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 for (i = 0; i < req->hcount; i++)
240 hdr = &req->headers[i];
241 if (0 == strcasecmp (name, hdr->name))
243 /* Replace existing header. */
244 release_header (hdr);
247 hdr->release_policy = release_policy;
252 /* Install new header. */
254 if (req->hcount >= req->hcount)
256 req->hcapacity <<= 1;
257 req->headers = xrealloc (req->headers,
258 req->hcapacity * sizeof (struct request_header));
260 hdr = &req->headers[req->hcount++];
263 hdr->release_policy = release_policy;
266 /* Like request_set_header, but sets the whole header line, as
267 provided by the user using the `--header' option. For example,
268 request_set_user_header (req, "Foo: bar") works just like
269 request_set_header (req, "Foo", "bar"). */
272 request_set_user_header (struct request *req, const char *header)
275 const char *p = strchr (header, ':');
278 BOUNDED_TO_ALLOCA (header, p, name);
282 request_set_header (req, xstrdup (name), (char *) p, rel_name);
285 #define APPEND(p, str) do { \
286 int A_len = strlen (str); \
287 memcpy (p, str, A_len); \
291 /* Construct the request and write it to FD using fd_write. */
294 request_send (const struct request *req, int fd)
296 char *request_string, *p;
297 int i, size, write_error;
299 /* Count the request size. */
302 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
303 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
305 for (i = 0; i < req->hcount; i++)
307 struct request_header *hdr = &req->headers[i];
308 /* NAME ": " VALUE "\r\n" */
309 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
315 p = request_string = alloca_array (char, size);
317 /* Generate the request. */
319 APPEND (p, req->method); *p++ = ' ';
320 APPEND (p, req->arg); *p++ = ' ';
321 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
323 for (i = 0; i < req->hcount; i++)
325 struct request_header *hdr = &req->headers[i];
326 APPEND (p, hdr->name);
327 *p++ = ':', *p++ = ' ';
328 APPEND (p, hdr->value);
329 *p++ = '\r', *p++ = '\n';
332 *p++ = '\r', *p++ = '\n', *p++ = '\0';
333 assert (p - request_string == size);
337 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
339 /* Send the request to the server. */
341 write_error = fd_write (fd, request_string, size - 1, -1);
343 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
348 /* Release the resources used by REQ. */
351 request_free (struct request *req)
354 xfree_null (req->arg);
355 for (i = 0; i < req->hcount; i++)
356 release_header (&req->headers[i]);
357 xfree_null (req->headers);
361 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
362 PROMISED_SIZE bytes are sent over the wire -- if the file is
363 longer, read only that much; if the file is shorter, report an error. */
366 post_file (int sock, const char *file_name, wgint promised_size)
368 static char chunk[8192];
373 DEBUGP (("[writing POST file %s ... ", file_name));
375 fp = fopen (file_name, "rb");
378 while (!feof (fp) && written < promised_size)
381 int length = fread (chunk, 1, sizeof (chunk), fp);
384 towrite = MIN (promised_size - written, length);
385 write_error = fd_write (sock, chunk, towrite, -1);
395 /* If we've written less than was promised, report a (probably
396 nonsensical) error rather than break the promise. */
397 if (written < promised_size)
403 assert (written == promised_size);
404 DEBUGP (("done]\n"));
409 response_head_terminator (const char *hunk, int oldlen, int peeklen)
411 const char *start, *end;
413 /* If at first peek, verify whether HUNK starts with "HTTP". If
414 not, this is a HTTP/0.9 request and we must bail out without
416 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
422 start = hunk + oldlen - 4;
423 end = hunk + oldlen + peeklen;
425 for (; start < end - 1; start++)
432 if (start[1] == '\n')
438 /* The maximum size of a single HTTP response we care to read. This
439 is not meant to impose an arbitrary limit, but to protect the user
440 from Wget slurping up available memory upon encountering malicious
441 or buggy server output. Define it to 0 to remove the limit. */
443 #define HTTP_RESPONSE_MAX_SIZE 65536
445 /* Read the HTTP request head from FD and return it. The error
446 conditions are the same as with fd_read_hunk.
448 To support HTTP/0.9 responses, this function tries to make sure
449 that the data begins with "HTTP". If this is not the case, no data
450 is read and an empty request is returned, so that the remaining
451 data can be treated as body. */
454 read_http_response_head (int fd)
456 return fd_read_hunk (fd, response_head_terminator, 512,
457 HTTP_RESPONSE_MAX_SIZE);
461 /* The response data. */
464 /* The array of pointers that indicate where each header starts.
465 For example, given this HTTP response:
472 The headers are located like this:
474 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
476 headers[0] headers[1] headers[2] headers[3]
478 I.e. headers[0] points to the beginning of the request,
479 headers[1] points to the end of the first header and the
480 beginning of the second one, etc. */
482 const char **headers;
485 /* Create a new response object from the text of the HTTP response,
486 available in HEAD. That text is automatically split into
487 constituent header lines for fast retrieval using
490 static struct response *
491 resp_new (const char *head)
496 struct response *resp = xnew0 (struct response);
501 /* Empty head means that we're dealing with a headerless
502 (HTTP/0.9) response. In that case, don't set HEADERS at
507 /* Split HEAD into header lines, so that resp_header_* functions
508 don't need to do this over and over again. */
514 DO_REALLOC (resp->headers, size, count + 1, const char *);
515 resp->headers[count++] = hdr;
517 /* Break upon encountering an empty line. */
518 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
521 /* Find the end of HDR, including continuations. */
524 const char *end = strchr (hdr, '\n');
530 while (*hdr == ' ' || *hdr == '\t');
532 DO_REALLOC (resp->headers, size, count + 1, const char *);
533 resp->headers[count] = NULL;
538 /* Locate the header named NAME in the request data, starting with
539 position START. This allows the code to loop through the request
540 data, filtering for all requests of a given name. Returns the
541 found position, or -1 for failure. The code that uses this
542 function typically looks like this:
544 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
545 ... do something with header ...
547 If you only care about one header, use resp_header_get instead of
551 resp_header_locate (const struct response *resp, const char *name, int start,
552 const char **begptr, const char **endptr)
555 const char **headers = resp->headers;
558 if (!headers || !headers[1])
561 name_len = strlen (name);
567 for (; headers[i + 1]; i++)
569 const char *b = headers[i];
570 const char *e = headers[i + 1];
572 && b[name_len] == ':'
573 && 0 == strncasecmp (b, name, name_len))
576 while (b < e && ISSPACE (*b))
578 while (b < e && ISSPACE (e[-1]))
588 /* Find and retrieve the header named NAME in the request data. If
589 found, set *BEGPTR to its starting, and *ENDPTR to its ending
590 position, and return 1. Otherwise return 0.
592 This function is used as a building block for resp_header_copy
593 and resp_header_strdup. */
596 resp_header_get (const struct response *resp, const char *name,
597 const char **begptr, const char **endptr)
599 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
603 /* Copy the response header named NAME to buffer BUF, no longer than
604 BUFSIZE (BUFSIZE includes the terminating 0). If the header
605 exists, 1 is returned, otherwise 0. If there should be no limit on
606 the size of the header, use resp_header_strdup instead.
