SSL */
581 if (!(*dt & HEAD_ONLY))
582 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
583 know the local filename so we can save to it. */
584 assert (u->local != NULL);
587 auth_tried_already = 0;
589 inhibit_keep_alive = (!opt.http_keep_alive || u->proxy != NULL);
592 /* We need to come back here when the initial attempt to retrieve
593 without authorization header fails. (Expected to happen at least
594 for the Digest authorization scheme.) */
597 http_keep_alive_1 = http_keep_alive_2 = 0;
600 cookies = build_cookies_request (u->host, u->port, u->path,
601 u->proto == URLHTTPS);
603 /* Initialize certain elements of struct http_stat. */
608 hs->remote_time = NULL;
611 /* Which structure to use to retrieve the original URL data. */
617 /* First: establish the connection. */
618 if (inhibit_keep_alive
621 !persistent_available_p (u->host, u->port)
623 !persistent_available_p (u->host, u->port, (u->proto==URLHTTPS ? 1 : 0))
624 #endif /* HAVE_SSL */
627 logprintf (LOG_VERBOSE, _("Connecting to %s:%hu... "), u->host, u->port);
628 err = make_connection (&sock, u->host, u->port);
632 logputs (LOG_VERBOSE, "\n");
633 logprintf (LOG_NOTQUIET, "%s: %s.\n", u->host, herrmsg (h_errno));
637 logputs (LOG_VERBOSE, "\n");
638 logprintf (LOG_NOTQUIET, "socket: %s\n", strerror (errno));
642 logputs (LOG_VERBOSE, "\n");
643 logprintf (LOG_NOTQUIET,
644 _("Connection to %s:%hu refused.\n"), u->host, u->port);
648 logputs (LOG_VERBOSE, "\n");
649 logprintf (LOG_NOTQUIET, "connect: %s\n", strerror (errno));
654 /* Everything is fine! */
655 logputs (LOG_VERBOSE, _("connected!\n"));
662 if (u->proto == URLHTTPS)
663 if (connect_ssl (&ssl, ssl_ctx,sock) != 0)
665 logputs (LOG_VERBOSE, "\n");
666 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
670 #endif /* HAVE_SSL */
674 logprintf (LOG_VERBOSE, _("Reusing connection to %s:%hu.\n"), u->host, u->port);
675 /* #### pc_last_fd should be accessed through an accessor
680 #endif /* HAVE_SSL */
681 DEBUGP (("Reusing fd %d.\n", sock));
685 path = u->proxy->url;
689 command = (*dt & HEAD_ONLY) ? "HEAD" : "GET";
693 referer = (char *)alloca (9 + strlen (ou->referer) + 3);
694 sprintf (referer, "Referer: %s\r\n", ou->referer);
696 if (*dt & SEND_NOCACHE)
697 pragma_h = "Pragma: no-cache\r\n";
702 range = (char *)alloca (13 + numdigit (hs->restval) + 4);
703 /* Gag me! Some servers (e.g. WebSitePro) have been known to
704 respond to the following `Range' format by generating a
705 multipart/x-byte-ranges MIME document! This MIME type was
706 present in an old draft of the byteranges specification.
707 HTTP/1.1 specifies a multipart/byte-ranges MIME type, but
708 only if multiple non-overlapping ranges are requested --
709 which Wget never does. */
710 sprintf (range, "Range: bytes=%ld-\r\n", hs->restval);
715 STRDUP_ALLOCA (useragent, opt.useragent);
718 useragent = (char *)alloca (10 + strlen (version_string));
719 sprintf (useragent, "Wget/%s", version_string);
721 /* Construct the authentication, if userid is present. */
724 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
725 user = user ? user : opt.http_user;
726 passwd = passwd ? passwd : opt.http_passwd;
733 /* We have the username and the password, but haven't tried
734 any authorization yet. Let's see if the "Basic" method
735 works. If not, we'll come back here and construct a
736 proper authorization method with the right challenges.
738 If we didn't employ this kind of logic, every URL that
739 requires authorization would have to be processed twice,
740 which is very suboptimal and generates a bunch of false
741 "unauthorized" errors in the server log.
743 #### But this logic also has a serious problem when used
744 with stronger authentications: we *first* transmit the
745 username and the password in clear text, and *then*
746 attempt a stronger authentication scheme. That cannot be
747 right! We are only fortunate that almost everyone still
748 uses the `Basic' scheme anyway.
