2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget 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 GNU Wget 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 Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
34 #include <sys/types.h>
45 #if TIME_WITH_SYS_TIME
46 # include <sys/time.h>
50 # include <sys/time.h>
67 # include "gen_sslfunc.h"
70 # include "http-ntlm.h"
78 extern char *version_string;
79 extern LARGE_INT total_downloaded_bytes;
81 extern FILE *output_stream;
82 extern int output_stream_regular;
85 # define MIN(x, y) ((x) > (y) ? (y) : (x))
89 static int cookies_loaded_p;
90 static struct cookie_jar *wget_cookie_jar;
92 #define TEXTHTML_S "text/html"
93 #define TEXTXHTML_S "application/xhtml+xml"
95 /* Some status code validation macros: */
96 #define H_20X(x) (((x) >= 200) && ((x) < 300))
97 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
98 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
99 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
100 || (x) == HTTP_STATUS_SEE_OTHER \
101 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
103 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
104 /* Successful 2xx. */
105 #define HTTP_STATUS_OK 200
106 #define HTTP_STATUS_CREATED 201
107 #define HTTP_STATUS_ACCEPTED 202
108 #define HTTP_STATUS_NO_CONTENT 204
109 #define HTTP_STATUS_PARTIAL_CONTENTS 206
111 /* Redirection 3xx. */
112 #define HTTP_STATUS_MULTIPLE_CHOICES 300
113 #define HTTP_STATUS_MOVED_PERMANENTLY 301
114 #define HTTP_STATUS_MOVED_TEMPORARILY 302
115 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
116 #define HTTP_STATUS_NOT_MODIFIED 304
117 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
119 /* Client error 4xx. */
120 #define HTTP_STATUS_BAD_REQUEST 400
121 #define HTTP_STATUS_UNAUTHORIZED 401
122 #define HTTP_STATUS_FORBIDDEN 403
123 #define HTTP_STATUS_NOT_FOUND 404
124 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
126 /* Server errors 5xx. */
127 #define HTTP_STATUS_INTERNAL 500
128 #define HTTP_STATUS_NOT_IMPLEMENTED 501
129 #define HTTP_STATUS_BAD_GATEWAY 502
130 #define HTTP_STATUS_UNAVAILABLE 503
133 rel_none, rel_name, rel_value, rel_both
140 struct request_header {
142 enum rp release_policy;
144 int hcount, hcapacity;
147 /* Create a new, empty request. At least request_set_method must be
148 called before the request can be used. */
150 static struct request *
153 struct request *req = xnew0 (struct request);
155 req->headers = xnew_array (struct request_header, req->hcapacity);
159 /* Set the request's method and its arguments. METH should be a
160 literal string (or it should outlive the request) because it will
161 not be freed. ARG will be freed by request_free. */
164 request_set_method (struct request *req, const char *meth, char *arg)
170 /* Return the method string passed with the last call to
171 request_set_method. */
174 request_method (const struct request *req)
179 /* Free one header according to the release policy specified with
180 request_set_header. */
183 release_header (struct request_header *hdr)
185 switch (hdr->release_policy)
202 /* Set the request named NAME to VALUE. Specifically, this means that
203 a "NAME: VALUE\r\n" header line will be used in the request. If a
204 header with the same name previously existed in the request, its
205 value will be replaced by this one.
207 RELEASE_POLICY determines whether NAME and VALUE should be released
208 (freed) with request_free. Allowed values are:
210 - rel_none - don't free NAME or VALUE
211 - rel_name - free NAME when done
212 - rel_value - free VALUE when done
213 - rel_both - free both NAME and VALUE when done
215 Setting release policy is useful when arguments come from different
216 sources. For example:
218 // Don't free literal strings!
219 request_set_header (req, "Pragma", "no-cache", rel_none);
221 // Don't free a global variable, we'll need it later.
222 request_set_header (req, "Referer", opt.referer, rel_none);
224 // Value freshly allocated, free it when done.
225 request_set_header (req, "Range",
226 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
231 request_set_header (struct request *req, char *name, char *value,
232 enum rp release_policy)
234 struct request_header *hdr;
238 /* A NULL value is a no-op; if freeing the name is requested,
239 free it now to avoid leaks. */
240 if (release_policy == rel_name || release_policy == rel_both)
244 for (i = 0; i < req->hcount; i++)
246 hdr = &req->headers[i];
247 if (0 == strcasecmp (name, hdr->name))
249 /* Replace existing header. */
250 release_header (hdr);
253 hdr->release_policy = release_policy;
258 /* Install new header. */
260 if (req->hcount >= req->hcapacity)
262 req->hcapacity <<= 1;
263 req->headers = xrealloc (req->headers,
264 req->hcapacity * sizeof (struct request_header));
266 hdr = &req->headers[req->hcount++];
269 hdr->release_policy = release_policy;
272 /* Like request_set_header, but sets the whole header line, as
273 provided by the user using the `--header' option. For example,
274 request_set_user_header (req, "Foo: bar") works just like
275 request_set_header (req, "Foo", "bar"). */
278 request_set_user_header (struct request *req, const char *header)
281 const char *p = strchr (header, ':');
284 BOUNDED_TO_ALLOCA (header, p, name);
288 request_set_header (req, xstrdup (name), (char *) p, rel_name);
291 #define APPEND(p, str) do { \
292 int A_len = strlen (str); \
293 memcpy (p, str, A_len); \
297 /* Construct the request and write it to FD using fd_write. */
300 request_send (const struct request *req, int fd)
302 char *request_string, *p;
303 int i, size, write_error;
305 /* Count the request size. */
308 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
309 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
311 for (i = 0; i < req->hcount; i++)
313 struct request_header *hdr = &req->headers[i];
314 /* NAME ": " VALUE "\r\n" */
315 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
321 p = request_string = alloca_array (char, size);
323 /* Generate the request. */
325 APPEND (p, req->method); *p++ = ' ';
326 APPEND (p, req->arg); *p++ = ' ';
327 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
329 for (i = 0; i < req->hcount; i++)
331 struct request_header *hdr = &req->headers[i];
332 APPEND (p, hdr->name);
333 *p++ = ':', *p++ = ' ';
334 APPEND (p, hdr->value);
335 *p++ = '\r', *p++ = '\n';
338 *p++ = '\r', *p++ = '\n', *p++ = '\0';
339 assert (p - request_string == size);
343 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
345 /* Send the request to the server. */
347 write_error = fd_write (fd, request_string, size - 1, -1);
349 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
354 /* Release the resources used by REQ. */
357 request_free (struct request *req)
360 xfree_null (req->arg);
361 for (i = 0; i < req->hcount; i++)
362 release_header (&req->headers[i]);
363 xfree_null (req->headers);
367 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
368 PROMISED_SIZE bytes are sent over the wire -- if the file is
369 longer, read only that much; if the file is shorter, report an error. */
372 post_file (int sock, const char *file_name, wgint promised_size)
374 static char chunk[8192];
379 DEBUGP (("[writing POST file %s ... ", file_name));
381 fp = fopen (file_name, "rb");
384 while (!feof (fp) && written < promised_size)
387 int length = fread (chunk, 1, sizeof (chunk), fp);
390 towrite = MIN (promised_size - written, length);
391 write_error = fd_write (sock, chunk, towrite, -1);
401 /* If we've written less than was promised, report a (probably
402 nonsensical) error rather than break the promise. */
403 if (written < promised_size)
409 assert (written == promised_size);
410 DEBUGP (("done]\n"));
415 response_head_terminator (const char *hunk, int oldlen, int peeklen)
417 const char *start, *end;
419 /* If at first peek, verify whether HUNK starts with "HTTP". If
420 not, this is a HTTP/0.9 request and we must bail out without
422 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
428 start = hunk + oldlen - 4;
429 end = hunk + oldlen + peeklen;
431 for (; start < end - 1; start++)
438 if (start[1] == '\n')
444 /* The maximum size of a single HTTP response we care to read. This
445 is not meant to impose an arbitrary limit, but to protect the user
446 from Wget slurping up available memory upon encountering malicious
447 or buggy server output. Define it to 0 to remove the limit. */
449 #define HTTP_RESPONSE_MAX_SIZE 65536
451 /* Read the HTTP request head from FD and return it. The error
452 conditions are the same as with fd_read_hunk.
