2 Copyright (C) 1996-2005 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 Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 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 servers implement 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 (Current implementation of test_socket_open has a nice side
1019 effect that it treats sockets with pending data as "closed".
1020 This is exactly what we want: if a broken server sends message
1021 body in response to HEAD, or if it sends more than conent-length
1022 data, we won't reuse the corrupted connection.) */
1024 if (!test_socket_open (pconn.socket))
1026 /* Oops, the socket is no longer open. Now that we know that,
1027 let's invalidate the persistent connection before returning
1029 invalidate_persistent ();
1036 /* The idea behind these two CLOSE macros is to distinguish between
1037 two cases: one when the job we've been doing is finished, and we
1038 want to close the connection and leave, and two when something is
1039 seriously wrong and we're closing the connection as part of
1042 In case of keep_alive, CLOSE_FINISH should leave the connection
1043 open, while CLOSE_INVALIDATE should still close it.
1045 Note that the semantics of the flag `keep_alive' is "this
1046 connection *will* be reused (the server has promised not to close
1047 the connection once we're done)", while the semantics of
1048 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1049 active, registered connection". */
1051 #define CLOSE_FINISH(fd) do { \
1054 if (pconn_active && (fd) == pconn.socket) \
1055 invalidate_persistent (); \
1064 #define CLOSE_INVALIDATE(fd) do { \
1065 if (pconn_active && (fd) == pconn.socket) \
1066 invalidate_persistent (); \
1074 wgint len; /* received length */
1075 wgint contlen; /* expected length */
1076 wgint restval; /* the restart value */
1077 int res; /* the result of last read */
1078 char *newloc; /* new location (redirection) */
1079 char *remote_time; /* remote time-stamp string */
1080 char *error; /* textual HTTP error */
1081 int statcode; /* status code */
1082 wgint rd_size; /* amount of data read from socket */
1083 double dltime; /* time it took to download the data */
1084 const char *referer; /* value of the referer header. */
1085 char **local_file; /* local file. */
1089 free_hstat (struct http_stat *hs)
1091 xfree_null (hs->newloc);
1092 xfree_null (hs->remote_time);
1093 xfree_null (hs->error);
1095 /* Guard against being called twice. */
1097 hs->remote_time = NULL;
1101 static char *create_authorization_line (const char *, const char *,
1102 const char *, const char *,
1103 const char *, bool *);
1104 static char *basic_authentication_encode (const char *, const char *);
1105 static bool known_authentication_scheme_p (const char *, const char *);
1107 #define BEGINS_WITH(line, string_constant) \
1108 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1109 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1110 || !line[sizeof (string_constant) - 1]))
1112 #define SET_USER_AGENT(req) do { \
1113 if (!opt.useragent) \
1114 request_set_header (req, "User-Agent", \
1115 aprintf ("Wget/%s", version_string), rel_value); \
1116 else if (*opt.useragent) \
1117 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1120 /* The flags that allow clobbering the file (opening with "wb").
1121 Defined here to avoid repetition later. #### This will require
1123 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1124 || opt.dirstruct || opt.output_document)
1126 /* Retrieve a document through HTTP protocol. It recognizes status
1127 code, and correctly handles redirections. It closes the network
1128 socket. If it receives an error from the functions below it, it
1129 will print it if there is enough information to do so (almost
1130 always), returning the error to the caller (i.e. http_loop).
1132 Various HTTP parameters are stored to hs.
1134 If PROXY is non-NULL, the connection will be made to the proxy
1135 server, and u->url will be requested. */
1137 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1139 struct request *req;
1142 char *user, *passwd;
1146 wgint contlen, contrange;
1153 /* Set to 1 when the authorization has failed permanently and should
1154 not be tried again. */
1155 bool auth_finished = false;
1157 /* Whether NTLM authentication is used for this request. */
1158 bool ntlm_seen = false;
1160 /* Whether our connection to the remote host is through SSL. */
1161 bool using_ssl = false;
1163 /* Whether a HEAD request will be issued (as opposed to GET or
1165 bool head_only = !!(*dt & HEAD_ONLY);
1168 struct response *resp;
1172 /* Whether this connection will be kept alive after the HTTP request
1176 /* Whether keep-alive should be inhibited.
1178 RFC 2068 requests that 1.0 clients not send keep-alive requests
1179 to proxies. This is because many 1.0 proxies do not interpret
1180 the Connection header and transfer it to the remote server,
1181 causing it to not close the connection and leave both the proxy
1182 and the client hanging. */
1183 bool inhibit_keep_alive =
1184 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1186 /* Headers sent when using POST. */
1187 wgint post_data_size = 0;
1189 bool host_lookup_failed = false;
1192 if (u->scheme == SCHEME_HTTPS)
1194 /* Initialize the SSL context. After this has once been done,
1195 it becomes a no-op. */
1198 scheme_disable (SCHEME_HTTPS);
1199 logprintf (LOG_NOTQUIET,
1200 _("Disabling SSL due to encountered errors.\n"));
1201 return SSLINITFAILED;
1204 #endif /* HAVE_SSL */
1207 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1208 know the local filename so we can save to it. */
1209 assert (*hs->local_file != NULL);
1211 /* Initialize certain elements of struct http_stat. */
1216 hs->remote_time = NULL;
1221 /* Prepare the request to send. */
1223 req = request_new ();
1226 const char *meth = "GET";
1229 else if (opt.post_file_name || opt.post_data)
1231 /* Use the full path, i.e. one that includes the leading slash and
1232 the query string. E.g. if u->path is "foo/bar" and u->query is
1233 "param=value", full_path will be "/foo/bar?param=value". */
1236 /* When using SSL over proxy, CONNECT establishes a direct
1237 connection to the HTTPS server. Therefore use the same
1238 argument as when talking to the server directly. */
1239 && u->scheme != SCHEME_HTTPS
1242 meth_arg = xstrdup (u->url);
1244 meth_arg = url_full_path (u);
1245 request_set_method (req, meth, meth_arg);
1248 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1249 if (*dt & SEND_NOCACHE)
1250 request_set_header (req, "Pragma", "no-cache", rel_none);
1252 request_set_header (req, "Range",
1253 aprintf ("bytes=%s-",
1254 number_to_static_string (hs->restval)),
1256 SET_USER_AGENT (req);
1257 request_set_header (req, "Accept", "*/*", rel_none);
1259 /* Find the username and password for authentication. */
1262 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1263 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1264 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1268 /* We have the username and the password, but haven't tried
1269 any authorization yet. Let's see if the "Basic" method
1270 works. If not, we'll come back here and construct a
1271 proper authorization method with the right challenges.