608 If BUFSIZE is 0, no data is copied, but the boolean indication of
609 whether the header is present is still returned. */
612 resp_header_copy (const struct response *resp, const char *name,
613 char *buf, int bufsize)
616 if (!resp_header_get (resp, name, &b, &e))
620 int len = MIN (e - b, bufsize - 1);
621 memcpy (buf, b, len);
627 /* Return the value of header named NAME in RESP, allocated with
628 malloc. If such a header does not exist in RESP, return NULL. */
631 resp_header_strdup (const struct response *resp, const char *name)
634 if (!resp_header_get (resp, name, &b, &e))
636 return strdupdelim (b, e);
639 /* Parse the HTTP status line, which is of format:
641 HTTP-Version SP Status-Code SP Reason-Phrase
643 The function returns the status-code, or -1 if the status line
644 appears malformed. The pointer to "reason-phrase" message is
645 returned in *MESSAGE. */
648 resp_status (const struct response *resp, char **message)
655 /* For a HTTP/0.9 response, assume status 200. */
657 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
661 p = resp->headers[0];
662 end = resp->headers[1];
668 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
672 /* Match the HTTP version. This is optional because Gnutella
673 servers have been reported to not specify HTTP version. */
674 if (p < end && *p == '/')
677 while (p < end && ISDIGIT (*p))
679 if (p < end && *p == '.')
681 while (p < end && ISDIGIT (*p))
685 while (p < end && ISSPACE (*p))
687 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
690 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
695 while (p < end && ISSPACE (*p))
697 while (p < end && ISSPACE (end[-1]))
699 *message = strdupdelim (p, end);
705 /* Release the resources used by RESP. */
708 resp_free (struct response *resp)
710 xfree_null (resp->headers);
714 /* Print [b, e) to the log, omitting the trailing CRLF. */
717 print_server_response_1 (const char *prefix, const char *b, const char *e)
720 if (b < e && e[-1] == '\n')
722 if (b < e && e[-1] == '\r')
724 BOUNDED_TO_ALLOCA (b, e, ln);
725 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint (ln));
728 /* Print the server response, line by line, omitting the trailing CR
729 characters, prefixed with PREFIX. */
732 print_server_response (const struct response *resp, const char *prefix)
737 for (i = 0; resp->headers[i + 1]; i++)
738 print_server_response_1 (prefix, resp->headers[i], resp->headers[i + 1]);
741 /* Parse the `Content-Range' header and extract the information it
742 contains. Returns 1 if successful, -1 otherwise. */
744 parse_content_range (const char *hdr, wgint *first_byte_ptr,
745 wgint *last_byte_ptr, wgint *entity_length_ptr)
749 /* Ancient versions of Netscape proxy server, presumably predating
750 rfc2068, sent out `Content-Range' without the "bytes"
752 if (!strncasecmp (hdr, "bytes", 5))
755 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
759 while (ISSPACE (*hdr))
766 for (num = 0; ISDIGIT (*hdr); hdr++)
767 num = 10 * num + (*hdr - '0');
768 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
770 *first_byte_ptr = num;
772 for (num = 0; ISDIGIT (*hdr); hdr++)
773 num = 10 * num + (*hdr - '0');
774 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
776 *last_byte_ptr = num;
778 for (num = 0; ISDIGIT (*hdr); hdr++)
779 num = 10 * num + (*hdr - '0');
780 *entity_length_ptr = num;
784 /* Read the body of the request, but don't store it anywhere and don't
785 display a progress gauge. This is useful for reading the error
786 responses whose bodies don't need to be displayed or logged, but
787 which need to be read anyway. */
790 skip_short_body (int fd, wgint contlen)
792 /* Skipping the body doesn't make sense if the content length is
793 unknown because, in that case, persistent connections cannot be
794 used. (#### This is not the case with HTTP/1.1 where they can
795 still be used with the magic of the "chunked" transfer!) */
798 DEBUGP (("Skipping %s bytes of body data... ", number_to_static_string (contlen)));
803 int ret = fd_read (fd, dlbuf, MIN (contlen, sizeof (dlbuf)), -1);
808 DEBUGP (("done.\n"));
811 /* Persistent connections. Currently, we cache the most recently used
812 connection as persistent, provided that the HTTP server agrees to
813 make it such. The persistence data is stored in the variables
814 below. Ideally, it should be possible to cache an arbitrary fixed
815 number of these connections. */
817 /* Whether a persistent connection is active. */
818 static int pconn_active;
821 /* The socket of the connection. */
824 /* Host and port of the currently active persistent connection. */
828 /* Whether a ssl handshake has occoured on this connection. */
832 /* NTLM data of the current connection. */
833 struct ntlmdata ntlm;
837 /* Mark the persistent connection as invalid and free the resources it
838 uses. This is used by the CLOSE_* macros after they forcefully
839 close a registered persistent connection. */
842 invalidate_persistent (void)
844 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
846 fd_close (pconn.socket);
851 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
852 persistent. This will enable someone to use the same connection
853 later. In the context of HTTP, this must be called only AFTER the
854 response has been received and the server has promised that the
855 connection will remain alive.
857 If a previous connection was persistent, it is closed. */
860 register_persistent (const char *host, int port, int fd, int ssl)
864 if (pconn.socket == fd)
866 /* The connection FD is already registered. */
871 /* The old persistent connection is still active; close it
872 first. This situation arises whenever a persistent
873 connection exists, but we then connect to a different
874 host, and try to register a persistent connection to that
876 invalidate_persistent ();
882 pconn.host = xstrdup (host);
886 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
889 /* Return non-zero if a persistent connection is available for
890 connecting to HOST:PORT. */
893 persistent_available_p (const char *host, int port, int ssl,
894 int *host_lookup_failed)
896 /* First, check whether a persistent connection is active at all. */
900 /* If we want SSL and the last connection wasn't or vice versa,
901 don't use it. Checking for host and port is not enough because
902 HTTP and HTTPS can apparently coexist on the same port. */
903 if (ssl != pconn.ssl)
906 /* If we're not connecting to the same port, we're not interested. */
907 if (port != pconn.port)
910 /* If the host is the same, we're in business. If not, there is
911 still hope -- read below. */
912 if (0 != strcasecmp (host, pconn.host))
914 /* If pconn.socket is already talking to HOST, we needn't
915 reconnect. This happens often when both sites are virtual
916 hosts distinguished only by name and served by the same
917 network interface, and hence the same web server (possibly
918 set up by the ISP and serving many different web sites).
919 This admittedly non-standard optimization does not contradict
920 HTTP and works well with popular server software. */
924 struct address_list *al;
927 /* Don't try to talk to two different SSL sites over the same
928 secure connection! (Besides, it's not clear if name-based
929 virtual hosting is even possible with SSL.) */
932 /* If pconn.socket's peer is one of the IP addresses HOST
933 resolves to, pconn.socket is for all intents and purposes
934 already talking to HOST. */
936 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
938 /* Can't get the peer's address -- something must be very
939 wrong with the connection. */
940 invalidate_persistent ();
943 al = lookup_host (host, 0);
946 *host_lookup_failed = 1;
950 found = address_list_contains (al, &ip);
951 address_list_release (al);
956 /* The persistent connection's peer address was found among the
957 addresses HOST resolved to; therefore, pconn.sock is in fact
958 already talking to HOST -- no need to reconnect. */
961 /* Finally, check whether the connection is still open. This is
962 important because most server implement a liberal (short) timeout
963 on persistent connections. Wget can of course always reconnect
964 if the connection doesn't work out, but it's nicer to know in
965 advance. This test is a logical followup of the first test, but
966 is "expensive" and therefore placed at the end of the list. */
968 if (!test_socket_open (pconn.socket))
970 /* Oops, the socket is no longer open. Now that we know that,
971 let's invalidate the persistent connection before returning
973 invalidate_persistent ();
980 /* The idea behind these two CLOSE macros is to distinguish between
981 two cases: one when the job we've been doing is finished, and we
982 want to close the connection and leave, and two when something is
983 seriously wrong and we're closing the connection as part of
986 In case of keep_alive, CLOSE_FINISH should leave the connection
987 open, while CLOSE_INVALIDATE should still close it.