750 There should be an option to prevent this from happening,
751 for those who use strong authentication schemes and value
753 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
757 wwwauth = create_authorization_line (authenticate_h, user, passwd,
765 char *proxy_user, *proxy_passwd;
766 /* For normal username and password, URL components override
767 command-line/wgetrc parameters. With proxy authentication,
768 it's the reverse, because proxy URLs are normally the
769 "permanent" ones, so command-line args should take
771 if (opt.proxy_user && opt.proxy_passwd)
773 proxy_user = opt.proxy_user;
774 proxy_passwd = opt.proxy_passwd;
778 proxy_user = u->user;
779 proxy_passwd = u->passwd;
781 /* #### This is junky. Can't the proxy request, say, `Digest'
783 if (proxy_user && proxy_passwd)
784 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd,
785 "Proxy-Authorization");
790 /* String of the form :PORT. Used only for non-standard ports. */
794 && remport != (u->proto == URLHTTPS
795 ? DEFAULT_HTTPS_PORT : DEFAULT_HTTP_PORT)
797 && remport != DEFAULT_HTTP_PORT
801 port_maybe = (char *)alloca (numdigit (remport) + 2);
802 sprintf (port_maybe, ":%d", remport);
805 if (!inhibit_keep_alive)
806 request_keep_alive = "Connection: Keep-Alive\r\n";
808 request_keep_alive = NULL;
810 /* Allocate the memory for the request. */
811 request = (char *)alloca (strlen (command) + strlen (path)
814 + (port_maybe ? strlen (port_maybe) : 0)
815 + strlen (HTTP_ACCEPT)
816 + (request_keep_alive
817 ? strlen (request_keep_alive) : 0)
818 + (referer ? strlen (referer) : 0)
819 + (cookies ? strlen (cookies) : 0)
820 + (wwwauth ? strlen (wwwauth) : 0)
821 + (proxyauth ? strlen (proxyauth) : 0)
822 + (range ? strlen (range) : 0)
824 + (opt.user_header ? strlen (opt.user_header) : 0)
826 /* Construct the request. */
832 %s%s%s%s%s%s%s%s\r\n",
833 command, path, useragent, remhost,
834 port_maybe ? port_maybe : "",
836 request_keep_alive ? request_keep_alive : "",
837 referer ? referer : "",
838 cookies ? cookies : "",
839 wwwauth ? wwwauth : "",
840 proxyauth ? proxyauth : "",
843 opt.user_header ? opt.user_header : "");
844 DEBUGP (("---request begin---\n%s---request end---\n", request));
845 /* Free the temporary memory. */
846 FREE_MAYBE (wwwauth);
847 FREE_MAYBE (proxyauth);
848 FREE_MAYBE (cookies);
850 /* Send the request to server. */
852 if (u->proto == URLHTTPS)
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 u->proxy ? "Proxy" : "HTTP");
867 contlen = contrange = -1;
872 /* Before reading anything, initialize the rbuf. */
873 rbuf_initialize (&rbuf, sock);
875 if (u->proto == URLHTTPS)
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 /* Print the header if requested. */
941 if (opt.server_response && hcount != 1)
942 logprintf (LOG_VERBOSE, "\n%d %s", hcount, hdr);
944 /* Check for status line. */
948 /* Parse the first line of server response. */
949 statcode = parse_http_status_line (hdr, &error);
950 hs->statcode = statcode;
951 /* Store the descriptive response. */
952 if (statcode == -1) /* malformed response */
954 /* A common reason for "malformed response" error is the
955 case when no data was actually received. Handle this
958 hs->error = xstrdup (_("No data received"));
960 hs->error = xstrdup (_("Malformed status line"));
965 hs->error = xstrdup (_("(no description)"));
967 hs->error = xstrdup (error);
974 logprintf (LOG_VERBOSE, "%d %s", statcode, error);
979 /* Exit on empty header. */
986 /* Try getting content-length. */
987 if (contlen == -1 && !opt.ignore_length)
988 if (header_process (hdr, "Content-Length", header_extract_number,
991 /* Try getting content-type. */
993 if (header_process (hdr, "Content-Type", http_process_type, &type))
995 /* Try getting location. */
997 if (header_process (hdr, "Location", header_strdup, &hs->newloc))
999 /* Try getting last-modified. */
1000 if (!hs->remote_time)
1001 if (header_process (hdr, "Last-Modified", header_strdup,
1004 /* Try getting cookies. */
1006 if (header_process (hdr, "Set-Cookie", set_cookie_header_cb, u))
1008 /* Try getting www-authentication. */
1009 if (!authenticate_h)
1010 if (header_process (hdr, "WWW-Authenticate", header_strdup,
1013 /* Check for accept-ranges header. If it contains the word
1014 `none', disable the ranges. */
1015 if (*dt & ACCEPTRANGES)
1018 if (header_process (hdr, "Accept-Ranges", http_process_none, &nonep))
1021 *dt &= ~ACCEPTRANGES;
1025 /* Try getting content-range. */
1026 if (contrange == -1)
1028 struct http_process_range_closure closure;
1029 if (header_process (hdr, "Content-Range", http_process_range, &closure))
1031 contrange = closure.first_byte_pos;
1035 /* Check for keep-alive related responses. */
1036 if (!inhibit_keep_alive)
1038 /* Check for the `Keep-Alive' header. */
1039 if (!http_keep_alive_1)
1041 if (header_process (hdr, "Keep-Alive", header_exists,
1042 &http_keep_alive_1))
1045 /* Check for `Connection: Keep-Alive'. */
1046 if (!http_keep_alive_2)
1048 if (header_process (hdr, "Connection", http_process_connection,
1049 &http_keep_alive_2))
1057 logputs (LOG_VERBOSE, "\n");
1060 && (http_keep_alive_1 || http_keep_alive_2))
1062 assert (inhibit_keep_alive == 0);
