2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
34 #include <sys/types.h>
45 #if TIME_WITH_SYS_TIME
46 # include <sys/time.h>
50 # include <sys/time.h>
70 # include "http-ntlm.h"
78 extern char *version_string;
79 extern LARGE_INT total_downloaded_bytes;
81 extern FILE *output_stream;
82 extern int output_stream_regular;
85 # define MIN(x, y) ((x) > (y) ? (y) : (x))
89 static int cookies_loaded_p;
90 static struct cookie_jar *wget_cookie_jar;
92 #define TEXTHTML_S "text/html"
93 #define TEXTXHTML_S "application/xhtml+xml"
95 /* Some status code validation macros: */
96 #define H_20X(x) (((x) >= 200) && ((x) < 300))
97 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
98 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
99 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
100 || (x) == HTTP_STATUS_SEE_OTHER \
101 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
103 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
104 /* Successful 2xx. */
105 #define HTTP_STATUS_OK 200
106 #define HTTP_STATUS_CREATED 201
107 #define HTTP_STATUS_ACCEPTED 202
108 #define HTTP_STATUS_NO_CONTENT 204
109 #define HTTP_STATUS_PARTIAL_CONTENTS 206
111 /* Redirection 3xx. */
112 #define HTTP_STATUS_MULTIPLE_CHOICES 300
113 #define HTTP_STATUS_MOVED_PERMANENTLY 301
114 #define HTTP_STATUS_MOVED_TEMPORARILY 302
115 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
116 #define HTTP_STATUS_NOT_MODIFIED 304
117 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
119 /* Client error 4xx. */
120 #define HTTP_STATUS_BAD_REQUEST 400
121 #define HTTP_STATUS_UNAUTHORIZED 401
122 #define HTTP_STATUS_FORBIDDEN 403
123 #define HTTP_STATUS_NOT_FOUND 404
124 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
126 /* Server errors 5xx. */
127 #define HTTP_STATUS_INTERNAL 500
128 #define HTTP_STATUS_NOT_IMPLEMENTED 501
129 #define HTTP_STATUS_BAD_GATEWAY 502
130 #define HTTP_STATUS_UNAVAILABLE 503
133 rel_none, rel_name, rel_value, rel_both
140 struct request_header {
142 enum rp release_policy;
144 int hcount, hcapacity;
147 /* Create a new, empty request. At least request_set_method must be
148 called before the request can be used. */
150 static struct request *
153 struct request *req = xnew0 (struct request);
155 req->headers = xnew_array (struct request_header, req->hcapacity);
159 /* Set the request's method and its arguments. METH should be a
160 literal string (or it should outlive the request) because it will
161 not be freed. ARG will be freed by request_free. */
164 request_set_method (struct request *req, const char *meth, char *arg)
170 /* Return the method string passed with the last call to
171 request_set_method. */
174 request_method (const struct request *req)
179 /* Free one header according to the release policy specified with
180 request_set_header. */
183 release_header (struct request_header *hdr)
185 switch (hdr->release_policy)
202 /* Set the request named NAME to VALUE. Specifically, this means that
203 a "NAME: VALUE\r\n" header line will be used in the request. If a
204 header with the same name previously existed in the request, its
205 value will be replaced by this one. A NULL value means do nothing.
207 RELEASE_POLICY determines whether NAME and VALUE should be released
208 (freed) with request_free. Allowed values are:
210 - rel_none - don't free NAME or VALUE
211 - rel_name - free NAME when done
212 - rel_value - free VALUE when done
213 - rel_both - free both NAME and VALUE when done
215 Setting release policy is useful when arguments come from different
216 sources. For example:
218 // Don't free literal strings!
219 request_set_header (req, "Pragma", "no-cache", rel_none);
221 // Don't free a global variable, we'll need it later.
222 request_set_header (req, "Referer", opt.referer, rel_none);
224 // Value freshly allocated, free it when done.