989 Note that the semantics of the flag `keep_alive' is "this
990 connection *will* be reused (the server has promised not to close
991 the connection once we're done)", while the semantics of
992 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
993 active, registered connection". */
995 #define CLOSE_FINISH(fd) do { \
998 if (pconn_active && (fd) == pconn.socket) \
999 invalidate_persistent (); \
1008 #define CLOSE_INVALIDATE(fd) do { \
1009 if (pconn_active && (fd) == pconn.socket) \
1010 invalidate_persistent (); \
1018 wgint len; /* received length */
1019 wgint contlen; /* expected length */
1020 wgint restval; /* the restart value */
1021 int res; /* the result of last read */
1022 char *newloc; /* new location (redirection) */
1023 char *remote_time; /* remote time-stamp string */
1024 char *error; /* textual HTTP error */
1025 int statcode; /* status code */
1026 wgint rd_size; /* amount of data read from socket */
1027 double dltime; /* time it took to download the data */
1028 const char *referer; /* value of the referer header. */
1029 char **local_file; /* local file. */
1033 free_hstat (struct http_stat *hs)
1035 xfree_null (hs->newloc);
1036 xfree_null (hs->remote_time);
1037 xfree_null (hs->error);
1039 /* Guard against being called twice. */
1041 hs->remote_time = NULL;
1045 static char *create_authorization_line PARAMS ((const char *, const char *,
1046 const char *, const char *,
1048 static char *basic_authentication_encode PARAMS ((const char *, const char *));
1049 static int known_authentication_scheme_p PARAMS ((const char *));
1051 time_t http_atotm PARAMS ((const char *));
1053 #define BEGINS_WITH(line, string_constant) \
1054 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1055 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1056 || !line[sizeof (string_constant) - 1]))
1058 /* Retrieve a document through HTTP protocol. It recognizes status
1059 code, and correctly handles redirections. It closes the network
1060 socket. If it receives an error from the functions below it, it
1061 will print it if there is enough information to do so (almost
1062 always), returning the error to the caller (i.e. http_loop).
1064 Various HTTP parameters are stored to hs.
1066 If PROXY is non-NULL, the connection will be made to the proxy
1067 server, and u->url will be requested. */
1069 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1071 struct request *req;
1074 char *user, *passwd;
1078 wgint contlen, contrange;
1085 /* Whether authorization has been already tried. */
1086 int auth_tried_already;
1088 /* Whether our connection to the remote host is through SSL. */
1092 struct response *resp;
1096 /* Whether this connection will be kept alive after the HTTP request
1100 /* Whether keep-alive should be inhibited.
1102 RFC 2068 requests that 1.0 clients not send keep-alive requests
1103 to proxies. This is because many 1.0 proxies do not interpret
1104 the Connection header and transfer it to the remote server,
1105 causing it to not close the connection and leave both the proxy
1106 and the client hanging. */
1107 int inhibit_keep_alive =
1108 !opt.http_keep_alive || opt.ignore_length /*|| proxy != NULL*/;
1110 /* Headers sent when using POST. */
1111 wgint post_data_size = 0;
1113 int host_lookup_failed = 0;
1116 if (u->scheme == SCHEME_HTTPS)
1118 /* Initialize the SSL context. After this has once been done,
1119 it becomes a no-op. */
1120 switch (ssl_init ())
1122 case SSLERRCTXCREATE:
1124 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
1125 return SSLERRCTXCREATE;
1126 case SSLERRCERTFILE:
1127 /* try without certfile */
1128 logprintf (LOG_NOTQUIET,
1129 _("Failed to load certificates from %s\n"),
1131 logprintf (LOG_NOTQUIET,
1132 _("Trying without the specified certificate\n"));
1135 logprintf (LOG_NOTQUIET,
1136 _("Failed to get certificate key from %s\n"),
1138 logprintf (LOG_NOTQUIET,
1139 _("Trying without the specified certificate\n"));
1145 #endif /* HAVE_SSL */
1147 if (!(*dt & HEAD_ONLY))
1148 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1149 know the local filename so we can save to it. */
1150 assert (*hs->local_file != NULL);
1152 auth_tried_already = 0;
1154 /* Initialize certain elements of struct http_stat. */
1159 hs->remote_time = NULL;
1164 /* Prepare the request to send. */
1166 req = request_new ();
1168 const char *meth = "GET";
1169 if (*dt & HEAD_ONLY)
1171 else if (opt.post_file_name || opt.post_data)
1173 /* Use the full path, i.e. one that includes the leading slash and
1174 the query string. E.g. if u->path is "foo/bar" and u->query is
1175 "param=value", full_path will be "/foo/bar?param=value". */
1176 request_set_method (req, meth,
1177 proxy ? xstrdup (u->url) : url_full_path (u));
1180 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1181 if (*dt & SEND_NOCACHE)
1182 request_set_header (req, "Pragma", "no-cache", rel_none);
1184 request_set_header (req, "Range",
1185 aprintf ("bytes=%s-",
1186 number_to_static_string (hs->restval)),
1189 request_set_header (req, "User-Agent", opt.useragent, rel_none);
1191 request_set_header (req, "User-Agent",
1192 aprintf ("Wget/%s", version_string), rel_value);
1193 request_set_header (req, "Accept", "*/*", rel_none);
1195 /* Find the username and password for authentication. */
1198 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1199 user = user ? user : opt.http_user;
1200 passwd = passwd ? passwd : opt.http_passwd;
1204 /* We have the username and the password, but haven't tried
1205 any authorization yet. Let's see if the "Basic" method
1206 works. If not, we'll come back here and construct a
1207 proper authorization method with the right challenges.
1209 If we didn't employ this kind of logic, every URL that
1210 requires authorization would have to be processed twice,
1211 which is very suboptimal and generates a bunch of false
1212 "unauthorized" errors in the server log.
1214 #### But this logic also has a serious problem when used
1215 with stronger authentications: we *first* transmit the
1216 username and the password in clear text, and *then* attempt a
1217 stronger authentication scheme. That cannot be right! We
1218 are only fortunate that almost everyone still uses the
1219 `Basic' scheme anyway.
