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. */
55 # include "http-ntlm.h"
63 extern char *version_string;
66 # define MIN(x, y) ((x) > (y) ? (y) : (x))
70 static bool cookies_loaded_p;
71 static struct cookie_jar *wget_cookie_jar;
73 #define TEXTHTML_S "text/html"
74 #define TEXTXHTML_S "application/xhtml+xml"
76 /* Some status code validation macros: */
77 #define H_20X(x) (((x) >= 200) && ((x) < 300))
78 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
79 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
80 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
81 || (x) == HTTP_STATUS_SEE_OTHER \
82 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
84 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
86 #define HTTP_STATUS_OK 200
87 #define HTTP_STATUS_CREATED 201
88 #define HTTP_STATUS_ACCEPTED 202
89 #define HTTP_STATUS_NO_CONTENT 204
90 #define HTTP_STATUS_PARTIAL_CONTENTS 206
92 /* Redirection 3xx. */
93 #define HTTP_STATUS_MULTIPLE_CHOICES 300
94 #define HTTP_STATUS_MOVED_PERMANENTLY 301
95 #define HTTP_STATUS_MOVED_TEMPORARILY 302
96 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
97 #define HTTP_STATUS_NOT_MODIFIED 304
98 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
100 /* Client error 4xx. */
101 #define HTTP_STATUS_BAD_REQUEST 400
102 #define HTTP_STATUS_UNAUTHORIZED 401
103 #define HTTP_STATUS_FORBIDDEN 403
104 #define HTTP_STATUS_NOT_FOUND 404
105 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
107 /* Server errors 5xx. */
108 #define HTTP_STATUS_INTERNAL 500
109 #define HTTP_STATUS_NOT_IMPLEMENTED 501
110 #define HTTP_STATUS_BAD_GATEWAY 502
111 #define HTTP_STATUS_UNAVAILABLE 503
114 rel_none, rel_name, rel_value, rel_both
121 struct request_header {
123 enum rp release_policy;
125 int hcount, hcapacity;
128 /* Create a new, empty request. At least request_set_method must be
129 called before the request can be used. */
131 static struct request *
134 struct request *req = xnew0 (struct request);
136 req->headers = xnew_array (struct request_header, req->hcapacity);
140 /* Set the request's method and its arguments. METH should be a
141 literal string (or it should outlive the request) because it will
142 not be freed. ARG will be freed by request_free. */
145 request_set_method (struct request *req, const char *meth, char *arg)
151 /* Return the method string passed with the last call to
152 request_set_method. */
155 request_method (const struct request *req)
160 /* Free one header according to the release policy specified with
161 request_set_header. */
164 release_header (struct request_header *hdr)
166 switch (hdr->release_policy)
183 /* Set the request named NAME to VALUE. Specifically, this means that
184 a "NAME: VALUE\r\n" header line will be used in the request. If a
185 header with the same name previously existed in the request, its
186 value will be replaced by this one. A NULL value means do nothing.
188 RELEASE_POLICY determines whether NAME and VALUE should be released
189 (freed) with request_free. Allowed values are:
191 - rel_none - don't free NAME or VALUE
192 - rel_name - free NAME when done
193 - rel_value - free VALUE when done
194 - rel_both - free both NAME and VALUE when done
196 Setting release policy is useful when arguments come from different
197 sources. For example:
199 // Don't free literal strings!
200 request_set_header (req, "Pragma", "no-cache", rel_none);
202 // Don't free a global variable, we'll need it later.
203 request_set_header (req, "Referer", opt.referer, rel_none);
205 // Value freshly allocated, free it when done.
206 request_set_header (req, "Range",
207 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
212 request_set_header (struct request *req, char *name, char *value,
213 enum rp release_policy)
215 struct request_header *hdr;
220 /* A NULL value is a no-op; if freeing the name is requested,
221 free it now to avoid leaks. */
222 if (release_policy == rel_name || release_policy == rel_both)
227 for (i = 0; i < req->hcount; i++)
229 hdr = &req->headers[i];
230 if (0 == strcasecmp (name, hdr->name))
232 /* Replace existing header. */
233 release_header (hdr);
236 hdr->release_policy = release_policy;
241 /* Install new header. */
243 if (req->hcount >= req->hcapacity)
245 req->hcapacity <<= 1;
246 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
248 hdr = &req->headers[req->hcount++];
251 hdr->release_policy = release_policy;
254 /* Like request_set_header, but sets the whole header line, as
255 provided by the user using the `--header' option. For example,
256 request_set_user_header (req, "Foo: bar") works just like
257 request_set_header (req, "Foo", "bar"). */
260 request_set_user_header (struct request *req, const char *header)
263 const char *p = strchr (header, ':');
266 BOUNDED_TO_ALLOCA (header, p, name);
270 request_set_header (req, xstrdup (name), (char *) p, rel_name);
273 /* Remove the header with specified name from REQ. Returns true if
274 the header was actually removed, false otherwise. */
277 request_remove_header (struct request *req, char *name)
280 for (i = 0; i < req->hcount; i++)
282 struct request_header *hdr = &req->headers[i];
283 if (0 == strcasecmp (name, hdr->name))
285 release_header (hdr);
286 /* Move the remaining headers by one. */
287 if (i < req->hcount - 1)
288 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
296 #define APPEND(p, str) do { \
297 int A_len = strlen (str); \
298 memcpy (p, str, A_len); \
302 /* Construct the request and write it to FD using fd_write. */
305 request_send (const struct request *req, int fd)
307 char *request_string, *p;
308 int i, size, write_error;
310 /* Count the request size. */
313 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
314 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
316 for (i = 0; i < req->hcount; i++)
318 struct request_header *hdr = &req->headers[i];
319 /* NAME ": " VALUE "\r\n" */
320 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
326 p = request_string = alloca_array (char, size);
328 /* Generate the request. */
330 APPEND (p, req->method); *p++ = ' ';
331 APPEND (p, req->arg); *p++ = ' ';
332 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
334 for (i = 0; i < req->hcount; i++)
336 struct request_header *hdr = &req->headers[i];
337 APPEND (p, hdr->name);
338 *p++ = ':', *p++ = ' ';
339 APPEND (p, hdr->value);
340 *p++ = '\r', *p++ = '\n';
343 *p++ = '\r', *p++ = '\n', *p++ = '\0';
344 assert (p - request_string == size);
348 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
350 /* Send the request to the server. */
352 write_error = fd_write (fd, request_string, size - 1, -1);
354 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
359 /* Release the resources used by REQ. */
362 request_free (struct request *req)
365 xfree_null (req->arg);
366 for (i = 0; i < req->hcount; i++)
367 release_header (&req->headers[i]);
368 xfree_null (req->headers);
372 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
373 PROMISED_SIZE bytes are sent over the wire -- if the file is
374 longer, read only that much; if the file is shorter, report an error. */
377 post_file (int sock, const char *file_name, wgint promised_size)
379 static char chunk[8192];
384 DEBUGP (("[writing POST file %s ... ", file_name));
386 fp = fopen (file_name, "rb");
389 while (!feof (fp) && written < promised_size)
392 int length = fread (chunk, 1, sizeof (chunk), fp);
395 towrite = MIN (promised_size - written, length);
396 write_error = fd_write (sock, chunk, towrite, -1);
406 /* If we've written less than was promised, report a (probably
407 nonsensical) error rather than break the promise. */
408 if (written < promised_size)
414 assert (written == promised_size);
415 DEBUGP (("done]\n"));
419 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
420 If so, return the pointer to the position after the line, otherwise
421 return NULL. This is used as callback to fd_read_hunk. The data
422 between START and PEEKED has been read and cannot be "unread"; the
423 data after PEEKED has only been peeked. */
426 response_head_terminator (const char *start, const char *peeked, int peeklen)
430 /* If at first peek, verify whether HUNK starts with "HTTP". If
431 not, this is a HTTP/0.9 request and we must bail out without
433 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
436 /* Look for "\n[\r]\n", and return the following position if found.
