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"));
420 response_head_terminator (const char *hunk, int oldlen, int peeklen)
422 const char *start, *end;
424 /* If at first peek, verify whether HUNK starts with "HTTP". If
425 not, this is a HTTP/0.9 request and we must bail out without
427 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
433 start = hunk + oldlen - 4;
434 end = hunk + oldlen + peeklen;
436 for (; start < end - 1; start++)
443 if (start[1] == '\n')
449 /* The maximum size of a single HTTP response we care to read. This
450 is not meant to impose an arbitrary limit, but to protect the user
451 from Wget slurping up available memory upon encountering malicious
452 or buggy server output. Define it to 0 to remove the limit. */
454 #define HTTP_RESPONSE_MAX_SIZE 65536
456 /* Read the HTTP request head from FD and return it. The error
457 conditions are the same as with fd_read_hunk.
459 To support HTTP/0.9 responses, this function tries to make sure
460 that the data begins with "HTTP". If this is not the case, no data
461 is read and an empty request is returned, so that the remaining
462 data can be treated as body. */
465 read_http_response_head (int fd)
467 return fd_read_hunk (fd, response_head_terminator, 512,
468 HTTP_RESPONSE_MAX_SIZE);
472 /* The response data. */
475 /* The array of pointers that indicate where each header starts.
476 For example, given this HTTP response:
483 The headers are located like this:
485 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
487 headers[0] headers[1] headers[2] headers[3]
489 I.e. headers[0] points to the beginning of the request,
490 headers[1] points to the end of the first header and the
491 beginning of the second one, etc. */
493 const char **headers;
496 /* Create a new response object from the text of the HTTP response,
497 available in HEAD. That text is automatically split into
498 constituent header lines for fast retrieval using
501 static struct response *
502 resp_new (const char *head)
507 struct response *resp = xnew0 (struct response);
512 /* Empty head means that we're dealing with a headerless
513 (HTTP/0.9) response. In that case, don't set HEADERS at
518 /* Split HEAD into header lines, so that resp_header_* functions
519 don't need to do this over and over again. */
525 DO_REALLOC (resp->headers, size, count + 1, const char *);
526 resp->headers[count++] = hdr;
528 /* Break upon encountering an empty line. */
529 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
532 /* Find the end of HDR, including continuations. */
535 const char *end = strchr (hdr, '\n');
541 while (*hdr == ' ' || *hdr == '\t');
543 DO_REALLOC (resp->headers, size, count + 1, const char *);
544 resp->headers[count] = NULL;
549 /* Locate the header named NAME in the request data, starting with
550 position START. This allows the code to loop through the request
551 data, filtering for all requests of a given name. Returns the
552 found position, or -1 for failure. The code that uses this
553 function typically looks like this:
555 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
556 ... do something with header ...
558 If you only care about one header, use resp_header_get instead of
562 resp_header_locate (const struct response *resp, const char *name, int start,
563 const char **begptr, const char **endptr)
566 const char **headers = resp->headers;
569 if (!headers || !headers[1])
572 name_len = strlen (name);
578 for (; headers[i + 1]; i++)
580 const char *b = headers[i];
581 const char *e = headers[i + 1];
583 && b[name_len] == ':'
584 && 0 == strncasecmp (b, name, name_len))
587 while (b < e && ISSPACE (*b))
589 while (b < e && ISSPACE (e[-1]))
599 /* Find and retrieve the header named NAME in the request data. If
600 found, set *BEGPTR to its starting, and *ENDPTR to its ending
601 position, and return true. Otherwise return false.
603 This function is used as a building block for resp_header_copy
604 and resp_header_strdup. */
607 resp_header_get (const struct response *resp, const char *name,
608 const char **begptr, const char **endptr)
610 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
614 /* Copy the response header named NAME to buffer BUF, no longer than
615 BUFSIZE (BUFSIZE includes the terminating 0). If the header
616 exists, true is returned, false otherwise. If there should be no
617 limit on the size of the header, use resp_header_strdup instead.
619 If BUFSIZE is 0, no data is copied, but the boolean indication of
620 whether the header is present is still returned. */
623 resp_header_copy (const struct response *resp, const char *name,
624 char *buf, int bufsize)
627 if (!resp_header_get (resp, name, &b, &e))
631 int len = MIN (e - b, bufsize - 1);
632 memcpy (buf, b, len);
638 /* Return the value of header named NAME in RESP, allocated with
639 malloc. If such a header does not exist in RESP, return NULL. */
642 resp_header_strdup (const struct response *resp, const char *name)
645 if (!resp_header_get (resp, name, &b, &e))
647 return strdupdelim (b, e);
650 /* Parse the HTTP status line, which is of format:
652 HTTP-Version SP Status-Code SP Reason-Phrase
654 The function returns the status-code, or -1 if the status line
655 appears malformed. The pointer to "reason-phrase" message is
656 returned in *MESSAGE. */
659 resp_status (const struct response *resp, char **message)
666 /* For a HTTP/0.9 response, assume status 200. */
668 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
672 p = resp->headers[0];
673 end = resp->headers[1];
679 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
683 /* Match the HTTP version. This is optional because Gnutella
684 servers have been reported to not specify HTTP version. */
685 if (p < end && *p == '/')
688 while (p < end && ISDIGIT (*p))
690 if (p < end && *p == '.')
692 while (p < end && ISDIGIT (*p))
696 while (p < end && ISSPACE (*p))
698 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
701 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
706 while (p < end && ISSPACE (*p))
708 while (p < end && ISSPACE (end[-1]))
710 *message = strdupdelim (p, end);
716 /* Release the resources used by RESP. */
719 resp_free (struct response *resp)
721 xfree_null (resp->headers);
725 /* Print the server response, line by line, omitting the trailing CRLF
726 from individual header lines, and prefixed with PREFIX. */
729 print_server_response (const struct response *resp, const char *prefix)
734 for (i = 0; resp->headers[i + 1]; i++)
736 const char *b = resp->headers[i];
737 const char *e = resp->headers[i + 1];
739 if (b < e && e[-1] == '\n')
741 if (b < e && e[-1] == '\r')
743 /* This is safe even on printfs with broken handling of "%.<n>s"
744 because resp->headers ends with \0. */
745 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
749 /* Parse the `Content-Range' header and extract the information it
750 contains. Returns true if successful, false otherwise. */
752 parse_content_range (const char *hdr, wgint *first_byte_ptr,
753 wgint *last_byte_ptr, wgint *entity_length_ptr)
757 /* Ancient versions of Netscape proxy server, presumably predating
758 rfc2068, sent out `Content-Range' without the "bytes"
760 if (0 == strncasecmp (hdr, "bytes", 5))
763 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
767 while (ISSPACE (*hdr))
774 for (num = 0; ISDIGIT (*hdr); hdr++)
775 num = 10 * num + (*hdr - '0');
776 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
778 *first_byte_ptr = num;
780 for (num = 0; ISDIGIT (*hdr); hdr++)
781 num = 10 * num + (*hdr - '0');
782 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
784 *last_byte_ptr = num;
786 for (num = 0; ISDIGIT (*hdr); hdr++)
787 num = 10 * num + (*hdr - '0');
788 *entity_length_ptr = num;
792 /* Read the body of the request, but don't store it anywhere and don't
793 display a progress gauge. This is useful for reading the bodies of
794 administrative responses to which we will soon issue another
795 request. The response is not useful to the user, but reading it
796 allows us to continue using the same connection to the server.