225 request_set_header (req, "Range",
226 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
231 request_set_header (struct request *req, char *name, char *value,
232 enum rp release_policy)
234 struct request_header *hdr;
239 /* A NULL value is a no-op; if freeing the name is requested,
240 free it now to avoid leaks. */
241 if (release_policy == rel_name || release_policy == rel_both)
246 for (i = 0; i < req->hcount; i++)
248 hdr = &req->headers[i];
249 if (0 == strcasecmp (name, hdr->name))
251 /* Replace existing header. */
252 release_header (hdr);
255 hdr->release_policy = release_policy;
260 /* Install new header. */
262 if (req->hcount >= req->hcapacity)
264 req->hcapacity <<= 1;
265 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
267 hdr = &req->headers[req->hcount++];
270 hdr->release_policy = release_policy;
273 /* Like request_set_header, but sets the whole header line, as
274 provided by the user using the `--header' option. For example,
275 request_set_user_header (req, "Foo: bar") works just like
276 request_set_header (req, "Foo", "bar"). */
279 request_set_user_header (struct request *req, const char *header)
282 const char *p = strchr (header, ':');
285 BOUNDED_TO_ALLOCA (header, p, name);
289 request_set_header (req, xstrdup (name), (char *) p, rel_name);
292 /* Remove the header with specified name from REQ. Returns 1 if the
293 header was actually removed, 0 otherwise. */
296 request_remove_header (struct request *req, char *name)
299 for (i = 0; i < req->hcount; i++)
301 struct request_header *hdr = &req->headers[i];
302 if (0 == strcasecmp (name, hdr->name))
304 release_header (hdr);
305 /* Move the remaining headers by one. */
306 if (i < req->hcount - 1)
307 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
315 #define APPEND(p, str) do { \
316 int A_len = strlen (str); \
317 memcpy (p, str, A_len); \
321 /* Construct the request and write it to FD using fd_write. */
324 request_send (const struct request *req, int fd)
326 char *request_string, *p;
327 int i, size, write_error;
329 /* Count the request size. */
332 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
333 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
335 for (i = 0; i < req->hcount; i++)
337 struct request_header *hdr = &req->headers[i];
338 /* NAME ": " VALUE "\r\n" */
339 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
345 p = request_string = alloca_array (char, size);
347 /* Generate the request. */
349 APPEND (p, req->method); *p++ = ' ';
350 APPEND (p, req->arg); *p++ = ' ';
351 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
353 for (i = 0; i < req->hcount; i++)
355 struct request_header *hdr = &req->headers[i];
356 APPEND (p, hdr->name);
357 *p++ = ':', *p++ = ' ';
358 APPEND (p, hdr->value);
359 *p++ = '\r', *p++ = '\n';
362 *p++ = '\r', *p++ = '\n', *p++ = '\0';
363 assert (p - request_string == size);
367 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
369 /* Send the request to the server. */
371 write_error = fd_write (fd, request_string, size - 1, -1.0);
373 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
378 /* Release the resources used by REQ. */
381 request_free (struct request *req)
384 xfree_null (req->arg);
385 for (i = 0; i < req->hcount; i++)
386 release_header (&req->headers[i]);
387 xfree_null (req->headers);
391 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
392 PROMISED_SIZE bytes are sent over the wire -- if the file is
393 longer, read only that much; if the file is shorter, report an error. */
396 post_file (int sock, const char *file_name, wgint promised_size)
398 static char chunk[8192];
403 DEBUGP (("[writing POST file %s ... ", file_name));
405 fp = fopen (file_name, "rb");
408 while (!feof (fp) && written < promised_size)
411 int length = fread (chunk, 1, sizeof (chunk), fp);
414 towrite = MIN (promised_size - written, length);
415 write_error = fd_write (sock, chunk, towrite, -1.0);
425 /* If we've written less than was promised, report a (probably
426 nonsensical) error rather than break the promise. */
427 if (written < promised_size)
433 assert (written == promised_size);
434 DEBUGP (("done]\n"));
439 response_head_terminator (const char *hunk, int oldlen, int peeklen)
441 const char *start, *end;
443 /* If at first peek, verify whether HUNK starts with "HTTP". If
444 not, this is a HTTP/0.9 request and we must bail out without
446 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
452 start = hunk + oldlen - 4;
453 end = hunk + oldlen + peeklen;
455 for (; start < end - 1; start++)
462 if (start[1] == '\n')
468 /* The maximum size of a single HTTP response we care to read. This
469 is not meant to impose an arbitrary limit, but to protect the user
470 from Wget slurping up available memory upon encountering malicious
471 or buggy server output. Define it to 0 to remove the limit. */
473 #define HTTP_RESPONSE_MAX_SIZE 65536
475 /* Read the HTTP request head from FD and return it. The error
476 conditions are the same as with fd_read_hunk.