437 Start two chars before the current to cover the possibility that
438 part of the terminator (e.g. "\n\r") arrived in the previous
440 p = peeked - start < 2 ? start : peeked - 2;
441 end = peeked + peeklen;
443 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
444 for (; p < end - 2; p++)
447 if (p[1] == '\r' && p[2] == '\n')
449 else if (p[1] == '\n')
452 /* p==end-2: check for \n\n directly preceding END. */
453 if (p[0] == '\n' && p[1] == '\n')
459 /* The maximum size of a single HTTP response we care to read. Rather
460 than being a limit of the reader implementation, this limit
461 prevents Wget from slurping all available memory upon encountering
462 malicious or buggy server output, thus protecting the user. Define
463 it to 0 to remove the limit. */
465 #define HTTP_RESPONSE_MAX_SIZE 65536
467 /* Read the HTTP request head from FD and return it. The error
468 conditions are the same as with fd_read_hunk.
470 To support HTTP/0.9 responses, this function tries to make sure
471 that the data begins with "HTTP". If this is not the case, no data
472 is read and an empty request is returned, so that the remaining
473 data can be treated as body. */
476 read_http_response_head (int fd)
478 return fd_read_hunk (fd, response_head_terminator, 512,
479 HTTP_RESPONSE_MAX_SIZE);
483 /* The response data. */
486 /* The array of pointers that indicate where each header starts.
487 For example, given this HTTP response:
494 The headers are located like this:
496 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
498 headers[0] headers[1] headers[2] headers[3]
500 I.e. headers[0] points to the beginning of the request,
501 headers[1] points to the end of the first header and the
502 beginning of the second one, etc. */
504 const char **headers;
507 /* Create a new response object from the text of the HTTP response,
508 available in HEAD. That text is automatically split into
509 constituent header lines for fast retrieval using
512 static struct response *
513 resp_new (const char *head)
518 struct response *resp = xnew0 (struct response);
523 /* Empty head means that we're dealing with a headerless
524 (HTTP/0.9) response. In that case, don't set HEADERS at
529 /* Split HEAD into header lines, so that resp_header_* functions
530 don't need to do this over and over again. */
536 DO_REALLOC (resp->headers, size, count + 1, const char *);
537 resp->headers[count++] = hdr;
539 /* Break upon encountering an empty line. */
540 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
543 /* Find the end of HDR, including continuations. */
546 const char *end = strchr (hdr, '\n');
552 while (*hdr == ' ' || *hdr == '\t');
554 DO_REALLOC (resp->headers, size, count + 1, const char *);
555 resp->headers[count] = NULL;
560 /* Locate the header named NAME in the request data, starting with
561 position START. This allows the code to loop through the request
562 data, filtering for all requests of a given name. Returns the
563 found position, or -1 for failure. The code that uses this
564 function typically looks like this:
566 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
567 ... do something with header ...
569 If you only care about one header, use resp_header_get instead of
573 resp_header_locate (const struct response *resp, const char *name, int start,
574 const char **begptr, const char **endptr)
577 const char **headers = resp->headers;
580 if (!headers || !headers[1])
583 name_len = strlen (name);
589 for (; headers[i + 1]; i++)
591 const char *b = headers[i];
592 const char *e = headers[i + 1];
594 && b[name_len] == ':'
595 && 0 == strncasecmp (b, name, name_len))
598 while (b < e && ISSPACE (*b))
600 while (b < e && ISSPACE (e[-1]))
610 /* Find and retrieve the header named NAME in the request data. If
611 found, set *BEGPTR to its starting, and *ENDPTR to its ending
612 position, and return true. Otherwise return false.
614 This function is used as a building block for resp_header_copy
615 and resp_header_strdup. */
618 resp_header_get (const struct response *resp, const char *name,
619 const char **begptr, const char **endptr)
621 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
625 /* Copy the response header named NAME to buffer BUF, no longer than
626 BUFSIZE (BUFSIZE includes the terminating 0). If the header
627 exists, true is returned, false otherwise. If there should be no
628 limit on the size of the header, use resp_header_strdup instead.
630 If BUFSIZE is 0, no data is copied, but the boolean indication of
631 whether the header is present is still returned. */
634 resp_header_copy (const struct response *resp, const char *name,
635 char *buf, int bufsize)
638 if (!resp_header_get (resp, name, &b, &e))
642 int len = MIN (e - b, bufsize - 1);
643 memcpy (buf, b, len);
649 /* Return the value of header named NAME in RESP, allocated with
650 malloc. If such a header does not exist in RESP, return NULL. */
653 resp_header_strdup (const struct response *resp, const char *name)
656 if (!resp_header_get (resp, name, &b, &e))
658 return strdupdelim (b, e);
661 /* Parse the HTTP status line, which is of format:
663 HTTP-Version SP Status-Code SP Reason-Phrase
665 The function returns the status-code, or -1 if the status line
666 appears malformed. The pointer to "reason-phrase" message is
667 returned in *MESSAGE. */
670 resp_status (const struct response *resp, char **message)
677 /* For a HTTP/0.9 response, assume status 200. */
679 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
683 p = resp->headers[0];
684 end = resp->headers[1];
690 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
694 /* Match the HTTP version. This is optional because Gnutella
695 servers have been reported to not specify HTTP version. */
696 if (p < end && *p == '/')
699 while (p < end && ISDIGIT (*p))
701 if (p < end && *p == '.')
703 while (p < end && ISDIGIT (*p))
707 while (p < end && ISSPACE (*p))
709 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
712 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
717 while (p < end && ISSPACE (*p))
719 while (p < end && ISSPACE (end[-1]))
721 *message = strdupdelim (p, end);
727 /* Release the resources used by RESP. */
730 resp_free (struct response *resp)
732 xfree_null (resp->headers);
736 /* Print the server response, line by line, omitting the trailing CRLF
737 from individual header lines, and prefixed with PREFIX. */
740 print_server_response (const struct response *resp, const char *prefix)
745 for (i = 0; resp->headers[i + 1]; i++)
747 const char *b = resp->headers[i];
748 const char *e = resp->headers[i + 1];
750 if (b < e && e[-1] == '\n')
752 if (b < e && e[-1] == '\r')
754 /* This is safe even on printfs with broken handling of "%.<n>s"
755 because resp->headers ends with \0. */
756 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
760 /* Parse the `Content-Range' header and extract the information it
761 contains. Returns true if successful, false otherwise. */
763 parse_content_range (const char *hdr, wgint *first_byte_ptr,
764 wgint *last_byte_ptr, wgint *entity_length_ptr)
768 /* Ancient versions of Netscape proxy server, presumably predating
769 rfc2068, sent out `Content-Range' without the "bytes"
771 if (0 == strncasecmp (hdr, "bytes", 5))
774 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
778 while (ISSPACE (*hdr))
785 for (num = 0; ISDIGIT (*hdr); hdr++)
786 num = 10 * num + (*hdr - '0');
787 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
789 *first_byte_ptr = num;
791 for (num = 0; ISDIGIT (*hdr); hdr++)
792 num = 10 * num + (*hdr - '0');
793 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
795 *last_byte_ptr = num;
797 for (num = 0; ISDIGIT (*hdr); hdr++)
798 num = 10 * num + (*hdr - '0');
799 *entity_length_ptr = num;
803 /* Read the body of the request, but don't store it anywhere and don't
804 display a progress gauge. This is useful for reading the bodies of
805 administrative responses to which we will soon issue another
806 request. The response is not useful to the user, but reading it
807 allows us to continue using the same connection to the server.