1273 If we didn't employ this kind of logic, every URL that
1274 requires authorization would have to be processed twice,
1275 which is very suboptimal and generates a bunch of false
1276 "unauthorized" errors in the server log.
1278 #### But this logic also has a serious problem when used
1279 with stronger authentications: we *first* transmit the
1280 username and the password in clear text, and *then* attempt a
1281 stronger authentication scheme. That cannot be right! We
1282 are only fortunate that almost everyone still uses the
1283 `Basic' scheme anyway.
1285 There should be an option to prevent this from happening, for
1286 those who use strong authentication schemes and value their
1288 request_set_header (req, "Authorization",
1289 basic_authentication_encode (user, passwd),
1296 char *proxy_user, *proxy_passwd;
1297 /* For normal username and password, URL components override
1298 command-line/wgetrc parameters. With proxy
1299 authentication, it's the reverse, because proxy URLs are
1300 normally the "permanent" ones, so command-line args
1301 should take precedence. */
1302 if (opt.proxy_user && opt.proxy_passwd)
1304 proxy_user = opt.proxy_user;
1305 proxy_passwd = opt.proxy_passwd;
1309 proxy_user = proxy->user;
1310 proxy_passwd = proxy->passwd;
1312 /* #### This does not appear right. Can't the proxy request,
1313 say, `Digest' authentication? */
1314 if (proxy_user && proxy_passwd)
1315 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1317 /* If we're using a proxy, we will be connecting to the proxy
1321 /* Proxy authorization over SSL is handled below. */
1323 if (u->scheme != SCHEME_HTTPS)
1325 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1328 /* Generate the Host header, HOST:PORT. Take into account that:
1330 - Broken server-side software often doesn't recognize the PORT
1331 argument, so we must generate "Host: www.server.com" instead of
1332 "Host: www.server.com:80" (and likewise for https port).
1334 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1335 becomes ambiguous and needs to be rewritten as "Host:
1336 [3ffe:8100:200:2::2]:1234". */
1338 /* Formats arranged for hfmt[add_port][add_squares]. */
1339 static const char *hfmt[][2] = {
1340 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1342 int add_port = u->port != scheme_default_port (u->scheme);
1343 int add_squares = strchr (u->host, ':') != NULL;
1344 request_set_header (req, "Host",
1345 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1349 if (!inhibit_keep_alive)
1350 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1353 request_set_header (req, "Cookie",
1354 cookie_header (wget_cookie_jar,
1355 u->host, u->port, u->path,
1357 u->scheme == SCHEME_HTTPS
1364 if (opt.post_data || opt.post_file_name)
1366 request_set_header (req, "Content-Type",
1367 "application/x-www-form-urlencoded", rel_none);
1369 post_data_size = strlen (opt.post_data);
1372 post_data_size = file_size (opt.post_file_name);
1373 if (post_data_size == -1)
1375 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1376 opt.post_file_name, strerror (errno));
1380 request_set_header (req, "Content-Length",
1381 xstrdup (number_to_static_string (post_data_size)),
1385 /* Add the user headers. */
1386 if (opt.user_headers)
1389 for (i = 0; opt.user_headers[i]; i++)
1390 request_set_user_header (req, opt.user_headers[i]);
1394 /* We need to come back here when the initial attempt to retrieve
1395 without authorization header fails. (Expected to happen at least
1396 for the Digest authorization scheme.) */
1400 /* Establish the connection. */
1402 if (!inhibit_keep_alive)
1404 /* Look for a persistent connection to target host, unless a
1405 proxy is used. The exception is when SSL is in use, in which
1406 case the proxy is nothing but a passthrough to the target
1407 host, registered as a connection to the latter. */
1408 struct url *relevant = conn;
1410 if (u->scheme == SCHEME_HTTPS)
1414 if (persistent_available_p (relevant->host, relevant->port,
1416 relevant->scheme == SCHEME_HTTPS,
1420 &host_lookup_failed))
1422 sock = pconn.socket;
1423 using_ssl = pconn.ssl;
1424 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1425 escnonprint (pconn.host), pconn.port);
1426 DEBUGP (("Reusing fd %d.\n", sock));
1427 if (pconn.authorized)
1428 /* If the connection is already authorized, the "Basic"
1429 authorization added by code above is unnecessary and
1431 request_remove_header (req, "Authorization");
1437 /* In its current implementation, persistent_available_p will
1438 look up conn->host in some cases. If that lookup failed, we
1439 don't need to bother with connect_to_host. */
1440 if (host_lookup_failed)
1446 sock = connect_to_host (conn->host, conn->port);
1455 return (retryable_socket_connect_error (errno)
1456 ? CONERROR : CONIMPOSSIBLE);
1460 if (proxy && u->scheme == SCHEME_HTTPS)
1462 /* When requesting SSL URLs through proxies, use the
1463 CONNECT method to request passthrough. */
1464 struct request *connreq = request_new ();
1465 request_set_method (connreq, "CONNECT",
1466 aprintf ("%s:%d", u->host, u->port));
1467 SET_USER_AGENT (connreq);
1470 request_set_header (connreq, "Proxy-Authorization",
1471 proxyauth, rel_value);
1472 /* Now that PROXYAUTH is part of the CONNECT request,
1473 zero it out so we don't send proxy authorization with