454 To support HTTP/0.9 responses, this function tries to make sure
455 that the data begins with "HTTP". If this is not the case, no data
456 is read and an empty request is returned, so that the remaining
457 data can be treated as body. */
460 read_http_response_head (int fd)
462 return fd_read_hunk (fd, response_head_terminator, 512,
463 HTTP_RESPONSE_MAX_SIZE);
467 /* The response data. */
470 /* The array of pointers that indicate where each header starts.
471 For example, given this HTTP response:
478 The headers are located like this:
480 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
482 headers[0] headers[1] headers[2] headers[3]
484 I.e. headers[0] points to the beginning of the request,
485 headers[1] points to the end of the first header and the
486 beginning of the second one, etc. */
488 const char **headers;
491 /* Create a new response object from the text of the HTTP response,
492 available in HEAD. That text is automatically split into
493 constituent header lines for fast retrieval using
496 static struct response *
497 resp_new (const char *head)
502 struct response *resp = xnew0 (struct response);
507 /* Empty head means that we're dealing with a headerless
508 (HTTP/0.9) response. In that case, don't set HEADERS at
513 /* Split HEAD into header lines, so that resp_header_* functions
514 don't need to do this over and over again. */
520 DO_REALLOC (resp->headers, size, count + 1, const char *);
521 resp->headers[count++] = hdr;
523 /* Break upon encountering an empty line. */
524 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
527 /* Find the end of HDR, including continuations. */
530 const char *end = strchr (hdr, '\n');
536 while (*hdr == ' ' || *hdr == '\t');
538 DO_REALLOC (resp->headers, size, count + 1, const char *);
539 resp->headers[count] = NULL;
544 /* Locate the header named NAME in the request data, starting with
545 position START. This allows the code to loop through the request
546 data, filtering for all requests of a given name. Returns the
547 found position, or -1 for failure. The code that uses this
548 function typically looks like this:
550 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
551 ... do something with header ...
553 If you only care about one header, use resp_header_get instead of
557 resp_header_locate (const struct response *resp, const char *name, int start,
558 const char **begptr, const char **endptr)
561 const char **headers = resp->headers;
564 if (!headers || !headers[1])
567 name_len = strlen (name);
573 for (; headers[i + 1]; i++)
575 const char *b = headers[i];
576 const char *e = headers[i + 1];
578 && b[name_len] == ':'
579 && 0 == strncasecmp (b, name, name_len))
582 while (b < e && ISSPACE (*b))
584 while (b < e && ISSPACE (e[-1]))
594 /* Find and retrieve the header named NAME in the request data. If
595 found, set *BEGPTR to its starting, and *ENDPTR to its ending
596 position, and return 1. Otherwise return 0.
598 This function is used as a building block for resp_header_copy
599 and resp_header_strdup. */
602 resp_header_get (const struct response *resp, const char *name,
603 const char **begptr, const char **endptr)
605 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
609 /* Copy the response header named NAME to buffer BUF, no longer than
610 BUFSIZE (BUFSIZE includes the terminating 0). If the header
611 exists, 1 is returned, otherwise 0. If there should be no limit on
612 the size of the header, use resp_header_strdup instead.
614 If BUFSIZE is 0, no data is copied, but the boolean indication of
615 whether the header is present is still returned. */
618 resp_header_copy (const struct response *resp, const char *name,
619 char *buf, int bufsize)
622 if (!resp_header_get (resp, name, &b, &e))
626 int len = MIN (e - b, bufsize - 1);
627 memcpy (buf, b, len);
633 /* Return the value of header named NAME in RESP, allocated with
634 malloc. If such a header does not exist in RESP, return NULL. */
637 resp_header_strdup (const struct response *resp, const char *name)
640 if (!resp_header_get (resp, name, &b, &e))
642 return strdupdelim (b, e);
645 /* Parse the HTTP status line, which is of format:
647 HTTP-Version SP Status-Code SP Reason-Phrase
649 The function returns the status-code, or -1 if the status line
650 appears malformed. The pointer to "reason-phrase" message is
651 returned in *MESSAGE. */
654 resp_status (const struct response *resp, char **message)
661 /* For a HTTP/0.9 response, assume status 200. */
663 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
667 p = resp->headers[0];
668 end = resp->headers[1];
674 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
678 /* Match the HTTP version. This is optional because Gnutella
679 servers have been reported to not specify HTTP version. */
680 if (p < end && *p == '/')
683 while (p < end && ISDIGIT (*p))
685 if (p < end && *p == '.')
687 while (p < end && ISDIGIT (*p))
691 while (p < end && ISSPACE (*p))
693 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
696 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
701 while (p < end && ISSPACE (*p))
703 while (p < end && ISSPACE (end[-1]))
705 *message = strdupdelim (p, end);
711 /* Release the resources used by RESP. */
714 resp_free (struct response *resp)
716 xfree_null (resp->headers);
720 /* Print the server response, line by line, omitting the trailing CRLF
721 from individual header lines, and prefixed with PREFIX. */
724 print_server_response (const struct response *resp, const char *prefix)
729 for (i = 0; resp->headers[i + 1]; i++)
731 const char *b = resp->headers[i];
732 const char *e = resp->headers[i + 1];
734 if (b < e && e[-1] == '\n')
736 if (b < e && e[-1] == '\r')
738 /* This is safe even on printfs with broken handling of "%.<n>s"
739 because resp->headers ends with \0. */
740 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
744 /* Parse the `Content-Range' header and extract the information it
745 contains. Returns 1 if successful, -1 otherwise. */
747 parse_content_range (const char *hdr, wgint *first_byte_ptr,
748 wgint *last_byte_ptr, wgint *entity_length_ptr)
752 /* Ancient versions of Netscape proxy server, presumably predating
753 rfc2068, sent out `Content-Range' without the "bytes"
755 if (!strncasecmp (hdr, "bytes", 5))
758 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
762 while (ISSPACE (*hdr))
769 for (num = 0; ISDIGIT (*hdr); hdr++)
770 num = 10 * num + (*hdr - '0');
771 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
773 *first_byte_ptr = num;
775 for (num = 0; ISDIGIT (*hdr); hdr++)
776 num = 10 * num + (*hdr - '0');
777 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
779 *last_byte_ptr = num;
781 for (num = 0; ISDIGIT (*hdr); hdr++)
782 num = 10 * num + (*hdr - '0');
783 *entity_length_ptr = num;
787 /* Read the body of the request, but don't store it anywhere and don't
788 display a progress gauge. This is useful for reading the bodies of
789 administrative responses to which we will soon issue another
790 request. The response is not useful to the user, but reading it
791 allows us to continue using the same connection to the server.
793 If reading fails, 0 is returned, non-zero otherwise. In debug
794 mode, the body is displayed for debugging purposes. */
797 skip_short_body (int fd, wgint contlen)
800 SKIP_SIZE = 512, /* size of the download buffer */
801 SKIP_THRESHOLD = 4096 /* the largest size we read */
803 char dlbuf[SKIP_SIZE + 1];
804 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
806 /* We shouldn't get here with unknown contlen. (This will change
807 with HTTP/1.1, which supports "chunked" transfer.) */
808 assert (contlen != -1);
810 /* If the body is too large, it makes more sense to simply close the
811 connection than to try to read the body. */
812 if (contlen > SKIP_THRESHOLD)
815 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
819 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
822 /* Don't normally report the error since this is an
823 optimization that should be invisible to the user. */
824 DEBUGP (("] aborting (%s).\n",
825 ret < 0 ? strerror (errno) : "EOF received"));
829 /* Safe even if %.*s bogusly expects terminating \0 because
830 we've zero-terminated dlbuf above. */
831 DEBUGP (("%.*s", ret, dlbuf));
834 DEBUGP (("] done.\n"));
838 /* Persistent connections. Currently, we cache the most recently used
839 connection as persistent, provided that the HTTP server agrees to
840 make it such. The persistence data is stored in the variables
841 below. Ideally, it should be possible to cache an arbitrary fixed
842 number of these connections. */
844 /* Whether a persistent connection is active. */
845 static int pconn_active;
848 /* The socket of the connection. */
851 /* Host and port of the currently active persistent connection. */
855 /* Whether a ssl handshake has occoured on this connection. */
859 /* NTLM data of the current connection. */
860 struct ntlmdata ntlm;
864 /* Mark the persistent connection as invalid and free the resources it
865 uses. This is used by the CLOSE_* macros after they forcefully
866 close a registered persistent connection. */
869 invalidate_persistent (void)
871 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
873 fd_close (pconn.socket);
878 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
879 persistent. This will enable someone to use the same connection
880 later. In the context of HTTP, this must be called only AFTER the
881 response has been received and the server has promised that the
882 connection will remain alive.
