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);
1312 /* Whether we need to print the host header with braces around
1313 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1314 usual "Host: symbolic-name:1234". */
1315 bool squares = strchr (u->host, ':') != NULL;
1316 if (u->port == scheme_default_port (u->scheme))
1317 request_set_header (req, "Host",
1318 aprintf (squares ? "[%s]" : "%s", u->host),
1321 request_set_header (req, "Host",
1322 aprintf (squares ? "[%s]:%d" : "%s:%d",
1327 if (!inhibit_keep_alive)
1328 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1331 request_set_header (req, "Cookie",
1332 cookie_header (wget_cookie_jar,
1333 u->host, u->port, u->path,
1335 u->scheme == SCHEME_HTTPS
1342 if (opt.post_data || opt.post_file_name)
1344 request_set_header (req, "Content-Type",
1345 "application/x-www-form-urlencoded", rel_none);
1347 post_data_size = strlen (opt.post_data);
1350 post_data_size = file_size (opt.post_file_name);
1351 if (post_data_size == -1)
1353 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1354 opt.post_file_name, strerror (errno));
1358 request_set_header (req, "Content-Length",
1359 xstrdup (number_to_static_string (post_data_size)),
1363 /* Add the user headers. */
1364 if (opt.user_headers)
1367 for (i = 0; opt.user_headers[i]; i++)
1368 request_set_user_header (req, opt.user_headers[i]);
1372 /* We need to come back here when the initial attempt to retrieve
1373 without authorization header fails. (Expected to happen at least
1374 for the Digest authorization scheme.) */
1378 /* Establish the connection. */
1380 if (!inhibit_keep_alive)
1382 /* Look for a persistent connection to target host, unless a
1383 proxy is used. The exception is when SSL is in use, in which
1384 case the proxy is nothing but a passthrough to the target
1385 host, registered as a connection to the latter. */
1386 struct url *relevant = conn;
1388 if (u->scheme == SCHEME_HTTPS)
1392 if (persistent_available_p (relevant->host, relevant->port,
1394 relevant->scheme == SCHEME_HTTPS,
1398 &host_lookup_failed))
1400 sock = pconn.socket;
1401 using_ssl = pconn.ssl;
1402 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1403 escnonprint (pconn.host), pconn.port);
1404 DEBUGP (("Reusing fd %d.\n", sock));
1405 if (pconn.authorized)
1406 /* If the connection is already authorized, the "Basic"
1407 authorization added by code above is unnecessary and
1409 request_remove_header (req, "Authorization");
1415 /* In its current implementation, persistent_available_p will
1416 look up conn->host in some cases. If that lookup failed, we
1417 don't need to bother with connect_to_host. */
1418 if (host_lookup_failed)
1424 sock = connect_to_host (conn->host, conn->port);
1433 return (retryable_socket_connect_error (errno)
1434 ? CONERROR : CONIMPOSSIBLE);
1438 if (proxy && u->scheme == SCHEME_HTTPS)
1440 /* When requesting SSL URLs through proxies, use the
1441 CONNECT method to request passthrough. */
1442 struct request *connreq = request_new ();
1443 request_set_method (connreq, "CONNECT",
1444 aprintf ("%s:%d", u->host, u->port));
1445 SET_USER_AGENT (connreq);
1448 request_set_header (connreq, "Proxy-Authorization",
1449 proxyauth, rel_value);
1450 /* Now that PROXYAUTH is part of the CONNECT request,
1451 zero it out so we don't send proxy authorization with
1452 the regular request below. */
1455 /* Examples in rfc2817 use the Host header in CONNECT
1456 requests. I don't see how that gains anything, given
1457 that the contents of Host would be exactly the same as
1458 the contents of CONNECT. */
1460 write_error = request_send (connreq, sock);
1461 request_free (connreq);
1462 if (write_error < 0)
1464 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1466 CLOSE_INVALIDATE (sock);
1470 head = read_http_response_head (sock);
1473 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1475 CLOSE_INVALIDATE (sock);
1484 DEBUGP (("proxy responded with: [%s]\n", head));
1486 resp = resp_new (head);
1487 statcode = resp_status (resp, &message);
1490 if (statcode != 200)
1493 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1494 message ? escnonprint (message) : "?");
1495 xfree_null (message);
1498 xfree_null (message);
1500 /* SOCK is now *really* connected to u->host, so update CONN
1501 to reflect this. That way register_persistent will
1502 register SOCK as being connected to u->host:u->port. */
1506 if (conn->scheme == SCHEME_HTTPS)
1508 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1515 #endif /* HAVE_SSL */
1518 /* Send the request to server. */
1519 write_error = request_send (req, sock);