1474 the regular request below. */
1477 /* Examples in rfc2817 use the Host header in CONNECT
1478 requests. I don't see how that gains anything, given
1479 that the contents of Host would be exactly the same as
1480 the contents of CONNECT. */
1482 write_error = request_send (connreq, sock);
1483 request_free (connreq);
1484 if (write_error < 0)
1486 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1488 CLOSE_INVALIDATE (sock);
1492 head = read_http_response_head (sock);
1495 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1497 CLOSE_INVALIDATE (sock);
1506 DEBUGP (("proxy responded with: [%s]\n", head));
1508 resp = resp_new (head);
1509 statcode = resp_status (resp, &message);
1512 if (statcode != 200)
1515 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1516 message ? escnonprint (message) : "?");
1517 xfree_null (message);
1520 xfree_null (message);
1522 /* SOCK is now *really* connected to u->host, so update CONN
1523 to reflect this. That way register_persistent will
1524 register SOCK as being connected to u->host:u->port. */
1528 if (conn->scheme == SCHEME_HTTPS)
1530 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1537 #endif /* HAVE_SSL */
1540 /* Send the request to server. */
1541 write_error = request_send (req, sock);
1543 if (write_error >= 0)
1547 DEBUGP (("[POST data: %s]\n", opt.post_data));
1548 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1550 else if (opt.post_file_name && post_data_size != 0)
1551 write_error = post_file (sock, opt.post_file_name, post_data_size);
1554 if (write_error < 0)
1556 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1558 CLOSE_INVALIDATE (sock);
1562 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1563 proxy ? "Proxy" : "HTTP");
1568 head = read_http_response_head (sock);
1573 logputs (LOG_NOTQUIET, _("No data received.\n"));
1574 CLOSE_INVALIDATE (sock);
1580 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1582 CLOSE_INVALIDATE (sock);
1587 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1589 resp = resp_new (head);
1591 /* Check for status line. */
1593 statcode = resp_status (resp, &message);
1594 if (!opt.server_response)
1595 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1596 message ? escnonprint (message) : "");
1599 logprintf (LOG_VERBOSE, "\n");
1600 print_server_response (resp, " ");
1603 if (!opt.ignore_length
1604 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1608 parsed = str_to_wgint (hdrval, NULL, 10);
1609 if (parsed == WGINT_MAX && errno == ERANGE)
1611 #### If Content-Length is out of range, it most likely
1612 means that the file is larger than 2G and that we're
1613 compiled without LFS. In that case we should probably
1614 refuse to even attempt to download the file. */
1620 /* Check for keep-alive related responses. */
1621 if (!inhibit_keep_alive && contlen != -1)
1623 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1625 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1627 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1632 /* The server has promised that it will not close the connection
1633 when we're done. This means that we can register it. */
1634 register_persistent (conn->host, conn->port, sock, using_ssl);
1636 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1638 /* Authorization is required. */
1639 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1640 CLOSE_FINISH (sock);
1642 CLOSE_INVALIDATE (sock);
1643 pconn.authorized = false;
1644 if (!auth_finished && (user && passwd))
1646 /* IIS sends multiple copies of WWW-Authenticate, one with
1647 the value "negotiate", and other(s) with data. Loop over
1648 all the occurrences and pick the one we recognize. */
1650 const char *wabeg, *waend;
1651 char *www_authenticate = NULL;
1653 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1654 &wabeg, &waend)) != -1;
1656 if (known_authentication_scheme_p (wabeg, waend))
1658 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1662 if (!www_authenticate)
1663 /* If the authentication header is missing or
1664 unrecognized, there's no sense in retrying. */
1665 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1666 else if (BEGINS_WITH (www_authenticate, "Basic"))
1667 /* If the authentication scheme is "Basic", which we send
1668 by default, there's no sense in retrying either. (This
1669 should be changed when we stop sending "Basic" data by
1675 pth = url_full_path (u);
1676 request_set_header (req, "Authorization",
1677 create_authorization_line (www_authenticate,
1679 request_method (req),
1683 if (BEGINS_WITH (www_authenticate, "NTLM"))
1686 goto retry_with_auth;
1689 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1693 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1695 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1697 pconn.authorized = true;
1701 hs->statcode = statcode;
1703 hs->error = xstrdup (_("Malformed status line"));
1705 hs->error = xstrdup (_("(no description)"));
1707 hs->error = xstrdup (message);
1710 type = resp_header_strdup (resp, "Content-Type");
1713 char *tmp = strchr (type, ';');
1716 while (tmp > type && ISSPACE (tmp[-1]))
1721 hs->newloc = resp_header_strdup (resp, "Location");
1722 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1724 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1728 const char *scbeg, *scend;