884 If a previous connection was persistent, it is closed. */
887 register_persistent (const char *host, int port, int fd, int ssl)
891 if (pconn.socket == fd)
893 /* The connection FD is already registered. */
898 /* The old persistent connection is still active; close it
899 first. This situation arises whenever a persistent
900 connection exists, but we then connect to a different
901 host, and try to register a persistent connection to that
903 invalidate_persistent ();
909 pconn.host = xstrdup (host);
913 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
916 /* Return non-zero if a persistent connection is available for
917 connecting to HOST:PORT. */
920 persistent_available_p (const char *host, int port, int ssl,
921 int *host_lookup_failed)
923 /* First, check whether a persistent connection is active at all. */
927 /* If we want SSL and the last connection wasn't or vice versa,
928 don't use it. Checking for host and port is not enough because
929 HTTP and HTTPS can apparently coexist on the same port. */
930 if (ssl != pconn.ssl)
933 /* If we're not connecting to the same port, we're not interested. */
934 if (port != pconn.port)
937 /* If the host is the same, we're in business. If not, there is
938 still hope -- read below. */
939 if (0 != strcasecmp (host, pconn.host))
941 /* If pconn.socket is already talking to HOST, we needn't
942 reconnect. This happens often when both sites are virtual
943 hosts distinguished only by name and served by the same
944 network interface, and hence the same web server (possibly
945 set up by the ISP and serving many different web sites).
946 This admittedly non-standard optimization does not contradict
947 HTTP and works well with popular server software. */
951 struct address_list *al;
954 /* Don't try to talk to two different SSL sites over the same
955 secure connection! (Besides, it's not clear if name-based
956 virtual hosting is even possible with SSL.) */
959 /* If pconn.socket's peer is one of the IP addresses HOST
960 resolves to, pconn.socket is for all intents and purposes
961 already talking to HOST. */
963 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
965 /* Can't get the peer's address -- something must be very
966 wrong with the connection. */
967 invalidate_persistent ();
970 al = lookup_host (host, 0);
973 *host_lookup_failed = 1;
977 found = address_list_contains (al, &ip);
978 address_list_release (al);
983 /* The persistent connection's peer address was found among the
984 addresses HOST resolved to; therefore, pconn.sock is in fact
985 already talking to HOST -- no need to reconnect. */
988 /* Finally, check whether the connection is still open. This is
989 important because most server implement a liberal (short) timeout
990 on persistent connections. Wget can of course always reconnect
991 if the connection doesn't work out, but it's nicer to know in
992 advance. This test is a logical followup of the first test, but
993 is "expensive" and therefore placed at the end of the list. */
995 if (!test_socket_open (pconn.socket))
997 /* Oops, the socket is no longer open. Now that we know that,
998 let's invalidate the persistent connection before returning
1000 invalidate_persistent ();
1007 /* The idea behind these two CLOSE macros is to distinguish between
1008 two cases: one when the job we've been doing is finished, and we
1009 want to close the connection and leave, and two when something is
1010 seriously wrong and we're closing the connection as part of
1013 In case of keep_alive, CLOSE_FINISH should leave the connection
1014 open, while CLOSE_INVALIDATE should still close it.
1016 Note that the semantics of the flag `keep_alive' is "this
1017 connection *will* be reused (the server has promised not to close
1018 the connection once we're done)", while the semantics of
1019 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1020 active, registered connection". */
1022 #define CLOSE_FINISH(fd) do { \
1025 if (pconn_active && (fd) == pconn.socket) \
1026 invalidate_persistent (); \
1035 #define CLOSE_INVALIDATE(fd) do { \
1036 if (pconn_active && (fd) == pconn.socket) \
1037 invalidate_persistent (); \
1045 wgint len; /* received length */
1046 wgint contlen; /* expected length */
1047 wgint restval; /* the restart value */
1048 int res; /* the result of last read */
1049 char *newloc; /* new location (redirection) */
1050 char *remote_time; /* remote time-stamp string */
1051 char *error; /* textual HTTP error */
1052 int statcode; /* status code */
1053 wgint rd_size; /* amount of data read from socket */
1054 double dltime; /* time it took to download the data */
1055 const char *referer; /* value of the referer header. */
1056 char **local_file; /* local file. */
1060 free_hstat (struct http_stat *hs)
1062 xfree_null (hs->newloc);
1063 xfree_null (hs->remote_time);
1064 xfree_null (hs->error);
1066 /* Guard against being called twice. */
1068 hs->remote_time = NULL;
1072 static char *create_authorization_line PARAMS ((const char *, const char *,
1073 const char *, const char *,
1074 const char *, int *));
1075 static char *basic_authentication_encode PARAMS ((const char *, const char *));
1076 static int known_authentication_scheme_p PARAMS ((const char *, const char *));
1078 time_t http_atotm PARAMS ((const char *));
1080 #define BEGINS_WITH(line, string_constant) \
1081 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1082 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1083 || !line[sizeof (string_constant) - 1]))
1085 /* Retrieve a document through HTTP protocol. It recognizes status
1086 code, and correctly handles redirections. It closes the network
1087 socket. If it receives an error from the functions below it, it
1088 will print it if there is enough information to do so (almost
1089 always), returning the error to the caller (i.e. http_loop).
1091 Various HTTP parameters are stored to hs.
1093 If PROXY is non-NULL, the connection will be made to the proxy
1094 server, and u->url will be requested. */
1096 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1098 struct request *req;
1101 char *user, *passwd;
1105 wgint contlen, contrange;
1112 /* Set to 1 when the authorization has failed permanently and should
1113 not be tried again. */
1114 int auth_finished = 0;
1116 /* Whether our connection to the remote host is through SSL. */
1120 struct response *resp;
1124 /* Whether this connection will be kept alive after the HTTP request
1128 /* Whether keep-alive should be inhibited.