798 If reading fails, false is returned, true otherwise. In debug
799 mode, the body is displayed for debugging purposes. */
802 skip_short_body (int fd, wgint contlen)
805 SKIP_SIZE = 512, /* size of the download buffer */
806 SKIP_THRESHOLD = 4096 /* the largest size we read */
808 char dlbuf[SKIP_SIZE + 1];
809 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
811 /* We shouldn't get here with unknown contlen. (This will change
812 with HTTP/1.1, which supports "chunked" transfer.) */
813 assert (contlen != -1);
815 /* If the body is too large, it makes more sense to simply close the
816 connection than to try to read the body. */
817 if (contlen > SKIP_THRESHOLD)
820 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
824 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
827 /* Don't normally report the error since this is an
828 optimization that should be invisible to the user. */
829 DEBUGP (("] aborting (%s).\n",
830 ret < 0 ? strerror (errno) : "EOF received"));
834 /* Safe even if %.*s bogusly expects terminating \0 because
835 we've zero-terminated dlbuf above. */
836 DEBUGP (("%.*s", ret, dlbuf));
839 DEBUGP (("] done.\n"));
843 /* Persistent connections. Currently, we cache the most recently used
844 connection as persistent, provided that the HTTP server agrees to
845 make it such. The persistence data is stored in the variables
846 below. Ideally, it should be possible to cache an arbitrary fixed
847 number of these connections. */
849 /* Whether a persistent connection is active. */
850 static bool pconn_active;
853 /* The socket of the connection. */
856 /* Host and port of the currently active persistent connection. */
860 /* Whether a ssl handshake has occoured on this connection. */
863 /* Whether the connection was authorized. This is only done by
864 NTLM, which authorizes *connections* rather than individual
865 requests. (That practice is peculiar for HTTP, but it is a
866 useful optimization.) */
870 /* NTLM data of the current connection. */
871 struct ntlmdata ntlm;
875 /* Mark the persistent connection as invalid and free the resources it
876 uses. This is used by the CLOSE_* macros after they forcefully
877 close a registered persistent connection. */
880 invalidate_persistent (void)
882 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
883 pconn_active = false;
884 fd_close (pconn.socket);
889 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
890 persistent. This will enable someone to use the same connection
891 later. In the context of HTTP, this must be called only AFTER the
892 response has been received and the server has promised that the
893 connection will remain alive.
895 If a previous connection was persistent, it is closed. */
898 register_persistent (const char *host, int port, int fd, bool ssl)
902 if (pconn.socket == fd)
904 /* The connection FD is already registered. */
909 /* The old persistent connection is still active; close it
910 first. This situation arises whenever a persistent
911 connection exists, but we then connect to a different
912 host, and try to register a persistent connection to that
914 invalidate_persistent ();
920 pconn.host = xstrdup (host);
923 pconn.authorized = false;
925 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
928 /* Return true if a persistent connection is available for connecting
932 persistent_available_p (const char *host, int port, bool ssl,
933 bool *host_lookup_failed)
935 /* First, check whether a persistent connection is active at all. */
939 /* If we want SSL and the last connection wasn't or vice versa,
940 don't use it. Checking for host and port is not enough because
941 HTTP and HTTPS can apparently coexist on the same port. */
942 if (ssl != pconn.ssl)
945 /* If we're not connecting to the same port, we're not interested. */
946 if (port != pconn.port)
949 /* If the host is the same, we're in business. If not, there is
950 still hope -- read below. */
951 if (0 != strcasecmp (host, pconn.host))
953 /* Check if pconn.socket is talking to HOST under another name.
954 This happens often when both sites are virtual hosts
955 distinguished only by name and served by the same network
956 interface, and hence the same web server (possibly set up by
957 the ISP and serving many different web sites). This
958 admittedly unconventional optimization does not contradict
959 HTTP and works well with popular server software. */
963 struct address_list *al;
966 /* Don't try to talk to two different SSL sites over the same
967 secure connection! (Besides, it's not clear that
968 name-based virtual hosting is even possible with SSL.) */
971 /* If pconn.socket's peer is one of the IP addresses HOST
972 resolves to, pconn.socket is for all intents and purposes
973 already talking to HOST. */
975 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
977 /* Can't get the peer's address -- something must be very
978 wrong with the connection. */
979 invalidate_persistent ();
982 al = lookup_host (host, 0);
985 *host_lookup_failed = true;
989 found = address_list_contains (al, &ip);
990 address_list_release (al);
995 /* The persistent connection's peer address was found among the
996 addresses HOST resolved to; therefore, pconn.sock is in fact
997 already talking to HOST -- no need to reconnect. */
1000 /* Finally, check whether the connection is still open. This is
1001 important because most server implement a liberal (short) timeout
1002 on persistent connections. Wget can of course always reconnect
1003 if the connection doesn't work out, but it's nicer to know in
1004 advance. This test is a logical followup of the first test, but
1005 is "expensive" and therefore placed at the end of the list. */
1007 if (!test_socket_open (pconn.socket))
1009 /* Oops, the socket is no longer open. Now that we know that,
1010 let's invalidate the persistent connection before returning
1012 invalidate_persistent ();
1019 /* The idea behind these two CLOSE macros is to distinguish between
1020 two cases: one when the job we've been doing is finished, and we
1021 want to close the connection and leave, and two when something is
1022 seriously wrong and we're closing the connection as part of
1025 In case of keep_alive, CLOSE_FINISH should leave the connection
1026 open, while CLOSE_INVALIDATE should still close it.
1028 Note that the semantics of the flag `keep_alive' is "this
1029 connection *will* be reused (the server has promised not to close
1030 the connection once we're done)", while the semantics of
1031 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1032 active, registered connection". */
1034 #define CLOSE_FINISH(fd) do { \
1037 if (pconn_active && (fd) == pconn.socket) \
1038 invalidate_persistent (); \
1047 #define CLOSE_INVALIDATE(fd) do { \
1048 if (pconn_active && (fd) == pconn.socket) \
1049 invalidate_persistent (); \
1057 wgint len; /* received length */
1058 wgint contlen; /* expected length */
1059 wgint restval; /* the restart value */
1060 int res; /* the result of last read */
1061 char *newloc; /* new location (redirection) */
1062 char *remote_time; /* remote time-stamp string */
1063 char *error; /* textual HTTP error */
1064 int statcode; /* status code */
1065 wgint rd_size; /* amount of data read from socket */
1066 double dltime; /* time it took to download the data */
1067 const char *referer; /* value of the referer header. */
1068 char **local_file; /* local file. */
1072 free_hstat (struct http_stat *hs)
1074 xfree_null (hs->newloc);
1075 xfree_null (hs->remote_time);
1076 xfree_null (hs->error);
1078 /* Guard against being called twice. */
1080 hs->remote_time = NULL;
1084 static char *create_authorization_line (const char *, const char *,
1085 const char *, const char *,
1086 const char *, bool *);
1087 static char *basic_authentication_encode (const char *, const char *);
1088 static bool known_authentication_scheme_p (const char *, const char *);
1090 #define BEGINS_WITH(line, string_constant) \
1091 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1092 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1093 || !line[sizeof (string_constant) - 1]))
1095 #define SET_USER_AGENT(req) do { \
1096 if (!opt.useragent) \
1097 request_set_header (req, "User-Agent", \
1098 aprintf ("Wget/%s", version_string), rel_value); \
1099 else if (*opt.useragent) \
1100 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1103 /* The flags that allow clobbering the file (opening with "wb").