1521 if (write_error >= 0)
1525 DEBUGP (("[POST data: %s]\n", opt.post_data));
1526 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1528 else if (opt.post_file_name && post_data_size != 0)
1529 write_error = post_file (sock, opt.post_file_name, post_data_size);
1532 if (write_error < 0)
1534 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1536 CLOSE_INVALIDATE (sock);
1540 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1541 proxy ? "Proxy" : "HTTP");
1546 head = read_http_response_head (sock);
1551 logputs (LOG_NOTQUIET, _("No data received.\n"));
1552 CLOSE_INVALIDATE (sock);
1558 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1560 CLOSE_INVALIDATE (sock);
1565 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1567 resp = resp_new (head);
1569 /* Check for status line. */
1571 statcode = resp_status (resp, &message);
1572 if (!opt.server_response)
1573 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1574 message ? escnonprint (message) : "");
1577 logprintf (LOG_VERBOSE, "\n");
1578 print_server_response (resp, " ");
1581 if (!opt.ignore_length
1582 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1586 parsed = str_to_wgint (hdrval, NULL, 10);
1587 if (parsed == WGINT_MAX && errno == ERANGE)
1589 #### If Content-Length is out of range, it most likely
1590 means that the file is larger than 2G and that we're
1591 compiled without LFS. In that case we should probably
1592 refuse to even attempt to download the file. */
1598 /* Check for keep-alive related responses. */
1599 if (!inhibit_keep_alive && contlen != -1)
1601 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1603 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1605 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1610 /* The server has promised that it will not close the connection
1611 when we're done. This means that we can register it. */
1612 register_persistent (conn->host, conn->port, sock, using_ssl);
1614 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1616 /* Authorization is required. */
1617 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1618 CLOSE_FINISH (sock);
1620 CLOSE_INVALIDATE (sock);
1621 pconn.authorized = false;
1622 if (!auth_finished && (user && passwd))
1624 /* IIS sends multiple copies of WWW-Authenticate, one with
1625 the value "negotiate", and other(s) with data. Loop over
1626 all the occurrences and pick the one we recognize. */
1628 const char *wabeg, *waend;
1629 char *www_authenticate = NULL;
1631 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1632 &wabeg, &waend)) != -1;
1634 if (known_authentication_scheme_p (wabeg, waend))
1636 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1640 if (!www_authenticate)
1641 /* If the authentication header is missing or
1642 unrecognized, there's no sense in retrying. */
1643 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1644 else if (BEGINS_WITH (www_authenticate, "Basic"))
1645 /* If the authentication scheme is "Basic", which we send
1646 by default, there's no sense in retrying either. (This
1647 should be changed when we stop sending "Basic" data by
1653 pth = url_full_path (u);
1654 request_set_header (req, "Authorization",
1655 create_authorization_line (www_authenticate,
1657 request_method (req),
1661 if (BEGINS_WITH (www_authenticate, "NTLM"))
1664 goto retry_with_auth;
1667 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1671 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1673 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1675 pconn.authorized = true;
1679 hs->statcode = statcode;
1681 hs->error = xstrdup (_("Malformed status line"));
1683 hs->error = xstrdup (_("(no description)"));
1685 hs->error = xstrdup (message);
1688 type = resp_header_strdup (resp, "Content-Type");
1691 char *tmp = strchr (type, ';');
1694 while (tmp > type && ISSPACE (tmp[-1]))
1699 hs->newloc = resp_header_strdup (resp, "Location");
1700 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1702 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1706 const char *scbeg, *scend;
1707 /* The jar should have been created by now. */
1708 assert (wget_cookie_jar != NULL);
1710 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1711 &scbeg, &scend)) != -1;
1714 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1715 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1716 u->path, set_cookie);
1720 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1722 wgint first_byte_pos, last_byte_pos, entity_length;
1723 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1725 contrange = first_byte_pos;