1130 RFC 2068 requests that 1.0 clients not send keep-alive requests
1131 to proxies. This is because many 1.0 proxies do not interpret
1132 the Connection header and transfer it to the remote server,
1133 causing it to not close the connection and leave both the proxy
1134 and the client hanging. */
1135 int inhibit_keep_alive =
1136 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1138 /* Headers sent when using POST. */
1139 wgint post_data_size = 0;
1141 int host_lookup_failed = 0;
1144 if (u->scheme == SCHEME_HTTPS)
1146 /* Initialize the SSL context. After this has once been done,
1147 it becomes a no-op. */
1148 switch (ssl_init ())
1150 case SSLERRCTXCREATE:
1152 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
1153 return SSLERRCTXCREATE;
1154 case SSLERRCERTFILE:
1155 /* try without certfile */
1156 logprintf (LOG_NOTQUIET,
1157 _("Failed to load certificates from %s\n"),
1159 logprintf (LOG_NOTQUIET,
1160 _("Trying without the specified certificate\n"));
1163 logprintf (LOG_NOTQUIET,
1164 _("Failed to get certificate key from %s\n"),
1166 logprintf (LOG_NOTQUIET,
1167 _("Trying without the specified certificate\n"));
1173 #endif /* HAVE_SSL */
1175 if (!(*dt & HEAD_ONLY))
1176 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1177 know the local filename so we can save to it. */
1178 assert (*hs->local_file != NULL);
1180 /* Initialize certain elements of struct http_stat. */
1185 hs->remote_time = NULL;
1190 /* Prepare the request to send. */
1192 req = request_new ();
1194 const char *meth = "GET";
1195 if (*dt & HEAD_ONLY)
1197 else if (opt.post_file_name || opt.post_data)
1199 /* Use the full path, i.e. one that includes the leading slash and
1200 the query string. E.g. if u->path is "foo/bar" and u->query is
1201 "param=value", full_path will be "/foo/bar?param=value". */
1202 request_set_method (req, meth,
1203 proxy ? xstrdup (u->url) : url_full_path (u));
1206 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1207 if (*dt & SEND_NOCACHE)
1208 request_set_header (req, "Pragma", "no-cache", rel_none);
1210 request_set_header (req, "Range",
1211 aprintf ("bytes=%s-",
1212 number_to_static_string (hs->restval)),
1215 request_set_header (req, "User-Agent", opt.useragent, rel_none);
1217 request_set_header (req, "User-Agent",
1218 aprintf ("Wget/%s", version_string), rel_value);
1219 request_set_header (req, "Accept", "*/*", rel_none);
1221 /* Find the username and password for authentication. */
1224 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1225 user = user ? user : opt.http_user;
1226 passwd = passwd ? passwd : opt.http_passwd;
1230 /* We have the username and the password, but haven't tried
1231 any authorization yet. Let's see if the "Basic" method
1232 works. If not, we'll come back here and construct a
1233 proper authorization method with the right challenges.
1235 If we didn't employ this kind of logic, every URL that
1236 requires authorization would have to be processed twice,
1237 which is very suboptimal and generates a bunch of false
1238 "unauthorized" errors in the server log.
1240 #### But this logic also has a serious problem when used
1241 with stronger authentications: we *first* transmit the
1242 username and the password in clear text, and *then* attempt a
1243 stronger authentication scheme. That cannot be right! We
1244 are only fortunate that almost everyone still uses the
1245 `Basic' scheme anyway.
1247 There should be an option to prevent this from happening, for
1248 those who use strong authentication schemes and value their
1250 request_set_header (req, "Authorization",
1251 basic_authentication_encode (user, passwd),
1258 char *proxy_user, *proxy_passwd;
1259 /* For normal username and password, URL components override
1260 command-line/wgetrc parameters. With proxy
1261 authentication, it's the reverse, because proxy URLs are
1262 normally the "permanent" ones, so command-line args
1263 should take precedence. */
1264 if (opt.proxy_user && opt.proxy_passwd)
1266 proxy_user = opt.proxy_user;
1267 proxy_passwd = opt.proxy_passwd;
1271 proxy_user = proxy->user;
1272 proxy_passwd = proxy->passwd;
1274 /* #### This does not appear right. Can't the proxy request,
1275 say, `Digest' authentication? */
1276 if (proxy_user && proxy_passwd)
1277 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1279 /* If we're using a proxy, we will be connecting to the proxy
1283 /* Proxy authorization over SSL is handled below. */
1285 if (u->scheme != SCHEME_HTTPS)
1287 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1291 /* Whether we need to print the host header with braces around
1292 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1293 usual "Host: symbolic-name:1234". */
1294 int squares = strchr (u->host, ':') != NULL;
1295 if (u->port == scheme_default_port (u->scheme))
1296 request_set_header (req, "Host",
1297 aprintf (squares ? "[%s]" : "%s", u->host),
1300 request_set_header (req, "Host",
1301 aprintf (squares ? "[%s]:%d" : "%s:%d",
1306 if (!inhibit_keep_alive)
1307 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1310 request_set_header (req, "Cookie",
1311 cookie_header (wget_cookie_jar,
1312 u->host, u->port, u->path,
1314 u->scheme == SCHEME_HTTPS
1321 if (opt.post_data || opt.post_file_name)
1323 request_set_header (req, "Content-Type",
1324 "application/x-www-form-urlencoded", rel_none);
1326 post_data_size = strlen (opt.post_data);
1329 post_data_size = file_size (opt.post_file_name);
1330 if (post_data_size == -1)
1332 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1333 opt.post_file_name);
1337 request_set_header (req, "Content-Length",
1338 xstrdup (number_to_static_string (post_data_size)),
1342 /* Add the user headers. */
1343 if (opt.user_headers)
1346 for (i = 0; opt.user_headers[i]; i++)
1347 request_set_user_header (req, opt.user_headers[i]);
1351 /* We need to come back here when the initial attempt to retrieve
1352 without authorization header fails. (Expected to happen at least
1353 for the Digest authorization scheme.) */
1357 /* Establish the connection. */
1359 if (!inhibit_keep_alive)
1361 /* Look for a persistent connection to target host, unless a
1362 proxy is used. The exception is when SSL is in use, in which
1363 case the proxy is nothing but a passthrough to the target
1364 host, registered as a connection to the latter. */
1365 struct url *relevant = conn;
1367 if (u->scheme == SCHEME_HTTPS)
1371 if (persistent_available_p (relevant->host, relevant->port,
1373 relevant->scheme == SCHEME_HTTPS,
1377 &host_lookup_failed))
1379 sock = pconn.socket;
1380 using_ssl = pconn.ssl;
1381 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1382 escnonprint (pconn.host), pconn.port);
1383 DEBUGP (("Reusing fd %d.\n", sock));
1389 /* In its current implementation, persistent_available_p will
1390 look up conn->host in some cases. If that lookup failed, we
1391 don't need to bother with connect_to_host. */
1392 if (host_lookup_failed)
1398 sock = connect_to_host (conn->host, conn->port);
1407 return (retryable_socket_connect_error (errno)
1408 ? CONERROR : CONIMPOSSIBLE);
1412 if (proxy && u->scheme == SCHEME_HTTPS)
1414 /* When requesting SSL URLs through proxies, use the
1415 CONNECT method to request passthrough. */
1416 struct request *connreq = request_new ();
1417 request_set_method (connreq, "CONNECT",
1418 aprintf ("%s:%d", u->host, u->port));
1421 request_set_header (connreq, "Proxy-Authorization",
1422 proxyauth, rel_value);
1423 /* Now that PROXYAUTH is part of the CONNECT request,
1424 zero it out so we don't send proxy authorization with
1425 the regular request below. */
1429 write_error = request_send (connreq, sock);
1430 request_free (connreq);
1431 if (write_error < 0)
1433 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1435 CLOSE_INVALIDATE (sock);
1439 head = read_http_response_head (sock);
1442 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1444 CLOSE_INVALIDATE (sock);
1453 DEBUGP (("proxy responded with: [%s]\n", head));
1455 resp = resp_new (head);
1456 statcode = resp_status (resp, &message);
1459 if (statcode != 200)
1462 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1463 message ? escnonprint (message) : "?");
1464 xfree_null (message);
1467 xfree_null (message);
1469 /* SOCK is now *really* connected to u->host, so update CONN
1470 to reflect this. That way register_persistent will
1471 register SOCK as being connected to u->host:u->port. */
1475 if (conn->scheme == SCHEME_HTTPS)
1477 if (!ssl_connect (sock))
1484 #endif /* HAVE_SSL */
1487 /* Send the request to server. */
1488 write_error = request_send (req, sock);
1490 if (write_error >= 0)