809 If reading fails, false is returned, true otherwise. In debug
810 mode, the body is displayed for debugging purposes. */
813 skip_short_body (int fd, wgint contlen)
816 SKIP_SIZE = 512, /* size of the download buffer */
817 SKIP_THRESHOLD = 4096 /* the largest size we read */
819 char dlbuf[SKIP_SIZE + 1];
820 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
822 /* We shouldn't get here with unknown contlen. (This will change
823 with HTTP/1.1, which supports "chunked" transfer.) */
824 assert (contlen != -1);
826 /* If the body is too large, it makes more sense to simply close the
827 connection than to try to read the body. */
828 if (contlen > SKIP_THRESHOLD)
831 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
835 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
838 /* Don't normally report the error since this is an
839 optimization that should be invisible to the user. */
840 DEBUGP (("] aborting (%s).\n",
841 ret < 0 ? strerror (errno) : "EOF received"));
845 /* Safe even if %.*s bogusly expects terminating \0 because
846 we've zero-terminated dlbuf above. */
847 DEBUGP (("%.*s", ret, dlbuf));
850 DEBUGP (("] done.\n"));
854 /* Persistent connections. Currently, we cache the most recently used
855 connection as persistent, provided that the HTTP server agrees to
856 make it such. The persistence data is stored in the variables
857 below. Ideally, it should be possible to cache an arbitrary fixed
858 number of these connections. */
860 /* Whether a persistent connection is active. */
861 static bool pconn_active;
864 /* The socket of the connection. */
867 /* Host and port of the currently active persistent connection. */
871 /* Whether a ssl handshake has occoured on this connection. */
874 /* Whether the connection was authorized. This is only done by
875 NTLM, which authorizes *connections* rather than individual
876 requests. (That practice is peculiar for HTTP, but it is a
877 useful optimization.) */
881 /* NTLM data of the current connection. */
882 struct ntlmdata ntlm;
886 /* Mark the persistent connection as invalid and free the resources it
887 uses. This is used by the CLOSE_* macros after they forcefully
888 close a registered persistent connection. */
891 invalidate_persistent (void)
893 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
894 pconn_active = false;
895 fd_close (pconn.socket);
900 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
901 persistent. This will enable someone to use the same connection
902 later. In the context of HTTP, this must be called only AFTER the
903 response has been received and the server has promised that the
904 connection will remain alive.
906 If a previous connection was persistent, it is closed. */
909 register_persistent (const char *host, int port, int fd, bool ssl)
913 if (pconn.socket == fd)
915 /* The connection FD is already registered. */
920 /* The old persistent connection is still active; close it
921 first. This situation arises whenever a persistent
922 connection exists, but we then connect to a different
923 host, and try to register a persistent connection to that
925 invalidate_persistent ();
931 pconn.host = xstrdup (host);
934 pconn.authorized = false;
936 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
939 /* Return true if a persistent connection is available for connecting
943 persistent_available_p (const char *host, int port, bool ssl,
944 bool *host_lookup_failed)
946 /* First, check whether a persistent connection is active at all. */
950 /* If we want SSL and the last connection wasn't or vice versa,
951 don't use it. Checking for host and port is not enough because
952 HTTP and HTTPS can apparently coexist on the same port. */
953 if (ssl != pconn.ssl)
956 /* If we're not connecting to the same port, we're not interested. */
957 if (port != pconn.port)
960 /* If the host is the same, we're in business. If not, there is
961 still hope -- read below. */
962 if (0 != strcasecmp (host, pconn.host))
964 /* Check if pconn.socket is talking to HOST under another name.
965 This happens often when both sites are virtual hosts
966 distinguished only by name and served by the same network
967 interface, and hence the same web server (possibly set up by
968 the ISP and serving many different web sites). This
969 admittedly unconventional optimization does not contradict
970 HTTP and works well with popular server software. */
974 struct address_list *al;
977 /* Don't try to talk to two different SSL sites over the same
978 secure connection! (Besides, it's not clear that
979 name-based virtual hosting is even possible with SSL.) */
982 /* If pconn.socket's peer is one of the IP addresses HOST
983 resolves to, pconn.socket is for all intents and purposes
984 already talking to HOST. */
986 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
988 /* Can't get the peer's address -- something must be very
989 wrong with the connection. */
990 invalidate_persistent ();
993 al = lookup_host (host, 0);
996 *host_lookup_failed = true;
1000 found = address_list_contains (al, &ip);
1001 address_list_release (al);
1006 /* The persistent connection's peer address was found among the
1007 addresses HOST resolved to; therefore, pconn.sock is in fact
1008 already talking to HOST -- no need to reconnect. */
1011 /* Finally, check whether the connection is still open. This is
1012 important because most server implement a liberal (short) timeout
1013 on persistent connections. Wget can of course always reconnect
1014 if the connection doesn't work out, but it's nicer to know in
1015 advance. This test is a logical followup of the first test, but
1016 is "expensive" and therefore placed at the end of the list. */
1018 if (!test_socket_open (pconn.socket))
1020 /* Oops, the socket is no longer open. Now that we know that,
1021 let's invalidate the persistent connection before returning
1023 invalidate_persistent ();
1030 /* The idea behind these two CLOSE macros is to distinguish between
1031 two cases: one when the job we've been doing is finished, and we
1032 want to close the connection and leave, and two when something is
1033 seriously wrong and we're closing the connection as part of
1036 In case of keep_alive, CLOSE_FINISH should leave the connection
1037 open, while CLOSE_INVALIDATE should still close it.
1039 Note that the semantics of the flag `keep_alive' is "this
1040 connection *will* be reused (the server has promised not to close
1041 the connection once we're done)", while the semantics of
1042 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1043 active, registered connection". */
1045 #define CLOSE_FINISH(fd) do { \
1048 if (pconn_active && (fd) == pconn.socket) \
1049 invalidate_persistent (); \
1058 #define CLOSE_INVALIDATE(fd) do { \
1059 if (pconn_active && (fd) == pconn.socket) \
1060 invalidate_persistent (); \
1068 wgint len; /* received length */
1069 wgint contlen; /* expected length */
1070 wgint restval; /* the restart value */
1071 int res; /* the result of last read */
1072 char *newloc; /* new location (redirection) */
1073 char *remote_time; /* remote time-stamp string */
1074 char *error; /* textual HTTP error */
1075 int statcode; /* status code */
1076 wgint rd_size; /* amount of data read from socket */
1077 double dltime; /* time it took to download the data */
1078 const char *referer; /* value of the referer header. */
1079 char **local_file; /* local file. */
1083 free_hstat (struct http_stat *hs)
1085 xfree_null (hs->newloc);
1086 xfree_null (hs->remote_time);
1087 xfree_null (hs->error);
1089 /* Guard against being called twice. */
1091 hs->remote_time = NULL;
1095 static char *create_authorization_line (const char *, const char *,
1096 const char *, const char *,
1097 const char *, bool *);
1098 static char *basic_authentication_encode (const char *, const char *);
1099 static bool known_authentication_scheme_p (const char *, const char *);
1101 #define BEGINS_WITH(line, string_constant) \
1102 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1103 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1104 || !line[sizeof (string_constant) - 1]))
1106 #define SET_USER_AGENT(req) do { \
1107 if (!opt.useragent) \
1108 request_set_header (req, "User-Agent", \
1109 aprintf ("Wget/%s", version_string), rel_value); \
1110 else if (*opt.useragent) \
1111 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1114 /* The flags that allow clobbering the file (opening with "wb").
1115 Defined here to avoid repetition later. #### This will require
1117 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1118 || opt.dirstruct || opt.output_document)
1120 /* Retrieve a document through HTTP protocol. It recognizes status
1121 code, and correctly handles redirections. It closes the network
1122 socket. If it receives an error from the functions below it, it
1123 will print it if there is enough information to do so (almost
1124 always), returning the error to the caller (i.e. http_loop).
1126 Various HTTP parameters are stored to hs.
1128 If PROXY is non-NULL, the connection will be made to the proxy
1129 server, and u->url will be requested. */
1131 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1133 struct request *req;
1136 char *user, *passwd;
1140 wgint contlen, contrange;
1147 /* Set to 1 when the authorization has failed permanently and should
1148 not be tried again. */
1149 bool auth_finished = false;
1151 /* Whether NTLM authentication is used for this request. */
1152 bool ntlm_seen = false;
1154 /* Whether our connection to the remote host is through SSL. */
1155 bool using_ssl = false;
1157 /* Whether a HEAD request will be issued (as opposed to GET or
1159 bool head_only = !!(*dt & HEAD_ONLY);
1162 struct response *resp;
1166 /* Whether this connection will be kept alive after the HTTP request
1170 /* Whether keep-alive should be inhibited.