1729 /* 20x responses are counted among successful by default. */
1730 if (H_20X (statcode))
1733 /* Return if redirected. */
1734 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1736 /* RFC2068 says that in case of the 300 (multiple choices)
1737 response, the server can output a preferred URL through
1738 `Location' header; otherwise, the request should be treated
1739 like GET. So, if the location is set, it will be a
1740 redirection; otherwise, just proceed normally. */
1741 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1745 logprintf (LOG_VERBOSE,
1746 _("Location: %s%s\n"),
1747 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1748 hs->newloc ? _(" [following]") : "");
1749 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1750 CLOSE_FINISH (sock);
1752 CLOSE_INVALIDATE (sock);
1758 /* If content-type is not given, assume text/html. This is because
1759 of the multitude of broken CGI's that "forget" to generate the
1762 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1763 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1768 if (opt.html_extension && (*dt & TEXTHTML))
1769 /* -E / --html-extension / html_extension = on was specified, and this is a
1770 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1771 already the file's suffix, tack on ".html". */
1773 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1775 if (last_period_in_local_filename == NULL
1776 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1777 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1779 int local_filename_len = strlen (*hs->local_file);
1780 /* Resize the local file, allowing for ".html" preceded by
1781 optional ".NUMBER". */
1782 *hs->local_file = xrealloc (*hs->local_file,
1783 local_filename_len + 24 + sizeof (".html"));
1784 strcpy(*hs->local_file + local_filename_len, ".html");
1785 /* If clobbering is not allowed and the file, as named,
1786 exists, tack on ".NUMBER.html" instead. */
1791 sprintf (*hs->local_file + local_filename_len,
1792 ".%d.html", ext_num++);
1793 while (file_exists_p (*hs->local_file));
1795 *dt |= ADDED_HTML_EXTENSION;
1799 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1801 /* If `-c' is in use and the file has been fully downloaded (or
1802 the remote file has shrunk), Wget effectively requests bytes
1803 after the end of file and the server response with 416. */
1804 logputs (LOG_VERBOSE, _("\
1805 \n The file is already fully retrieved; nothing to do.\n\n"));
1806 /* In case the caller inspects. */
1809 /* Mark as successfully retrieved. */
1812 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1813 might be more bytes in the body. */
1814 return RETRUNNEEDED;
1816 if ((contrange != 0 && contrange != hs->restval)
1817 || (H_PARTIAL (statcode) && !contrange))
1819 /* The Range request was somehow misunderstood by the server.
1822 CLOSE_INVALIDATE (sock);
1825 hs->contlen = contlen + contrange;
1831 /* No need to print this output if the body won't be
1832 downloaded at all, or if the original server response is
1834 logputs (LOG_VERBOSE, _("Length: "));
1837 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1838 if (contlen + contrange >= 1024)
1839 logprintf (LOG_VERBOSE, " (%s)",
1840 human_readable (contlen + contrange));
1843 if (contlen >= 1024)
1844 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1845 number_to_static_string (contlen),
1846 human_readable (contlen));
1848 logprintf (LOG_VERBOSE, _(", %s remaining"),
1849 number_to_static_string (contlen));
1853 logputs (LOG_VERBOSE,
1854 opt.ignore_length ? _("ignored") : _("unspecified"));
1856 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1858 logputs (LOG_VERBOSE, "\n");
1862 type = NULL; /* We don't need it any more. */
1864 /* Return if we have no intention of further downloading. */
1865 if (!(*dt & RETROKF) || head_only)
1867 /* In case the caller cares to look... */
1871 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1872 servers not to send body in response to a HEAD request. If
1873 you encounter such a server (more likely a broken CGI), use
1874 `--no-http-keep-alive'. */
1875 CLOSE_FINISH (sock);
1876 return RETRFINISHED;
1879 /* Open the local file. */
1882 mkalldirs (*hs->local_file);
1884 rotate_backups (*hs->local_file);
1886 fp = fopen (*hs->local_file, "ab");
1887 else if (ALLOW_CLOBBER)
1888 fp = fopen (*hs->local_file, "wb");
1891 fp = fopen_excl (*hs->local_file, true);
1892 if (!fp && errno == EEXIST)
1894 /* We cannot just invent a new name and use it (which is
1895 what functions like unique_create typically do)
1896 because we told the user we'd use this name.