1729 /* The jar should have been created by now. */
1730 assert (wget_cookie_jar != NULL);
1732 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1733 &scbeg, &scend)) != -1;
1736 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1737 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1738 u->path, set_cookie);
1742 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1744 wgint first_byte_pos, last_byte_pos, entity_length;
1745 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1747 contrange = first_byte_pos;
1751 /* 20x responses are counted among successful by default. */
1752 if (H_20X (statcode))
1755 /* Return if redirected. */
1756 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1758 /* RFC2068 says that in case of the 300 (multiple choices)
1759 response, the server can output a preferred URL through
1760 `Location' header; otherwise, the request should be treated
1761 like GET. So, if the location is set, it will be a
1762 redirection; otherwise, just proceed normally. */
1763 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1767 logprintf (LOG_VERBOSE,
1768 _("Location: %s%s\n"),
1769 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1770 hs->newloc ? _(" [following]") : "");
1771 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1772 CLOSE_FINISH (sock);
1774 CLOSE_INVALIDATE (sock);
1780 /* If content-type is not given, assume text/html. This is because
1781 of the multitude of broken CGI's that "forget" to generate the
1784 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1785 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1790 if (opt.html_extension && (*dt & TEXTHTML))
1791 /* -E / --html-extension / html_extension = on was specified, and this is a
1792 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1793 already the file's suffix, tack on ".html". */
1795 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1797 if (last_period_in_local_filename == NULL
1798 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1799 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1801 int local_filename_len = strlen (*hs->local_file);
1802 /* Resize the local file, allowing for ".html" preceded by
1803 optional ".NUMBER". */
1804 *hs->local_file = xrealloc (*hs->local_file,
1805 local_filename_len + 24 + sizeof (".html"));
1806 strcpy(*hs->local_file + local_filename_len, ".html");
1807 /* If clobbering is not allowed and the file, as named,
1808 exists, tack on ".NUMBER.html" instead. */
1813 sprintf (*hs->local_file + local_filename_len,
1814 ".%d.html", ext_num++);
1815 while (file_exists_p (*hs->local_file));
1817 *dt |= ADDED_HTML_EXTENSION;
1821 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1823 /* If `-c' is in use and the file has been fully downloaded (or
1824 the remote file has shrunk), Wget effectively requests bytes
1825 after the end of file and the server response with 416. */
1826 logputs (LOG_VERBOSE, _("\
1827 \n The file is already fully retrieved; nothing to do.\n\n"));
1828 /* In case the caller inspects. */
1831 /* Mark as successfully retrieved. */
1834 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1835 might be more bytes in the body. */
1836 return RETRUNNEEDED;
1838 if ((contrange != 0 && contrange != hs->restval)
1839 || (H_PARTIAL (statcode) && !contrange))
1841 /* The Range request was somehow misunderstood by the server.
1844 CLOSE_INVALIDATE (sock);
1847 hs->contlen = contlen + contrange;
1853 /* No need to print this output if the body won't be
1854 downloaded at all, or if the original server response is
1856 logputs (LOG_VERBOSE, _("Length: "));
1859 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1860 if (contlen + contrange >= 1024)
1861 logprintf (LOG_VERBOSE, " (%s)",
1862 human_readable (contlen + contrange));
1865 if (contlen >= 1024)
1866 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1867 number_to_static_string (contlen),
1868 human_readable (contlen));
1870 logprintf (LOG_VERBOSE, _(", %s remaining"),
1871 number_to_static_string (contlen));
1875 logputs (LOG_VERBOSE,
1876 opt.ignore_length ? _("ignored") : _("unspecified"));
1878 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1880 logputs (LOG_VERBOSE, "\n");
1884 type = NULL; /* We don't need it any more. */
1886 /* Return if we have no intention of further downloading. */
1887 if (!(*dt & RETROKF) || head_only)
1889 /* In case the caller cares to look... */
1894 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1895 servers not to send body in response to a HEAD request. If
1896 you encounter such a server (more likely a broken CGI), use
1897 `--no-http-keep-alive'. */
1898 CLOSE_FINISH (sock);
1899 else if (keep_alive && skip_short_body (sock, contlen))
1900 /* Successfully skipped the body; also keep using the socket. */
1901 CLOSE_FINISH (sock);
1903 CLOSE_INVALIDATE (sock);
1904 return RETRFINISHED;
1907 /* Open the local file. */
1910 mkalldirs (*hs->local_file);
1912 rotate_backups (*hs->local_file);
1914 fp = fopen (*hs->local_file, "ab");
1915 else if (ALLOW_CLOBBER)
1916 fp = fopen (*hs->local_file, "wb");
1919 fp = fopen_excl (*hs->local_file, true);
1920 if (!fp && errno == EEXIST)
1922 /* We cannot just invent a new name and use it (which is
1923 what functions like unique_create typically do)
1924 because we told the user we'd use this name.