1104 Defined here to avoid repetition later. #### This will require
1106 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1107 || opt.dirstruct || opt.output_document)
1109 /* Retrieve a document through HTTP protocol. It recognizes status
1110 code, and correctly handles redirections. It closes the network
1111 socket. If it receives an error from the functions below it, it
1112 will print it if there is enough information to do so (almost
1113 always), returning the error to the caller (i.e. http_loop).
1115 Various HTTP parameters are stored to hs.
1117 If PROXY is non-NULL, the connection will be made to the proxy
1118 server, and u->url will be requested. */
1120 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1122 struct request *req;
1125 char *user, *passwd;
1129 wgint contlen, contrange;
1136 /* Set to 1 when the authorization has failed permanently and should
1137 not be tried again. */
1138 bool auth_finished = false;
1140 /* Whether NTLM authentication is used for this request. */
1141 bool ntlm_seen = false;
1143 /* Whether our connection to the remote host is through SSL. */
1144 bool using_ssl = false;
1146 /* Whether a HEAD request will be issued (as opposed to GET or
1148 bool head_only = !!(*dt & HEAD_ONLY);
1151 struct response *resp;
1155 /* Whether this connection will be kept alive after the HTTP request
1159 /* Whether keep-alive should be inhibited.
1161 RFC 2068 requests that 1.0 clients not send keep-alive requests
1162 to proxies. This is because many 1.0 proxies do not interpret
1163 the Connection header and transfer it to the remote server,
1164 causing it to not close the connection and leave both the proxy
1165 and the client hanging. */
1166 bool inhibit_keep_alive =
1167 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1169 /* Headers sent when using POST. */
1170 wgint post_data_size = 0;
1172 bool host_lookup_failed = false;
1175 if (u->scheme == SCHEME_HTTPS)
1177 /* Initialize the SSL context. After this has once been done,
1178 it becomes a no-op. */
1181 scheme_disable (SCHEME_HTTPS);
1182 logprintf (LOG_NOTQUIET,
1183 _("Disabling SSL due to encountered errors.\n"));
1184 return SSLINITFAILED;
1187 #endif /* HAVE_SSL */
1190 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1191 know the local filename so we can save to it. */
1192 assert (*hs->local_file != NULL);
1194 /* Initialize certain elements of struct http_stat. */
1199 hs->remote_time = NULL;
1204 /* Prepare the request to send. */
1206 req = request_new ();
1209 const char *meth = "GET";
1212 else if (opt.post_file_name || opt.post_data)
1214 /* Use the full path, i.e. one that includes the leading slash and
1215 the query string. E.g. if u->path is "foo/bar" and u->query is
1216 "param=value", full_path will be "/foo/bar?param=value". */
1219 /* When using SSL over proxy, CONNECT establishes a direct
1220 connection to the HTTPS server. Therefore use the same
1221 argument as when talking to the server directly. */
1222 && u->scheme != SCHEME_HTTPS
1225 meth_arg = xstrdup (u->url);
1227 meth_arg = url_full_path (u);
1228 request_set_method (req, meth, meth_arg);
1231 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1232 if (*dt & SEND_NOCACHE)
1233 request_set_header (req, "Pragma", "no-cache", rel_none);
1235 request_set_header (req, "Range",
1236 aprintf ("bytes=%s-",
1237 number_to_static_string (hs->restval)),
1239 SET_USER_AGENT (req);
1240 request_set_header (req, "Accept", "*/*", rel_none);
1242 /* Find the username and password for authentication. */
1245 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1246 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1247 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1251 /* We have the username and the password, but haven't tried
1252 any authorization yet. Let's see if the "Basic" method
1253 works. If not, we'll come back here and construct a
1254 proper authorization method with the right challenges.
1256 If we didn't employ this kind of logic, every URL that
1257 requires authorization would have to be processed twice,
1258 which is very suboptimal and generates a bunch of false
1259 "unauthorized" errors in the server log.
1261 #### But this logic also has a serious problem when used
1262 with stronger authentications: we *first* transmit the
1263 username and the password in clear text, and *then* attempt a
1264 stronger authentication scheme. That cannot be right! We
1265 are only fortunate that almost everyone still uses the
1266 `Basic' scheme anyway.
1268 There should be an option to prevent this from happening, for
1269 those who use strong authentication schemes and value their
1271 request_set_header (req, "Authorization",
1272 basic_authentication_encode (user, passwd),
1279 char *proxy_user, *proxy_passwd;
1280 /* For normal username and password, URL components override
1281 command-line/wgetrc parameters. With proxy
1282 authentication, it's the reverse, because proxy URLs are
1283 normally the "permanent" ones, so command-line args
1284 should take precedence. */
1285 if (opt.proxy_user && opt.proxy_passwd)
1287 proxy_user = opt.proxy_user;
1288 proxy_passwd = opt.proxy_passwd;
1292 proxy_user = proxy->user;
1293 proxy_passwd = proxy->passwd;
1295 /* #### This does not appear right. Can't the proxy request,
1296 say, `Digest' authentication? */
1297 if (proxy_user && proxy_passwd)
1298 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1300 /* If we're using a proxy, we will be connecting to the proxy
1304 /* Proxy authorization over SSL is handled below. */
1306 if (u->scheme != SCHEME_HTTPS)
1308 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1311 /* Generate the Host header, HOST:PORT. Take into account that:
1313 - Broken server-side software often doesn't recognize the PORT
1314 argument, so we must generate "Host: www.server.com" instead of
1315 "Host: www.server.com:80" (and likewise for https port).