1172 RFC 2068 requests that 1.0 clients not send keep-alive requests
1173 to proxies. This is because many 1.0 proxies do not interpret
1174 the Connection header and transfer it to the remote server,
1175 causing it to not close the connection and leave both the proxy
1176 and the client hanging. */
1177 bool inhibit_keep_alive =
1178 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1180 /* Headers sent when using POST. */
1181 wgint post_data_size = 0;
1183 bool host_lookup_failed = false;
1186 if (u->scheme == SCHEME_HTTPS)
1188 /* Initialize the SSL context. After this has once been done,
1189 it becomes a no-op. */
1192 scheme_disable (SCHEME_HTTPS);
1193 logprintf (LOG_NOTQUIET,
1194 _("Disabling SSL due to encountered errors.\n"));
1195 return SSLINITFAILED;
1198 #endif /* HAVE_SSL */
1201 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1202 know the local filename so we can save to it. */
1203 assert (*hs->local_file != NULL);
1205 /* Initialize certain elements of struct http_stat. */
1210 hs->remote_time = NULL;
1215 /* Prepare the request to send. */
1217 req = request_new ();
1220 const char *meth = "GET";
1223 else if (opt.post_file_name || opt.post_data)
1225 /* Use the full path, i.e. one that includes the leading slash and
1226 the query string. E.g. if u->path is "foo/bar" and u->query is
1227 "param=value", full_path will be "/foo/bar?param=value". */
1230 /* When using SSL over proxy, CONNECT establishes a direct
1231 connection to the HTTPS server. Therefore use the same
1232 argument as when talking to the server directly. */
1233 && u->scheme != SCHEME_HTTPS
1236 meth_arg = xstrdup (u->url);
1238 meth_arg = url_full_path (u);
1239 request_set_method (req, meth, meth_arg);
1242 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1243 if (*dt & SEND_NOCACHE)
1244 request_set_header (req, "Pragma", "no-cache", rel_none);
1246 request_set_header (req, "Range",
1247 aprintf ("bytes=%s-",
1248 number_to_static_string (hs->restval)),
1250 SET_USER_AGENT (req);
1251 request_set_header (req, "Accept", "*/*", rel_none);
1253 /* Find the username and password for authentication. */
1256 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1257 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1258 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1262 /* We have the username and the password, but haven't tried
1263 any authorization yet. Let's see if the "Basic" method
1264 works. If not, we'll come back here and construct a
1265 proper authorization method with the right challenges.
1267 If we didn't employ this kind of logic, every URL that
1268 requires authorization would have to be processed twice,
1269 which is very suboptimal and generates a bunch of false
1270 "unauthorized" errors in the server log.
1272 #### But this logic also has a serious problem when used
1273 with stronger authentications: we *first* transmit the
1274 username and the password in clear text, and *then* attempt a
1275 stronger authentication scheme. That cannot be right! We
1276 are only fortunate that almost everyone still uses the
1277 `Basic' scheme anyway.
1279 There should be an option to prevent this from happening, for
1280 those who use strong authentication schemes and value their
1282 request_set_header (req, "Authorization",
1283 basic_authentication_encode (user, passwd),
1290 char *proxy_user, *proxy_passwd;
1291 /* For normal username and password, URL components override
1292 command-line/wgetrc parameters. With proxy
1293 authentication, it's the reverse, because proxy URLs are
1294 normally the "permanent" ones, so command-line args
1295 should take precedence. */
1296 if (opt.proxy_user && opt.proxy_passwd)
1298 proxy_user = opt.proxy_user;
1299 proxy_passwd = opt.proxy_passwd;
1303 proxy_user = proxy->user;
1304 proxy_passwd = proxy->passwd;
1306 /* #### This does not appear right. Can't the proxy request,
1307 say, `Digest' authentication? */
1308 if (proxy_user && proxy_passwd)
1309 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1311 /* If we're using a proxy, we will be connecting to the proxy
1315 /* Proxy authorization over SSL is handled below. */
1317 if (u->scheme != SCHEME_HTTPS)
1319 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1322 /* Generate the Host header, HOST:PORT. Take into account that:
1324 - Broken server-side software often doesn't recognize the PORT
1325 argument, so we must generate "Host: www.server.com" instead of
1326 "Host: www.server.com:80" (and likewise for https port).
1328 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1329 becomes ambiguous and needs to be rewritten as "Host:
1330 [3ffe:8100:200:2::2]:1234". */
1332 /* Formats arranged for hfmt[add_port][add_squares]. */
1333 static const char *hfmt[][2] = {
1334 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1336 int add_port = u->port != scheme_default_port (u->scheme);
1337 int add_squares = strchr (u->host, ':') != NULL;
1338 request_set_header (req, "Host",
1339 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1343 if (!inhibit_keep_alive)
1344 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1347 request_set_header (req, "Cookie",
1348 cookie_header (wget_cookie_jar,
1349 u->host, u->port, u->path,
1351 u->scheme == SCHEME_HTTPS
1358 if (opt.post_data || opt.post_file_name)
1360 request_set_header (req, "Content-Type",
1361 "application/x-www-form-urlencoded", rel_none);
1363 post_data_size = strlen (opt.post_data);
1366 post_data_size = file_size (opt.post_file_name);
1367 if (post_data_size == -1)
1369 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1370 opt.post_file_name, strerror (errno));
1374 request_set_header (req, "Content-Length",
1375 xstrdup (number_to_static_string (post_data_size)),
1379 /* Add the user headers. */
1380 if (opt.user_headers)
1383 for (i = 0; opt.user_headers[i]; i++)
1384 request_set_user_header (req, opt.user_headers[i]);
1388 /* We need to come back here when the initial attempt to retrieve
1389 without authorization header fails. (Expected to happen at least
1390 for the Digest authorization scheme.) */
1394 /* Establish the connection. */
1396 if (!inhibit_keep_alive)
1398 /* Look for a persistent connection to target host, unless a
1399 proxy is used. The exception is when SSL is in use, in which
1400 case the proxy is nothing but a passthrough to the target
1401 host, registered as a connection to the latter. */
1402 struct url *relevant = conn;
1404 if (u->scheme == SCHEME_HTTPS)
1408 if (persistent_available_p (relevant->host, relevant->port,
1410 relevant->scheme == SCHEME_HTTPS,
1414 &host_lookup_failed))
1416 sock = pconn.socket;
1417 using_ssl = pconn.ssl;
1418 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1419 escnonprint (pconn.host), pconn.port);
1420 DEBUGP (("Reusing fd %d.\n", sock));
1421 if (pconn.authorized)
1422 /* If the connection is already authorized, the "Basic"
1423 authorization added by code above is unnecessary and
1425 request_remove_header (req, "Authorization");
1431 /* In its current implementation, persistent_available_p will
1432 look up conn->host in some cases. If that lookup failed, we
1433 don't need to bother with connect_to_host. */
1434 if (host_lookup_failed)
1440 sock = connect_to_host (conn->host, conn->port);
1449 return (retryable_socket_connect_error (errno)
1450 ? CONERROR : CONIMPOSSIBLE);
1454 if (proxy && u->scheme == SCHEME_HTTPS)
1456 /* When requesting SSL URLs through proxies, use the
1457 CONNECT method to request passthrough. */
1458 struct request *connreq = request_new ();
1459 request_set_method (connreq, "CONNECT",
1460 aprintf ("%s:%d", u->host, u->port));
1461 SET_USER_AGENT (connreq);
1464 request_set_header (connreq, "Proxy-Authorization",
1465 proxyauth, rel_value);
1466 /* Now that PROXYAUTH is part of the CONNECT request,
1467 zero it out so we don't send proxy authorization with
1468 the regular request below. */
1471 /* Examples in rfc2817 use the Host header in CONNECT
1472 requests. I don't see how that gains anything, given
1473 that the contents of Host would be exactly the same as
1474 the contents of CONNECT. */
1476 write_error = request_send (connreq, sock);
1477 request_free (connreq);
1478 if (write_error < 0)
1480 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1482 CLOSE_INVALIDATE (sock);
1486 head = read_http_response_head (sock);
1489 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1491 CLOSE_INVALIDATE (sock);
1500 DEBUGP (("proxy responded with: [%s]\n", head));
1502 resp = resp_new (head);
1503 statcode = resp_status (resp, &message);
1506 if (statcode != 200)
1509 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1510 message ? escnonprint (message) : "?");
1511 xfree_null (message);
1514 xfree_null (message);
1516 /* SOCK is now *really* connected to u->host, so update CONN
1517 to reflect this. That way register_persistent will
1518 register SOCK as being connected to u->host:u->port. */
1522 if (conn->scheme == SCHEME_HTTPS)
1524 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1531 #endif /* HAVE_SSL */
1534 /* Send the request to server. */
1535 write_error = request_send (req, sock);
1537 if (write_error >= 0)