1925 Instead, return and retry the download. */
1926 logprintf (LOG_NOTQUIET,
1927 _("%s has sprung into existence.\n"),
1929 CLOSE_INVALIDATE (sock);
1930 return FOPEN_EXCL_ERR;
1935 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1936 CLOSE_INVALIDATE (sock);
1943 /* #### This confuses the timestamping code that checks for file
1944 size. Maybe we should save some additional information? */
1945 if (opt.save_headers)
1946 fwrite (head, 1, strlen (head), fp);
1948 /* Now we no longer need to store the response header. */
1951 /* Download the request body. */
1954 flags |= rb_read_exactly;
1955 if (hs->restval > 0 && contrange == 0)
1956 /* If the server ignored our range request, instruct fd_read_body
1957 to skip the first RESTVAL bytes of body. */
1958 flags |= rb_skip_startpos;
1959 hs->len = hs->restval;
1961 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1962 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1966 CLOSE_FINISH (sock);
1968 CLOSE_INVALIDATE (sock);
1971 /* Close or flush the file. We have to be careful to check for
1972 error here. Checking the result of fwrite() is not enough --
1973 errors could go unnoticed! */
1976 flush_res = fclose (fp);
1978 flush_res = fflush (fp);
1979 if (flush_res == EOF)
1984 return RETRFINISHED;
1987 /* The genuine HTTP loop! This is the part where the retrieval is
1988 retried, and retried, and retried, and... */
1990 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1991 int *dt, struct url *proxy)
1994 bool use_ts, got_head = false;/* time-stamping info */
1995 char *filename_plus_orig_suffix;
1996 char *local_filename = NULL;
2000 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
2001 wgint local_size = 0; /* the size of the local file */
2002 size_t filename_len;
2003 struct http_stat hstat; /* HTTP status */
2007 /* This used to be done in main(), but it's a better idea to do it
2008 here so that we don't go through the hoops if we're just using
2012 if (!wget_cookie_jar)
2013 wget_cookie_jar = cookie_jar_new ();
2014 if (opt.cookies_input && !cookies_loaded_p)
2016 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2017 cookies_loaded_p = true;
2023 /* Warn on (likely bogus) wildcard usage in HTTP. */
2024 if (opt.ftp_glob && has_wildcards_p (u->path))
2025 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2029 /* Determine the local filename. */
2030 if (local_file && *local_file)
2031 hstat.local_file = local_file;
2032 else if (local_file && !opt.output_document)
2034 *local_file = url_file_name (u);
2035 hstat.local_file = local_file;
2039 dummy = url_file_name (u);
2040 hstat.local_file = &dummy;
2041 /* be honest about where we will save the file */
2042 if (local_file && opt.output_document)
2043 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2046 if (!opt.output_document)
2047 locf = *hstat.local_file;
2049 locf = opt.output_document;
2051 hstat.referer = referer;
2053 filename_len = strlen (*hstat.local_file);
2054 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2056 if (opt.noclobber && file_exists_p (*hstat.local_file))
2058 /* If opt.noclobber is turned on and file already exists, do not
2059 retrieve the file */
2060 logprintf (LOG_VERBOSE, _("\
2061 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2062 /* If the file is there, we suppose it's retrieved OK. */
2065 /* #### Bogusness alert. */
2066 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2067 if (has_html_suffix_p (*hstat.local_file))
2075 if (opt.timestamping)
2077 bool local_dot_orig_file_exists = false;
2079 if (opt.backup_converted)
2080 /* If -K is specified, we'll act on the assumption that it was specified
2081 last time these files were downloaded as well, and instead of just
2082 comparing local file X against server file X, we'll compare local
2083 file X.orig (if extant, else X) against server file X. If -K
2084 _wasn't_ specified last time, or the server contains files called
2085 *.orig, -N will be back to not operating correctly with -k. */
2087 /* Would a single s[n]printf() call be faster? --dan
2089 Definitely not. sprintf() is horribly slow. It's a
2090 different question whether the difference between the two
2091 affects a program. Usually I'd say "no", but at one
2092 point I profiled Wget, and found that a measurable and
2093 non-negligible amount of time was lost calling sprintf()
2094 in url.c. Replacing sprintf with inline calls to
2095 strcpy() and number_to_string() made a difference.
2097 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2098 memcpy (filename_plus_orig_suffix + filename_len,
2099 ".orig", sizeof (".orig"));
2101 /* Try to stat() the .orig file. */
2102 if (stat (filename_plus_orig_suffix, &st) == 0)
2104 local_dot_orig_file_exists = 1;
2105 local_filename = filename_plus_orig_suffix;
2109 if (!local_dot_orig_file_exists)
2110 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2111 if (stat (*hstat.local_file, &st) == 0)
2112 local_filename = *hstat.local_file;
2114 if (local_filename != NULL)
2115 /* There was a local file, so we'll check later to see if the version
2116 the server has is the same version we already have, allowing us to
2122 /* Modification time granularity is 2 seconds for Windows, so
2123 increase local time by 1 second for later comparison. */
2126 local_size = st.st_size;
2130 /* Reset the counter. */
2136 /* Increment the pass counter. */
2138 sleep_between_retrievals (count);
2139 /* Get the current time string. */
2140 tms = time_str (NULL);
2141 /* Print fetch message, if opt.verbose. */
2144 char *hurl = url_string (u, true);
2148 sprintf (tmp, _("(try:%2d)"), count);
2149 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2150 tms, hurl, tmp, locf);
2152 ws_changetitle (hurl);
2157 /* Default document type is empty. However, if spider mode is
2158 on or time-stamping is employed, HEAD_ONLY commands is
2159 encoded within *dt. */
2160 if (opt.spider || (use_ts && !got_head))
2165 /* Decide whether or not to restart. */
2167 && stat (locf, &st) == 0
2168 && S_ISREG (st.st_mode))
2169 /* When -c is used, continue from on-disk size. (Can't use
2170 hstat.len even if count>1 because we don't want a failed
2171 first attempt to clobber existing data.) */
2172 hstat.restval = st.st_size;
2174 /* otherwise, continue where the previous try left off */
2175 hstat.restval = hstat.len;
2179 /* Decide whether to send the no-cache directive. We send it in
2181 a) we're using a proxy, and we're past our first retrieval.
2182 Some proxies are notorious for caching incomplete data, so
2183 we require a fresh get.