478 To support HTTP/0.9 responses, this function tries to make sure
479 that the data begins with "HTTP". If this is not the case, no data
480 is read and an empty request is returned, so that the remaining
481 data can be treated as body. */
484 read_http_response_head (int fd)
486 return fd_read_hunk (fd, response_head_terminator, 512,
487 HTTP_RESPONSE_MAX_SIZE);
491 /* The response data. */
494 /* The array of pointers that indicate where each header starts.
495 For example, given this HTTP response:
502 The headers are located like this:
504 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
506 headers[0] headers[1] headers[2] headers[3]
508 I.e. headers[0] points to the beginning of the request,
509 headers[1] points to the end of the first header and the
510 beginning of the second one, etc. */
512 const char **headers;
515 /* Create a new response object from the text of the HTTP response,
516 available in HEAD. That text is automatically split into
517 constituent header lines for fast retrieval using
520 static struct response *
521 resp_new (const char *head)
526 struct response *resp = xnew0 (struct response);
531 /* Empty head means that we're dealing with a headerless
532 (HTTP/0.9) response. In that case, don't set HEADERS at
537 /* Split HEAD into header lines, so that resp_header_* functions
538 don't need to do this over and over again. */
544 DO_REALLOC (resp->headers, size, count + 1, const char *);
545 resp->headers[count++] = hdr;
547 /* Break upon encountering an empty line. */
548 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
551 /* Find the end of HDR, including continuations. */
554 const char *end = strchr (hdr, '\n');
560 while (*hdr == ' ' || *hdr == '\t');
562 DO_REALLOC (resp->headers, size, count + 1, const char *);
563 resp->headers[count] = NULL;
568 /* Locate the header named NAME in the request data, starting with
569 position START. This allows the code to loop through the request
570 data, filtering for all requests of a given name. Returns the
571 found position, or -1 for failure. The code that uses this
572 function typically looks like this:
574 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
575 ... do something with header ...
577 If you only care about one header, use resp_header_get instead of
581 resp_header_locate (const struct response *resp, const char *name, int start,
582 const char **begptr, const char **endptr)
585 const char **headers = resp->headers;
588 if (!headers || !headers[1])
591 name_len = strlen (name);
597 for (; headers[i + 1]; i++)
599 const char *b = headers[i];
600 const char *e = headers[i + 1];
602 && b[name_len] == ':'
603 && 0 == strncasecmp (b, name, name_len))
606 while (b < e && ISSPACE (*b))
608 while (b < e && ISSPACE (e[-1]))
618 /* Find and retrieve the header named NAME in the request data. If
619 found, set *BEGPTR to its starting, and *ENDPTR to its ending
620 position, and return 1. Otherwise return 0.
622 This function is used as a building block for resp_header_copy
623 and resp_header_strdup. */
626 resp_header_get (const struct response *resp, const char *name,
627 const char **begptr, const char **endptr)
629 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
633 /* Copy the response header named NAME to buffer BUF, no longer than
634 BUFSIZE (BUFSIZE includes the terminating 0). If the header
635 exists, 1 is returned, otherwise 0. If there should be no limit on
636 the size of the header, use resp_header_strdup instead.