1221 There should be an option to prevent this from happening, for
1222 those who use strong authentication schemes and value their
1224 request_set_header (req, "Authorization",
1225 basic_authentication_encode (user, passwd),
1232 char *proxy_user, *proxy_passwd;
1233 /* For normal username and password, URL components override
1234 command-line/wgetrc parameters. With proxy
1235 authentication, it's the reverse, because proxy URLs are
1236 normally the "permanent" ones, so command-line args
1237 should take precedence. */
1238 if (opt.proxy_user && opt.proxy_passwd)
1240 proxy_user = opt.proxy_user;
1241 proxy_passwd = opt.proxy_passwd;
1245 proxy_user = proxy->user;
1246 proxy_passwd = proxy->passwd;
1248 /* #### This does not appear right. Can't the proxy request,
1249 say, `Digest' authentication? */
1250 if (proxy_user && proxy_passwd)
1251 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1253 /* If we're using a proxy, we will be connecting to the proxy
1257 /* Proxy authorization over SSL is handled below. */
1259 if (u->scheme != SCHEME_HTTPS)
1261 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1265 /* Whether we need to print the host header with braces around
1266 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1267 usual "Host: symbolic-name:1234". */
1268 int squares = strchr (u->host, ':') != NULL;
1269 if (u->port == scheme_default_port (u->scheme))
1270 request_set_header (req, "Host",
1271 aprintf (squares ? "[%s]" : "%s", u->host),
1274 request_set_header (req, "Host",
1275 aprintf (squares ? "[%s]:%d" : "%s:%d",
1280 if (!inhibit_keep_alive)
1281 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1284 request_set_header (req, "Cookie",
1285 cookie_header (wget_cookie_jar,
1286 u->host, u->port, u->path,
1288 u->scheme == SCHEME_HTTPS
1295 if (opt.post_data || opt.post_file_name)
1297 request_set_header (req, "Content-Type",
1298 "application/x-www-form-urlencoded", rel_none);
1300 post_data_size = strlen (opt.post_data);
1303 post_data_size = file_size (opt.post_file_name);
1304 if (post_data_size == -1)
1306 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1307 opt.post_file_name);
1311 request_set_header (req, "Content-Length",
1312 xstrdup (number_to_static_string (post_data_size)),
1316 /* Add the user headers. */
1317 if (opt.user_headers)
1320 for (i = 0; opt.user_headers[i]; i++)
1321 request_set_user_header (req, opt.user_headers[i]);
1325 /* We need to come back here when the initial attempt to retrieve
1326 without authorization header fails. (Expected to happen at least
1327 for the Digest authorization scheme.) */
1331 /* Establish the connection. */
1333 if (!inhibit_keep_alive)
1335 /* Look for a persistent connection to target host, unless a
1336 proxy is used. The exception is when SSL is in use, in which
1337 case the proxy is nothing but a passthrough to the target
1338 host, registered as a connection to the latter. */
1339 struct url *relevant = conn;
1341 if (u->scheme == SCHEME_HTTPS)
1345 if (persistent_available_p (relevant->host, relevant->port,
1347 relevant->scheme == SCHEME_HTTPS,
1351 &host_lookup_failed))
1353 sock = pconn.socket;
1354 using_ssl = pconn.ssl;
1355 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1356 escnonprint (pconn.host), pconn.port);
1357 DEBUGP (("Reusing fd %d.\n", sock));
1363 /* In its current implementation, persistent_available_p will
1364 look up conn->host in some cases. If that lookup failed, we
1365 don't need to bother with connect_to_host. */
1366 if (host_lookup_failed)
1372 sock = connect_to_host (conn->host, conn->port);
1381 return (retryable_socket_connect_error (errno)
1382 ? CONERROR : CONIMPOSSIBLE);
1386 if (proxy && u->scheme == SCHEME_HTTPS)
1388 /* When requesting SSL URLs through proxies, use the
1389 CONNECT method to request passthrough. */
1390 struct request *connreq = request_new ();
1391 request_set_method (connreq, "CONNECT",
1392 aprintf ("%s:%d", u->host, u->port));
1395 request_set_header (connreq, "Proxy-Authorization",
1396 proxyauth, rel_value);
1397 /* Now that PROXYAUTH is part of the CONNECT request,
1398 zero it out so we don't send proxy authorization with
1399 the regular request below. */
1403 write_error = request_send (connreq, sock);
1404 request_free (connreq);
1405 if (write_error < 0)
1407 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1409 CLOSE_INVALIDATE (sock);
1413 head = read_http_response_head (sock);
1416 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1418 CLOSE_INVALIDATE (sock);
1427 DEBUGP (("proxy responded with: [%s]\n", head));
1429 resp = resp_new (head);
1430 statcode = resp_status (resp, &message);
1433 if (statcode != 200)
1436 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1437 message ? escnonprint (message) : "?");
1438 xfree_null (message);
1441 xfree_null (message);
1443 /* SOCK is now *really* connected to u->host, so update CONN
1444 to reflect this. That way register_persistent will
1445 register SOCK as being connected to u->host:u->port. */
1449 if (conn->scheme == SCHEME_HTTPS)
1451 if (!ssl_connect (sock))
1458 #endif /* HAVE_SSL */
1461 /* Send the request to server. */
1462 write_error = request_send (req, sock);
1464 if (write_error >= 0)
1468 DEBUGP (("[POST data: %s]\n", opt.post_data));
1469 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1471 else if (opt.post_file_name && post_data_size != 0)
1472 write_error = post_file (sock, opt.post_file_name, post_data_size);
1475 if (write_error < 0)
1477 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1479 CLOSE_INVALIDATE (sock);
1483 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1484 proxy ? "Proxy" : "HTTP");
1489 head = read_http_response_head (sock);
1494 logputs (LOG_NOTQUIET, _("No data received.\n"));
1495 CLOSE_INVALIDATE (sock);
1501 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1503 CLOSE_INVALIDATE (sock);
1508 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1510 resp = resp_new (head);
1512 /* Check for status line. */
1514 statcode = resp_status (resp, &message);
1515 if (!opt.server_response)
1516 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1517 message ? escnonprint (message) : "");
1520 logprintf (LOG_VERBOSE, "\n");
1521 print_server_response (resp, " ");
1524 if (!opt.ignore_length
1525 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1529 parsed = str_to_wgint (hdrval, NULL, 10);
1530 if (parsed == WGINT_MAX && errno == ERANGE)
1532 #### If Content-Length is out of range, it most likely
1533 means that the file is larger than 2G and that we're
1534 compiled without LFS. In that case we should probably
1535 refuse to even attempt to download the file. */
1541 /* Check for keep-alive related responses. */
1542 if (!inhibit_keep_alive && contlen != -1)
1544 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1546 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1548 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1553 /* The server has promised that it will not close the connection
1554 when we're done. This means that we can register it. */
1555 register_persistent (conn->host, conn->port, sock, using_ssl);
1557 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1559 /* Authorization is required. */
1560 skip_short_body (sock, contlen);
1561 CLOSE_FINISH (sock);
1562 if (auth_tried_already || !(user && passwd))
1564 /* If we have tried it already, then there is not point
1566 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1570 char *www_authenticate = resp_header_strdup (resp,
1571 "WWW-Authenticate");
1572 /* If the authentication scheme is unknown or if it's the
1573 "Basic" authentication (which we try by default), there's
1574 no sense in retrying. */
1575 if (!www_authenticate
1576 || !known_authentication_scheme_p (www_authenticate)
1577 || BEGINS_WITH (www_authenticate, "Basic"))
1579 xfree_null (www_authenticate);
1580 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1585 auth_tried_already = 1;
1586 pth = url_full_path (u);
1587 request_set_header (req, "Authorization",
1588 create_authorization_line (www_authenticate,
1590 request_method (req),
1594 xfree (www_authenticate);
1595 goto retry_with_auth;
1603 hs->statcode = statcode;
1605 hs->error = xstrdup (_("Malformed status line"));
1607 hs->error = xstrdup (_("(no description)"));
1609 hs->error = xstrdup (message);
1612 type = resp_header_strdup (resp, "Content-Type");
1615 char *tmp = strchr (type, ';');
1618 while (tmp > type && ISSPACE (tmp[-1]))
1623 hs->newloc = resp_header_strdup (resp, "Location");
1624 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1626 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1629 const char *scbeg, *scend;
1630 /* The jar should have been created by now. */
1631 assert (wget_cookie_jar != NULL);
1633 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1634 &scbeg, &scend)) != -1;
1637 char *set_cookie = strdupdelim (scbeg, scend);
1638 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1644 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1646 wgint first_byte_pos, last_byte_pos, entity_length;