1066 /* The server has promised that it will not close the connection
1067 when we're done. This means that we can register it. */
1069 register_persistent (u->host, u->port, sock);
1071 register_persistent (u->host, u->port, sock, ssl);
1072 #endif /* HAVE_SSL */
1074 if ((statcode == HTTP_STATUS_UNAUTHORIZED)
1077 /* Authorization is required. */
1081 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1082 might be more bytes in the body. */
1083 if (auth_tried_already)
1085 /* If we have tried it already, then there is not point
1088 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1089 xfree (authenticate_h);
1092 else if (!known_authentication_scheme_p (authenticate_h))
1094 xfree (authenticate_h);
1095 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1098 else if (BEGINS_WITH (authenticate_h, "Basic"))
1100 /* The authentication scheme is basic, the one we try by
1101 default, and it failed. There's no sense in trying
1107 auth_tried_already = 1;
1111 /* We do not need this anymore. */
1114 xfree (authenticate_h);
1115 authenticate_h = NULL;
1118 /* 20x responses are counted among successful by default. */
1119 if (H_20X (statcode))
1122 /* Return if redirected. */
1123 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1125 /* RFC2068 says that in case of the 300 (multiple choices)
1126 response, the server can output a preferred URL through
1127 `Location' header; otherwise, the request should be treated
1128 like GET. So, if the location is set, it will be a
1129 redirection; otherwise, just proceed normally. */
1130 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1134 logprintf (LOG_VERBOSE,
1135 _("Location: %s%s\n"),
1136 hs->newloc ? hs->newloc : _("unspecified"),
1137 hs->newloc ? _(" [following]") : "");
1138 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1139 might be more bytes in the body. */
1141 FREE_MAYBE (all_headers);
1146 if (type && !strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)))
1149 /* We don't assume text/html by default. */
1152 if (opt.html_extension && (*dt & TEXTHTML))
1153 /* -E / --html-extension / html_extension = on was specified, and this is a
1154 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1155 already the file's suffix, tack on ".html". */
1157 char* last_period_in_local_filename = strrchr(u->local, '.');
1159 if (last_period_in_local_filename == NULL ||
1160 !(strcasecmp(last_period_in_local_filename, ".htm") == EQ ||
1161 strcasecmp(last_period_in_local_filename, ".html") == EQ))
1163 size_t local_filename_len = strlen(u->local);
1165 u->local = xrealloc(u->local, local_filename_len + sizeof(".html"));
1166 strcpy(u->local + local_filename_len, ".html");
1168 *dt |= ADDED_HTML_EXTENSION;
1172 if (contrange == -1)
1174 /* We did not get a content-range header. This means that the
1175 server did not honor our `Range' request. Normally, this
1176 means we should reset hs->restval and continue normally. */
1178 /* However, if `-c' is used, we need to be a bit more careful:
1180 1. If `-c' is specified and the file already existed when
1181 Wget was started, it would be a bad idea for us to start
1182 downloading it from scratch, effectively truncating it. I
1183 believe this cannot happen unless `-c' was specified.
1185 2. If `-c' is used on a file that is already fully
1186 downloaded, we're requesting bytes after the end of file,
1187 which can result in server not honoring `Range'. If this is
1188 the case, `Content-Length' will be equal to the length of the
1190 if (opt.always_rest)
1192 /* Check for condition #2. */
1193 if (hs->restval >= contlen)
1195 logputs (LOG_VERBOSE, _("\
1196 \n The file is already fully retrieved; nothing to do.\n\n"));
1197 /* In case the caller inspects. */
1201 FREE_MAYBE (all_headers);
1202 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1203 might be more bytes in the body. */
1204 return RETRUNNEEDED;
1207 /* Check for condition #1. */
1208 if (hs->no_truncate)
1210 logprintf (LOG_NOTQUIET,
1213 The server does not support continued downloads, which conflicts with `-c'.\n\
1214 Refusing to truncate `%s'.\n\n"), u->local);
1216 FREE_MAYBE (all_headers);
1217 CLOSE_INVALIDATE (sock);
1218 return CONTNOTSUPPORTED;
1226 else if (contrange != hs->restval ||
1227 (H_PARTIAL (statcode) && contrange == -1))
1229 /* This means the whole request was somehow misunderstood by the
1230 server. Bail out. */
1232 FREE_MAYBE (all_headers);
1233 CLOSE_INVALIDATE (sock);
1240 contlen += contrange;
1242 contrange = -1; /* If conent-length was not sent,
1243 content-range will be ignored. */
1245 hs->contlen = contlen;
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);
1336 timer = wtimer_new ();
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 = wtimer_elapsed (timer);
1342 wtimer_delete (timer);
1344 /* Close or flush the file. We have to be careful to check for
1345 error here. Checking the result of fwrite() is not enough --
1346 errors could go unnoticed! */
1349 flush_res = fclose (fp);
1351 flush_res = fflush (fp);
1352 if (flush_res == EOF)
1355 FREE_MAYBE (all_headers);
1356 CLOSE_FINISH (sock);
1359 return RETRFINISHED;