638 If BUFSIZE is 0, no data is copied, but the boolean indication of
639 whether the header is present is still returned. */
642 resp_header_copy (const struct response *resp, const char *name,
643 char *buf, int bufsize)
646 if (!resp_header_get (resp, name, &b, &e))
650 int len = MIN (e - b, bufsize - 1);
651 memcpy (buf, b, len);
657 /* Return the value of header named NAME in RESP, allocated with
658 malloc. If such a header does not exist in RESP, return NULL. */
661 resp_header_strdup (const struct response *resp, const char *name)
664 if (!resp_header_get (resp, name, &b, &e))
666 return strdupdelim (b, e);
669 /* Parse the HTTP status line, which is of format:
671 HTTP-Version SP Status-Code SP Reason-Phrase
673 The function returns the status-code, or -1 if the status line
674 appears malformed. The pointer to "reason-phrase" message is
675 returned in *MESSAGE. */
678 resp_status (const struct response *resp, char **message)
685 /* For a HTTP/0.9 response, assume status 200. */
687 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
691 p = resp->headers[0];
692 end = resp->headers[1];
698 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
702 /* Match the HTTP version. This is optional because Gnutella
703 servers have been reported to not specify HTTP version. */
704 if (p < end && *p == '/')
707 while (p < end && ISDIGIT (*p))
709 if (p < end && *p == '.')
711 while (p < end && ISDIGIT (*p))
715 while (p < end && ISSPACE (*p))
717 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
720 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
725 while (p < end && ISSPACE (*p))
727 while (p < end && ISSPACE (end[-1]))
729 *message = strdupdelim (p, end);
735 /* Release the resources used by RESP. */
738 resp_free (struct response *resp)
740 xfree_null (resp->headers);
744 /* Print the server response, line by line, omitting the trailing CRLF
745 from individual header lines, and prefixed with PREFIX. */
748 print_server_response (const struct response *resp, const char *prefix)
753 for (i = 0; resp->headers[i + 1]; i++)
755 const char *b = resp->headers[i];
756 const char *e = resp->headers[i + 1];
758 if (b < e && e[-1] == '\n')
760 if (b < e && e[-1] == '\r')
762 /* This is safe even on printfs with broken handling of "%.<n>s"
763 because resp->headers ends with \0. */
764 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
768 /* Parse the `Content-Range' header and extract the information it
769 contains. Returns 1 if successful, -1 otherwise. */
771 parse_content_range (const char *hdr, wgint *first_byte_ptr,
772 wgint *last_byte_ptr, wgint *entity_length_ptr)
776 /* Ancient versions of Netscape proxy server, presumably predating
777 rfc2068, sent out `Content-Range' without the "bytes"
779 if (!strncasecmp (hdr, "bytes", 5))
782 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
786 while (ISSPACE (*hdr))
793 for (num = 0; ISDIGIT (*hdr); hdr++)
794 num = 10 * num + (*hdr - '0');
795 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
797 *first_byte_ptr = num;
799 for (num = 0; ISDIGIT (*hdr); hdr++)
800 num = 10 * num + (*hdr - '0');
801 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
803 *last_byte_ptr = num;
805 for (num = 0; ISDIGIT (*hdr); hdr++)
806 num = 10 * num + (*hdr - '0');
807 *entity_length_ptr = num;
811 /* Read the body of the request, but don't store it anywhere and don't
812 display a progress gauge. This is useful for reading the bodies of
813 administrative responses to which we will soon issue another
814 request. The response is not useful to the user, but reading it
815 allows us to continue using the same connection to the server.