1647 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1649 contrange = first_byte_pos;
1654 /* 20x responses are counted among successful by default. */
1655 if (H_20X (statcode))
1658 /* Return if redirected. */
1659 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1661 /* RFC2068 says that in case of the 300 (multiple choices)
1662 response, the server can output a preferred URL through
1663 `Location' header; otherwise, the request should be treated
1664 like GET. So, if the location is set, it will be a
1665 redirection; otherwise, just proceed normally. */
1666 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1670 logprintf (LOG_VERBOSE,
1671 _("Location: %s%s\n"),
1672 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1673 hs->newloc ? _(" [following]") : "");
1675 skip_short_body (sock, contlen);
1676 CLOSE_FINISH (sock);
1682 /* If content-type is not given, assume text/html. This is because
1683 of the multitude of broken CGI's that "forget" to generate the
1686 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1687 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1692 if (opt.html_extension && (*dt & TEXTHTML))
1693 /* -E / --html-extension / html_extension = on was specified, and this is a
1694 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1695 already the file's suffix, tack on ".html". */
1697 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1699 if (last_period_in_local_filename == NULL
1700 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1701 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1703 size_t local_filename_len = strlen(*hs->local_file);
1705 *hs->local_file = xrealloc(*hs->local_file,
1706 local_filename_len + sizeof(".html"));
1707 strcpy(*hs->local_file + local_filename_len, ".html");
1709 *dt |= ADDED_HTML_EXTENSION;
1713 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1715 /* If `-c' is in use and the file has been fully downloaded (or
1716 the remote file has shrunk), Wget effectively requests bytes
1717 after the end of file and the server response with 416. */
1718 logputs (LOG_VERBOSE, _("\
1719 \n The file is already fully retrieved; nothing to do.\n\n"));
1720 /* In case the caller inspects. */
1723 /* Mark as successfully retrieved. */
1726 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1727 might be more bytes in the body. */
1728 return RETRUNNEEDED;
1730 if ((contrange != 0 && contrange != hs->restval)
1731 || (H_PARTIAL (statcode) && !contrange))
1733 /* The Range request was somehow misunderstood by the server.
1736 CLOSE_INVALIDATE (sock);
1739 hs->contlen = contlen + contrange;
1745 /* No need to print this output if the body won't be
1746 downloaded at all, or if the original server response is
1748 logputs (LOG_VERBOSE, _("Length: "));
1751 logputs (LOG_VERBOSE, with_thousand_seps (contlen + contrange));
1752 if (contlen + contrange >= 1024)
1753 logprintf (LOG_VERBOSE, " (%s)",
1754 human_readable (contlen + contrange));
1757 if (contlen >= 1024)
1758 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1759 with_thousand_seps (contlen),
1760 human_readable (contlen));
1762 logprintf (LOG_VERBOSE, _(", %s remaining"),
1763 with_thousand_seps (contlen));
1767 logputs (LOG_VERBOSE,
1768 opt.ignore_length ? _("ignored") : _("unspecified"));
1770 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1772 logputs (LOG_VERBOSE, "\n");
1776 type = NULL; /* We don't need it any more. */
1778 /* Return if we have no intention of further downloading. */
1779 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1781 /* In case the caller cares to look... */
1785 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1786 servers not to send body in response to a HEAD request. If
1787 you encounter such a server (more likely a broken CGI), use
1788 `--no-http-keep-alive'. */
1789 CLOSE_FINISH (sock);
1790 return RETRFINISHED;
1793 /* Open the local file. */
1796 mkalldirs (*hs->local_file);
1798 rotate_backups (*hs->local_file);
1800 fp = fopen (*hs->local_file, "ab");
1801 else if (opt.noclobber || opt.always_rest || opt.timestamping || opt.dirstruct
1802 || opt.output_document)
1803 fp = fopen (*hs->local_file, "wb");
1806 fp = fopen_excl (*hs->local_file, 0);
1807 if (!fp && errno == EEXIST)
1809 /* We cannot just invent a new name and use it (which is
1810 what functions like unique_create typically do)
1811 because we told the user we'd use this name.
1812 Instead, return and retry the download. */
1813 logprintf (LOG_NOTQUIET,
1814 _("%s has sprung into existence.\n"),
1816 CLOSE_INVALIDATE (sock);
1817 return FOPEN_EXCL_ERR;
1822 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1823 CLOSE_INVALIDATE (sock);
1830 /* #### This confuses the timestamping code that checks for file
1831 size. Maybe we should save some additional information? */
1832 if (opt.save_headers)
1833 fwrite (head, 1, strlen (head), fp);
1835 /* Download the request body. */
1838 flags |= rb_read_exactly;
1839 if (hs->restval > 0 && contrange == 0)
1840 /* If the server ignored our range request, instruct fd_read_body
1841 to skip the first RESTVAL bytes of body. */
1842 flags |= rb_skip_startpos;
1843 hs->len = hs->restval;
1845 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1846 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1850 CLOSE_FINISH (sock);
1852 CLOSE_INVALIDATE (sock);
1855 /* Close or flush the file. We have to be careful to check for
1856 error here. Checking the result of fwrite() is not enough --
1857 errors could go unnoticed! */
1860 flush_res = fclose (fp);
1862 flush_res = fflush (fp);
1863 if (flush_res == EOF)
1868 return RETRFINISHED;
1871 /* The genuine HTTP loop! This is the part where the retrieval is
1872 retried, and retried, and retried, and... */
1874 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1875 int *dt, struct url *proxy)
1878 int use_ts, got_head = 0; /* time-stamping info */
1879 char *filename_plus_orig_suffix;
1880 char *local_filename = NULL;
1881 char *tms, *locf, *tmrate;
1883 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1884 wgint local_size = 0; /* the size of the local file */
1885 size_t filename_len;
1886 struct http_stat hstat; /* HTTP status */
1890 /* This used to be done in main(), but it's a better idea to do it
1891 here so that we don't go through the hoops if we're just using
1895 if (!wget_cookie_jar)
1896 wget_cookie_jar = cookie_jar_new ();
1897 if (opt.cookies_input && !cookies_loaded_p)
1899 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1900 cookies_loaded_p = 1;
1906 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1907 has_wildcards_p because it would also warn on `?', and we know that
1908 shows up in CGI paths a *lot*. */
1909 if (strchr (u->url, '*'))
1910 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1914 /* Determine the local filename. */
1915 if (local_file && *local_file)
1916 hstat.local_file = local_file;
1917 else if (local_file && !opt.output_document)
1919 *local_file = url_file_name (u);
1920 hstat.local_file = local_file;
1924 dummy = url_file_name (u);
1925 hstat.local_file = &dummy;
1926 /* be honest about where we will save the file */
1927 if (local_file && opt.output_document)
1928 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
1931 if (!opt.output_document)
1932 locf = *hstat.local_file;
1934 locf = opt.output_document;
1936 hstat.referer = referer;
1938 filename_len = strlen (*hstat.local_file);
1939 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1941 if (opt.noclobber && file_exists_p (*hstat.local_file))
1943 /* If opt.noclobber is turned on and file already exists, do not
1944 retrieve the file */
1945 logprintf (LOG_VERBOSE, _("\
1946 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1947 /* If the file is there, we suppose it's retrieved OK. */
1950 /* #### Bogusness alert. */
1951 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1952 if (has_html_suffix_p (*hstat.local_file))
1960 if (opt.timestamping)
1962 int local_dot_orig_file_exists = 0;
1964 if (opt.backup_converted)
1965 /* If -K is specified, we'll act on the assumption that it was specified
1966 last time these files were downloaded as well, and instead of just
1967 comparing local file X against server file X, we'll compare local
1968 file X.orig (if extant, else X) against server file X. If -K
1969 _wasn't_ specified last time, or the server contains files called
1970 *.orig, -N will be back to not operating correctly with -k. */
1972 /* Would a single s[n]printf() call be faster? --dan
1974 Definitely not. sprintf() is horribly slow. It's a
1975 different question whether the difference between the two
1976 affects a program. Usually I'd say "no", but at one
1977 point I profiled Wget, and found that a measurable and
1978 non-negligible amount of time was lost calling sprintf()
1979 in url.c. Replacing sprintf with inline calls to
1980 strcpy() and number_to_string() made a difference.