1362 /* The genuine HTTP loop! This is the part where the retrieval is
1363 retried, and retried, and retried, and... */
1365 http_loop (struct urlinfo *u, char **newloc, int *dt)
1368 int use_ts, got_head = 0; /* time-stamping info */
1369 char *filename_plus_orig_suffix;
1370 char *local_filename = NULL;
1371 char *tms, *suf, *locf, *tmrate;
1373 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1374 long local_size = 0; /* the size of the local file */
1375 size_t filename_len;
1376 struct http_stat hstat; /* HTTP status */
1379 /* This used to be done in main(), but it's a better idea to do it
1380 here so that we don't go through the hoops if we're just using
1382 if (opt.cookies && opt.cookies_input && !cookies_loaded_p)
1384 load_cookies (opt.cookies_input);
1385 cookies_loaded_p = 1;
1390 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1391 has_wildcards_p because it would also warn on `?', and we know that
1392 shows up in CGI paths a *lot*. */
1393 if (strchr (u->url, '*'))
1394 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1396 /* Determine the local filename. */
1398 u->local = url_filename (u->proxy ? u->proxy : u);
1400 if (!opt.output_document)
1403 locf = opt.output_document;
1405 /* Yuck. Multiple returns suck. We need to remember to free() the space we
1406 xmalloc() here before EACH return. This is one reason it's better to set
1407 flags that influence flow control and then return once at the end. */
1408 filename_len = strlen(u->local);
1409 filename_plus_orig_suffix = xmalloc(filename_len + sizeof(".orig"));
1411 if (opt.noclobber && file_exists_p (u->local))
1413 /* If opt.noclobber is turned on and file already exists, do not
1414 retrieve the file */
1415 logprintf (LOG_VERBOSE, _("\
1416 File `%s' already there, will not retrieve.\n"), u->local);
1417 /* If the file is there, we suppose it's retrieved OK. */
1420 /* #### Bogusness alert. */
1421 /* If its suffix is "html" or (yuck!) "htm", we suppose it's
1422 text/html, a harmless lie. */
1423 if (((suf = suffix (u->local)) != NULL)
1424 && (!strcmp (suf, "html") || !strcmp (suf, "htm")))
1427 xfree (filename_plus_orig_suffix); /* must precede every return! */
1428 /* Another harmless lie: */
1433 if (opt.timestamping)
1435 boolean local_dot_orig_file_exists = FALSE;
1437 if (opt.backup_converted)
1438 /* If -K is specified, we'll act on the assumption that it was specified
1439 last time these files were downloaded as well, and instead of just
1440 comparing local file X against server file X, we'll compare local
1441 file X.orig (if extant, else X) against server file X. If -K
1442 _wasn't_ specified last time, or the server contains files called
1443 *.orig, -N will be back to not operating correctly with -k. */
1445 /* Would a single s[n]printf() call be faster? --dan
1447 Definitely not. sprintf() is horribly slow. It's a
1448 different question whether the difference between the two
1449 affects a program. Usually I'd say "no", but at one
1450 point I profiled Wget, and found that a measurable and
1451 non-negligible amount of time was lost calling sprintf()
1452 in url.c. Replacing sprintf with inline calls to
1453 strcpy() and long_to_string() made a difference.
1455 strcpy(filename_plus_orig_suffix, u->local);
1456 strcpy(filename_plus_orig_suffix + filename_len, ".orig");
1458 /* Try to stat() the .orig file. */
1459 if (stat(filename_plus_orig_suffix, &st) == 0)
1461 local_dot_orig_file_exists = TRUE;
1462 local_filename = filename_plus_orig_suffix;
1466 if (!local_dot_orig_file_exists)
1467 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1468 if (stat (u->local, &st) == 0)
1469 local_filename = u->local;
1471 if (local_filename != NULL)
1472 /* There was a local file, so we'll check later to see if the version
1473 the server has is the same version we already have, allowing us to
1478 local_size = st.st_size;
1482 /* Reset the counter. */
1484 *dt = 0 | ACCEPTRANGES;
1488 /* Increment the pass counter. */
1490 sleep_between_retrievals (count);
1491 /* Get the current time string. */
1492 tms = time_str (NULL);
1493 /* Print fetch message, if opt.verbose. */
1496 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1500 sprintf (tmp, _("(try:%2d)"), count);
1501 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1502 tms, hurl, tmp, locf);
1504 ws_changetitle (hurl, 1);
1509 /* Default document type is empty. However, if spider mode is
1510 on or time-stamping is employed, HEAD_ONLY commands is
1511 encoded within *dt. */
1512 if (opt.spider || (use_ts && !got_head))
1516 /* Assume no restarting. */
1518 /* Decide whether or not to restart. */
1519 if (((count > 1 && (*dt & ACCEPTRANGES)) || opt.always_rest)
1520 && file_exists_p (locf))
1521 if (stat (locf, &st) == 0 && S_ISREG (st.st_mode))
1522 hstat.restval = st.st_size;
1524 /* In `-c' is used and the file is existing and non-empty,
1525 refuse to truncate it if the server doesn't support continued
1527 if (opt.always_rest && hstat.restval)
1528 hstat.no_truncate = file_exists_p (locf);
1530 /* Decide whether to send the no-cache directive. We send it in
1532 a) we're using a proxy, and we're past our first retrieval.