1897 Instead, return and retry the download. */
1898 logprintf (LOG_NOTQUIET,
1899 _("%s has sprung into existence.\n"),
1901 CLOSE_INVALIDATE (sock);
1902 return FOPEN_EXCL_ERR;
1907 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1908 CLOSE_INVALIDATE (sock);
1915 /* #### This confuses the timestamping code that checks for file
1916 size. Maybe we should save some additional information? */
1917 if (opt.save_headers)
1918 fwrite (head, 1, strlen (head), fp);
1920 /* Now we no longer need to store the response header. */
1923 /* Download the request body. */
1926 flags |= rb_read_exactly;
1927 if (hs->restval > 0 && contrange == 0)
1928 /* If the server ignored our range request, instruct fd_read_body
1929 to skip the first RESTVAL bytes of body. */
1930 flags |= rb_skip_startpos;
1931 hs->len = hs->restval;
1933 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1934 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1938 CLOSE_FINISH (sock);
1940 CLOSE_INVALIDATE (sock);
1943 /* Close or flush the file. We have to be careful to check for
1944 error here. Checking the result of fwrite() is not enough --
1945 errors could go unnoticed! */
1948 flush_res = fclose (fp);
1950 flush_res = fflush (fp);
1951 if (flush_res == EOF)
1956 return RETRFINISHED;
1959 /* The genuine HTTP loop! This is the part where the retrieval is
1960 retried, and retried, and retried, and... */
1962 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1963 int *dt, struct url *proxy)
1966 bool use_ts, got_head = false;/* time-stamping info */
1967 char *filename_plus_orig_suffix;
1968 char *local_filename = NULL;
1972 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1973 wgint local_size = 0; /* the size of the local file */
1974 size_t filename_len;
1975 struct http_stat hstat; /* HTTP status */
1979 /* This used to be done in main(), but it's a better idea to do it
1980 here so that we don't go through the hoops if we're just using
1984 if (!wget_cookie_jar)
1985 wget_cookie_jar = cookie_jar_new ();
1986 if (opt.cookies_input && !cookies_loaded_p)
1988 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1989 cookies_loaded_p = true;
1995 /* Warn on (likely bogus) wildcard usage in HTTP. */
1996 if (opt.ftp_glob && has_wildcards_p (u->path))
1997 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2001 /* Determine the local filename. */
2002 if (local_file && *local_file)
2003 hstat.local_file = local_file;
2004 else if (local_file && !opt.output_document)
2006 *local_file = url_file_name (u);
2007 hstat.local_file = local_file;
2011 dummy = url_file_name (u);
2012 hstat.local_file = &dummy;
2013 /* be honest about where we will save the file */
2014 if (local_file && opt.output_document)
2015 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2018 if (!opt.output_document)
2019 locf = *hstat.local_file;
2021 locf = opt.output_document;
2023 hstat.referer = referer;
2025 filename_len = strlen (*hstat.local_file);
2026 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2028 if (opt.noclobber && file_exists_p (*hstat.local_file))
2030 /* If opt.noclobber is turned on and file already exists, do not
2031 retrieve the file */
2032 logprintf (LOG_VERBOSE, _("\
2033 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2034 /* If the file is there, we suppose it's retrieved OK. */
2037 /* #### Bogusness alert. */
2038 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2039 if (has_html_suffix_p (*hstat.local_file))
2047 if (opt.timestamping)
2049 bool local_dot_orig_file_exists = false;
2051 if (opt.backup_converted)
2052 /* If -K is specified, we'll act on the assumption that it was specified
2053 last time these files were downloaded as well, and instead of just
2054 comparing local file X against server file X, we'll compare local
2055 file X.orig (if extant, else X) against server file X. If -K
2056 _wasn't_ specified last time, or the server contains files called
2057 *.orig, -N will be back to not operating correctly with -k. */
2059 /* Would a single s[n]printf() call be faster? --dan
2061 Definitely not. sprintf() is horribly slow. It's a
2062 different question whether the difference between the two
2063 affects a program. Usually I'd say "no", but at one
2064 point I profiled Wget, and found that a measurable and
2065 non-negligible amount of time was lost calling sprintf()
2066 in url.c. Replacing sprintf with inline calls to
2067 strcpy() and number_to_string() made a difference.
2069 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2070 memcpy (filename_plus_orig_suffix + filename_len,
2071 ".orig", sizeof (".orig"));
2073 /* Try to stat() the .orig file. */
2074 if (stat (filename_plus_orig_suffix, &st) == 0)
2076 local_dot_orig_file_exists = 1;
2077 local_filename = filename_plus_orig_suffix;
2081 if (!local_dot_orig_file_exists)
2082 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2083 if (stat (*hstat.local_file, &st) == 0)
2084 local_filename = *hstat.local_file;
2086 if (local_filename != NULL)
2087 /* There was a local file, so we'll check later to see if the version
2088 the server has is the same version we already have, allowing us to
2094 /* Modification time granularity is 2 seconds for Windows, so
2095 increase local time by 1 second for later comparison. */
2098 local_size = st.st_size;
2102 /* Reset the counter. */
2108 /* Increment the pass counter. */
2110 sleep_between_retrievals (count);
2111 /* Get the current time string. */
2112 tms = time_str (NULL);
2113 /* Print fetch message, if opt.verbose. */
2116 char *hurl = url_string (u, true);
2120 sprintf (tmp, _("(try:%2d)"), count);
2121 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2122 tms, hurl, tmp, locf);
2124 ws_changetitle (hurl);
2129 /* Default document type is empty. However, if spider mode is
2130 on or time-stamping is employed, HEAD_ONLY commands is
2131 encoded within *dt. */
2132 if (opt.spider || (use_ts && !got_head))
2137 /* Decide whether or not to restart. */
2139 && stat (locf, &st) == 0
2140 && S_ISREG (st.st_mode))
2141 /* When -c is used, continue from on-disk size. (Can't use
2142 hstat.len even if count>1 because we don't want a failed
2143 first attempt to clobber existing data.) */
2144 hstat.restval = st.st_size;
2146 /* otherwise, continue where the previous try left off */
2147 hstat.restval = hstat.len;
2151 /* Decide whether to send the no-cache directive. We send it in
2153 a) we're using a proxy, and we're past our first retrieval.