817 If reading fails, 0 is returned, non-zero otherwise. In debug
818 mode, the body is displayed for debugging purposes. */
821 skip_short_body (int fd, wgint contlen)
824 SKIP_SIZE = 512, /* size of the download buffer */
825 SKIP_THRESHOLD = 4096 /* the largest size we read */
827 char dlbuf[SKIP_SIZE + 1];
828 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
830 /* We shouldn't get here with unknown contlen. (This will change
831 with HTTP/1.1, which supports "chunked" transfer.) */
832 assert (contlen != -1);
834 /* If the body is too large, it makes more sense to simply close the
835 connection than to try to read the body. */
836 if (contlen > SKIP_THRESHOLD)
839 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
843 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1.0);
846 /* Don't normally report the error since this is an
847 optimization that should be invisible to the user. */
848 DEBUGP (("] aborting (%s).\n",
849 ret < 0 ? strerror (errno) : "EOF received"));
853 /* Safe even if %.*s bogusly expects terminating \0 because
854 we've zero-terminated dlbuf above. */
855 DEBUGP (("%.*s", ret, dlbuf));
858 DEBUGP (("] done.\n"));
862 /* Persistent connections. Currently, we cache the most recently used
863 connection as persistent, provided that the HTTP server agrees to
864 make it such. The persistence data is stored in the variables
865 below. Ideally, it should be possible to cache an arbitrary fixed
866 number of these connections. */
868 /* Whether a persistent connection is active. */
869 static int pconn_active;
872 /* The socket of the connection. */
875 /* Host and port of the currently active persistent connection. */
879 /* Whether a ssl handshake has occoured on this connection. */
882 /* Whether the connection was authorized. This is only done by
883 NTLM, which authorizes *connections* rather than individual
884 requests. (That practice is peculiar for HTTP, but it is a
885 useful optimization.) */
889 /* NTLM data of the current connection. */
890 struct ntlmdata ntlm;
894 /* Mark the persistent connection as invalid and free the resources it
895 uses. This is used by the CLOSE_* macros after they forcefully
896 close a registered persistent connection. */
899 invalidate_persistent (void)
901 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
903 fd_close (pconn.socket);
908 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
909 persistent. This will enable someone to use the same connection
910 later. In the context of HTTP, this must be called only AFTER the
911 response has been received and the server has promised that the
912 connection will remain alive.
914 If a previous connection was persistent, it is closed. */
917 register_persistent (const char *host, int port, int fd, int ssl)
921 if (pconn.socket == fd)
923 /* The connection FD is already registered. */
928 /* The old persistent connection is still active; close it
929 first. This situation arises whenever a persistent
930 connection exists, but we then connect to a different
931 host, and try to register a persistent connection to that
933 invalidate_persistent ();
939 pconn.host = xstrdup (host);
942 pconn.authorized = 0;
944 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
947 /* Return non-zero if a persistent connection is available for
948 connecting to HOST:PORT. */
951 persistent_available_p (const char *host, int port, int ssl,
952 int *host_lookup_failed)
954 /* First, check whether a persistent connection is active at all. */
958 /* If we want SSL and the last connection wasn't or vice versa,
959 don't use it. Checking for host and port is not enough because
960 HTTP and HTTPS can apparently coexist on the same port. */
961 if (ssl != pconn.ssl)
964 /* If we're not connecting to the same port, we're not interested. */
965 if (port != pconn.port)
968 /* If the host is the same, we're in business. If not, there is
969 still hope -- read below. */
970 if (0 != strcasecmp (host, pconn.host))
972 /* Check if pconn.socket is talking to HOST under another name.