1982 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1983 memcpy (filename_plus_orig_suffix + filename_len,
1984 ".orig", sizeof (".orig"));
1986 /* Try to stat() the .orig file. */
1987 if (stat (filename_plus_orig_suffix, &st) == 0)
1989 local_dot_orig_file_exists = 1;
1990 local_filename = filename_plus_orig_suffix;
1994 if (!local_dot_orig_file_exists)
1995 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1996 if (stat (*hstat.local_file, &st) == 0)
1997 local_filename = *hstat.local_file;
1999 if (local_filename != NULL)
2000 /* There was a local file, so we'll check later to see if the version
2001 the server has is the same version we already have, allowing us to
2007 /* Modification time granularity is 2 seconds for Windows, so
2008 increase local time by 1 second for later comparison. */
2011 local_size = st.st_size;
2015 /* Reset the counter. */
2021 /* Increment the pass counter. */
2023 sleep_between_retrievals (count);
2024 /* Get the current time string. */
2025 tms = time_str (NULL);
2026 /* Print fetch message, if opt.verbose. */
2029 char *hurl = url_string (u, 1);
2033 sprintf (tmp, _("(try:%2d)"), count);
2034 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2035 tms, hurl, tmp, locf);
2037 ws_changetitle (hurl);
2042 /* Default document type is empty. However, if spider mode is
2043 on or time-stamping is employed, HEAD_ONLY commands is
2044 encoded within *dt. */
2045 if (opt.spider || (use_ts && !got_head))
2050 /* Decide whether or not to restart. */
2053 hstat.restval = hstat.len; /* continue where we left off */
2054 else if (opt.always_rest
2055 && stat (locf, &st) == 0
2056 && S_ISREG (st.st_mode))
2057 hstat.restval = st.st_size;
2059 /* Decide whether to send the no-cache directive. We send it in
2061 a) we're using a proxy, and we're past our first retrieval.
2062 Some proxies are notorious for caching incomplete data, so
2063 we require a fresh get.
2064 b) caching is explicitly inhibited. */
2065 if ((proxy && count > 1) /* a */
2066 || !opt.allow_cache /* b */
2068 *dt |= SEND_NOCACHE;
2070 *dt &= ~SEND_NOCACHE;
2072 /* Try fetching the document, or at least its head. */
2073 err = gethttp (u, &hstat, dt, proxy);
2075 /* It's unfortunate that wget determines the local filename before finding
2076 out the Content-Type of the file. Barring a major restructuring of the
2077 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2078 *hstat.local_file to tack on ".html". */
2079 if (!opt.output_document)
2080 locf = *hstat.local_file;
2083 tms = time_str (NULL);
2084 /* Get the new location (with or without the redirection). */
2086 *newloc = xstrdup (hstat.newloc);
2089 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2090 case CONERROR: case READERR: case WRITEFAILED:
2091 case RANGEERR: case FOPEN_EXCL_ERR:
2092 /* Non-fatal errors continue executing the loop, which will
2093 bring them to "while" statement at the end, to judge
2094 whether the number of tries was exceeded. */
2095 free_hstat (&hstat);
2096 printwhat (count, opt.ntry);
2097 if (err == FOPEN_EXCL_ERR)
2099 /* Re-determine the file name. */
2100 if (local_file && *local_file)
2102 xfree (*local_file);
2103 *local_file = url_file_name (u);
2104 hstat.local_file = local_file;
2109 dummy = url_file_name (u);
2110 hstat.local_file = &dummy;
2112 /* be honest about where we will save the file */
2113 if (local_file && opt.output_document)
2114 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2115 if (!opt.output_document)
2116 locf = *hstat.local_file;
2118 locf = opt.output_document;
2122 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2123 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
2124 /* Fatal errors just return from the function. */
2125 free_hstat (&hstat);
2129 case FWRITEERR: case FOPENERR:
2130 /* Another fatal error. */
2131 logputs (LOG_VERBOSE, "\n");
2132 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2133 *hstat.local_file, strerror (errno));
2134 free_hstat (&hstat);
2139 /* Another fatal error. */
2140 logputs (LOG_VERBOSE, "\n");
2141 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2142 free_hstat (&hstat);
2147 /* Return the new location to the caller. */
2150 logprintf (LOG_NOTQUIET,
2151 _("ERROR: Redirection (%d) without location.\n"),
2153 free_hstat (&hstat);
2157 free_hstat (&hstat);
2162 /* The file was already fully retrieved. */
2163 free_hstat (&hstat);
2168 /* Deal with you later. */
2171 /* All possibilities should have been exhausted. */
2174 if (!(*dt & RETROKF))
2178 /* #### Ugly ugly ugly! */
2179 char *hurl = url_string (u, 1);
2180 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2183 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2184 tms, hstat.statcode, escnonprint (hstat.error));
2185 logputs (LOG_VERBOSE, "\n");
2186 free_hstat (&hstat);
2191 /* Did we get the time-stamp? */
2194 if (opt.timestamping && !hstat.remote_time)
2196 logputs (LOG_NOTQUIET, _("\
2197 Last-modified header missing -- time-stamps turned off.\n"));
2199 else if (hstat.remote_time)
2201 /* Convert the date-string into struct tm. */
2202 tmr = http_atotm (hstat.remote_time);
2203 if (tmr == (time_t) (-1))
2204 logputs (LOG_VERBOSE, _("\
2205 Last-modified header invalid -- time-stamp ignored.\n"));
2209 /* The time-stamping section. */
2214 use_ts = 0; /* no more time-stamping */
2215 count = 0; /* the retrieve count for HEAD is
2217 if (hstat.remote_time && tmr != (time_t) (-1))
2219 /* Now time-stamping can be used validly. Time-stamping
2220 means that if the sizes of the local and remote file
2221 match, and local file is newer than the remote file,
2222 it will not be retrieved. Otherwise, the normal
2223 download procedure is resumed. */
2225 (hstat.contlen == -1 || local_size == hstat.contlen))
2227 logprintf (LOG_VERBOSE, _("\
2228 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2230 free_hstat (&hstat);
2234 else if (tml >= tmr)
2235 logprintf (LOG_VERBOSE, _("\
2236 The sizes do not match (local %s) -- retrieving.\n"),
2237 number_to_static_string (local_size));
2239 logputs (LOG_VERBOSE,
2240 _("Remote file is newer, retrieving.\n"));
2242 free_hstat (&hstat);
2245 if ((tmr != (time_t) (-1))
2247 && ((hstat.len == hstat.contlen) ||
2248 ((hstat.res == 0) &&
2249 ((hstat.contlen == -1) ||
2250 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2252 /* #### This code repeats in http.c and ftp.c. Move it to a
2254 const char *fl = NULL;
2255 if (opt.output_document)
2257 if (output_stream_regular)
2258 fl = opt.output_document;
2261 fl = *hstat.local_file;
2265 /* End of time-stamping section. */
2269 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2270 escnonprint (hstat.error));
2275 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2277 if (hstat.len == hstat.contlen)