2184 b) caching is explicitly inhibited. */
2185 if ((proxy && count > 1) /* a */
2186 || !opt.allow_cache /* b */
2188 *dt |= SEND_NOCACHE;
2190 *dt &= ~SEND_NOCACHE;
2192 /* Try fetching the document, or at least its head. */
2193 err = gethttp (u, &hstat, dt, proxy);
2195 /* It's unfortunate that wget determines the local filename before finding
2196 out the Content-Type of the file. Barring a major restructuring of the
2197 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2198 *hstat.local_file to tack on ".html". */
2199 if (!opt.output_document)
2200 locf = *hstat.local_file;
2203 tms = time_str (NULL);
2204 /* Get the new location (with or without the redirection). */
2206 *newloc = xstrdup (hstat.newloc);
2209 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2210 case CONERROR: case READERR: case WRITEFAILED:
2211 case RANGEERR: case FOPEN_EXCL_ERR:
2212 /* Non-fatal errors continue executing the loop, which will
2213 bring them to "while" statement at the end, to judge
2214 whether the number of tries was exceeded. */
2215 free_hstat (&hstat);
2216 printwhat (count, opt.ntry);
2217 if (err == FOPEN_EXCL_ERR)
2219 /* Re-determine the file name. */
2220 if (local_file && *local_file)
2222 xfree (*local_file);
2223 *local_file = url_file_name (u);
2224 hstat.local_file = local_file;
2229 dummy = url_file_name (u);
2230 hstat.local_file = &dummy;
2232 /* be honest about where we will save the file */
2233 if (local_file && opt.output_document)
2234 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2235 if (!opt.output_document)
2236 locf = *hstat.local_file;
2238 locf = opt.output_document;
2241 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2242 case SSLINITFAILED: case CONTNOTSUPPORTED:
2243 /* Fatal errors just return from the function. */
2244 free_hstat (&hstat);
2247 case FWRITEERR: case FOPENERR:
2248 /* Another fatal error. */
2249 logputs (LOG_VERBOSE, "\n");
2250 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2251 *hstat.local_file, strerror (errno));
2252 free_hstat (&hstat);
2256 /* Another fatal error. */
2257 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2258 free_hstat (&hstat);
2262 /* Return the new location to the caller. */
2265 logprintf (LOG_NOTQUIET,
2266 _("ERROR: Redirection (%d) without location.\n"),
2268 free_hstat (&hstat);
2272 free_hstat (&hstat);
2276 /* The file was already fully retrieved. */
2277 free_hstat (&hstat);
2281 /* Deal with you later. */
2284 /* All possibilities should have been exhausted. */
2287 if (!(*dt & RETROKF))
2291 /* #### Ugly ugly ugly! */
2292 char *hurl = url_string (u, true);
2293 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2296 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2297 tms, hstat.statcode, escnonprint (hstat.error));
2298 logputs (LOG_VERBOSE, "\n");
2299 free_hstat (&hstat);
2304 /* Did we get the time-stamp? */
2307 if (opt.timestamping && !hstat.remote_time)
2309 logputs (LOG_NOTQUIET, _("\
2310 Last-modified header missing -- time-stamps turned off.\n"));
2312 else if (hstat.remote_time)
2314 /* Convert the date-string into struct tm. */
2315 tmr = http_atotm (hstat.remote_time);
2316 if (tmr == (time_t) (-1))
2317 logputs (LOG_VERBOSE, _("\
2318 Last-modified header invalid -- time-stamp ignored.\n"));
2322 /* The time-stamping section. */
2327 use_ts = false; /* no more time-stamping */
2328 count = 0; /* the retrieve count for HEAD is
2330 if (hstat.remote_time && tmr != (time_t) (-1))
2332 /* Now time-stamping can be used validly. Time-stamping
2333 means that if the sizes of the local and remote file
2334 match, and local file is newer than the remote file,
2335 it will not be retrieved. Otherwise, the normal
2336 download procedure is resumed. */
2338 (hstat.contlen == -1 || local_size == hstat.contlen))
2340 logprintf (LOG_VERBOSE, _("\
2341 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2343 free_hstat (&hstat);
2347 else if (tml >= tmr)
2348 logprintf (LOG_VERBOSE, _("\
2349 The sizes do not match (local %s) -- retrieving.\n"),
2350 number_to_static_string (local_size));
2352 logputs (LOG_VERBOSE,
2353 _("Remote file is newer, retrieving.\n"));
2355 free_hstat (&hstat);
2358 if ((tmr != (time_t) (-1))
2360 && ((hstat.len == hstat.contlen) ||
2361 ((hstat.res == 0) &&
2362 ((hstat.contlen == -1) ||
2363 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2365 /* #### This code repeats in http.c and ftp.c. Move it to a
2367 const char *fl = NULL;
2368 if (opt.output_document)
2370 if (output_stream_regular)
2371 fl = opt.output_document;
2374 fl = *hstat.local_file;
2378 /* End of time-stamping section. */
2382 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2383 escnonprint (hstat.error));
2388 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2389 total_download_time += hstat.dltime;
2391 if (hstat.len == hstat.contlen)
2395 logprintf (LOG_VERBOSE,
2396 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2398 number_to_static_string (hstat.len),
2399 number_to_static_string (hstat.contlen));
2400 logprintf (LOG_NONVERBOSE,
2401 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2403 number_to_static_string (hstat.len),
2404 number_to_static_string (hstat.contlen),
2408 total_downloaded_bytes += hstat.len;
2410 /* Remember that we downloaded the file for later ".orig" code. */
2411 if (*dt & ADDED_HTML_EXTENSION)
2412 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2414 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2416 free_hstat (&hstat);
2420 else if (hstat.res == 0) /* No read error */
2422 if (hstat.contlen == -1) /* We don't know how much we were supposed
2423 to get, so assume we succeeded. */
2427 logprintf (LOG_VERBOSE,
2428 _("%s (%s) - `%s' saved [%s]\n\n"),
2430 number_to_static_string (hstat.len));
2431 logprintf (LOG_NONVERBOSE,
2432 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2433 tms, u->url, number_to_static_string (hstat.len),
2437 total_downloaded_bytes += hstat.len;
2439 /* Remember that we downloaded the file for later ".orig" code. */
2440 if (*dt & ADDED_HTML_EXTENSION)