1494 DEBUGP (("[POST data: %s]\n", opt.post_data));
1495 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1497 else if (opt.post_file_name && post_data_size != 0)
1498 write_error = post_file (sock, opt.post_file_name, post_data_size);
1501 if (write_error < 0)
1503 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1505 CLOSE_INVALIDATE (sock);
1509 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1510 proxy ? "Proxy" : "HTTP");
1515 head = read_http_response_head (sock);
1520 logputs (LOG_NOTQUIET, _("No data received.\n"));
1521 CLOSE_INVALIDATE (sock);
1527 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1529 CLOSE_INVALIDATE (sock);
1534 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1536 resp = resp_new (head);
1538 /* Check for status line. */
1540 statcode = resp_status (resp, &message);
1541 if (!opt.server_response)
1542 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1543 message ? escnonprint (message) : "");
1546 logprintf (LOG_VERBOSE, "\n");
1547 print_server_response (resp, " ");
1550 if (!opt.ignore_length
1551 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1555 parsed = str_to_wgint (hdrval, NULL, 10);
1556 if (parsed == WGINT_MAX && errno == ERANGE)
1558 #### If Content-Length is out of range, it most likely
1559 means that the file is larger than 2G and that we're
1560 compiled without LFS. In that case we should probably
1561 refuse to even attempt to download the file. */
1567 /* Check for keep-alive related responses. */
1568 if (!inhibit_keep_alive && contlen != -1)
1570 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1572 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1574 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1579 /* The server has promised that it will not close the connection
1580 when we're done. This means that we can register it. */
1581 register_persistent (conn->host, conn->port, sock, using_ssl);
1583 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1585 /* Authorization is required. */
1586 if (skip_short_body (sock, contlen))
1587 CLOSE_FINISH (sock);
1589 CLOSE_INVALIDATE (sock);
1590 if (auth_finished || !(user && passwd))
1592 /* If we have tried it already, then there is not point
1594 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1598 /* IIS sometimes sends two instances of WWW-Authenticate
1599 header, one with the keyword "negotiate", and other with
1600 useful data. Loop over all occurrences of this header
1601 and use the one we recognize. */
1603 const char *wabeg, *waend;
1604 char *www_authenticate = NULL;
1606 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1607 &wabeg, &waend)) != -1;
1609 if (known_authentication_scheme_p (wabeg, waend))
1611 www_authenticate = strdupdelim (wabeg, waend);
1614 /* If the authentication header is missing or recognized, or
1615 if the authentication scheme is "Basic" (which we send by
1616 default), there's no sense in retrying. */
1617 if (!www_authenticate
1618 || BEGINS_WITH (www_authenticate, "Basic"))
1620 xfree_null (www_authenticate);
1621 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1626 pth = url_full_path (u);
1627 request_set_header (req, "Authorization",
1628 create_authorization_line (www_authenticate,
1630 request_method (req),
1635 xfree (www_authenticate);
1636 goto retry_with_auth;
1644 hs->statcode = statcode;
1646 hs->error = xstrdup (_("Malformed status line"));
1648 hs->error = xstrdup (_("(no description)"));
1650 hs->error = xstrdup (message);
1653 type = resp_header_strdup (resp, "Content-Type");
1656 char *tmp = strchr (type, ';');
1659 while (tmp > type && ISSPACE (tmp[-1]))
1664 hs->newloc = resp_header_strdup (resp, "Location");
1665 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1667 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1670 const char *scbeg, *scend;
1671 /* The jar should have been created by now. */
1672 assert (wget_cookie_jar != NULL);
1674 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1675 &scbeg, &scend)) != -1;
1678 char *set_cookie = strdupdelim (scbeg, scend);
1679 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1685 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1687 wgint first_byte_pos, last_byte_pos, entity_length;
1688 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1690 contrange = first_byte_pos;
1695 /* 20x responses are counted among successful by default. */
1696 if (H_20X (statcode))
1699 /* Return if redirected. */
1700 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1702 /* RFC2068 says that in case of the 300 (multiple choices)
1703 response, the server can output a preferred URL through
1704 `Location' header; otherwise, the request should be treated
1705 like GET. So, if the location is set, it will be a
1706 redirection; otherwise, just proceed normally. */
1707 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1711 logprintf (LOG_VERBOSE,
1712 _("Location: %s%s\n"),
1713 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1714 hs->newloc ? _(" [following]") : "");
1717 if (skip_short_body (sock, contlen))
1718 CLOSE_FINISH (sock);
1720 CLOSE_INVALIDATE (sock);
1727 /* If content-type is not given, assume text/html. This is because
1728 of the multitude of broken CGI's that "forget" to generate the
1731 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1732 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1737 if (opt.html_extension && (*dt & TEXTHTML))
1738 /* -E / --html-extension / html_extension = on was specified, and this is a
1739 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1740 already the file's suffix, tack on ".html". */
1742 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1744 if (last_period_in_local_filename == NULL
1745 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1746 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1748 size_t local_filename_len = strlen(*hs->local_file);
1750 *hs->local_file = xrealloc(*hs->local_file,
1751 local_filename_len + sizeof(".html"));
1752 strcpy(*hs->local_file + local_filename_len, ".html");
1754 *dt |= ADDED_HTML_EXTENSION;
1758 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1760 /* If `-c' is in use and the file has been fully downloaded (or
1761 the remote file has shrunk), Wget effectively requests bytes
1762 after the end of file and the server response with 416. */
1763 logputs (LOG_VERBOSE, _("\
1764 \n The file is already fully retrieved; nothing to do.\n\n"));
1765 /* In case the caller inspects. */
1768 /* Mark as successfully retrieved. */
1771 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1772 might be more bytes in the body. */
1773 return RETRUNNEEDED;
1775 if ((contrange != 0 && contrange != hs->restval)
1776 || (H_PARTIAL (statcode) && !contrange))
1778 /* The Range request was somehow misunderstood by the server.
1781 CLOSE_INVALIDATE (sock);
1784 hs->contlen = contlen + contrange;
1790 /* No need to print this output if the body won't be
1791 downloaded at all, or if the original server response is
1793 logputs (LOG_VERBOSE, _("Length: "));
1796 logputs (LOG_VERBOSE, with_thousand_seps (contlen + contrange));
1797 if (contlen + contrange >= 1024)
1798 logprintf (LOG_VERBOSE, " (%s)",
1799 human_readable (contlen + contrange));
1802 if (contlen >= 1024)
1803 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1804 with_thousand_seps (contlen),
1805 human_readable (contlen));
1807 logprintf (LOG_VERBOSE, _(", %s remaining"),
1808 with_thousand_seps (contlen));
1812 logputs (LOG_VERBOSE,
1813 opt.ignore_length ? _("ignored") : _("unspecified"));
1815 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1817 logputs (LOG_VERBOSE, "\n");
1821 type = NULL; /* We don't need it any more. */
1823 /* Return if we have no intention of further downloading. */
1824 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1826 /* In case the caller cares to look... */
1830 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1831 servers not to send body in response to a HEAD request. If
1832 you encounter such a server (more likely a broken CGI), use
1833 `--no-http-keep-alive'. */
1834 CLOSE_FINISH (sock);
1835 return RETRFINISHED;
1838 /* Open the local file. */
1841 mkalldirs (*hs->local_file);
1843 rotate_backups (*hs->local_file);
1845 fp = fopen (*hs->local_file, "ab");
1846 else if (opt.noclobber || opt.always_rest || opt.timestamping || opt.dirstruct
1847 || opt.output_document)
1848 fp = fopen (*hs->local_file, "wb");
1851 fp = fopen_excl (*hs->local_file, 1);
1852 if (!fp && errno == EEXIST)
1854 /* We cannot just invent a new name and use it (which is
1855 what functions like unique_create typically do)
1856 because we told the user we'd use this name.