2281 logprintf (LOG_VERBOSE,
2282 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2284 number_to_static_string (hstat.len),
2285 number_to_static_string (hstat.contlen));
2286 logprintf (LOG_NONVERBOSE,
2287 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2289 number_to_static_string (hstat.len),
2290 number_to_static_string (hstat.contlen),
2294 total_downloaded_bytes += hstat.len;
2296 /* Remember that we downloaded the file for later ".orig" code. */
2297 if (*dt & ADDED_HTML_EXTENSION)
2298 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2300 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2302 free_hstat (&hstat);
2306 else if (hstat.res == 0) /* No read error */
2308 if (hstat.contlen == -1) /* We don't know how much we were supposed
2309 to get, so assume we succeeded. */
2313 logprintf (LOG_VERBOSE,
2314 _("%s (%s) - `%s' saved [%s]\n\n"),
2316 number_to_static_string (hstat.len));
2317 logprintf (LOG_NONVERBOSE,
2318 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2319 tms, u->url, number_to_static_string (hstat.len),
2323 total_downloaded_bytes += hstat.len;
2325 /* Remember that we downloaded the file for later ".orig" code. */
2326 if (*dt & ADDED_HTML_EXTENSION)
2327 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2329 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2331 free_hstat (&hstat);
2335 else if (hstat.len < hstat.contlen) /* meaning we lost the
2336 connection too soon */
2338 logprintf (LOG_VERBOSE,
2339 _("%s (%s) - Connection closed at byte %s. "),
2340 tms, tmrate, number_to_static_string (hstat.len));
2341 printwhat (count, opt.ntry);
2342 free_hstat (&hstat);
2345 else if (!opt.kill_longer) /* meaning we got more than expected */
2347 logprintf (LOG_VERBOSE,
2348 _("%s (%s) - `%s' saved [%s/%s])\n\n"),
2350 number_to_static_string (hstat.len),
2351 number_to_static_string (hstat.contlen));
2352 logprintf (LOG_NONVERBOSE,
2353 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2355 number_to_static_string (hstat.len),
2356 number_to_static_string (hstat.contlen),
2359 total_downloaded_bytes += hstat.len;
2361 /* Remember that we downloaded the file for later ".orig" code. */
2362 if (*dt & ADDED_HTML_EXTENSION)
2363 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2365 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2367 free_hstat (&hstat);
2371 else /* the same, but not accepted */
2373 logprintf (LOG_VERBOSE,
2374 _("%s (%s) - Connection closed at byte %s/%s. "),
2376 number_to_static_string (hstat.len),
2377 number_to_static_string (hstat.contlen));
2378 printwhat (count, opt.ntry);
2379 free_hstat (&hstat);
2383 else /* now hstat.res can only be -1 */
2385 if (hstat.contlen == -1)
2387 logprintf (LOG_VERBOSE,
2388 _("%s (%s) - Read error at byte %s (%s)."),
2389 tms, tmrate, number_to_static_string (hstat.len),
2391 printwhat (count, opt.ntry);
2392 free_hstat (&hstat);
2395 else /* hstat.res == -1 and contlen is given */
2397 logprintf (LOG_VERBOSE,
2398 _("%s (%s) - Read error at byte %s/%s (%s). "),
2400 number_to_static_string (hstat.len),
2401 number_to_static_string (hstat.contlen),
2403 printwhat (count, opt.ntry);
2404 free_hstat (&hstat);
2411 while (!opt.ntry || (count < opt.ntry));
2415 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2416 than local timezone.
2418 mktime is similar but assumes struct tm, also known as the
2419 "broken-down" form of time, is in local time zone. mktime_from_utc
2420 uses mktime to make the conversion understanding that an offset
2421 will be introduced by the local time assumption.
2423 mktime_from_utc then measures the introduced offset by applying
2424 gmtime to the initial result and applying mktime to the resulting
2425 "broken-down" form. The difference between the two mktime results
2426 is the measured offset which is then subtracted from the initial
2427 mktime result to yield a calendar time which is the value returned.
2429 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2430 consistent offset (the non DST offset) since tm and tm+o might be
2431 on opposite sides of a DST change.
2433 Some implementations of mktime return -1 for the nonexistent
2434 localtime hour at the beginning of DST. In this event, use
2435 mktime(tm - 1hr) + 3600.
2439 gmtime(t+o) --> tm+o
2440 mktime(tm+o) --> t+2o
2441 t+o - (t+2o - t+o) = t
2443 Note that glibc contains a function of the same purpose named
2444 `timegm' (reverse of gmtime). But obviously, it is not universally
2445 available, and unfortunately it is not straightforwardly
2446 extractable for use here. Perhaps configure should detect timegm
2447 and use it where available.
2449 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2450 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2451 Further improved by Roger with assistance from Edward J. Sabol
2452 based on input by Jamie Zawinski. */
2455 mktime_from_utc (struct tm *t)
2466 return -1; /* can't deal with output from strptime */
2477 return -1; /* can't deal with output from gmtime */
2480 return (tl - (tb - tl));
2483 /* Check whether the result of strptime() indicates success.
2484 strptime() returns the pointer to how far it got to in the string.
2485 The processing has been successful if the string is at `GMT' or
2486 `+X', or at the end of the string.
2488 In extended regexp parlance, the function returns 1 if P matches
2489 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2490 can return) is considered a failure and 0 is returned. */
2492 check_end (const char *p)
2496 while (ISSPACE (*p))
2499 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2500 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2506 /* Convert the textual specification of time in TIME_STRING to the
2507 number of seconds since the Epoch.
2509 TIME_STRING can be in any of the three formats RFC2068 allows the
2510 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2511 Timezones are ignored, and should be GMT.
2513 Return the computed time_t representation, or -1 if the conversion
2516 This function uses strptime with various string formats for parsing
2517 TIME_STRING. This results in a parser that is not as lenient in
2518 interpreting TIME_STRING as I would like it to be. Being based on
2519 strptime, it always allows shortened months, one-digit days, etc.,
2520 but due to the multitude of formats in which time can be
2521 represented, an ideal HTTP time parser would be even more
2522 forgiving. It should completely ignore things like week days and
2523 concentrate only on the various forms of representing years,
2524 months, days, hours, minutes, and seconds. For example, it would
2525 be nice if it accepted ISO 8601 out of the box.