1541 DEBUGP (("[POST data: %s]\n", opt.post_data));
1542 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1544 else if (opt.post_file_name && post_data_size != 0)
1545 write_error = post_file (sock, opt.post_file_name, post_data_size);
1548 if (write_error < 0)
1550 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1552 CLOSE_INVALIDATE (sock);
1556 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1557 proxy ? "Proxy" : "HTTP");
1562 head = read_http_response_head (sock);
1567 logputs (LOG_NOTQUIET, _("No data received.\n"));
1568 CLOSE_INVALIDATE (sock);
1574 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1576 CLOSE_INVALIDATE (sock);
1581 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1583 resp = resp_new (head);
1585 /* Check for status line. */
1587 statcode = resp_status (resp, &message);
1588 if (!opt.server_response)
1589 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1590 message ? escnonprint (message) : "");
1593 logprintf (LOG_VERBOSE, "\n");
1594 print_server_response (resp, " ");
1597 if (!opt.ignore_length
1598 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1602 parsed = str_to_wgint (hdrval, NULL, 10);
1603 if (parsed == WGINT_MAX && errno == ERANGE)
1605 #### If Content-Length is out of range, it most likely
1606 means that the file is larger than 2G and that we're
1607 compiled without LFS. In that case we should probably
1608 refuse to even attempt to download the file. */
1614 /* Check for keep-alive related responses. */
1615 if (!inhibit_keep_alive && contlen != -1)
1617 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1619 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1621 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1626 /* The server has promised that it will not close the connection
1627 when we're done. This means that we can register it. */
1628 register_persistent (conn->host, conn->port, sock, using_ssl);
1630 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1632 /* Authorization is required. */
1633 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1634 CLOSE_FINISH (sock);
1636 CLOSE_INVALIDATE (sock);
1637 pconn.authorized = false;
1638 if (!auth_finished && (user && passwd))
1640 /* IIS sends multiple copies of WWW-Authenticate, one with
1641 the value "negotiate", and other(s) with data. Loop over
1642 all the occurrences and pick the one we recognize. */
1644 const char *wabeg, *waend;
1645 char *www_authenticate = NULL;
1647 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1648 &wabeg, &waend)) != -1;
1650 if (known_authentication_scheme_p (wabeg, waend))
1652 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1656 if (!www_authenticate)
1657 /* If the authentication header is missing or
1658 unrecognized, there's no sense in retrying. */
1659 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1660 else if (BEGINS_WITH (www_authenticate, "Basic"))
1661 /* If the authentication scheme is "Basic", which we send
1662 by default, there's no sense in retrying either. (This
1663 should be changed when we stop sending "Basic" data by
1669 pth = url_full_path (u);
1670 request_set_header (req, "Authorization",
1671 create_authorization_line (www_authenticate,
1673 request_method (req),
1677 if (BEGINS_WITH (www_authenticate, "NTLM"))
1680 goto retry_with_auth;
1683 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1687 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1689 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1691 pconn.authorized = true;
1695 hs->statcode = statcode;
1697 hs->error = xstrdup (_("Malformed status line"));
1699 hs->error = xstrdup (_("(no description)"));
1701 hs->error = xstrdup (message);
1704 type = resp_header_strdup (resp, "Content-Type");
1707 char *tmp = strchr (type, ';');
1710 while (tmp > type && ISSPACE (tmp[-1]))
1715 hs->newloc = resp_header_strdup (resp, "Location");
1716 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1718 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1722 const char *scbeg, *scend;
1723 /* The jar should have been created by now. */
1724 assert (wget_cookie_jar != NULL);
1726 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1727 &scbeg, &scend)) != -1;
1730 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1731 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1732 u->path, set_cookie);
1736 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1738 wgint first_byte_pos, last_byte_pos, entity_length;
1739 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1741 contrange = first_byte_pos;
1745 /* 20x responses are counted among successful by default. */
1746 if (H_20X (statcode))
1749 /* Return if redirected. */
1750 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1752 /* RFC2068 says that in case of the 300 (multiple choices)
1753 response, the server can output a preferred URL through
1754 `Location' header; otherwise, the request should be treated
1755 like GET. So, if the location is set, it will be a
1756 redirection; otherwise, just proceed normally. */
1757 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1761 logprintf (LOG_VERBOSE,
1762 _("Location: %s%s\n"),
1763 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1764 hs->newloc ? _(" [following]") : "");
1765 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1766 CLOSE_FINISH (sock);
1768 CLOSE_INVALIDATE (sock);
1774 /* If content-type is not given, assume text/html. This is because
1775 of the multitude of broken CGI's that "forget" to generate the
1778 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1779 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1784 if (opt.html_extension && (*dt & TEXTHTML))
1785 /* -E / --html-extension / html_extension = on was specified, and this is a
1786 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1787 already the file's suffix, tack on ".html". */
1789 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1791 if (last_period_in_local_filename == NULL
1792 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1793 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1795 int local_filename_len = strlen (*hs->local_file);
1796 /* Resize the local file, allowing for ".html" preceded by
1797 optional ".NUMBER". */
1798 *hs->local_file = xrealloc (*hs->local_file,
1799 local_filename_len + 24 + sizeof (".html"));
1800 strcpy(*hs->local_file + local_filename_len, ".html");
1801 /* If clobbering is not allowed and the file, as named,
1802 exists, tack on ".NUMBER.html" instead. */
1807 sprintf (*hs->local_file + local_filename_len,
1808 ".%d.html", ext_num++);
1809 while (file_exists_p (*hs->local_file));
1811 *dt |= ADDED_HTML_EXTENSION;
1815 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1817 /* If `-c' is in use and the file has been fully downloaded (or
1818 the remote file has shrunk), Wget effectively requests bytes
1819 after the end of file and the server response with 416. */
1820 logputs (LOG_VERBOSE, _("\
1821 \n The file is already fully retrieved; nothing to do.\n\n"));
1822 /* In case the caller inspects. */
1825 /* Mark as successfully retrieved. */
1828 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1829 might be more bytes in the body. */
1830 return RETRUNNEEDED;
1832 if ((contrange != 0 && contrange != hs->restval)
1833 || (H_PARTIAL (statcode) && !contrange))
1835 /* The Range request was somehow misunderstood by the server.
1838 CLOSE_INVALIDATE (sock);
1841 hs->contlen = contlen + contrange;
1847 /* No need to print this output if the body won't be
1848 downloaded at all, or if the original server response is
1850 logputs (LOG_VERBOSE, _("Length: "));
1853 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1854 if (contlen + contrange >= 1024)
1855 logprintf (LOG_VERBOSE, " (%s)",
1856 human_readable (contlen + contrange));
1859 if (contlen >= 1024)
1860 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1861 number_to_static_string (contlen),
1862 human_readable (contlen));
1864 logprintf (LOG_VERBOSE, _(", %s remaining"),
1865 number_to_static_string (contlen));
1869 logputs (LOG_VERBOSE,
1870 opt.ignore_length ? _("ignored") : _("unspecified"));
1872 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1874 logputs (LOG_VERBOSE, "\n");
1878 type = NULL; /* We don't need it any more. */
1880 /* Return if we have no intention of further downloading. */
1881 if (!(*dt & RETROKF) || head_only)
1883 /* In case the caller cares to look... */
1887 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1888 servers not to send body in response to a HEAD request. If
1889 you encounter such a server (more likely a broken CGI), use
1890 `--no-http-keep-alive'. */
1891 CLOSE_FINISH (sock);
1892 return RETRFINISHED;
1895 /* Open the local file. */
1898 mkalldirs (*hs->local_file);
1900 rotate_backups (*hs->local_file);
1902 fp = fopen (*hs->local_file, "ab");
1903 else if (ALLOW_CLOBBER)
1904 fp = fopen (*hs->local_file, "wb");
1907 fp = fopen_excl (*hs->local_file, true);
1908 if (!fp && errno == EEXIST)
1910 /* We cannot just invent a new name and use it (which is
1911 what functions like unique_create typically do)
1912 because we told the user we'd use this name.