1317 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1318 becomes ambiguous and needs to be rewritten as "Host:
1319 [3ffe:8100:200:2::2]:1234". */
1321 /* Formats arranged for hfmt[add_port][add_squares]. */
1322 static const char *hfmt[][2] = {
1323 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1325 int add_port = u->port != scheme_default_port (u->scheme);
1326 int add_squares = strchr (u->host, ':') != NULL;
1327 request_set_header (req, "Host",
1328 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1332 if (!inhibit_keep_alive)
1333 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1336 request_set_header (req, "Cookie",
1337 cookie_header (wget_cookie_jar,
1338 u->host, u->port, u->path,
1340 u->scheme == SCHEME_HTTPS
1347 if (opt.post_data || opt.post_file_name)
1349 request_set_header (req, "Content-Type",
1350 "application/x-www-form-urlencoded", rel_none);
1352 post_data_size = strlen (opt.post_data);
1355 post_data_size = file_size (opt.post_file_name);
1356 if (post_data_size == -1)
1358 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1359 opt.post_file_name, strerror (errno));
1363 request_set_header (req, "Content-Length",
1364 xstrdup (number_to_static_string (post_data_size)),
1368 /* Add the user headers. */
1369 if (opt.user_headers)
1372 for (i = 0; opt.user_headers[i]; i++)
1373 request_set_user_header (req, opt.user_headers[i]);
1377 /* We need to come back here when the initial attempt to retrieve
1378 without authorization header fails. (Expected to happen at least
1379 for the Digest authorization scheme.) */
1383 /* Establish the connection. */
1385 if (!inhibit_keep_alive)
1387 /* Look for a persistent connection to target host, unless a
1388 proxy is used. The exception is when SSL is in use, in which
1389 case the proxy is nothing but a passthrough to the target
1390 host, registered as a connection to the latter. */
1391 struct url *relevant = conn;
1393 if (u->scheme == SCHEME_HTTPS)
1397 if (persistent_available_p (relevant->host, relevant->port,
1399 relevant->scheme == SCHEME_HTTPS,
1403 &host_lookup_failed))
1405 sock = pconn.socket;
1406 using_ssl = pconn.ssl;
1407 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1408 escnonprint (pconn.host), pconn.port);
1409 DEBUGP (("Reusing fd %d.\n", sock));
1410 if (pconn.authorized)
1411 /* If the connection is already authorized, the "Basic"
1412 authorization added by code above is unnecessary and
1414 request_remove_header (req, "Authorization");
1420 /* In its current implementation, persistent_available_p will
1421 look up conn->host in some cases. If that lookup failed, we
1422 don't need to bother with connect_to_host. */
1423 if (host_lookup_failed)
1429 sock = connect_to_host (conn->host, conn->port);
1438 return (retryable_socket_connect_error (errno)
1439 ? CONERROR : CONIMPOSSIBLE);
1443 if (proxy && u->scheme == SCHEME_HTTPS)
1445 /* When requesting SSL URLs through proxies, use the
1446 CONNECT method to request passthrough. */
1447 struct request *connreq = request_new ();
1448 request_set_method (connreq, "CONNECT",
1449 aprintf ("%s:%d", u->host, u->port));
1450 SET_USER_AGENT (connreq);
1453 request_set_header (connreq, "Proxy-Authorization",
1454 proxyauth, rel_value);
1455 /* Now that PROXYAUTH is part of the CONNECT request,
1456 zero it out so we don't send proxy authorization with
1457 the regular request below. */
1460 /* Examples in rfc2817 use the Host header in CONNECT
1461 requests. I don't see how that gains anything, given
1462 that the contents of Host would be exactly the same as
1463 the contents of CONNECT. */
1465 write_error = request_send (connreq, sock);
1466 request_free (connreq);
1467 if (write_error < 0)
1469 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1471 CLOSE_INVALIDATE (sock);
1475 head = read_http_response_head (sock);
1478 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1480 CLOSE_INVALIDATE (sock);
1489 DEBUGP (("proxy responded with: [%s]\n", head));
1491 resp = resp_new (head);
1492 statcode = resp_status (resp, &message);
1495 if (statcode != 200)
1498 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1499 message ? escnonprint (message) : "?");
1500 xfree_null (message);
1503 xfree_null (message);
1505 /* SOCK is now *really* connected to u->host, so update CONN
1506 to reflect this. That way register_persistent will
1507 register SOCK as being connected to u->host:u->port. */
1511 if (conn->scheme == SCHEME_HTTPS)
1513 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1520 #endif /* HAVE_SSL */
1523 /* Send the request to server. */
1524 write_error = request_send (req, sock);
1526 if (write_error >= 0)
1530 DEBUGP (("[POST data: %s]\n", opt.post_data));
1531 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1533 else if (opt.post_file_name && post_data_size != 0)
1534 write_error = post_file (sock, opt.post_file_name, post_data_size);
1537 if (write_error < 0)
1539 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1541 CLOSE_INVALIDATE (sock);
1545 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1546 proxy ? "Proxy" : "HTTP");
1551 head = read_http_response_head (sock);
1556 logputs (LOG_NOTQUIET, _("No data received.\n"));
1557 CLOSE_INVALIDATE (sock);
1563 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1565 CLOSE_INVALIDATE (sock);
1570 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1572 resp = resp_new (head);
1574 /* Check for status line. */
1576 statcode = resp_status (resp, &message);
1577 if (!opt.server_response)
1578 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1579 message ? escnonprint (message) : "");
1582 logprintf (LOG_VERBOSE, "\n");
1583 print_server_response (resp, " ");
1586 if (!opt.ignore_length
1587 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1591 parsed = str_to_wgint (hdrval, NULL, 10);
1592 if (parsed == WGINT_MAX && errno == ERANGE)
1594 #### If Content-Length is out of range, it most likely
1595 means that the file is larger than 2G and that we're
1596 compiled without LFS. In that case we should probably
1597 refuse to even attempt to download the file. */
1603 /* Check for keep-alive related responses. */
1604 if (!inhibit_keep_alive && contlen != -1)
1606 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1608 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1610 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1615 /* The server has promised that it will not close the connection
1616 when we're done. This means that we can register it. */
1617 register_persistent (conn->host, conn->port, sock, using_ssl);
1619 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1621 /* Authorization is required. */
1622 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1623 CLOSE_FINISH (sock);
1625 CLOSE_INVALIDATE (sock);
1626 pconn.authorized = false;
1627 if (!auth_finished && (user && passwd))
1629 /* IIS sends multiple copies of WWW-Authenticate, one with
1630 the value "negotiate", and other(s) with data. Loop over
1631 all the occurrences and pick the one we recognize. */
1633 const char *wabeg, *waend;
1634 char *www_authenticate = NULL;
1636 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1637 &wabeg, &waend)) != -1;
1639 if (known_authentication_scheme_p (wabeg, waend))
1641 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1645 if (!www_authenticate)
1646 /* If the authentication header is missing or
1647 unrecognized, there's no sense in retrying. */
1648 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1649 else if (BEGINS_WITH (www_authenticate, "Basic"))
1650 /* If the authentication scheme is "Basic", which we send
1651 by default, there's no sense in retrying either. (This
1652 should be changed when we stop sending "Basic" data by
1658 pth = url_full_path (u);
1659 request_set_header (req, "Authorization",
1660 create_authorization_line (www_authenticate,
1662 request_method (req),
1666 if (BEGINS_WITH (www_authenticate, "NTLM"))
1669 goto retry_with_auth;
1672 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1676 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1678 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1680 pconn.authorized = true;
1684 hs->statcode = statcode;
1686 hs->error = xstrdup (_("Malformed status line"));
1688 hs->error = xstrdup (_("(no description)"));
1690 hs->error = xstrdup (message);
1693 type = resp_header_strdup (resp, "Content-Type");
1696 char *tmp = strchr (type, ';');
1699 while (tmp > type && ISSPACE (tmp[-1]))
1704 hs->newloc = resp_header_strdup (resp, "Location");
1705 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1707 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1711 const char *scbeg, *scend;
1712 /* The jar should have been created by now. */
1713 assert (wget_cookie_jar != NULL);
1715 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1716 &scbeg, &scend)) != -1;
1719 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1720 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1721 u->path, set_cookie);
1725 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1727 wgint first_byte_pos, last_byte_pos, entity_length;