1857 Instead, return and retry the download. */
1858 logprintf (LOG_NOTQUIET,
1859 _("%s has sprung into existence.\n"),
1861 CLOSE_INVALIDATE (sock);
1862 return FOPEN_EXCL_ERR;
1867 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1868 CLOSE_INVALIDATE (sock);
1875 /* #### This confuses the timestamping code that checks for file
1876 size. Maybe we should save some additional information? */
1877 if (opt.save_headers)
1878 fwrite (head, 1, strlen (head), fp);
1880 /* Download the request body. */
1883 flags |= rb_read_exactly;
1884 if (hs->restval > 0 && contrange == 0)
1885 /* If the server ignored our range request, instruct fd_read_body
1886 to skip the first RESTVAL bytes of body. */
1887 flags |= rb_skip_startpos;
1888 hs->len = hs->restval;
1890 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1891 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1895 CLOSE_FINISH (sock);
1897 CLOSE_INVALIDATE (sock);
1900 /* Close or flush the file. We have to be careful to check for
1901 error here. Checking the result of fwrite() is not enough --
1902 errors could go unnoticed! */
1905 flush_res = fclose (fp);
1907 flush_res = fflush (fp);
1908 if (flush_res == EOF)
1913 return RETRFINISHED;
1916 /* The genuine HTTP loop! This is the part where the retrieval is
1917 retried, and retried, and retried, and... */
1919 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1920 int *dt, struct url *proxy)
1923 int use_ts, got_head = 0; /* time-stamping info */
1924 char *filename_plus_orig_suffix;
1925 char *local_filename = NULL;
1926 char *tms, *locf, *tmrate;
1928 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1929 wgint local_size = 0; /* the size of the local file */
1930 size_t filename_len;
1931 struct http_stat hstat; /* HTTP status */
1935 /* This used to be done in main(), but it's a better idea to do it
1936 here so that we don't go through the hoops if we're just using
1940 if (!wget_cookie_jar)
1941 wget_cookie_jar = cookie_jar_new ();
1942 if (opt.cookies_input && !cookies_loaded_p)
1944 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1945 cookies_loaded_p = 1;
1951 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1952 has_wildcards_p because it would also warn on `?', and we know that
1953 shows up in CGI paths a *lot*. */
1954 if (strchr (u->url, '*'))
1955 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1959 /* Determine the local filename. */
1960 if (local_file && *local_file)
1961 hstat.local_file = local_file;
1962 else if (local_file && !opt.output_document)
1964 *local_file = url_file_name (u);
1965 hstat.local_file = local_file;
1969 dummy = url_file_name (u);
1970 hstat.local_file = &dummy;
1971 /* be honest about where we will save the file */
1972 if (local_file && opt.output_document)
1973 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
1976 if (!opt.output_document)
1977 locf = *hstat.local_file;
1979 locf = opt.output_document;
1981 hstat.referer = referer;
1983 filename_len = strlen (*hstat.local_file);
1984 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1986 if (opt.noclobber && file_exists_p (*hstat.local_file))
1988 /* If opt.noclobber is turned on and file already exists, do not
1989 retrieve the file */
1990 logprintf (LOG_VERBOSE, _("\
1991 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1992 /* If the file is there, we suppose it's retrieved OK. */
1995 /* #### Bogusness alert. */
1996 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1997 if (has_html_suffix_p (*hstat.local_file))
2005 if (opt.timestamping)
2007 int local_dot_orig_file_exists = 0;
2009 if (opt.backup_converted)
2010 /* If -K is specified, we'll act on the assumption that it was specified
2011 last time these files were downloaded as well, and instead of just
2012 comparing local file X against server file X, we'll compare local
2013 file X.orig (if extant, else X) against server file X. If -K
2014 _wasn't_ specified last time, or the server contains files called
2015 *.orig, -N will be back to not operating correctly with -k. */
2017 /* Would a single s[n]printf() call be faster? --dan
2019 Definitely not. sprintf() is horribly slow. It's a
2020 different question whether the difference between the two
2021 affects a program. Usually I'd say "no", but at one
2022 point I profiled Wget, and found that a measurable and
2023 non-negligible amount of time was lost calling sprintf()
2024 in url.c. Replacing sprintf with inline calls to
2025 strcpy() and number_to_string() made a difference.
2027 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2028 memcpy (filename_plus_orig_suffix + filename_len,
2029 ".orig", sizeof (".orig"));
2031 /* Try to stat() the .orig file. */
2032 if (stat (filename_plus_orig_suffix, &st) == 0)
2034 local_dot_orig_file_exists = 1;
2035 local_filename = filename_plus_orig_suffix;
2039 if (!local_dot_orig_file_exists)
2040 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2041 if (stat (*hstat.local_file, &st) == 0)
2042 local_filename = *hstat.local_file;
2044 if (local_filename != NULL)
2045 /* There was a local file, so we'll check later to see if the version
2046 the server has is the same version we already have, allowing us to
2052 /* Modification time granularity is 2 seconds for Windows, so
2053 increase local time by 1 second for later comparison. */
2056 local_size = st.st_size;
2060 /* Reset the counter. */
2066 /* Increment the pass counter. */
2068 sleep_between_retrievals (count);
2069 /* Get the current time string. */
2070 tms = time_str (NULL);
2071 /* Print fetch message, if opt.verbose. */
2074 char *hurl = url_string (u, 1);
2078 sprintf (tmp, _("(try:%2d)"), count);
2079 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2080 tms, hurl, tmp, locf);
2082 ws_changetitle (hurl);
2087 /* Default document type is empty. However, if spider mode is
2088 on or time-stamping is employed, HEAD_ONLY commands is
2089 encoded within *dt. */
2090 if (opt.spider || (use_ts && !got_head))
2095 /* Decide whether or not to restart. */
2098 hstat.restval = hstat.len; /* continue where we left off */
2099 else if (opt.always_rest
2100 && stat (locf, &st) == 0
2101 && S_ISREG (st.st_mode))
2102 hstat.restval = st.st_size;
2104 /* Decide whether to send the no-cache directive. We send it in
2106 a) we're using a proxy, and we're past our first retrieval.
2107 Some proxies are notorious for caching incomplete data, so
2108 we require a fresh get.
2109 b) caching is explicitly inhibited. */
2110 if ((proxy && count > 1) /* a */
2111 || !opt.allow_cache /* b */
2113 *dt |= SEND_NOCACHE;
2115 *dt &= ~SEND_NOCACHE;
2117 /* Try fetching the document, or at least its head. */
2118 err = gethttp (u, &hstat, dt, proxy);
2120 /* It's unfortunate that wget determines the local filename before finding
2121 out the Content-Type of the file. Barring a major restructuring of the
2122 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2123 *hstat.local_file to tack on ".html". */
2124 if (!opt.output_document)
2125 locf = *hstat.local_file;
2128 tms = time_str (NULL);
2129 /* Get the new location (with or without the redirection). */
2131 *newloc = xstrdup (hstat.newloc);
2134 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2135 case CONERROR: case READERR: case WRITEFAILED:
2136 case RANGEERR: case FOPEN_EXCL_ERR:
2137 /* Non-fatal errors continue executing the loop, which will
2138 bring them to "while" statement at the end, to judge
2139 whether the number of tries was exceeded. */
2140 free_hstat (&hstat);
2141 printwhat (count, opt.ntry);
2142 if (err == FOPEN_EXCL_ERR)
2144 /* Re-determine the file name. */
2145 if (local_file && *local_file)
2147 xfree (*local_file);
2148 *local_file = url_file_name (u);
2149 hstat.local_file = local_file;
2154 dummy = url_file_name (u);
2155 hstat.local_file = &dummy;
2157 /* be honest about where we will save the file */
2158 if (local_file && opt.output_document)
2159 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2160 if (!opt.output_document)
2161 locf = *hstat.local_file;
2163 locf = opt.output_document;
2167 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2168 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
2169 /* Fatal errors just return from the function. */
2170 free_hstat (&hstat);
2174 case FWRITEERR: case FOPENERR:
2175 /* Another fatal error. */
2176 logputs (LOG_VERBOSE, "\n");
2177 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2178 *hstat.local_file, strerror (errno));
2179 free_hstat (&hstat);
2184 /* Another fatal error. */
2185 logputs (LOG_VERBOSE, "\n");
2186 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2187 free_hstat (&hstat);
2192 /* Return the new location to the caller. */
2195 logprintf (LOG_NOTQUIET,
2196 _("ERROR: Redirection (%d) without location.\n"),
2198 free_hstat (&hstat);
2202 free_hstat (&hstat);
2207 /* The file was already fully retrieved. */
2208 free_hstat (&hstat);
2213 /* Deal with you later. */
2216 /* All possibilities should have been exhausted. */