1913 Instead, return and retry the download. */
1914 logprintf (LOG_NOTQUIET,
1915 _("%s has sprung into existence.\n"),
1917 CLOSE_INVALIDATE (sock);
1918 return FOPEN_EXCL_ERR;
1923 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1924 CLOSE_INVALIDATE (sock);
1931 /* #### This confuses the timestamping code that checks for file
1932 size. Maybe we should save some additional information? */
1933 if (opt.save_headers)
1934 fwrite (head, 1, strlen (head), fp);
1936 /* Now we no longer need to store the response header. */
1939 /* Download the request body. */
1942 flags |= rb_read_exactly;
1943 if (hs->restval > 0 && contrange == 0)
1944 /* If the server ignored our range request, instruct fd_read_body
1945 to skip the first RESTVAL bytes of body. */
1946 flags |= rb_skip_startpos;
1947 hs->len = hs->restval;
1949 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1950 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1954 CLOSE_FINISH (sock);
1956 CLOSE_INVALIDATE (sock);
1959 /* Close or flush the file. We have to be careful to check for
1960 error here. Checking the result of fwrite() is not enough --
1961 errors could go unnoticed! */
1964 flush_res = fclose (fp);
1966 flush_res = fflush (fp);
1967 if (flush_res == EOF)
1972 return RETRFINISHED;
1975 /* The genuine HTTP loop! This is the part where the retrieval is
1976 retried, and retried, and retried, and... */
1978 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1979 int *dt, struct url *proxy)
1982 bool use_ts, got_head = false;/* time-stamping info */
1983 char *filename_plus_orig_suffix;
1984 char *local_filename = NULL;
1988 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1989 wgint local_size = 0; /* the size of the local file */
1990 size_t filename_len;
1991 struct http_stat hstat; /* HTTP status */
1995 /* This used to be done in main(), but it's a better idea to do it
1996 here so that we don't go through the hoops if we're just using
2000 if (!wget_cookie_jar)
2001 wget_cookie_jar = cookie_jar_new ();
2002 if (opt.cookies_input && !cookies_loaded_p)
2004 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2005 cookies_loaded_p = true;
2011 /* Warn on (likely bogus) wildcard usage in HTTP. */
2012 if (opt.ftp_glob && has_wildcards_p (u->path))
2013 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2017 /* Determine the local filename. */
2018 if (local_file && *local_file)
2019 hstat.local_file = local_file;
2020 else if (local_file && !opt.output_document)
2022 *local_file = url_file_name (u);
2023 hstat.local_file = local_file;
2027 dummy = url_file_name (u);
2028 hstat.local_file = &dummy;
2029 /* be honest about where we will save the file */
2030 if (local_file && opt.output_document)
2031 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2034 if (!opt.output_document)
2035 locf = *hstat.local_file;
2037 locf = opt.output_document;
2039 hstat.referer = referer;
2041 filename_len = strlen (*hstat.local_file);
2042 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2044 if (opt.noclobber && file_exists_p (*hstat.local_file))
2046 /* If opt.noclobber is turned on and file already exists, do not
2047 retrieve the file */
2048 logprintf (LOG_VERBOSE, _("\
2049 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2050 /* If the file is there, we suppose it's retrieved OK. */
2053 /* #### Bogusness alert. */
2054 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2055 if (has_html_suffix_p (*hstat.local_file))
2063 if (opt.timestamping)
2065 bool local_dot_orig_file_exists = false;
2067 if (opt.backup_converted)
2068 /* If -K is specified, we'll act on the assumption that it was specified
2069 last time these files were downloaded as well, and instead of just
2070 comparing local file X against server file X, we'll compare local
2071 file X.orig (if extant, else X) against server file X. If -K
2072 _wasn't_ specified last time, or the server contains files called
2073 *.orig, -N will be back to not operating correctly with -k. */
2075 /* Would a single s[n]printf() call be faster? --dan
2077 Definitely not. sprintf() is horribly slow. It's a
2078 different question whether the difference between the two
2079 affects a program. Usually I'd say "no", but at one
2080 point I profiled Wget, and found that a measurable and
2081 non-negligible amount of time was lost calling sprintf()
2082 in url.c. Replacing sprintf with inline calls to
2083 strcpy() and number_to_string() made a difference.
2085 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2086 memcpy (filename_plus_orig_suffix + filename_len,
2087 ".orig", sizeof (".orig"));
2089 /* Try to stat() the .orig file. */
2090 if (stat (filename_plus_orig_suffix, &st) == 0)
2092 local_dot_orig_file_exists = 1;
2093 local_filename = filename_plus_orig_suffix;
2097 if (!local_dot_orig_file_exists)
2098 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2099 if (stat (*hstat.local_file, &st) == 0)
2100 local_filename = *hstat.local_file;
2102 if (local_filename != NULL)
2103 /* There was a local file, so we'll check later to see if the version
2104 the server has is the same version we already have, allowing us to
2110 /* Modification time granularity is 2 seconds for Windows, so
2111 increase local time by 1 second for later comparison. */
2114 local_size = st.st_size;
2118 /* Reset the counter. */
2124 /* Increment the pass counter. */
2126 sleep_between_retrievals (count);
2127 /* Get the current time string. */
2128 tms = time_str (NULL);
2129 /* Print fetch message, if opt.verbose. */
2132 char *hurl = url_string (u, true);
2136 sprintf (tmp, _("(try:%2d)"), count);
2137 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2138 tms, hurl, tmp, locf);
2140 ws_changetitle (hurl);
2145 /* Default document type is empty. However, if spider mode is
2146 on or time-stamping is employed, HEAD_ONLY commands is
2147 encoded within *dt. */
2148 if (opt.spider || (use_ts && !got_head))
2153 /* Decide whether or not to restart. */
2155 && stat (locf, &st) == 0
2156 && S_ISREG (st.st_mode))
2157 /* When -c is used, continue from on-disk size. (Can't use
2158 hstat.len even if count>1 because we don't want a failed
2159 first attempt to clobber existing data.) */
2160 hstat.restval = st.st_size;
2162 /* otherwise, continue where the previous try left off */
2163 hstat.restval = hstat.len;
2167 /* Decide whether to send the no-cache directive. We send it in
2169 a) we're using a proxy, and we're past our first retrieval.
2170 Some proxies are notorious for caching incomplete data, so
2171 we require a fresh get.