2154 Some proxies are notorious for caching incomplete data, so
2155 we require a fresh get.
2156 b) caching is explicitly inhibited. */
2157 if ((proxy && count > 1) /* a */
2158 || !opt.allow_cache /* b */
2160 *dt |= SEND_NOCACHE;
2162 *dt &= ~SEND_NOCACHE;
2164 /* Try fetching the document, or at least its head. */
2165 err = gethttp (u, &hstat, dt, proxy);
2167 /* It's unfortunate that wget determines the local filename before finding
2168 out the Content-Type of the file. Barring a major restructuring of the
2169 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2170 *hstat.local_file to tack on ".html". */
2171 if (!opt.output_document)
2172 locf = *hstat.local_file;
2175 tms = time_str (NULL);
2176 /* Get the new location (with or without the redirection). */
2178 *newloc = xstrdup (hstat.newloc);
2181 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2182 case CONERROR: case READERR: case WRITEFAILED:
2183 case RANGEERR: case FOPEN_EXCL_ERR:
2184 /* Non-fatal errors continue executing the loop, which will
2185 bring them to "while" statement at the end, to judge
2186 whether the number of tries was exceeded. */
2187 free_hstat (&hstat);
2188 printwhat (count, opt.ntry);
2189 if (err == FOPEN_EXCL_ERR)
2191 /* Re-determine the file name. */
2192 if (local_file && *local_file)
2194 xfree (*local_file);
2195 *local_file = url_file_name (u);
2196 hstat.local_file = local_file;
2201 dummy = url_file_name (u);
2202 hstat.local_file = &dummy;
2204 /* be honest about where we will save the file */
2205 if (local_file && opt.output_document)
2206 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2207 if (!opt.output_document)
2208 locf = *hstat.local_file;
2210 locf = opt.output_document;
2213 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2214 case SSLINITFAILED: case CONTNOTSUPPORTED:
2215 /* Fatal errors just return from the function. */
2216 free_hstat (&hstat);
2219 case FWRITEERR: case FOPENERR:
2220 /* Another fatal error. */
2221 logputs (LOG_VERBOSE, "\n");
2222 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2223 *hstat.local_file, strerror (errno));
2224 free_hstat (&hstat);
2228 /* Another fatal error. */
2229 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2230 free_hstat (&hstat);
2234 /* Return the new location to the caller. */
2237 logprintf (LOG_NOTQUIET,
2238 _("ERROR: Redirection (%d) without location.\n"),
2240 free_hstat (&hstat);
2244 free_hstat (&hstat);
2248 /* The file was already fully retrieved. */
2249 free_hstat (&hstat);
2253 /* Deal with you later. */
2256 /* All possibilities should have been exhausted. */
2259 if (!(*dt & RETROKF))
2263 /* #### Ugly ugly ugly! */
2264 char *hurl = url_string (u, true);
2265 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2268 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2269 tms, hstat.statcode, escnonprint (hstat.error));
2270 logputs (LOG_VERBOSE, "\n");
2271 free_hstat (&hstat);
2276 /* Did we get the time-stamp? */
2279 if (opt.timestamping && !hstat.remote_time)
2281 logputs (LOG_NOTQUIET, _("\
2282 Last-modified header missing -- time-stamps turned off.\n"));
2284 else if (hstat.remote_time)
2286 /* Convert the date-string into struct tm. */
2287 tmr = http_atotm (hstat.remote_time);
2288 if (tmr == (time_t) (-1))
2289 logputs (LOG_VERBOSE, _("\
2290 Last-modified header invalid -- time-stamp ignored.\n"));
2294 /* The time-stamping section. */
2299 use_ts = false; /* no more time-stamping */
2300 count = 0; /* the retrieve count for HEAD is
2302 if (hstat.remote_time && tmr != (time_t) (-1))
2304 /* Now time-stamping can be used validly. Time-stamping
2305 means that if the sizes of the local and remote file
2306 match, and local file is newer than the remote file,
2307 it will not be retrieved. Otherwise, the normal
2308 download procedure is resumed. */
2310 (hstat.contlen == -1 || local_size == hstat.contlen))
2312 logprintf (LOG_VERBOSE, _("\
2313 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2315 free_hstat (&hstat);
2319 else if (tml >= tmr)
2320 logprintf (LOG_VERBOSE, _("\
2321 The sizes do not match (local %s) -- retrieving.\n"),