1533 Some proxies are notorious for caching incomplete data, so
1534 we require a fresh get.
1535 b) caching is explicitly inhibited. */
1536 if ((u->proxy && count > 1) /* a */
1537 || !opt.allow_cache /* b */
1539 *dt |= SEND_NOCACHE;
1541 *dt &= ~SEND_NOCACHE;
1543 /* Try fetching the document, or at least its head. :-) */
1544 err = gethttp (u, &hstat, dt);
1546 /* It's unfortunate that wget determines the local filename before finding
1547 out the Content-Type of the file. Barring a major restructuring of the
1548 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1549 u->local to tack on ".html". */
1550 if (!opt.output_document)
1553 locf = opt.output_document;
1556 tms = time_str (NULL);
1557 /* Get the new location (with or without the redirection). */
1559 *newloc = xstrdup (hstat.newloc);
1562 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1563 case CONERROR: case READERR: case WRITEFAILED:
1565 /* Non-fatal errors continue executing the loop, which will
1566 bring them to "while" statement at the end, to judge
1567 whether the number of tries was exceeded. */
1569 printwhat (count, opt.ntry);
1572 case HOSTERR: case CONREFUSED: case PROXERR: case AUTHFAILED:
1573 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1574 /* Fatal errors just return from the function. */
1576 xfree (filename_plus_orig_suffix); /* must precede every return! */
1579 case FWRITEERR: case FOPENERR:
1580 /* Another fatal error. */
1581 logputs (LOG_VERBOSE, "\n");
1582 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1583 u->local, strerror (errno));
1588 /* Another fatal error. */
1589 logputs (LOG_VERBOSE, "\n");
1590 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
1592 xfree (filename_plus_orig_suffix); /* must precede every return! */
1596 /* Return the new location to the caller. */
1599 logprintf (LOG_NOTQUIET,
1600 _("ERROR: Redirection (%d) without location.\n"),
1602 xfree (filename_plus_orig_suffix); /* must precede every return! */
1606 xfree (filename_plus_orig_suffix); /* must precede every return! */
1610 /* The file was already fully retrieved. */
1612 xfree (filename_plus_orig_suffix); /* must precede every return! */
1616 /* Deal with you later. */
1619 /* All possibilities should have been exhausted. */
1622 if (!(*dt & RETROKF))
1626 /* #### Ugly ugly ugly! */
1627 char *hurl = str_url (u->proxy ? u->proxy : u, 1);
1628 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
1631 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
1632 tms, hstat.statcode, hstat.error);
1633 logputs (LOG_VERBOSE, "\n");
1635 xfree (filename_plus_orig_suffix); /* must precede every return! */
1639 /* Did we get the time-stamp? */
1642 if (opt.timestamping && !hstat.remote_time)
1644 logputs (LOG_NOTQUIET, _("\
1645 Last-modified header missing -- time-stamps turned off.\n"));
1647 else if (hstat.remote_time)
1649 /* Convert the date-string into struct tm. */
1650 tmr = http_atotm (hstat.remote_time);
1651 if (tmr == (time_t) (-1))
1652 logputs (LOG_VERBOSE, _("\
1653 Last-modified header invalid -- time-stamp ignored.\n"));
1657 /* The time-stamping section. */
1662 use_ts = 0; /* no more time-stamping */
1663 count = 0; /* the retrieve count for HEAD is
1665 if (hstat.remote_time && tmr != (time_t) (-1))
1667 /* Now time-stamping can be used validly. Time-stamping
1668 means that if the sizes of the local and remote file
1669 match, and local file is newer than the remote file,
1670 it will not be retrieved. Otherwise, the normal
1671 download procedure is resumed. */
1673 (hstat.contlen == -1 || local_size == hstat.contlen))
1675 logprintf (LOG_VERBOSE, _("\
1676 Server file no newer than local file `%s' -- not retrieving.\n\n"),
1679 xfree (filename_plus_orig_suffix); /*must precede every return!*/
1682 else if (tml >= tmr)
1683 logprintf (LOG_VERBOSE, _("\
1684 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
1686 logputs (LOG_VERBOSE,
1687 _("Remote file is newer, retrieving.\n"));
1692 if ((tmr != (time_t) (-1))
1694 && ((hstat.len == hstat.contlen) ||
1695 ((hstat.res == 0) &&
1696 ((hstat.contlen == -1) ||
1697 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
1699 /* #### This code repeats in http.c and ftp.c. Move it to a
1701 const char *fl = NULL;
1702 if (opt.output_document)
1704 if (opt.od_known_regular)
1705 fl = opt.output_document;
1712 /* End of time-stamping section. */
1716 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
1717 xfree (filename_plus_orig_suffix); /* must precede every return! */
1721 /* It is now safe to free the remainder of hstat, since the
1722 strings within it will no longer be used. */
1725 tmrate = rate (hstat.len - hstat.restval, hstat.dltime, 0);
1727 if (hstat.len == hstat.contlen)
1731 logprintf (LOG_VERBOSE,
1732 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
1733 tms, tmrate, locf, hstat.len, hstat.contlen);