2219 if (!(*dt & RETROKF))
2223 /* #### Ugly ugly ugly! */
2224 char *hurl = url_string (u, 1);
2225 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2228 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2229 tms, hstat.statcode, escnonprint (hstat.error));
2230 logputs (LOG_VERBOSE, "\n");
2231 free_hstat (&hstat);
2236 /* Did we get the time-stamp? */
2239 if (opt.timestamping && !hstat.remote_time)
2241 logputs (LOG_NOTQUIET, _("\
2242 Last-modified header missing -- time-stamps turned off.\n"));
2244 else if (hstat.remote_time)
2246 /* Convert the date-string into struct tm. */
2247 tmr = http_atotm (hstat.remote_time);
2248 if (tmr == (time_t) (-1))
2249 logputs (LOG_VERBOSE, _("\
2250 Last-modified header invalid -- time-stamp ignored.\n"));
2254 /* The time-stamping section. */
2259 use_ts = 0; /* no more time-stamping */
2260 count = 0; /* the retrieve count for HEAD is
2262 if (hstat.remote_time && tmr != (time_t) (-1))
2264 /* Now time-stamping can be used validly. Time-stamping
2265 means that if the sizes of the local and remote file
2266 match, and local file is newer than the remote file,
2267 it will not be retrieved. Otherwise, the normal
2268 download procedure is resumed. */
2270 (hstat.contlen == -1 || local_size == hstat.contlen))
2272 logprintf (LOG_VERBOSE, _("\
2273 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2275 free_hstat (&hstat);
2279 else if (tml >= tmr)
2280 logprintf (LOG_VERBOSE, _("\
2281 The sizes do not match (local %s) -- retrieving.\n"),
2282 number_to_static_string (local_size));
2284 logputs (LOG_VERBOSE,
2285 _("Remote file is newer, retrieving.\n"));
2287 free_hstat (&hstat);
2290 if ((tmr != (time_t) (-1))
2292 && ((hstat.len == hstat.contlen) ||
2293 ((hstat.res == 0) &&
2294 ((hstat.contlen == -1) ||
2295 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2297 /* #### This code repeats in http.c and ftp.c. Move it to a
2299 const char *fl = NULL;
2300 if (opt.output_document)
2302 if (output_stream_regular)
2303 fl = opt.output_document;
2306 fl = *hstat.local_file;
2310 /* End of time-stamping section. */
2314 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2315 escnonprint (hstat.error));
2320 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2322 if (hstat.len == hstat.contlen)
2326 logprintf (LOG_VERBOSE,
2327 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2329 number_to_static_string (hstat.len),
2330 number_to_static_string (hstat.contlen));
2331 logprintf (LOG_NONVERBOSE,
2332 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2334 number_to_static_string (hstat.len),
2335 number_to_static_string (hstat.contlen),
2339 total_downloaded_bytes += hstat.len;
2341 /* Remember that we downloaded the file for later ".orig" code. */
2342 if (*dt & ADDED_HTML_EXTENSION)
2343 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2345 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2347 free_hstat (&hstat);
2351 else if (hstat.res == 0) /* No read error */
2353 if (hstat.contlen == -1) /* We don't know how much we were supposed
2354 to get, so assume we succeeded. */
2358 logprintf (LOG_VERBOSE,
2359 _("%s (%s) - `%s' saved [%s]\n\n"),
2361 number_to_static_string (hstat.len));
2362 logprintf (LOG_NONVERBOSE,
2363 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2364 tms, u->url, number_to_static_string (hstat.len),
2368 total_downloaded_bytes += hstat.len;
2370 /* Remember that we downloaded the file for later ".orig" code. */
2371 if (*dt & ADDED_HTML_EXTENSION)
2372 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2374 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2376 free_hstat (&hstat);
2380 else if (hstat.len < hstat.contlen) /* meaning we lost the
2381 connection too soon */
2383 logprintf (LOG_VERBOSE,
2384 _("%s (%s) - Connection closed at byte %s. "),
2385 tms, tmrate, number_to_static_string (hstat.len));
2386 printwhat (count, opt.ntry);
2387 free_hstat (&hstat);
2390 else if (!opt.kill_longer) /* meaning we got more than expected */
2392 logprintf (LOG_VERBOSE,
2393 _("%s (%s) - `%s' saved [%s/%s])\n\n"),
2395 number_to_static_string (hstat.len),
2396 number_to_static_string (hstat.contlen));
2397 logprintf (LOG_NONVERBOSE,
2398 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2400 number_to_static_string (hstat.len),
2401 number_to_static_string (hstat.contlen),
2404 total_downloaded_bytes += hstat.len;
2406 /* Remember that we downloaded the file for later ".orig" code. */
2407 if (*dt & ADDED_HTML_EXTENSION)
2408 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2410 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2412 free_hstat (&hstat);
2416 else /* the same, but not accepted */
2418 logprintf (LOG_VERBOSE,
2419 _("%s (%s) - Connection closed at byte %s/%s. "),
2421 number_to_static_string (hstat.len),
2422 number_to_static_string (hstat.contlen));
2423 printwhat (count, opt.ntry);
2424 free_hstat (&hstat);
2428 else /* now hstat.res can only be -1 */
2430 if (hstat.contlen == -1)
2432 logprintf (LOG_VERBOSE,
2433 _("%s (%s) - Read error at byte %s (%s)."),
2434 tms, tmrate, number_to_static_string (hstat.len),
2436 printwhat (count, opt.ntry);
2437 free_hstat (&hstat);
2440 else /* hstat.res == -1 and contlen is given */
2442 logprintf (LOG_VERBOSE,
2443 _("%s (%s) - Read error at byte %s/%s (%s). "),
2445 number_to_static_string (hstat.len),
2446 number_to_static_string (hstat.contlen),
2448 printwhat (count, opt.ntry);
2449 free_hstat (&hstat);
2456 while (!opt.ntry || (count < opt.ntry));
2460 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2461 than local timezone.
2463 mktime is similar but assumes struct tm, also known as the
2464 "broken-down" form of time, is in local time zone. mktime_from_utc
2465 uses mktime to make the conversion understanding that an offset
2466 will be introduced by the local time assumption.
2468 mktime_from_utc then measures the introduced offset by applying
2469 gmtime to the initial result and applying mktime to the resulting
2470 "broken-down" form. The difference between the two mktime results
2471 is the measured offset which is then subtracted from the initial
2472 mktime result to yield a calendar time which is the value returned.
2474 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2475 consistent offset (the non DST offset) since tm and tm+o might be
2476 on opposite sides of a DST change.
2478 Some implementations of mktime return -1 for the nonexistent
2479 localtime hour at the beginning of DST. In this event, use
2480 mktime(tm - 1hr) + 3600.
2484 gmtime(t+o) --> tm+o
2485 mktime(tm+o) --> t+2o
2486 t+o - (t+2o - t+o) = t
2488 Note that glibc contains a function of the same purpose named
2489 `timegm' (reverse of gmtime). But obviously, it is not universally
2490 available, and unfortunately it is not straightforwardly
2491 extractable for use here. Perhaps configure should detect timegm
2492 and use it where available.
2494 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2495 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2496 Further improved by Roger with assistance from Edward J. Sabol
2497 based on input by Jamie Zawinski. */
2500 mktime_from_utc (struct tm *t)
2511 return -1; /* can't deal with output from strptime */
2522 return -1; /* can't deal with output from gmtime */
2525 return (tl - (tb - tl));
2528 /* Check whether the result of strptime() indicates success.
2529 strptime() returns the pointer to how far it got to in the string.
2530 The processing has been successful if the string is at `GMT' or
2531 `+X', or at the end of the string.
2533 In extended regexp parlance, the function returns 1 if P matches
2534 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2535 can return) is considered a failure and 0 is returned. */
2537 check_end (const char *p)
2541 while (ISSPACE (*p))
2544 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2545 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2551 /* Convert the textual specification of time in TIME_STRING to the
2552 number of seconds since the Epoch.
2554 TIME_STRING can be in any of the three formats RFC2068 allows the
2555 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2556 Timezones are ignored, and should be GMT.
2558 Return the computed time_t representation, or -1 if the conversion
2561 This function uses strptime with various string formats for parsing
2562 TIME_STRING. This results in a parser that is not as lenient in
2563 interpreting TIME_STRING as I would like it to be. Being based on
2564 strptime, it always allows shortened months, one-digit days, etc.,
2565 but due to the multitude of formats in which time can be
2566 represented, an ideal HTTP time parser would be even more
2567 forgiving. It should completely ignore things like week days and
2568 concentrate only on the various forms of representing years,
2569 months, days, hours, minutes, and seconds. For example, it would
2570 be nice if it accepted ISO 8601 out of the box.