2322 number_to_static_string (local_size));
2324 logputs (LOG_VERBOSE,
2325 _("Remote file is newer, retrieving.\n"));
2327 free_hstat (&hstat);
2330 if ((tmr != (time_t) (-1))
2332 && ((hstat.len == hstat.contlen) ||
2333 ((hstat.res == 0) &&
2334 ((hstat.contlen == -1) ||
2335 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2337 /* #### This code repeats in http.c and ftp.c. Move it to a
2339 const char *fl = NULL;
2340 if (opt.output_document)
2342 if (output_stream_regular)
2343 fl = opt.output_document;
2346 fl = *hstat.local_file;
2350 /* End of time-stamping section. */
2354 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2355 escnonprint (hstat.error));
2360 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2361 total_download_time += hstat.dltime;
2363 if (hstat.len == hstat.contlen)
2367 logprintf (LOG_VERBOSE,
2368 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2370 number_to_static_string (hstat.len),
2371 number_to_static_string (hstat.contlen));
2372 logprintf (LOG_NONVERBOSE,
2373 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2375 number_to_static_string (hstat.len),
2376 number_to_static_string (hstat.contlen),
2380 total_downloaded_bytes += hstat.len;
2382 /* Remember that we downloaded the file for later ".orig" code. */
2383 if (*dt & ADDED_HTML_EXTENSION)
2384 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2386 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2388 free_hstat (&hstat);
2392 else if (hstat.res == 0) /* No read error */
2394 if (hstat.contlen == -1) /* We don't know how much we were supposed
2395 to get, so assume we succeeded. */
2399 logprintf (LOG_VERBOSE,
2400 _("%s (%s) - `%s' saved [%s]\n\n"),
2402 number_to_static_string (hstat.len));
2403 logprintf (LOG_NONVERBOSE,
2404 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2405 tms, u->url, number_to_static_string (hstat.len),
2409 total_downloaded_bytes += hstat.len;
2411 /* Remember that we downloaded the file for later ".orig" code. */
2412 if (*dt & ADDED_HTML_EXTENSION)
2413 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2415 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2417 free_hstat (&hstat);
2421 else if (hstat.len < hstat.contlen) /* meaning we lost the
2422 connection too soon */
2424 logprintf (LOG_VERBOSE,
2425 _("%s (%s) - Connection closed at byte %s. "),
2426 tms, tmrate, number_to_static_string (hstat.len));
2427 printwhat (count, opt.ntry);
2428 free_hstat (&hstat);
2431 else if (!opt.kill_longer) /* meaning we got more than expected */
2433 logprintf (LOG_VERBOSE,
2434 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2436 number_to_static_string (hstat.len),
2437 number_to_static_string (hstat.contlen));
2438 logprintf (LOG_NONVERBOSE,
2439 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2441 number_to_static_string (hstat.len),
2442 number_to_static_string (hstat.contlen),
2445 total_downloaded_bytes += hstat.len;
2447 /* Remember that we downloaded the file for later ".orig" code. */
2448 if (*dt & ADDED_HTML_EXTENSION)
2449 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2451 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2453 free_hstat (&hstat);
2457 else /* the same, but not accepted */
2459 logprintf (LOG_VERBOSE,
2460 _("%s (%s) - Connection closed at byte %s/%s. "),
2462 number_to_static_string (hstat.len),
2463 number_to_static_string (hstat.contlen));
2464 printwhat (count, opt.ntry);
2465 free_hstat (&hstat);
2469 else /* now hstat.res can only be -1 */
2471 if (hstat.contlen == -1)
2473 logprintf (LOG_VERBOSE,
2474 _("%s (%s) - Read error at byte %s (%s)."),
2475 tms, tmrate, number_to_static_string (hstat.len),
2477 printwhat (count, opt.ntry);
2478 free_hstat (&hstat);
2481 else /* hstat.res == -1 and contlen is given */
2483 logprintf (LOG_VERBOSE,
2484 _("%s (%s) - Read error at byte %s/%s (%s). "),
2486 number_to_static_string (hstat.len),
2487 number_to_static_string (hstat.contlen),
2489 printwhat (count, opt.ntry);
2490 free_hstat (&hstat);
2496 while (!opt.ntry || (count < opt.ntry));
2500 /* Check whether the result of strptime() indicates success.