1734 logprintf (LOG_NONVERBOSE,
1735 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1736 tms, u->url, hstat.len, hstat.contlen, locf, count);
1739 downloaded_increase (hstat.len);
1741 /* Remember that we downloaded the file for later ".orig" code. */
1742 if (*dt & ADDED_HTML_EXTENSION)
1743 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1745 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1747 xfree(filename_plus_orig_suffix); /* must precede every return! */
1750 else if (hstat.res == 0) /* No read error */
1752 if (hstat.contlen == -1) /* We don't know how much we were supposed
1753 to get, so assume we succeeded. */
1757 logprintf (LOG_VERBOSE,
1758 _("%s (%s) - `%s' saved [%ld]\n\n"),
1759 tms, tmrate, locf, hstat.len);
1760 logprintf (LOG_NONVERBOSE,
1761 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
1762 tms, u->url, hstat.len, locf, count);
1765 downloaded_increase (hstat.len);
1767 /* Remember that we downloaded the file for later ".orig" code. */
1768 if (*dt & ADDED_HTML_EXTENSION)
1769 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1771 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1773 xfree (filename_plus_orig_suffix); /* must precede every return! */
1776 else if (hstat.len < hstat.contlen) /* meaning we lost the
1777 connection too soon */
1779 logprintf (LOG_VERBOSE,
1780 _("%s (%s) - Connection closed at byte %ld. "),
1781 tms, tmrate, hstat.len);
1782 printwhat (count, opt.ntry);
1785 else if (!opt.kill_longer) /* meaning we got more than expected */
1787 logprintf (LOG_VERBOSE,
1788 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
1789 tms, tmrate, locf, hstat.len, hstat.contlen);
1790 logprintf (LOG_NONVERBOSE,
1791 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
1792 tms, u->url, hstat.len, hstat.contlen, locf, count);
1794 downloaded_increase (hstat.len);
1796 /* Remember that we downloaded the file for later ".orig" code. */
1797 if (*dt & ADDED_HTML_EXTENSION)
1798 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
1800 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
1802 xfree (filename_plus_orig_suffix); /* must precede every return! */
1805 else /* the same, but not accepted */
1807 logprintf (LOG_VERBOSE,
1808 _("%s (%s) - Connection closed at byte %ld/%ld. "),
1809 tms, tmrate, hstat.len, hstat.contlen);
1810 printwhat (count, opt.ntry);
1814 else /* now hstat.res can only be -1 */
1816 if (hstat.contlen == -1)
1818 logprintf (LOG_VERBOSE,
1819 _("%s (%s) - Read error at byte %ld (%s)."),
1820 tms, tmrate, hstat.len, strerror (errno));
1821 printwhat (count, opt.ntry);
1824 else /* hstat.res == -1 and contlen is given */
1826 logprintf (LOG_VERBOSE,
1827 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
1828 tms, tmrate, hstat.len, hstat.contlen,
1830 printwhat (count, opt.ntry);
1837 while (!opt.ntry || (count < opt.ntry));
1838 xfree (filename_plus_orig_suffix); /* must precede every return! */
1842 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
1843 than local timezone.
1845 mktime is similar but assumes struct tm, also known as the
1846 "broken-down" form of time, is in local time zone. mktime_from_utc
1847 uses mktime to make the conversion understanding that an offset
1848 will be introduced by the local time assumption.
1850 mktime_from_utc then measures the introduced offset by applying
1851 gmtime to the initial result and applying mktime to the resulting
1852 "broken-down" form. The difference between the two mktime results
1853 is the measured offset which is then subtracted from the initial
1854 mktime result to yield a calendar time which is the value returned.
1856 tm_isdst in struct tm is set to 0 to force mktime to introduce a
1857 consistent offset (the non DST offset) since tm and tm+o might be
1858 on opposite sides of a DST change.
1860 Some implementations of mktime return -1 for the nonexistent
1861 localtime hour at the beginning of DST. In this event, use
1862 mktime(tm - 1hr) + 3600.
1866 gmtime(t+o) --> tm+o
1867 mktime(tm+o) --> t+2o
1868 t+o - (t+2o - t+o) = t
1870 Note that glibc contains a function of the same purpose named
1871 `timegm' (reverse of gmtime). But obviously, it is not universally
1872 available, and unfortunately it is not straightforwardly
1873 extractable for use here. Perhaps configure should detect timegm
1874 and use it where available.
1876 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
1877 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
1878 Further improved by Roger with assistance from Edward J. Sabol
1879 based on input by Jamie Zawinski. */
1882 mktime_from_utc (struct tm *t)
1893 return -1; /* can't deal with output from strptime */
1904 return -1; /* can't deal with output from gmtime */
1907 return (tl - (tb - tl));
1910 /* Check whether the result of strptime() indicates success.