973 This happens often when both sites are virtual hosts
974 distinguished only by name and served by the same network
975 interface, and hence the same web server (possibly set up by
976 the ISP and serving many different web sites). This
977 admittedly unconventional optimization does not contradict
978 HTTP and works well with popular server software. */
982 struct address_list *al;
985 /* Don't try to talk to two different SSL sites over the same
986 secure connection! (Besides, it's not clear that
987 name-based virtual hosting is even possible with SSL.) */
990 /* If pconn.socket's peer is one of the IP addresses HOST
991 resolves to, pconn.socket is for all intents and purposes
992 already talking to HOST. */
994 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
996 /* Can't get the peer's address -- something must be very
997 wrong with the connection. */
998 invalidate_persistent ();
1001 al = lookup_host (host, 0);
1004 *host_lookup_failed = 1;
1008 found = address_list_contains (al, &ip);
1009 address_list_release (al);
1014 /* The persistent connection's peer address was found among the
1015 addresses HOST resolved to; therefore, pconn.sock is in fact
1016 already talking to HOST -- no need to reconnect. */
1019 /* Finally, check whether the connection is still open. This is
1020 important because most server implement a liberal (short) timeout
1021 on persistent connections. Wget can of course always reconnect
1022 if the connection doesn't work out, but it's nicer to know in
1023 advance. This test is a logical followup of the first test, but
1024 is "expensive" and therefore placed at the end of the list. */
1026 if (!test_socket_open (pconn.socket))
1028 /* Oops, the socket is no longer open. Now that we know that,
1029 let's invalidate the persistent connection before returning
1031 invalidate_persistent ();
1038 /* The idea behind these two CLOSE macros is to distinguish between
1039 two cases: one when the job we've been doing is finished, and we
1040 want to close the connection and leave, and two when something is
1041 seriously wrong and we're closing the connection as part of
1044 In case of keep_alive, CLOSE_FINISH should leave the connection
1045 open, while CLOSE_INVALIDATE should still close it.
1047 Note that the semantics of the flag `keep_alive' is "this
1048 connection *will* be reused (the server has promised not to close
1049 the connection once we're done)", while the semantics of
1050 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1051 active, registered connection". */
1053 #define CLOSE_FINISH(fd) do { \
1056 if (pconn_active && (fd) == pconn.socket) \
1057 invalidate_persistent (); \
1066 #define CLOSE_INVALIDATE(fd) do { \
1067 if (pconn_active && (fd) == pconn.socket) \
1068 invalidate_persistent (); \
1076 wgint len; /* received length */
1077 wgint contlen; /* expected length */
1078 wgint restval; /* the restart value */
1079 int res; /* the result of last read */
1080 char *newloc; /* new location (redirection) */
1081 char *remote_time; /* remote time-stamp string */
1082 char *error; /* textual HTTP error */
1083 int statcode; /* status code */
1084 wgint rd_size; /* amount of data read from socket */
1085 double dltime; /* time it took to download the data */
1086 const char *referer; /* value of the referer header. */
1087 char **local_file; /* local file. */
1091 free_hstat (struct http_stat *hs)
1093 xfree_null (hs->newloc);
1094 xfree_null (hs->remote_time);
1095 xfree_null (hs->error);
1097 /* Guard against being called twice. */
1099 hs->remote_time = NULL;
1103 static char *create_authorization_line PARAMS ((const char *, const char *,
1104 const char *, const char *,
1105 const char *, int *));
1106 static char *basic_authentication_encode PARAMS ((const char *, const char *));
1107 static int known_authentication_scheme_p PARAMS ((const char *, const char *));
1109 time_t http_atotm PARAMS ((const char *));
1111 #define BEGINS_WITH(line, string_constant) \
1112 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1113 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1114 || !line[sizeof (string_constant) - 1]))
1116 #define SET_USER_AGENT(req) do { \
1117 if (!opt.useragent) \
1118 request_set_header (req, "User-Agent", \
1119 aprintf ("Wget/%s", version_string), rel_value); \
1120 else if (*opt.useragent) \
1121 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1124 /* The flags that allow clobbering the file (opening with "wb").
1125 Defined here to avoid repetition later. #### This will require
1127 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1128 || opt.dirstruct || opt.output_document)
1130 /* Retrieve a document through HTTP protocol. It recognizes status
1131 code, and correctly handles redirections. It closes the network
1132 socket. If it receives an error from the functions below it, it
1133 will print it if there is enough information to do so (almost
1134 always), returning the error to the caller (i.e. http_loop).