2172 b) caching is explicitly inhibited. */
2173 if ((proxy && count > 1) /* a */
2174 || !opt.allow_cache /* b */
2176 *dt |= SEND_NOCACHE;
2178 *dt &= ~SEND_NOCACHE;
2180 /* Try fetching the document, or at least its head. */
2181 err = gethttp (u, &hstat, dt, proxy);
2183 /* It's unfortunate that wget determines the local filename before finding
2184 out the Content-Type of the file. Barring a major restructuring of the
2185 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2186 *hstat.local_file to tack on ".html". */
2187 if (!opt.output_document)
2188 locf = *hstat.local_file;
2191 tms = time_str (NULL);
2192 /* Get the new location (with or without the redirection). */
2194 *newloc = xstrdup (hstat.newloc);
2197 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2198 case CONERROR: case READERR: case WRITEFAILED:
2199 case RANGEERR: case FOPEN_EXCL_ERR:
2200 /* Non-fatal errors continue executing the loop, which will
2201 bring them to "while" statement at the end, to judge
2202 whether the number of tries was exceeded. */
2203 free_hstat (&hstat);
2204 printwhat (count, opt.ntry);
2205 if (err == FOPEN_EXCL_ERR)
2207 /* Re-determine the file name. */
2208 if (local_file && *local_file)
2210 xfree (*local_file);
2211 *local_file = url_file_name (u);
2212 hstat.local_file = local_file;
2217 dummy = url_file_name (u);
2218 hstat.local_file = &dummy;
2220 /* be honest about where we will save the file */
2221 if (local_file && opt.output_document)
2222 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2223 if (!opt.output_document)
2224 locf = *hstat.local_file;
2226 locf = opt.output_document;
2229 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2230 case SSLINITFAILED: case CONTNOTSUPPORTED:
2231 /* Fatal errors just return from the function. */
2232 free_hstat (&hstat);
2235 case FWRITEERR: case FOPENERR:
2236 /* Another fatal error. */
2237 logputs (LOG_VERBOSE, "\n");
2238 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2239 *hstat.local_file, strerror (errno));
2240 free_hstat (&hstat);
2244 /* Another fatal error. */
2245 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2246 free_hstat (&hstat);
2250 /* Return the new location to the caller. */
2253 logprintf (LOG_NOTQUIET,
2254 _("ERROR: Redirection (%d) without location.\n"),
2256 free_hstat (&hstat);
2260 free_hstat (&hstat);
2264 /* The file was already fully retrieved. */
2265 free_hstat (&hstat);
2269 /* Deal with you later. */
2272 /* All possibilities should have been exhausted. */
2275 if (!(*dt & RETROKF))
2279 /* #### Ugly ugly ugly! */
2280 char *hurl = url_string (u, true);
2281 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2284 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2285 tms, hstat.statcode, escnonprint (hstat.error));
2286 logputs (LOG_VERBOSE, "\n");
2287 free_hstat (&hstat);
2292 /* Did we get the time-stamp? */
2295 if (opt.timestamping && !hstat.remote_time)
2297 logputs (LOG_NOTQUIET, _("\
2298 Last-modified header missing -- time-stamps turned off.\n"));
2300 else if (hstat.remote_time)
2302 /* Convert the date-string into struct tm. */
2303 tmr = http_atotm (hstat.remote_time);
2304 if (tmr == (time_t) (-1))
2305 logputs (LOG_VERBOSE, _("\
2306 Last-modified header invalid -- time-stamp ignored.\n"));
2310 /* The time-stamping section. */
2315 use_ts = false; /* no more time-stamping */
2316 count = 0; /* the retrieve count for HEAD is
2318 if (hstat.remote_time && tmr != (time_t) (-1))
2320 /* Now time-stamping can be used validly. Time-stamping
2321 means that if the sizes of the local and remote file
2322 match, and local file is newer than the remote file,
2323 it will not be retrieved. Otherwise, the normal
2324 download procedure is resumed. */
2326 (hstat.contlen == -1 || local_size == hstat.contlen))
2328 logprintf (LOG_VERBOSE, _("\
2329 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2331 free_hstat (&hstat);
2335 else if (tml >= tmr)
2336 logprintf (LOG_VERBOSE, _("\
2337 The sizes do not match (local %s) -- retrieving.\n"),
2338 number_to_static_string (local_size));
2340 logputs (LOG_VERBOSE,
2341 _("Remote file is newer, retrieving.\n"));
2343 free_hstat (&hstat);
2346 if ((tmr != (time_t) (-1))
2348 && ((hstat.len == hstat.contlen) ||
2349 ((hstat.res == 0) &&
2350 ((hstat.contlen == -1) ||
2351 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2353 /* #### This code repeats in http.c and ftp.c. Move it to a
2355 const char *fl = NULL;
2356 if (opt.output_document)
2358 if (output_stream_regular)
2359 fl = opt.output_document;
2362 fl = *hstat.local_file;
2366 /* End of time-stamping section. */
2370 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2371 escnonprint (hstat.error));
2376 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2377 total_download_time += hstat.dltime;
2379 if (hstat.len == hstat.contlen)
2383 logprintf (LOG_VERBOSE,
2384 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2386 number_to_static_string (hstat.len),
2387 number_to_static_string (hstat.contlen));
2388 logprintf (LOG_NONVERBOSE,
2389 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2391 number_to_static_string (hstat.len),
2392 number_to_static_string (hstat.contlen),
2396 total_downloaded_bytes += hstat.len;
2398 /* Remember that we downloaded the file for later ".orig" code. */
2399 if (*dt & ADDED_HTML_EXTENSION)
2400 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2402 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2404 free_hstat (&hstat);
2408 else if (hstat.res == 0) /* No read error */
2410 if (hstat.contlen == -1) /* We don't know how much we were supposed
2411 to get, so assume we succeeded. */
2415 logprintf (LOG_VERBOSE,
2416 _("%s (%s) - `%s' saved [%s]\n\n"),
2418 number_to_static_string (hstat.len));
2419 logprintf (LOG_NONVERBOSE,
2420 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2421 tms, u->url, number_to_static_string (hstat.len),
2425 total_downloaded_bytes += hstat.len;
2427 /* Remember that we downloaded the file for later ".orig" code. */
2428 if (*dt & ADDED_HTML_EXTENSION)
2429 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2431 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2433 free_hstat (&hstat);
2437 else if (hstat.len < hstat.contlen) /* meaning we lost the
2438 connection too soon */
2440 logprintf (LOG_VERBOSE,
2441 _("%s (%s) - Connection closed at byte %s. "),
2442 tms, tmrate, number_to_static_string (hstat.len));
2443 printwhat (count, opt.ntry);
2444 free_hstat (&hstat);
2447 else if (!opt.kill_longer) /* meaning we got more than expected */
2449 logprintf (LOG_VERBOSE,
2450 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2452 number_to_static_string (hstat.len),
2453 number_to_static_string (hstat.contlen));
2454 logprintf (LOG_NONVERBOSE,
2455 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2457 number_to_static_string (hstat.len),
2458 number_to_static_string (hstat.contlen),
2461 total_downloaded_bytes += hstat.len;
2463 /* Remember that we downloaded the file for later ".orig" code. */
2464 if (*dt & ADDED_HTML_EXTENSION)
2465 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2467 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2469 free_hstat (&hstat);
2473 else /* the same, but not accepted */
2475 logprintf (LOG_VERBOSE,
2476 _("%s (%s) - Connection closed at byte %s/%s. "),
2478 number_to_static_string (hstat.len),
2479 number_to_static_string (hstat.contlen));
2480 printwhat (count, opt.ntry);
2481 free_hstat (&hstat);
2485 else /* now hstat.res can only be -1 */
2487 if (hstat.contlen == -1)
2489 logprintf (LOG_VERBOSE,
2490 _("%s (%s) - Read error at byte %s (%s)."),
2491 tms, tmrate, number_to_static_string (hstat.len),
2493 printwhat (count, opt.ntry);
2494 free_hstat (&hstat);
2497 else /* hstat.res == -1 and contlen is given */
2499 logprintf (LOG_VERBOSE,
2500 _("%s (%s) - Read error at byte %s/%s (%s). "),
2502 number_to_static_string (hstat.len),
2503 number_to_static_string (hstat.contlen),
2505 printwhat (count, opt.ntry);
2506 free_hstat (&hstat);
2512 while (!opt.ntry || (count < opt.ntry));
2516 /* Check whether the result of strptime() indicates success.
2517 strptime() returns the pointer to how far it got to in the string.
2518 The processing has been successful if the string is at `GMT' or
2519 `+X', or at the end of the string.
2521 In extended regexp parlance, the function returns 1 if P matches
2522 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2523 can return) is considered a failure and 0 is returned. */
2525 check_end (const char *p)
2529 while (ISSPACE (*p))
2532 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2533 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2539 /* Convert the textual specification of time in TIME_STRING to the
2540 number of seconds since the Epoch.
2542 TIME_STRING can be in any of the three formats RFC2616 allows the
2543 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2544 as well as the time format used in the Set-Cookie header.