1728 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1730 contrange = first_byte_pos;
1734 /* 20x responses are counted among successful by default. */
1735 if (H_20X (statcode))
1738 /* Return if redirected. */
1739 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1741 /* RFC2068 says that in case of the 300 (multiple choices)
1742 response, the server can output a preferred URL through
1743 `Location' header; otherwise, the request should be treated
1744 like GET. So, if the location is set, it will be a
1745 redirection; otherwise, just proceed normally. */
1746 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1750 logprintf (LOG_VERBOSE,
1751 _("Location: %s%s\n"),
1752 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1753 hs->newloc ? _(" [following]") : "");
1754 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1755 CLOSE_FINISH (sock);
1757 CLOSE_INVALIDATE (sock);
1763 /* If content-type is not given, assume text/html. This is because
1764 of the multitude of broken CGI's that "forget" to generate the
1767 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1768 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1773 if (opt.html_extension && (*dt & TEXTHTML))
1774 /* -E / --html-extension / html_extension = on was specified, and this is a
1775 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1776 already the file's suffix, tack on ".html". */
1778 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1780 if (last_period_in_local_filename == NULL
1781 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1782 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1784 int local_filename_len = strlen (*hs->local_file);
1785 /* Resize the local file, allowing for ".html" preceded by
1786 optional ".NUMBER". */
1787 *hs->local_file = xrealloc (*hs->local_file,
1788 local_filename_len + 24 + sizeof (".html"));
1789 strcpy(*hs->local_file + local_filename_len, ".html");
1790 /* If clobbering is not allowed and the file, as named,
1791 exists, tack on ".NUMBER.html" instead. */
1796 sprintf (*hs->local_file + local_filename_len,
1797 ".%d.html", ext_num++);
1798 while (file_exists_p (*hs->local_file));
1800 *dt |= ADDED_HTML_EXTENSION;
1804 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1806 /* If `-c' is in use and the file has been fully downloaded (or
1807 the remote file has shrunk), Wget effectively requests bytes
1808 after the end of file and the server response with 416. */
1809 logputs (LOG_VERBOSE, _("\
1810 \n The file is already fully retrieved; nothing to do.\n\n"));
1811 /* In case the caller inspects. */
1814 /* Mark as successfully retrieved. */
1817 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1818 might be more bytes in the body. */
1819 return RETRUNNEEDED;
1821 if ((contrange != 0 && contrange != hs->restval)
1822 || (H_PARTIAL (statcode) && !contrange))
1824 /* The Range request was somehow misunderstood by the server.
1827 CLOSE_INVALIDATE (sock);
1830 hs->contlen = contlen + contrange;
1836 /* No need to print this output if the body won't be
1837 downloaded at all, or if the original server response is
1839 logputs (LOG_VERBOSE, _("Length: "));
1842 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1843 if (contlen + contrange >= 1024)
1844 logprintf (LOG_VERBOSE, " (%s)",
1845 human_readable (contlen + contrange));
1848 if (contlen >= 1024)
1849 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1850 number_to_static_string (contlen),
1851 human_readable (contlen));
1853 logprintf (LOG_VERBOSE, _(", %s remaining"),
1854 number_to_static_string (contlen));
1858 logputs (LOG_VERBOSE,
1859 opt.ignore_length ? _("ignored") : _("unspecified"));
1861 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1863 logputs (LOG_VERBOSE, "\n");
1867 type = NULL; /* We don't need it any more. */
1869 /* Return if we have no intention of further downloading. */
1870 if (!(*dt & RETROKF) || head_only)
1872 /* In case the caller cares to look... */
1876 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1877 servers not to send body in response to a HEAD request. If
1878 you encounter such a server (more likely a broken CGI), use
1879 `--no-http-keep-alive'. */
1880 CLOSE_FINISH (sock);
1881 return RETRFINISHED;
1884 /* Open the local file. */
1887 mkalldirs (*hs->local_file);
1889 rotate_backups (*hs->local_file);
1891 fp = fopen (*hs->local_file, "ab");
1892 else if (ALLOW_CLOBBER)
1893 fp = fopen (*hs->local_file, "wb");
1896 fp = fopen_excl (*hs->local_file, true);
1897 if (!fp && errno == EEXIST)
1899 /* We cannot just invent a new name and use it (which is
1900 what functions like unique_create typically do)
1901 because we told the user we'd use this name.
1902 Instead, return and retry the download. */
1903 logprintf (LOG_NOTQUIET,
1904 _("%s has sprung into existence.\n"),
1906 CLOSE_INVALIDATE (sock);
1907 return FOPEN_EXCL_ERR;
1912 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1913 CLOSE_INVALIDATE (sock);
1920 /* #### This confuses the timestamping code that checks for file
1921 size. Maybe we should save some additional information? */
1922 if (opt.save_headers)
1923 fwrite (head, 1, strlen (head), fp);
1925 /* Now we no longer need to store the response header. */
1928 /* Download the request body. */
1931 flags |= rb_read_exactly;
1932 if (hs->restval > 0 && contrange == 0)
1933 /* If the server ignored our range request, instruct fd_read_body
1934 to skip the first RESTVAL bytes of body. */
1935 flags |= rb_skip_startpos;
1936 hs->len = hs->restval;
1938 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1939 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1943 CLOSE_FINISH (sock);
1945 CLOSE_INVALIDATE (sock);
1948 /* Close or flush the file. We have to be careful to check for
1949 error here. Checking the result of fwrite() is not enough --
1950 errors could go unnoticed! */
1953 flush_res = fclose (fp);
1955 flush_res = fflush (fp);
1956 if (flush_res == EOF)
1961 return RETRFINISHED;
1964 /* The genuine HTTP loop! This is the part where the retrieval is
1965 retried, and retried, and retried, and... */
1967 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1968 int *dt, struct url *proxy)
1971 bool use_ts, got_head = false;/* time-stamping info */
1972 char *filename_plus_orig_suffix;
1973 char *local_filename = NULL;
1977 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1978 wgint local_size = 0; /* the size of the local file */
1979 size_t filename_len;
1980 struct http_stat hstat; /* HTTP status */
1984 /* This used to be done in main(), but it's a better idea to do it
1985 here so that we don't go through the hoops if we're just using
1989 if (!wget_cookie_jar)
1990 wget_cookie_jar = cookie_jar_new ();
1991 if (opt.cookies_input && !cookies_loaded_p)
1993 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1994 cookies_loaded_p = true;
2000 /* Warn on (likely bogus) wildcard usage in HTTP. */
2001 if (opt.ftp_glob && has_wildcards_p (u->path))
2002 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2006 /* Determine the local filename. */
2007 if (local_file && *local_file)
2008 hstat.local_file = local_file;
2009 else if (local_file && !opt.output_document)
2011 *local_file = url_file_name (u);
2012 hstat.local_file = local_file;
2016 dummy = url_file_name (u);
2017 hstat.local_file = &dummy;
2018 /* be honest about where we will save the file */
2019 if (local_file && opt.output_document)
2020 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2023 if (!opt.output_document)
2024 locf = *hstat.local_file;
2026 locf = opt.output_document;
2028 hstat.referer = referer;
2030 filename_len = strlen (*hstat.local_file);
2031 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2033 if (opt.noclobber && file_exists_p (*hstat.local_file))
2035 /* If opt.noclobber is turned on and file already exists, do not
2036 retrieve the file */
2037 logprintf (LOG_VERBOSE, _("\
2038 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2039 /* If the file is there, we suppose it's retrieved OK. */
2042 /* #### Bogusness alert. */
2043 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2044 if (has_html_suffix_p (*hstat.local_file))
2052 if (opt.timestamping)
2054 bool local_dot_orig_file_exists = false;
2056 if (opt.backup_converted)
2057 /* If -K is specified, we'll act on the assumption that it was specified
2058 last time these files were downloaded as well, and instead of just
2059 comparing local file X against server file X, we'll compare local
2060 file X.orig (if extant, else X) against server file X. If -K
2061 _wasn't_ specified last time, or the server contains files called
2062 *.orig, -N will be back to not operating correctly with -k. */
2064 /* Would a single s[n]printf() call be faster? --dan
2066 Definitely not. sprintf() is horribly slow. It's a
2067 different question whether the difference between the two
2068 affects a program. Usually I'd say "no", but at one
2069 point I profiled Wget, and found that a measurable and
2070 non-negligible amount of time was lost calling sprintf()
2071 in url.c. Replacing sprintf with inline calls to
2072 strcpy() and number_to_string() made a difference.