2441 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2443 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2445 free_hstat (&hstat);
2449 else if (hstat.len < hstat.contlen) /* meaning we lost the
2450 connection too soon */
2452 logprintf (LOG_VERBOSE,
2453 _("%s (%s) - Connection closed at byte %s. "),
2454 tms, tmrate, number_to_static_string (hstat.len));
2455 printwhat (count, opt.ntry);
2456 free_hstat (&hstat);
2459 else if (!opt.kill_longer) /* meaning we got more than expected */
2461 logprintf (LOG_VERBOSE,
2462 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2464 number_to_static_string (hstat.len),
2465 number_to_static_string (hstat.contlen));
2466 logprintf (LOG_NONVERBOSE,
2467 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2469 number_to_static_string (hstat.len),
2470 number_to_static_string (hstat.contlen),
2473 total_downloaded_bytes += hstat.len;
2475 /* Remember that we downloaded the file for later ".orig" code. */
2476 if (*dt & ADDED_HTML_EXTENSION)
2477 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2479 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2481 free_hstat (&hstat);
2485 else /* the same, but not accepted */
2487 logprintf (LOG_VERBOSE,
2488 _("%s (%s) - Connection closed at byte %s/%s. "),
2490 number_to_static_string (hstat.len),
2491 number_to_static_string (hstat.contlen));
2492 printwhat (count, opt.ntry);
2493 free_hstat (&hstat);
2497 else /* now hstat.res can only be -1 */
2499 if (hstat.contlen == -1)
2501 logprintf (LOG_VERBOSE,
2502 _("%s (%s) - Read error at byte %s (%s)."),
2503 tms, tmrate, number_to_static_string (hstat.len),
2505 printwhat (count, opt.ntry);
2506 free_hstat (&hstat);
2509 else /* hstat.res == -1 and contlen is given */
2511 logprintf (LOG_VERBOSE,
2512 _("%s (%s) - Read error at byte %s/%s (%s). "),
2514 number_to_static_string (hstat.len),
2515 number_to_static_string (hstat.contlen),
2517 printwhat (count, opt.ntry);
2518 free_hstat (&hstat);
2524 while (!opt.ntry || (count < opt.ntry));
2528 /* Check whether the result of strptime() indicates success.
2529 strptime() returns the pointer to how far it got to in the string.
2530 The processing has been successful if the string is at `GMT' or
2531 `+X', or at the end of the string.
2533 In extended regexp parlance, the function returns 1 if P matches
2534 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2535 can return) is considered a failure and 0 is returned. */
2537 check_end (const char *p)
2541 while (ISSPACE (*p))
2544 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2545 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2551 /* Convert the textual specification of time in TIME_STRING to the
2552 number of seconds since the Epoch.
2554 TIME_STRING can be in any of the three formats RFC2616 allows the
2555 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2556 as well as the time format used in the Set-Cookie header.
2557 Timezones are ignored, and should be GMT.
2559 Return the computed time_t representation, or -1 if the conversion
2562 This function uses strptime with various string formats for parsing
2563 TIME_STRING. This results in a parser that is not as lenient in
2564 interpreting TIME_STRING as I would like it to be. Being based on
2565 strptime, it always allows shortened months, one-digit days, etc.,
2566 but due to the multitude of formats in which time can be
2567 represented, an ideal HTTP time parser would be even more
2568 forgiving. It should completely ignore things like week days and
2569 concentrate only on the various forms of representing years,
2570 months, days, hours, minutes, and seconds. For example, it would
2571 be nice if it accepted ISO 8601 out of the box.
2573 I've investigated free and PD code for this purpose, but none was
2574 usable. getdate was big and unwieldy, and had potential copyright
2575 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2576 distributed with phttpd, is excellent, but we cannot use it because
2577 it is not assigned to the FSF. So I stuck it with strptime. */
2580 http_atotm (const char *time_string)
2582 /* NOTE: Solaris strptime man page claims that %n and %t match white
2583 space, but that's not universally available. Instead, we simply
2584 use ` ' to mean "skip all WS", which works under all strptime
2585 implementations I've tested. */
2587 static const char *time_formats[] = {
2588 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2589 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2590 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2591 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2592 (used in Set-Cookie, defined in the
2593 Netscape cookie specification.) */
2595 const char *oldlocale;
2597 time_t ret = (time_t) -1;
2599 /* Solaris strptime fails to recognize English month names in
2600 non-English locales, which we work around by temporarily setting
2601 locale to C before invoking strptime. */
2602 oldlocale = setlocale (LC_TIME, NULL);
2603 setlocale (LC_TIME, "C");
2605 for (i = 0; i < countof (time_formats); i++)
2609 /* Some versions of strptime use the existing contents of struct
2610 tm to recalculate the date according to format. Zero it out
2611 to prevent stack garbage from influencing strptime. */
2614 if (check_end (strptime (time_string, time_formats[i], &t)))
2621 /* Restore the previous locale. */
2622 setlocale (LC_TIME, oldlocale);
2627 /* Authorization support: We support three authorization schemes:
2629 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2631 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2632 consisting of answering to the server's challenge with the proper
2635 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2636 Stenberg for libcurl. Like digest, NTLM is based on a
2637 challenge-response mechanism, but unlike digest, it is non-standard
2638 (authenticates TCP connections rather than requests), undocumented
2639 and Microsoft-specific. */
2641 /* Create the authentication header contents for the `Basic' scheme.