1911 strptime() returns the pointer to how far it got to in the string.
1912 The processing has been successful if the string is at `GMT' or
1913 `+X', or at the end of the string.
1915 In extended regexp parlance, the function returns 1 if P matches
1916 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
1917 can return) is considered a failure and 0 is returned. */
1919 check_end (const char *p)
1923 while (ISSPACE (*p))
1926 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
1927 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
1933 /* Convert the textual specification of time in TIME_STRING to the
1934 number of seconds since the Epoch.
1936 TIME_STRING can be in any of the three formats RFC2068 allows the
1937 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
1938 Timezones are ignored, and should be GMT.
1940 Return the computed time_t representation, or -1 if the conversion
1943 This function uses strptime with various string formats for parsing
1944 TIME_STRING. This results in a parser that is not as lenient in
1945 interpreting TIME_STRING as I would like it to be. Being based on
1946 strptime, it always allows shortened months, one-digit days, etc.,
1947 but due to the multitude of formats in which time can be
1948 represented, an ideal HTTP time parser would be even more
1949 forgiving. It should completely ignore things like week days and
1950 concentrate only on the various forms of representing years,
1951 months, days, hours, minutes, and seconds. For example, it would
1952 be nice if it accepted ISO 8601 out of the box.
1954 I've investigated free and PD code for this purpose, but none was
1955 usable. getdate was big and unwieldy, and had potential copyright
1956 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
1957 distributed with phttpd, is excellent, but we cannot use it because
1958 it is not assigned to the FSF. So I stuck it with strptime. */
1961 http_atotm (char *time_string)
1963 /* NOTE: Solaris strptime man page claims that %n and %t match white
1964 space, but that's not universally available. Instead, we simply
1965 use ` ' to mean "skip all WS", which works under all strptime
1966 implementations I've tested. */
1968 static const char *time_formats[] = {
1969 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
1970 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
1971 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
1972 (google.com uses this for their cookies.) */
1973 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
1979 /* According to Roger Beeman, we need to initialize tm_isdst, since
1980 strptime won't do it. */
1983 /* Note that under foreign locales Solaris strptime() fails to
1984 recognize English dates, which renders this function useless. We
1985 solve this by being careful not to affect LC_TIME when
1986 initializing locale.
1988 Another solution would be to temporarily set locale to C, invoke
1989 strptime(), and restore it back. This is slow and dirty,
1990 however, and locale support other than LC_MESSAGES can mess other
1991 things, so I rather chose to stick with just setting LC_MESSAGES.
1993 GNU strptime does not have this problem because it recognizes
1994 both international and local dates. */
1996 for (i = 0; i < ARRAY_SIZE (time_formats); i++)
1997 if (check_end (strptime (time_string, time_formats[i], &t)))
1998 return mktime_from_utc (&t);
2000 /* All formats have failed. */
2004 /* Authorization support: We support two authorization schemes:
2006 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2008 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2009 consisting of answering to the server's challenge with the proper
2012 /* How many bytes it will take to store LEN bytes in base64. */
2013 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2015 /* Encode the string S of length LENGTH to base64 format and place it
2016 to STORE. STORE will be 0-terminated, and must point to a writable
2017 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2019 base64_encode (const char *s, char *store, int length)
2021 /* Conversion table. */
2022 static char tbl[64] = {
2023 'A','B','C','D','E','F','G','H',
2024 'I','J','K','L','M','N','O','P',
2025 'Q','R','S','T','U','V','W','X',
2026 'Y','Z','a','b','c','d','e','f',
2027 'g','h','i','j','k','l','m','n',
2028 'o','p','q','r','s','t','u','v',
2029 'w','x','y','z','0','1','2','3',
2030 '4','5','6','7','8','9','+','/'
2033 unsigned char *p = (unsigned char *)store;
2035 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2036 for (i = 0; i < length; i += 3)
2038 *p++ = tbl[s[0] >> 2];
2039 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2040 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2041 *p++ = tbl[s[2] & 0x3f];
2044 /* Pad the result if necessary... */
2045 if (i == length + 1)
2047 else if (i == length + 2)
2048 *(p - 1) = *(p - 2) = '=';
2049 /* ...and zero-terminate it. */
2053 /* Create the authentication header contents for the `Basic' scheme.