2074 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2075 memcpy (filename_plus_orig_suffix + filename_len,
2076 ".orig", sizeof (".orig"));
2078 /* Try to stat() the .orig file. */
2079 if (stat (filename_plus_orig_suffix, &st) == 0)
2081 local_dot_orig_file_exists = 1;
2082 local_filename = filename_plus_orig_suffix;
2086 if (!local_dot_orig_file_exists)
2087 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2088 if (stat (*hstat.local_file, &st) == 0)
2089 local_filename = *hstat.local_file;
2091 if (local_filename != NULL)
2092 /* There was a local file, so we'll check later to see if the version
2093 the server has is the same version we already have, allowing us to
2099 /* Modification time granularity is 2 seconds for Windows, so
2100 increase local time by 1 second for later comparison. */
2103 local_size = st.st_size;
2107 /* Reset the counter. */
2113 /* Increment the pass counter. */
2115 sleep_between_retrievals (count);
2116 /* Get the current time string. */
2117 tms = time_str (NULL);
2118 /* Print fetch message, if opt.verbose. */
2121 char *hurl = url_string (u, true);
2125 sprintf (tmp, _("(try:%2d)"), count);
2126 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2127 tms, hurl, tmp, locf);
2129 ws_changetitle (hurl);
2134 /* Default document type is empty. However, if spider mode is
2135 on or time-stamping is employed, HEAD_ONLY commands is
2136 encoded within *dt. */
2137 if (opt.spider || (use_ts && !got_head))
2142 /* Decide whether or not to restart. */
2144 && stat (locf, &st) == 0
2145 && S_ISREG (st.st_mode))
2146 /* When -c is used, continue from on-disk size. (Can't use
2147 hstat.len even if count>1 because we don't want a failed
2148 first attempt to clobber existing data.) */
2149 hstat.restval = st.st_size;
2151 /* otherwise, continue where the previous try left off */
2152 hstat.restval = hstat.len;
2156 /* Decide whether to send the no-cache directive. We send it in
2158 a) we're using a proxy, and we're past our first retrieval.
2159 Some proxies are notorious for caching incomplete data, so
2160 we require a fresh get.
2161 b) caching is explicitly inhibited. */
2162 if ((proxy && count > 1) /* a */
2163 || !opt.allow_cache /* b */
2165 *dt |= SEND_NOCACHE;
2167 *dt &= ~SEND_NOCACHE;
2169 /* Try fetching the document, or at least its head. */
2170 err = gethttp (u, &hstat, dt, proxy);
2172 /* It's unfortunate that wget determines the local filename before finding
2173 out the Content-Type of the file. Barring a major restructuring of the
2174 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2175 *hstat.local_file to tack on ".html". */
2176 if (!opt.output_document)
2177 locf = *hstat.local_file;
2180 tms = time_str (NULL);
2181 /* Get the new location (with or without the redirection). */
2183 *newloc = xstrdup (hstat.newloc);
2186 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2187 case CONERROR: case READERR: case WRITEFAILED:
2188 case RANGEERR: case FOPEN_EXCL_ERR:
2189 /* Non-fatal errors continue executing the loop, which will
2190 bring them to "while" statement at the end, to judge
2191 whether the number of tries was exceeded. */
2192 free_hstat (&hstat);
2193 printwhat (count, opt.ntry);
2194 if (err == FOPEN_EXCL_ERR)
2196 /* Re-determine the file name. */
2197 if (local_file && *local_file)
2199 xfree (*local_file);
2200 *local_file = url_file_name (u);
2201 hstat.local_file = local_file;
2206 dummy = url_file_name (u);
2207 hstat.local_file = &dummy;
2209 /* be honest about where we will save the file */
2210 if (local_file && opt.output_document)
2211 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2212 if (!opt.output_document)
2213 locf = *hstat.local_file;
2215 locf = opt.output_document;
2218 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2219 case SSLINITFAILED: case CONTNOTSUPPORTED:
2220 /* Fatal errors just return from the function. */
2221 free_hstat (&hstat);
2224 case FWRITEERR: case FOPENERR:
2225 /* Another fatal error. */
2226 logputs (LOG_VERBOSE, "\n");
2227 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2228 *hstat.local_file, strerror (errno));
2229 free_hstat (&hstat);
2233 /* Another fatal error. */
2234 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2235 free_hstat (&hstat);
2239 /* Return the new location to the caller. */
2242 logprintf (LOG_NOTQUIET,
2243 _("ERROR: Redirection (%d) without location.\n"),
2245 free_hstat (&hstat);
2249 free_hstat (&hstat);
2253 /* The file was already fully retrieved. */
2254 free_hstat (&hstat);
2258 /* Deal with you later. */
2261 /* All possibilities should have been exhausted. */
2264 if (!(*dt & RETROKF))
2268 /* #### Ugly ugly ugly! */
2269 char *hurl = url_string (u, true);
2270 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2273 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2274 tms, hstat.statcode, escnonprint (hstat.error));
2275 logputs (LOG_VERBOSE, "\n");
2276 free_hstat (&hstat);
2281 /* Did we get the time-stamp? */
2284 if (opt.timestamping && !hstat.remote_time)
2286 logputs (LOG_NOTQUIET, _("\
2287 Last-modified header missing -- time-stamps turned off.\n"));
2289 else if (hstat.remote_time)
2291 /* Convert the date-string into struct tm. */
2292 tmr = http_atotm (hstat.remote_time);
2293 if (tmr == (time_t) (-1))
2294 logputs (LOG_VERBOSE, _("\
2295 Last-modified header invalid -- time-stamp ignored.\n"));
2299 /* The time-stamping section. */
2304 use_ts = false; /* no more time-stamping */
2305 count = 0; /* the retrieve count for HEAD is
2307 if (hstat.remote_time && tmr != (time_t) (-1))
2309 /* Now time-stamping can be used validly. Time-stamping
2310 means that if the sizes of the local and remote file
2311 match, and local file is newer than the remote file,
2312 it will not be retrieved. Otherwise, the normal
2313 download procedure is resumed. */
2315 (hstat.contlen == -1 || local_size == hstat.contlen))
2317 logprintf (LOG_VERBOSE, _("\