2545 Timezones are ignored, and should be GMT.
2547 Return the computed time_t representation, or -1 if the conversion
2550 This function uses strptime with various string formats for parsing
2551 TIME_STRING. This results in a parser that is not as lenient in
2552 interpreting TIME_STRING as I would like it to be. Being based on
2553 strptime, it always allows shortened months, one-digit days, etc.,
2554 but due to the multitude of formats in which time can be
2555 represented, an ideal HTTP time parser would be even more
2556 forgiving. It should completely ignore things like week days and
2557 concentrate only on the various forms of representing years,
2558 months, days, hours, minutes, and seconds. For example, it would
2559 be nice if it accepted ISO 8601 out of the box.
2561 I've investigated free and PD code for this purpose, but none was
2562 usable. getdate was big and unwieldy, and had potential copyright
2563 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2564 distributed with phttpd, is excellent, but we cannot use it because
2565 it is not assigned to the FSF. So I stuck it with strptime. */
2568 http_atotm (const char *time_string)
2570 /* NOTE: Solaris strptime man page claims that %n and %t match white
2571 space, but that's not universally available. Instead, we simply
2572 use ` ' to mean "skip all WS", which works under all strptime
2573 implementations I've tested. */
2575 static const char *time_formats[] = {
2576 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2577 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2578 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2579 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2580 (used in Set-Cookie, defined in the
2581 Netscape cookie specification.) */
2583 const char *oldlocale;
2585 time_t ret = (time_t) -1;
2587 /* Solaris strptime fails to recognize English month names in
2588 non-English locales, which we work around by temporarily setting
2589 locale to C before invoking strptime. */
2590 oldlocale = setlocale (LC_TIME, NULL);
2591 setlocale (LC_TIME, "C");
2593 for (i = 0; i < countof (time_formats); i++)
2597 /* Some versions of strptime use the existing contents of struct
2598 tm to recalculate the date according to format. Zero it out
2599 to prevent stack garbage from influencing strptime. */
2602 if (check_end (strptime (time_string, time_formats[i], &t)))
2609 /* Restore the previous locale. */
2610 setlocale (LC_TIME, oldlocale);
2615 /* Authorization support: We support three authorization schemes:
2617 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2619 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2620 consisting of answering to the server's challenge with the proper
2623 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2624 Stenberg for libcurl. Like digest, NTLM is based on a
2625 challenge-response mechanism, but unlike digest, it is non-standard
2626 (authenticates TCP connections rather than requests), undocumented
2627 and Microsoft-specific. */
2629 /* Create the authentication header contents for the `Basic' scheme.
2630 This is done by encoding the string "USER:PASS" to base64 and
2631 prepending the string "Basic " in front of it. */
2634 basic_authentication_encode (const char *user, const char *passwd)
2637 int len1 = strlen (user) + 1 + strlen (passwd);
2639 t1 = (char *)alloca (len1 + 1);
2640 sprintf (t1, "%s:%s", user, passwd);
2642 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2643 base64_encode (t1, len1, t2);
2645 return concat_strings ("Basic ", t2, (char *) 0);
2648 #define SKIP_WS(x) do { \
2649 while (ISSPACE (*(x))) \
2653 #ifdef ENABLE_DIGEST
2654 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2655 of a field in such a header. If the field is the one specified by
2656 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2657 digest authorization code), extract its value in the (char*)
2658 variable pointed by RET. Returns negative on a malformed header,
2659 or number of bytes that have been parsed by this call. */
2661 extract_header_attr (const char *au, const char *attr_name, char **ret)
2664 const char *cp = au;
2666 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2668 cp += strlen (attr_name);
2681 for (ep = cp; *ep && *ep != '\"'; ep++)
2686 *ret = strdupdelim (cp, ep);
2693 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2694 an array of 16 bytes containing the hash keys, and BUF should be a
2695 buffer of 33 writable characters (32 for hex digits plus one for
2696 zero termination). */
2698 dump_hash (char *buf, const unsigned char *hash)
2702 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2704 *buf++ = XNUM_TO_digit (*hash >> 4);
2705 *buf++ = XNUM_TO_digit (*hash & 0xf);
2710 /* Take the line apart to find the challenge, and compose a digest
2711 authorization header. See RFC2069 section 2.1.2. */
2713 digest_authentication_encode (const char *au, const char *user,
2714 const char *passwd, const char *method,
2717 static char *realm, *opaque, *nonce;
2722 { "realm", &realm },
2723 { "opaque", &opaque },
2728 realm = opaque = nonce = NULL;
2730 au += 6; /* skip over `Digest' */
2736 for (i = 0; i < countof (options); i++)
2738 int skip = extract_header_attr (au, options[i].name,
2739 options[i].variable);
2743 xfree_null (opaque);
2753 if (i == countof (options))
2755 while (*au && *au != '=')
2763 while (*au && *au != '\"')
2770 while (*au && *au != ',')
2775 if (!realm || !nonce || !user || !passwd || !path || !method)
2778 xfree_null (opaque);
2783 /* Calculate the digest value. */
2785 ALLOCA_MD5_CONTEXT (ctx);
2786 unsigned char hash[MD5_HASHLEN];
2787 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2788 char response_digest[MD5_HASHLEN * 2 + 1];
2790 /* A1BUF = H(user ":" realm ":" password) */
2792 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2793 gen_md5_update ((unsigned char *)":", 1, ctx);
2794 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2795 gen_md5_update ((unsigned char *)":", 1, ctx);
2796 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2797 gen_md5_finish (ctx, hash);
2798 dump_hash (a1buf, hash);
2800 /* A2BUF = H(method ":" path) */
2802 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2803 gen_md5_update ((unsigned char *)":", 1, ctx);
2804 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2805 gen_md5_finish (ctx, hash);
2806 dump_hash (a2buf, hash);
2808 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2810 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2811 gen_md5_update ((unsigned char *)":", 1, ctx);
2812 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2813 gen_md5_update ((unsigned char *)":", 1, ctx);
2814 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2815 gen_md5_finish (ctx, hash);
2816 dump_hash (response_digest, hash);
2818 res = xmalloc (strlen (user)
2823 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2824 + (opaque ? strlen (opaque) : 0)
2826 sprintf (res, "Digest \
2827 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2828 user, realm, nonce, path, response_digest);
2831 char *p = res + strlen (res);
2832 strcat (p, ", opaque=\"");
2839 #endif /* ENABLE_DIGEST */
2841 /* Computing the size of a string literal must take into account that
2842 value returned by sizeof includes the terminating \0. */
2843 #define STRSIZE(literal) (sizeof (literal) - 1)
2845 /* Whether chars in [b, e) begin with the literal string provided as
2846 first argument and are followed by whitespace or terminating \0.
2847 The comparison is case-insensitive. */
2848 #define STARTS(literal, b, e) \
2849 ((e) - (b) >= STRSIZE (literal) \
2850 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2851 && ((e) - (b) == STRSIZE (literal) \
2852 || ISSPACE (b[STRSIZE (literal)])))
2855 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2857 return STARTS ("Basic", hdrbeg, hdrend)
2858 #ifdef ENABLE_DIGEST
2859 || STARTS ("Digest", hdrbeg, hdrend)
2862 || STARTS ("NTLM", hdrbeg, hdrend)
2869 /* Create the HTTP authorization request header. When the
2870 `WWW-Authenticate' response header is seen, according to the
2871 authorization scheme specified in that header (`Basic' and `Digest'
2872 are supported by the current implementation), produce an
2873 appropriate HTTP authorization request header. */
2875 create_authorization_line (const char *au, const char *user,
2876 const char *passwd, const char *method,
2877 const char *path, bool *finished)
2879 /* We are called only with known schemes, so we can dispatch on the
2881 switch (TOUPPER (*au))
2883 case 'B': /* Basic */
2885 return basic_authentication_encode (user, passwd);
2886 #ifdef ENABLE_DIGEST
2887 case 'D': /* Digest */
2889 return digest_authentication_encode (au, user, passwd, method, path);
2892 case 'N': /* NTLM */
2893 if (!ntlm_input (&pconn.ntlm, au))
2898 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2901 /* We shouldn't get here -- this function should be only called
2902 with values approved by known_authentication_scheme_p. */
2910 if (wget_cookie_jar)
2911 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2917 xfree_null (pconn.host);
2918 if (wget_cookie_jar)
2919 cookie_jar_delete (wget_cookie_jar);