2501 strptime() returns the pointer to how far it got to in the string.
2502 The processing has been successful if the string is at `GMT' or
2503 `+X', or at the end of the string.
2505 In extended regexp parlance, the function returns 1 if P matches
2506 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2507 can return) is considered a failure and 0 is returned. */
2509 check_end (const char *p)
2513 while (ISSPACE (*p))
2516 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2517 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2523 /* Convert the textual specification of time in TIME_STRING to the
2524 number of seconds since the Epoch.
2526 TIME_STRING can be in any of the three formats RFC2616 allows the
2527 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2528 as well as the time format used in the Set-Cookie header.
2529 Timezones are ignored, and should be GMT.
2531 Return the computed time_t representation, or -1 if the conversion
2534 This function uses strptime with various string formats for parsing
2535 TIME_STRING. This results in a parser that is not as lenient in
2536 interpreting TIME_STRING as I would like it to be. Being based on
2537 strptime, it always allows shortened months, one-digit days, etc.,
2538 but due to the multitude of formats in which time can be
2539 represented, an ideal HTTP time parser would be even more
2540 forgiving. It should completely ignore things like week days and
2541 concentrate only on the various forms of representing years,
2542 months, days, hours, minutes, and seconds. For example, it would
2543 be nice if it accepted ISO 8601 out of the box.
2545 I've investigated free and PD code for this purpose, but none was
2546 usable. getdate was big and unwieldy, and had potential copyright
2547 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2548 distributed with phttpd, is excellent, but we cannot use it because
2549 it is not assigned to the FSF. So I stuck it with strptime. */
2552 http_atotm (const char *time_string)
2554 /* NOTE: Solaris strptime man page claims that %n and %t match white
2555 space, but that's not universally available. Instead, we simply
2556 use ` ' to mean "skip all WS", which works under all strptime
2557 implementations I've tested. */
2559 static const char *time_formats[] = {
2560 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2561 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2562 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2563 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2564 (used in Set-Cookie, defined in the
2565 Netscape cookie specification.) */
2567 const char *oldlocale;
2569 time_t ret = (time_t) -1;
2571 /* Solaris strptime fails to recognize English month names in
2572 non-English locales, which we work around by temporarily setting
2573 locale to C before invoking strptime. */
2574 oldlocale = setlocale (LC_TIME, NULL);
2575 setlocale (LC_TIME, "C");
2577 for (i = 0; i < countof (time_formats); i++)
2581 /* Some versions of strptime use the existing contents of struct
2582 tm to recalculate the date according to format. Zero it out
2583 to prevent stack garbage from influencing strptime. */
2586 if (check_end (strptime (time_string, time_formats[i], &t)))
2593 /* Restore the previous locale. */
2594 setlocale (LC_TIME, oldlocale);
2599 /* Authorization support: We support three authorization schemes:
2601 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2603 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2604 consisting of answering to the server's challenge with the proper
2607 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2608 Stenberg for libcurl. Like digest, NTLM is based on a
2609 challenge-response mechanism, but unlike digest, it is non-standard
2610 (authenticates TCP connections rather than requests), undocumented
2611 and Microsoft-specific. */
2613 /* Create the authentication header contents for the `Basic' scheme.