2054 This is done by encoding the string `USER:PASS' in base64 and
2055 prepending `HEADER: Basic ' to it. */
2057 basic_authentication_encode (const char *user, const char *passwd,
2060 char *t1, *t2, *res;
2061 int len1 = strlen (user) + 1 + strlen (passwd);
2062 int len2 = BASE64_LENGTH (len1);
2064 t1 = (char *)alloca (len1 + 1);
2065 sprintf (t1, "%s:%s", user, passwd);
2066 t2 = (char *)alloca (1 + len2);
2067 base64_encode (t1, t2, len1);
2068 res = (char *)xmalloc (len2 + 11 + strlen (header));
2069 sprintf (res, "%s: Basic %s\r\n", header, t2);
2075 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2076 of a field in such a header. If the field is the one specified by
2077 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2078 digest authorization code), extract its value in the (char*)
2079 variable pointed by RET. Returns negative on a malformed header,
2080 or number of bytes that have been parsed by this call. */
2082 extract_header_attr (const char *au, const char *attr_name, char **ret)
2084 const char *cp, *ep;
2088 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2090 cp += strlen (attr_name);
2093 cp += skip_lws (cp);
2098 cp += skip_lws (cp);
2103 for (ep = cp; *ep && *ep != '\"'; ep++)
2108 *ret = strdupdelim (cp, ep);
2115 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2116 an array of 16 bytes containing the hash keys, and BUF should be a
2117 buffer of 33 writable characters (32 for hex digits plus one for
2118 zero termination). */
2120 dump_hash (unsigned char *buf, const unsigned char *hash)
2124 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2126 *buf++ = XDIGIT_TO_xchar (*hash >> 4);
2127 *buf++ = XDIGIT_TO_xchar (*hash & 0xf);
2132 /* Take the line apart to find the challenge, and compose a digest
2133 authorization header. See RFC2069 section 2.1.2. */
2135 digest_authentication_encode (const char *au, const char *user,
2136 const char *passwd, const char *method,
2139 static char *realm, *opaque, *nonce;
2144 { "realm", &realm },
2145 { "opaque", &opaque },
2150 realm = opaque = nonce = NULL;
2152 au += 6; /* skip over `Digest' */
2157 au += skip_lws (au);
2158 for (i = 0; i < ARRAY_SIZE (options); i++)
2160 int skip = extract_header_attr (au, options[i].name,
2161 options[i].variable);
2165 FREE_MAYBE (opaque);
2175 if (i == ARRAY_SIZE (options))
2177 while (*au && *au != '=')
2181 au += skip_lws (au);
2185 while (*au && *au != '\"')
2192 while (*au && *au != ',')
2197 if (!realm || !nonce || !user || !passwd || !path || !method)
2200 FREE_MAYBE (opaque);
2205 /* Calculate the digest value. */
2208 unsigned char hash[MD5_HASHLEN];
2209 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2210 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2212 /* A1BUF = H(user ":" realm ":" password) */
2213 md5_init_ctx (&ctx);
2214 md5_process_bytes (user, strlen (user), &ctx);
2215 md5_process_bytes (":", 1, &ctx);
2216 md5_process_bytes (realm, strlen (realm), &ctx);
2217 md5_process_bytes (":", 1, &ctx);
2218 md5_process_bytes (passwd, strlen (passwd), &ctx);
2219 md5_finish_ctx (&ctx, hash);
2220 dump_hash (a1buf, hash);
2222 /* A2BUF = H(method ":" path) */
2223 md5_init_ctx (&ctx);
2224 md5_process_bytes (method, strlen (method), &ctx);
2225 md5_process_bytes (":", 1, &ctx);
2226 md5_process_bytes (path, strlen (path), &ctx);
2227 md5_finish_ctx (&ctx, hash);
2228 dump_hash (a2buf, hash);
2230 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2231 md5_init_ctx (&ctx);
2232 md5_process_bytes (a1buf, MD5_HASHLEN * 2, &ctx);
2233 md5_process_bytes (":", 1, &ctx);
2234 md5_process_bytes (nonce, strlen (nonce), &ctx);
2235 md5_process_bytes (":", 1, &ctx);
2236 md5_process_bytes (a2buf, MD5_HASHLEN * 2, &ctx);
2237 md5_finish_ctx (&ctx, hash);
2238 dump_hash (response_digest, hash);
2240 res = (char*) xmalloc (strlen (user)
2245 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2246 + (opaque ? strlen (opaque) : 0)
2248 sprintf (res, "Authorization: Digest \
2249 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2250 user, realm, nonce, path, response_digest);
2253 char *p = res + strlen (res);
2254 strcat (p, ", opaque=\"");
2258 strcat (res, "\r\n");
2262 #endif /* USE_DIGEST */
2265 #define BEGINS_WITH(line, string_constant) \
2266 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2267 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2268 || !line[sizeof (string_constant) - 1]))
2271 known_authentication_scheme_p (const char *au)
2273 return BEGINS_WITH (au, "Basic")
2274 || BEGINS_WITH (au, "Digest")
2275 || BEGINS_WITH (au, "NTLM");
2280 /* Create the HTTP authorization request header. When the
2281 `WWW-Authenticate' response header is seen, according to the
2282 authorization scheme specified in that header (`Basic' and `Digest'
2283 are supported by the current implementation), produce an
2284 appropriate HTTP authorization request header. */
2286 create_authorization_line (const char *au, const char *user,
2287 const char *passwd, const char *method,
2290 char *wwwauth = NULL;
2292 if (!strncasecmp (au, "Basic", 5))
2293 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2294 if (!strncasecmp (au, "NTLM", 4))
2295 wwwauth = basic_authentication_encode (user, passwd, "Authorization");
2297 else if (!strncasecmp (au, "Digest", 6))
2298 wwwauth = digest_authentication_encode (au, user, passwd, method, path);
2299 #endif /* USE_DIGEST */