2318 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2320 free_hstat (&hstat);
2324 else if (tml >= tmr)
2325 logprintf (LOG_VERBOSE, _("\
2326 The sizes do not match (local %s) -- retrieving.\n"),
2327 number_to_static_string (local_size));
2329 logputs (LOG_VERBOSE,
2330 _("Remote file is newer, retrieving.\n"));
2332 free_hstat (&hstat);
2335 if ((tmr != (time_t) (-1))
2337 && ((hstat.len == hstat.contlen) ||
2338 ((hstat.res == 0) &&
2339 ((hstat.contlen == -1) ||
2340 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2342 /* #### This code repeats in http.c and ftp.c. Move it to a
2344 const char *fl = NULL;
2345 if (opt.output_document)
2347 if (output_stream_regular)
2348 fl = opt.output_document;
2351 fl = *hstat.local_file;
2355 /* End of time-stamping section. */
2359 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2360 escnonprint (hstat.error));
2365 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2366 total_download_time += hstat.dltime;
2368 if (hstat.len == hstat.contlen)
2372 logprintf (LOG_VERBOSE,
2373 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2375 number_to_static_string (hstat.len),
2376 number_to_static_string (hstat.contlen));
2377 logprintf (LOG_NONVERBOSE,
2378 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2380 number_to_static_string (hstat.len),
2381 number_to_static_string (hstat.contlen),
2385 total_downloaded_bytes += hstat.len;
2387 /* Remember that we downloaded the file for later ".orig" code. */
2388 if (*dt & ADDED_HTML_EXTENSION)
2389 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2391 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2393 free_hstat (&hstat);
2397 else if (hstat.res == 0) /* No read error */
2399 if (hstat.contlen == -1) /* We don't know how much we were supposed
2400 to get, so assume we succeeded. */
2404 logprintf (LOG_VERBOSE,
2405 _("%s (%s) - `%s' saved [%s]\n\n"),
2407 number_to_static_string (hstat.len));
2408 logprintf (LOG_NONVERBOSE,
2409 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2410 tms, u->url, number_to_static_string (hstat.len),
2414 total_downloaded_bytes += hstat.len;
2416 /* Remember that we downloaded the file for later ".orig" code. */
2417 if (*dt & ADDED_HTML_EXTENSION)
2418 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2420 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2422 free_hstat (&hstat);
2426 else if (hstat.len < hstat.contlen) /* meaning we lost the
2427 connection too soon */
2429 logprintf (LOG_VERBOSE,
2430 _("%s (%s) - Connection closed at byte %s. "),
2431 tms, tmrate, number_to_static_string (hstat.len));
2432 printwhat (count, opt.ntry);
2433 free_hstat (&hstat);
2436 else if (!opt.kill_longer) /* meaning we got more than expected */
2438 logprintf (LOG_VERBOSE,
2439 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2441 number_to_static_string (hstat.len),
2442 number_to_static_string (hstat.contlen));
2443 logprintf (LOG_NONVERBOSE,
2444 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2446 number_to_static_string (hstat.len),
2447 number_to_static_string (hstat.contlen),
2450 total_downloaded_bytes += hstat.len;
2452 /* Remember that we downloaded the file for later ".orig" code. */
2453 if (*dt & ADDED_HTML_EXTENSION)
2454 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2456 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2458 free_hstat (&hstat);
2462 else /* the same, but not accepted */
2464 logprintf (LOG_VERBOSE,
2465 _("%s (%s) - Connection closed at byte %s/%s. "),
2467 number_to_static_string (hstat.len),
2468 number_to_static_string (hstat.contlen));
2469 printwhat (count, opt.ntry);
2470 free_hstat (&hstat);
2474 else /* now hstat.res can only be -1 */
2476 if (hstat.contlen == -1)
2478 logprintf (LOG_VERBOSE,
2479 _("%s (%s) - Read error at byte %s (%s)."),
2480 tms, tmrate, number_to_static_string (hstat.len),
2482 printwhat (count, opt.ntry);
2483 free_hstat (&hstat);
2486 else /* hstat.res == -1 and contlen is given */
2488 logprintf (LOG_VERBOSE,
2489 _("%s (%s) - Read error at byte %s/%s (%s). "),
2491 number_to_static_string (hstat.len),
2492 number_to_static_string (hstat.contlen),
2494 printwhat (count, opt.ntry);
2495 free_hstat (&hstat);
2501 while (!opt.ntry || (count < opt.ntry));
2505 /* Check whether the result of strptime() indicates success.
2506 strptime() returns the pointer to how far it got to in the string.
2507 The processing has been successful if the string is at `GMT' or
2508 `+X', or at the end of the string.
2510 In extended regexp parlance, the function returns 1 if P matches
2511 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2512 can return) is considered a failure and 0 is returned. */
2514 check_end (const char *p)
2518 while (ISSPACE (*p))
2521 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2522 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2528 /* Convert the textual specification of time in TIME_STRING to the
2529 number of seconds since the Epoch.
2531 TIME_STRING can be in any of the three formats RFC2616 allows the
2532 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2533 as well as the time format used in the Set-Cookie header.
2534 Timezones are ignored, and should be GMT.
2536 Return the computed time_t representation, or -1 if the conversion
2539 This function uses strptime with various string formats for parsing
2540 TIME_STRING. This results in a parser that is not as lenient in
2541 interpreting TIME_STRING as I would like it to be. Being based on
2542 strptime, it always allows shortened months, one-digit days, etc.,
2543 but due to the multitude of formats in which time can be
2544 represented, an ideal HTTP time parser would be even more
2545 forgiving. It should completely ignore things like week days and
2546 concentrate only on the various forms of representing years,
2547 months, days, hours, minutes, and seconds. For example, it would
2548 be nice if it accepted ISO 8601 out of the box.