2527 I've investigated free and PD code for this purpose, but none was
2528 usable. getdate was big and unwieldy, and had potential copyright
2529 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2530 distributed with phttpd, is excellent, but we cannot use it because
2531 it is not assigned to the FSF. So I stuck it with strptime. */
2534 http_atotm (const char *time_string)
2536 /* NOTE: Solaris strptime man page claims that %n and %t match white
2537 space, but that's not universally available. Instead, we simply
2538 use ` ' to mean "skip all WS", which works under all strptime
2539 implementations I've tested. */
2541 static const char *time_formats[] = {
2542 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2543 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2544 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2545 (google.com uses this for their cookies.) */
2546 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2552 /* According to Roger Beeman, we need to initialize tm_isdst, since
2553 strptime won't do it. */
2556 /* Note that under foreign locales Solaris strptime() fails to
2557 recognize English dates, which renders this function useless. We
2558 solve this by being careful not to affect LC_TIME when
2559 initializing locale.
2561 Another solution would be to temporarily set locale to C, invoke
2562 strptime(), and restore it back. This is slow and dirty,
2563 however, and locale support other than LC_MESSAGES can mess other
2564 things, so I rather chose to stick with just setting LC_MESSAGES.
2566 GNU strptime does not have this problem because it recognizes
2567 both international and local dates. */
2569 for (i = 0; i < countof (time_formats); i++)
2570 if (check_end (strptime (time_string, time_formats[i], &t)))
2571 return mktime_from_utc (&t);
2573 /* All formats have failed. */
2577 /* Authorization support: We support two authorization schemes:
2579 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2581 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2582 consisting of answering to the server's challenge with the proper
2585 /* Create the authentication header contents for the `Basic' scheme.
2586 This is done by encoding the string `USER:PASS' in base64 and
2587 prepending `HEADER: Basic ' to it. */
2589 basic_authentication_encode (const char *user, const char *passwd)
2592 int len1 = strlen (user) + 1 + strlen (passwd);
2593 int len2 = BASE64_LENGTH (len1);
2595 t1 = (char *)alloca (len1 + 1);
2596 sprintf (t1, "%s:%s", user, passwd);
2598 t2 = (char *)alloca (len2 + 1);
2599 base64_encode (t1, t2, len1);
2601 return concat_strings ("Basic ", t2, (char *) 0);
2604 #define SKIP_WS(x) do { \
2605 while (ISSPACE (*(x))) \
2609 #ifdef ENABLE_DIGEST
2610 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2611 of a field in such a header. If the field is the one specified by
2612 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2613 digest authorization code), extract its value in the (char*)
2614 variable pointed by RET. Returns negative on a malformed header,
2615 or number of bytes that have been parsed by this call. */
2617 extract_header_attr (const char *au, const char *attr_name, char **ret)
2620 const char *cp = au;
2622 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2624 cp += strlen (attr_name);
2637 for (ep = cp; *ep && *ep != '\"'; ep++)
2642 *ret = strdupdelim (cp, ep);
2649 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2650 an array of 16 bytes containing the hash keys, and BUF should be a
2651 buffer of 33 writable characters (32 for hex digits plus one for
2652 zero termination). */
2654 dump_hash (unsigned char *buf, const unsigned char *hash)
2658 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2660 *buf++ = XNUM_TO_digit (*hash >> 4);
2661 *buf++ = XNUM_TO_digit (*hash & 0xf);
2666 /* Take the line apart to find the challenge, and compose a digest
2667 authorization header. See RFC2069 section 2.1.2. */
2669 digest_authentication_encode (const char *au, const char *user,
2670 const char *passwd, const char *method,
2673 static char *realm, *opaque, *nonce;
2678 { "realm", &realm },
2679 { "opaque", &opaque },
2684 realm = opaque = nonce = NULL;
2686 au += 6; /* skip over `Digest' */
2692 for (i = 0; i < countof (options); i++)
2694 int skip = extract_header_attr (au, options[i].name,
2695 options[i].variable);
2699 xfree_null (opaque);
2709 if (i == countof (options))
2711 while (*au && *au != '=')
2719 while (*au && *au != '\"')
2726 while (*au && *au != ',')
2731 if (!realm || !nonce || !user || !passwd || !path || !method)
2734 xfree_null (opaque);
2739 /* Calculate the digest value. */
2741 ALLOCA_MD5_CONTEXT (ctx);
2742 unsigned char hash[MD5_HASHLEN];
2743 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2744 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2746 /* A1BUF = H(user ":" realm ":" password) */
2748 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2749 gen_md5_update ((unsigned char *)":", 1, ctx);
2750 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2751 gen_md5_update ((unsigned char *)":", 1, ctx);
2752 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2753 gen_md5_finish (ctx, hash);
2754 dump_hash (a1buf, hash);
2756 /* A2BUF = H(method ":" path) */
2758 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2759 gen_md5_update ((unsigned char *)":", 1, ctx);
2760 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2761 gen_md5_finish (ctx, hash);
2762 dump_hash (a2buf, hash);
2764 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2766 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2767 gen_md5_update ((unsigned char *)":", 1, ctx);
2768 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2769 gen_md5_update ((unsigned char *)":", 1, ctx);
2770 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2771 gen_md5_finish (ctx, hash);
2772 dump_hash (response_digest, hash);
2774 res = (char*) xmalloc (strlen (user)
2779 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2780 + (opaque ? strlen (opaque) : 0)
2782 sprintf (res, "Digest \
2783 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2784 user, realm, nonce, path, response_digest);
2787 char *p = res + strlen (res);
2788 strcat (p, ", opaque=\"");
2795 #endif /* ENABLE_DIGEST */
2798 #define BEGINS_WITH(line, string_constant) \
2799 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2800 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2801 || !line[sizeof (string_constant) - 1]))
2804 known_authentication_scheme_p (const char *au)
2806 return BEGINS_WITH (au, "Basic")
2807 #ifdef ENABLE_DIGEST
2808 || BEGINS_WITH (au, "Digest")
2811 || BEGINS_WITH (au, "NTLM")
2818 /* Create the HTTP authorization request header. When the
2819 `WWW-Authenticate' response header is seen, according to the
2820 authorization scheme specified in that header (`Basic' and `Digest'
2821 are supported by the current implementation), produce an
2822 appropriate HTTP authorization request header. */
2824 create_authorization_line (const char *au, const char *user,
2825 const char *passwd, const char *method,
2828 if (0 == strncasecmp (au, "Basic", 5))
2829 return basic_authentication_encode (user, passwd);
2830 #ifdef ENABLE_DIGEST
2831 if (0 == strncasecmp (au, "Digest", 6))
2832 return digest_authentication_encode (au, user, passwd, method, path);
2835 if (0 == strncasecmp (au, "NTLM", 4))
2837 int ok = ntlm_input (&pconn.ntlm, au);
2840 /* #### we shouldn't ignore the OK that ntlm_output returns. */
2841 return ntlm_output (&pconn.ntlm, user, passwd, &ok);
2850 if (wget_cookie_jar)
2851 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2857 xfree_null (pconn.host);
2858 if (wget_cookie_jar)
2859 cookie_jar_delete (wget_cookie_jar);