2642 This is done by encoding the string "USER:PASS" to base64 and
2643 prepending the string "Basic " in front of it. */
2646 basic_authentication_encode (const char *user, const char *passwd)
2649 int len1 = strlen (user) + 1 + strlen (passwd);
2651 t1 = (char *)alloca (len1 + 1);
2652 sprintf (t1, "%s:%s", user, passwd);
2654 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2655 base64_encode (t1, len1, t2);
2657 return concat_strings ("Basic ", t2, (char *) 0);
2660 #define SKIP_WS(x) do { \
2661 while (ISSPACE (*(x))) \
2665 #ifdef ENABLE_DIGEST
2666 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2667 of a field in such a header. If the field is the one specified by
2668 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2669 digest authorization code), extract its value in the (char*)
2670 variable pointed by RET. Returns negative on a malformed header,
2671 or number of bytes that have been parsed by this call. */
2673 extract_header_attr (const char *au, const char *attr_name, char **ret)
2676 const char *cp = au;
2678 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2680 cp += strlen (attr_name);
2693 for (ep = cp; *ep && *ep != '\"'; ep++)
2698 *ret = strdupdelim (cp, ep);
2705 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2706 an array of 16 bytes containing the hash keys, and BUF should be a
2707 buffer of 33 writable characters (32 for hex digits plus one for
2708 zero termination). */
2710 dump_hash (char *buf, const unsigned char *hash)
2714 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2716 *buf++ = XNUM_TO_digit (*hash >> 4);
2717 *buf++ = XNUM_TO_digit (*hash & 0xf);
2722 /* Take the line apart to find the challenge, and compose a digest
2723 authorization header. See RFC2069 section 2.1.2. */
2725 digest_authentication_encode (const char *au, const char *user,
2726 const char *passwd, const char *method,
2729 static char *realm, *opaque, *nonce;
2734 { "realm", &realm },
2735 { "opaque", &opaque },
2740 realm = opaque = nonce = NULL;
2742 au += 6; /* skip over `Digest' */
2748 for (i = 0; i < countof (options); i++)
2750 int skip = extract_header_attr (au, options[i].name,
2751 options[i].variable);
2755 xfree_null (opaque);
2765 if (i == countof (options))
2767 while (*au && *au != '=')
2775 while (*au && *au != '\"')
2782 while (*au && *au != ',')
2787 if (!realm || !nonce || !user || !passwd || !path || !method)
2790 xfree_null (opaque);
2795 /* Calculate the digest value. */
2797 ALLOCA_MD5_CONTEXT (ctx);
2798 unsigned char hash[MD5_HASHLEN];
2799 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2800 char response_digest[MD5_HASHLEN * 2 + 1];
2802 /* A1BUF = H(user ":" realm ":" password) */
2804 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2805 gen_md5_update ((unsigned char *)":", 1, ctx);
2806 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2807 gen_md5_update ((unsigned char *)":", 1, ctx);
2808 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2809 gen_md5_finish (ctx, hash);
2810 dump_hash (a1buf, hash);
2812 /* A2BUF = H(method ":" path) */
2814 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2815 gen_md5_update ((unsigned char *)":", 1, ctx);
2816 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2817 gen_md5_finish (ctx, hash);
2818 dump_hash (a2buf, hash);
2820 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2822 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2823 gen_md5_update ((unsigned char *)":", 1, ctx);
2824 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2825 gen_md5_update ((unsigned char *)":", 1, ctx);
2826 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2827 gen_md5_finish (ctx, hash);
2828 dump_hash (response_digest, hash);
2830 res = xmalloc (strlen (user)
2835 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2836 + (opaque ? strlen (opaque) : 0)
2838 sprintf (res, "Digest \
2839 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2840 user, realm, nonce, path, response_digest);
2843 char *p = res + strlen (res);
2844 strcat (p, ", opaque=\"");
2851 #endif /* ENABLE_DIGEST */
2853 /* Computing the size of a string literal must take into account that
2854 value returned by sizeof includes the terminating \0. */
2855 #define STRSIZE(literal) (sizeof (literal) - 1)
2857 /* Whether chars in [b, e) begin with the literal string provided as
2858 first argument and are followed by whitespace or terminating \0.
2859 The comparison is case-insensitive. */
2860 #define STARTS(literal, b, e) \
2861 ((e) - (b) >= STRSIZE (literal) \
2862 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2863 && ((e) - (b) == STRSIZE (literal) \
2864 || ISSPACE (b[STRSIZE (literal)])))
2867 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2869 return STARTS ("Basic", hdrbeg, hdrend)
2870 #ifdef ENABLE_DIGEST
2871 || STARTS ("Digest", hdrbeg, hdrend)
2874 || STARTS ("NTLM", hdrbeg, hdrend)
2881 /* Create the HTTP authorization request header. When the
2882 `WWW-Authenticate' response header is seen, according to the
2883 authorization scheme specified in that header (`Basic' and `Digest'
2884 are supported by the current implementation), produce an
2885 appropriate HTTP authorization request header. */
2887 create_authorization_line (const char *au, const char *user,
2888 const char *passwd, const char *method,
2889 const char *path, bool *finished)
2891 /* We are called only with known schemes, so we can dispatch on the
2893 switch (TOUPPER (*au))
2895 case 'B': /* Basic */
2897 return basic_authentication_encode (user, passwd);
2898 #ifdef ENABLE_DIGEST
2899 case 'D': /* Digest */
2901 return digest_authentication_encode (au, user, passwd, method, path);
2904 case 'N': /* NTLM */
2905 if (!ntlm_input (&pconn.ntlm, au))
2910 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2913 /* We shouldn't get here -- this function should be only called
2914 with values approved by known_authentication_scheme_p. */
2922 if (wget_cookie_jar)
2923 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2929 xfree_null (pconn.host);
2930 if (wget_cookie_jar)
2931 cookie_jar_delete (wget_cookie_jar);