2614 This is done by encoding the string "USER:PASS" to base64 and
2615 prepending the string "Basic " in front of it. */
2618 basic_authentication_encode (const char *user, const char *passwd)
2621 int len1 = strlen (user) + 1 + strlen (passwd);
2623 t1 = (char *)alloca (len1 + 1);
2624 sprintf (t1, "%s:%s", user, passwd);
2626 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2627 base64_encode (t1, len1, t2);
2629 return concat_strings ("Basic ", t2, (char *) 0);
2632 #define SKIP_WS(x) do { \
2633 while (ISSPACE (*(x))) \
2637 #ifdef ENABLE_DIGEST
2638 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2639 of a field in such a header. If the field is the one specified by
2640 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2641 digest authorization code), extract its value in the (char*)
2642 variable pointed by RET. Returns negative on a malformed header,
2643 or number of bytes that have been parsed by this call. */
2645 extract_header_attr (const char *au, const char *attr_name, char **ret)
2648 const char *cp = au;
2650 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2652 cp += strlen (attr_name);
2665 for (ep = cp; *ep && *ep != '\"'; ep++)
2670 *ret = strdupdelim (cp, ep);
2677 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2678 an array of 16 bytes containing the hash keys, and BUF should be a
2679 buffer of 33 writable characters (32 for hex digits plus one for
2680 zero termination). */
2682 dump_hash (char *buf, const unsigned char *hash)
2686 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2688 *buf++ = XNUM_TO_digit (*hash >> 4);
2689 *buf++ = XNUM_TO_digit (*hash & 0xf);
2694 /* Take the line apart to find the challenge, and compose a digest
2695 authorization header. See RFC2069 section 2.1.2. */
2697 digest_authentication_encode (const char *au, const char *user,
2698 const char *passwd, const char *method,
2701 static char *realm, *opaque, *nonce;
2706 { "realm", &realm },
2707 { "opaque", &opaque },
2712 realm = opaque = nonce = NULL;
2714 au += 6; /* skip over `Digest' */
2720 for (i = 0; i < countof (options); i++)
2722 int skip = extract_header_attr (au, options[i].name,
2723 options[i].variable);
2727 xfree_null (opaque);
2737 if (i == countof (options))
2739 while (*au && *au != '=')
2747 while (*au && *au != '\"')
2754 while (*au && *au != ',')
2759 if (!realm || !nonce || !user || !passwd || !path || !method)
2762 xfree_null (opaque);
2767 /* Calculate the digest value. */
2769 ALLOCA_MD5_CONTEXT (ctx);
2770 unsigned char hash[MD5_HASHLEN];
2771 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2772 char response_digest[MD5_HASHLEN * 2 + 1];
2774 /* A1BUF = H(user ":" realm ":" password) */
2776 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2777 gen_md5_update ((unsigned char *)":", 1, ctx);
2778 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2779 gen_md5_update ((unsigned char *)":", 1, ctx);
2780 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2781 gen_md5_finish (ctx, hash);
2782 dump_hash (a1buf, hash);
2784 /* A2BUF = H(method ":" path) */
2786 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2787 gen_md5_update ((unsigned char *)":", 1, ctx);
2788 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2789 gen_md5_finish (ctx, hash);
2790 dump_hash (a2buf, hash);
2792 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2794 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2795 gen_md5_update ((unsigned char *)":", 1, ctx);
2796 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2797 gen_md5_update ((unsigned char *)":", 1, ctx);
2798 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2799 gen_md5_finish (ctx, hash);
2800 dump_hash (response_digest, hash);
2802 res = xmalloc (strlen (user)
2807 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2808 + (opaque ? strlen (opaque) : 0)
2810 sprintf (res, "Digest \
2811 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2812 user, realm, nonce, path, response_digest);
2815 char *p = res + strlen (res);
2816 strcat (p, ", opaque=\"");
2823 #endif /* ENABLE_DIGEST */
2825 /* Computing the size of a string literal must take into account that
2826 value returned by sizeof includes the terminating \0. */
2827 #define STRSIZE(literal) (sizeof (literal) - 1)
2829 /* Whether chars in [b, e) begin with the literal string provided as
2830 first argument and are followed by whitespace or terminating \0.
2831 The comparison is case-insensitive. */
2832 #define STARTS(literal, b, e) \
2833 ((e) - (b) >= STRSIZE (literal) \
2834 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2835 && ((e) - (b) == STRSIZE (literal) \
2836 || ISSPACE (b[STRSIZE (literal)])))
2839 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2841 return STARTS ("Basic", hdrbeg, hdrend)
2842 #ifdef ENABLE_DIGEST
2843 || STARTS ("Digest", hdrbeg, hdrend)
2846 || STARTS ("NTLM", hdrbeg, hdrend)
2853 /* Create the HTTP authorization request header. When the
2854 `WWW-Authenticate' response header is seen, according to the
2855 authorization scheme specified in that header (`Basic' and `Digest'
2856 are supported by the current implementation), produce an
2857 appropriate HTTP authorization request header. */
2859 create_authorization_line (const char *au, const char *user,
2860 const char *passwd, const char *method,
2861 const char *path, bool *finished)
2863 /* We are called only with known schemes, so we can dispatch on the
2865 switch (TOUPPER (*au))
2867 case 'B': /* Basic */
2869 return basic_authentication_encode (user, passwd);
2870 #ifdef ENABLE_DIGEST
2871 case 'D': /* Digest */
2873 return digest_authentication_encode (au, user, passwd, method, path);
2876 case 'N': /* NTLM */
2877 if (!ntlm_input (&pconn.ntlm, au))
2882 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2885 /* We shouldn't get here -- this function should be only called
2886 with values approved by known_authentication_scheme_p. */
2894 if (wget_cookie_jar)
2895 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2901 xfree_null (pconn.host);
2902 if (wget_cookie_jar)
2903 cookie_jar_delete (wget_cookie_jar);