2572 I've investigated free and PD code for this purpose, but none was
2573 usable. getdate was big and unwieldy, and had potential copyright
2574 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2575 distributed with phttpd, is excellent, but we cannot use it because
2576 it is not assigned to the FSF. So I stuck it with strptime. */
2579 http_atotm (const char *time_string)
2581 /* NOTE: Solaris strptime man page claims that %n and %t match white
2582 space, but that's not universally available. Instead, we simply
2583 use ` ' to mean "skip all WS", which works under all strptime
2584 implementations I've tested. */
2586 static const char *time_formats[] = {
2587 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2588 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2589 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2590 (google.com uses this for their cookies.) */
2591 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2597 /* According to Roger Beeman, we need to initialize tm_isdst, since
2598 strptime won't do it. */
2601 /* Note that under foreign locales Solaris strptime() fails to
2602 recognize English dates, which renders this function useless. We
2603 solve this by being careful not to affect LC_TIME when
2604 initializing locale.
2606 Another solution would be to temporarily set locale to C, invoke
2607 strptime(), and restore it back. This is slow and dirty,
2608 however, and locale support other than LC_MESSAGES can mess other
2609 things, so I rather chose to stick with just setting LC_MESSAGES.
2611 GNU strptime does not have this problem because it recognizes
2612 both international and local dates. */
2614 for (i = 0; i < countof (time_formats); i++)
2615 if (check_end (strptime (time_string, time_formats[i], &t)))
2616 return mktime_from_utc (&t);
2618 /* All formats have failed. */
2622 /* Authorization support: We support three authorization schemes:
2624 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2626 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2627 consisting of answering to the server's challenge with the proper
2630 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2631 Stenberg for libcurl. Like digest, NTLM is based on a
2632 challenge-response mechanism, but unlike digest, it is non-standard
2633 (authenticates TCP connections rather than requests), undocumented
2634 and Microsoft-specific. */
2636 /* Create the authentication header contents for the `Basic' scheme.
2637 This is done by encoding the string `USER:PASS' in base64 and
2638 prepending `HEADER: Basic ' to it. */
2641 basic_authentication_encode (const char *user, const char *passwd)
2644 int len1 = strlen (user) + 1 + strlen (passwd);
2646 t1 = (char *)alloca (len1 + 1);
2647 sprintf (t1, "%s:%s", user, passwd);
2649 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2650 base64_encode (t1, len1, t2);
2652 return concat_strings ("Basic ", t2, (char *) 0);
2655 #define SKIP_WS(x) do { \
2656 while (ISSPACE (*(x))) \
2660 #ifdef ENABLE_DIGEST
2661 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2662 of a field in such a header. If the field is the one specified by
2663 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2664 digest authorization code), extract its value in the (char*)
2665 variable pointed by RET. Returns negative on a malformed header,
2666 or number of bytes that have been parsed by this call. */
2668 extract_header_attr (const char *au, const char *attr_name, char **ret)
2671 const char *cp = au;
2673 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2675 cp += strlen (attr_name);
2688 for (ep = cp; *ep && *ep != '\"'; ep++)
2693 *ret = strdupdelim (cp, ep);
2700 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2701 an array of 16 bytes containing the hash keys, and BUF should be a
2702 buffer of 33 writable characters (32 for hex digits plus one for
2703 zero termination). */
2705 dump_hash (unsigned char *buf, const unsigned char *hash)
2709 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2711 *buf++ = XNUM_TO_digit (*hash >> 4);
2712 *buf++ = XNUM_TO_digit (*hash & 0xf);
2717 /* Take the line apart to find the challenge, and compose a digest
2718 authorization header. See RFC2069 section 2.1.2. */
2720 digest_authentication_encode (const char *au, const char *user,
2721 const char *passwd, const char *method,
2724 static char *realm, *opaque, *nonce;
2729 { "realm", &realm },
2730 { "opaque", &opaque },
2735 realm = opaque = nonce = NULL;
2737 au += 6; /* skip over `Digest' */
2743 for (i = 0; i < countof (options); i++)
2745 int skip = extract_header_attr (au, options[i].name,
2746 options[i].variable);
2750 xfree_null (opaque);
2760 if (i == countof (options))
2762 while (*au && *au != '=')
2770 while (*au && *au != '\"')
2777 while (*au && *au != ',')
2782 if (!realm || !nonce || !user || !passwd || !path || !method)
2785 xfree_null (opaque);
2790 /* Calculate the digest value. */
2792 ALLOCA_MD5_CONTEXT (ctx);
2793 unsigned char hash[MD5_HASHLEN];
2794 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2795 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2797 /* A1BUF = H(user ":" realm ":" password) */
2799 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2800 gen_md5_update ((unsigned char *)":", 1, ctx);
2801 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2802 gen_md5_update ((unsigned char *)":", 1, ctx);
2803 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2804 gen_md5_finish (ctx, hash);
2805 dump_hash (a1buf, hash);
2807 /* A2BUF = H(method ":" path) */
2809 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2810 gen_md5_update ((unsigned char *)":", 1, ctx);
2811 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2812 gen_md5_finish (ctx, hash);
2813 dump_hash (a2buf, hash);
2815 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2817 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2818 gen_md5_update ((unsigned char *)":", 1, ctx);
2819 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2820 gen_md5_update ((unsigned char *)":", 1, ctx);
2821 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2822 gen_md5_finish (ctx, hash);
2823 dump_hash (response_digest, hash);
2825 res = (char*) xmalloc (strlen (user)
2830 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2831 + (opaque ? strlen (opaque) : 0)
2833 sprintf (res, "Digest \
2834 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2835 user, realm, nonce, path, response_digest);
2838 char *p = res + strlen (res);
2839 strcat (p, ", opaque=\"");
2846 #endif /* ENABLE_DIGEST */
2848 /* Computing the size of a string literal must take into account that
2849 value returned by sizeof includes the terminating \0. */
2850 #define STRSIZE(literal) (sizeof (literal) - 1)
2852 /* Whether chars in [b, e) begin with the literal string provided as
2853 first argument and are followed by whitespace or terminating \0.
2854 The comparison is case-insensitive. */
2855 #define STARTS(literal, b, e) \
2856 ((e) - (b) >= STRSIZE (literal) \
2857 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2858 && ((e) - (b) == STRSIZE (literal) \
2859 || ISSPACE (b[STRSIZE (literal)])))
2862 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2864 return STARTS ("Basic", hdrbeg, hdrend)
2865 #ifdef ENABLE_DIGEST
2866 || STARTS ("Digest", hdrbeg, hdrend)
2869 || STARTS ("NTLM", hdrbeg, hdrend)
2876 /* Create the HTTP authorization request header. When the
2877 `WWW-Authenticate' response header is seen, according to the
2878 authorization scheme specified in that header (`Basic' and `Digest'
2879 are supported by the current implementation), produce an
2880 appropriate HTTP authorization request header. */
2882 create_authorization_line (const char *au, const char *user,
2883 const char *passwd, const char *method,
2884 const char *path, int *finished)
2886 /* We are called only with known schemes, so we can dispatch on the
2888 switch (TOUPPER (*au))
2890 case 'B': /* Basic */
2892 return basic_authentication_encode (user, passwd);
2893 #ifdef ENABLE_DIGEST
2894 case 'D': /* Digest */
2896 return digest_authentication_encode (au, user, passwd, method, path);
2899 case 'N': /* NTLM */
2900 if (!ntlm_input (&pconn.ntlm, au))
2905 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2908 /* We shouldn't get here -- this function should be only called
2909 with values approved by known_authentication_scheme_p. */
2917 if (wget_cookie_jar)
2918 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2924 xfree_null (pconn.host);
2925 if (wget_cookie_jar)
2926 cookie_jar_delete (wget_cookie_jar);