2550 I've investigated free and PD code for this purpose, but none was
2551 usable. getdate was big and unwieldy, and had potential copyright
2552 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2553 distributed with phttpd, is excellent, but we cannot use it because
2554 it is not assigned to the FSF. So I stuck it with strptime. */
2557 http_atotm (const char *time_string)
2559 /* NOTE: Solaris strptime man page claims that %n and %t match white
2560 space, but that's not universally available. Instead, we simply
2561 use ` ' to mean "skip all WS", which works under all strptime
2562 implementations I've tested. */
2564 static const char *time_formats[] = {
2565 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2566 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2567 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2568 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2569 (used in Set-Cookie, defined in the
2570 Netscape cookie specification.) */
2572 const char *oldlocale;
2574 time_t ret = (time_t) -1;
2576 /* Solaris strptime fails to recognize English month names in
2577 non-English locales, which we work around by temporarily setting
2578 locale to C before invoking strptime. */
2579 oldlocale = setlocale (LC_TIME, NULL);
2580 setlocale (LC_TIME, "C");
2582 for (i = 0; i < countof (time_formats); i++)
2586 /* Some versions of strptime use the existing contents of struct
2587 tm to recalculate the date according to format. Zero it out
2588 to prevent stack garbage from influencing strptime. */
2591 if (check_end (strptime (time_string, time_formats[i], &t)))
2598 /* Restore the previous locale. */
2599 setlocale (LC_TIME, oldlocale);
2604 /* Authorization support: We support three authorization schemes:
2606 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2608 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2609 consisting of answering to the server's challenge with the proper
2612 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2613 Stenberg for libcurl. Like digest, NTLM is based on a
2614 challenge-response mechanism, but unlike digest, it is non-standard
2615 (authenticates TCP connections rather than requests), undocumented
2616 and Microsoft-specific. */
2618 /* Create the authentication header contents for the `Basic' scheme.
2619 This is done by encoding the string "USER:PASS" to base64 and
2620 prepending the string "Basic " in front of it. */
2623 basic_authentication_encode (const char *user, const char *passwd)
2626 int len1 = strlen (user) + 1 + strlen (passwd);
2628 t1 = (char *)alloca (len1 + 1);
2629 sprintf (t1, "%s:%s", user, passwd);
2631 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2632 base64_encode (t1, len1, t2);
2634 return concat_strings ("Basic ", t2, (char *) 0);
2637 #define SKIP_WS(x) do { \
2638 while (ISSPACE (*(x))) \
2642 #ifdef ENABLE_DIGEST
2643 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2644 of a field in such a header. If the field is the one specified by
2645 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2646 digest authorization code), extract its value in the (char*)
2647 variable pointed by RET. Returns negative on a malformed header,
2648 or number of bytes that have been parsed by this call. */
2650 extract_header_attr (const char *au, const char *attr_name, char **ret)
2653 const char *cp = au;
2655 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2657 cp += strlen (attr_name);
2670 for (ep = cp; *ep && *ep != '\"'; ep++)
2675 *ret = strdupdelim (cp, ep);
2682 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2683 an array of 16 bytes containing the hash keys, and BUF should be a
2684 buffer of 33 writable characters (32 for hex digits plus one for
2685 zero termination). */
2687 dump_hash (char *buf, const unsigned char *hash)
2691 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2693 *buf++ = XNUM_TO_digit (*hash >> 4);
2694 *buf++ = XNUM_TO_digit (*hash & 0xf);
2699 /* Take the line apart to find the challenge, and compose a digest
2700 authorization header. See RFC2069 section 2.1.2. */
2702 digest_authentication_encode (const char *au, const char *user,
2703 const char *passwd, const char *method,
2706 static char *realm, *opaque, *nonce;
2711 { "realm", &realm },
2712 { "opaque", &opaque },
2717 realm = opaque = nonce = NULL;
2719 au += 6; /* skip over `Digest' */
2725 for (i = 0; i < countof (options); i++)
2727 int skip = extract_header_attr (au, options[i].name,
2728 options[i].variable);
2732 xfree_null (opaque);
2742 if (i == countof (options))
2744 while (*au && *au != '=')
2752 while (*au && *au != '\"')
2759 while (*au && *au != ',')
2764 if (!realm || !nonce || !user || !passwd || !path || !method)
2767 xfree_null (opaque);
2772 /* Calculate the digest value. */
2774 ALLOCA_MD5_CONTEXT (ctx);
2775 unsigned char hash[MD5_HASHLEN];
2776 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2777 char response_digest[MD5_HASHLEN * 2 + 1];
2779 /* A1BUF = H(user ":" realm ":" password) */
2781 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2782 gen_md5_update ((unsigned char *)":", 1, ctx);
2783 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2784 gen_md5_update ((unsigned char *)":", 1, ctx);
2785 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2786 gen_md5_finish (ctx, hash);
2787 dump_hash (a1buf, hash);
2789 /* A2BUF = H(method ":" path) */
2791 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2792 gen_md5_update ((unsigned char *)":", 1, ctx);
2793 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2794 gen_md5_finish (ctx, hash);
2795 dump_hash (a2buf, hash);
2797 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2799 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2800 gen_md5_update ((unsigned char *)":", 1, ctx);
2801 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2802 gen_md5_update ((unsigned char *)":", 1, ctx);
2803 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2804 gen_md5_finish (ctx, hash);
2805 dump_hash (response_digest, hash);
2807 res = xmalloc (strlen (user)
2812 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2813 + (opaque ? strlen (opaque) : 0)
2815 sprintf (res, "Digest \
2816 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2817 user, realm, nonce, path, response_digest);
2820 char *p = res + strlen (res);
2821 strcat (p, ", opaque=\"");
2828 #endif /* ENABLE_DIGEST */
2830 /* Computing the size of a string literal must take into account that
2831 value returned by sizeof includes the terminating \0. */
2832 #define STRSIZE(literal) (sizeof (literal) - 1)
2834 /* Whether chars in [b, e) begin with the literal string provided as
2835 first argument and are followed by whitespace or terminating \0.
2836 The comparison is case-insensitive. */
2837 #define STARTS(literal, b, e) \
2838 ((e) - (b) >= STRSIZE (literal) \
2839 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2840 && ((e) - (b) == STRSIZE (literal) \
2841 || ISSPACE (b[STRSIZE (literal)])))
2844 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2846 return STARTS ("Basic", hdrbeg, hdrend)
2847 #ifdef ENABLE_DIGEST
2848 || STARTS ("Digest", hdrbeg, hdrend)
2851 || STARTS ("NTLM", hdrbeg, hdrend)
2858 /* Create the HTTP authorization request header. When the
2859 `WWW-Authenticate' response header is seen, according to the
2860 authorization scheme specified in that header (`Basic' and `Digest'
2861 are supported by the current implementation), produce an
2862 appropriate HTTP authorization request header. */
2864 create_authorization_line (const char *au, const char *user,
2865 const char *passwd, const char *method,
2866 const char *path, bool *finished)
2868 /* We are called only with known schemes, so we can dispatch on the
2870 switch (TOUPPER (*au))
2872 case 'B': /* Basic */
2874 return basic_authentication_encode (user, passwd);
2875 #ifdef ENABLE_DIGEST
2876 case 'D': /* Digest */
2878 return digest_authentication_encode (au, user, passwd, method, path);
2881 case 'N': /* NTLM */
2882 if (!ntlm_input (&pconn.ntlm, au))
2887 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2890 /* We shouldn't get here -- this function should be only called
2891 with values approved by known_authentication_scheme_p. */
2899 if (wget_cookie_jar)
2900 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2906 xfree_null (pconn.host);
2907 if (wget_cookie_jar)
2908 cookie_jar_delete (wget_cookie_jar);