1136 Various HTTP parameters are stored to hs.
1138 If PROXY is non-NULL, the connection will be made to the proxy
1139 server, and u->url will be requested. */
1141 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1143 struct request *req;
1146 char *user, *passwd;
1150 wgint contlen, contrange;
1157 /* Set to 1 when the authorization has failed permanently and should
1158 not be tried again. */
1159 int auth_finished = 0;
1161 /* Whether NTLM authentication is used for this request. */
1164 /* Whether our connection to the remote host is through SSL. */
1167 /* Whether a HEAD request will be issued (as opposed to GET or
1169 int head_only = *dt & HEAD_ONLY;
1172 struct response *resp;
1176 /* Whether this connection will be kept alive after the HTTP request
1180 /* Whether keep-alive should be inhibited.
1182 RFC 2068 requests that 1.0 clients not send keep-alive requests
1183 to proxies. This is because many 1.0 proxies do not interpret
1184 the Connection header and transfer it to the remote server,
1185 causing it to not close the connection and leave both the proxy
1186 and the client hanging. */
1187 int inhibit_keep_alive =
1188 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1190 /* Headers sent when using POST. */
1191 wgint post_data_size = 0;
1193 int host_lookup_failed = 0;
1196 if (u->scheme == SCHEME_HTTPS)
1198 /* Initialize the SSL context. After this has once been done,
1199 it becomes a no-op. */
1202 scheme_disable (SCHEME_HTTPS);
1203 logprintf (LOG_NOTQUIET,
1204 _("Disabling SSL due to encountered errors.\n"));
1205 return SSLINITFAILED;
1208 #endif /* HAVE_SSL */
1211 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1212 know the local filename so we can save to it. */
1213 assert (*hs->local_file != NULL);
1215 /* Initialize certain elements of struct http_stat. */
1220 hs->remote_time = NULL;
1225 /* Prepare the request to send. */
1227 req = request_new ();
1230 const char *meth = "GET";
1233 else if (opt.post_file_name || opt.post_data)
1235 /* Use the full path, i.e. one that includes the leading slash and
1236 the query string. E.g. if u->path is "foo/bar" and u->query is
1237 "param=value", full_path will be "/foo/bar?param=value". */
1240 /* When using SSL over proxy, CONNECT establishes a direct
1241 connection to the HTTPS server. Therefore use the same
1242 argument as when talking to the server directly. */
1243 && u->scheme != SCHEME_HTTPS
1246 meth_arg = xstrdup (u->url);
1248 meth_arg = url_full_path (u);
1249 request_set_method (req, meth, meth_arg);
1252 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1253 if (*dt & SEND_NOCACHE)
1254 request_set_header (req, "Pragma", "no-cache", rel_none);
1256 request_set_header (req, "Range",
1257 aprintf ("bytes=%s-",
1258 number_to_static_string (hs->restval)),
1260 SET_USER_AGENT (req);
1261 request_set_header (req, "Accept", "*/*", rel_none);
1263 /* Find the username and password for authentication. */
1266 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1267 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1268 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1272 /* We have the username and the password, but haven't tried
1273 any authorization yet. Let's see if the "Basic" method
1274 works. If not, we'll come back here and construct a
1275 proper authorization method with the right challenges.
1277 If we didn't employ this kind of logic, every URL that
1278 requires authorization would have to be processed twice,
1279 which is very suboptimal and generates a bunch of false
1280 "unauthorized" errors in the server log.
1282 #### But this logic also has a serious problem when used
1283 with stronger authentications: we *first* transmit the
1284 username and the password in clear text, and *then* attempt a
1285 stronger authentication scheme. That cannot be right! We
1286 are only fortunate that almost everyone still uses the
1287 `Basic' scheme anyway.
1289 There should be an option to prevent this from happening, for
1290 those who use strong authentication schemes and value their
1292 request_set_header (req, "Authorization",
1293 basic_authentication_encode (user, passwd),
1300 char *proxy_user, *proxy_passwd;
1301 /* For normal username and password, URL components override
1302 command-line/wgetrc parameters. With proxy
1303 authentication, it's the reverse, because proxy URLs are
1304 normally the "permanent" ones, so command-line args
1305 should take precedence. */
1306 if (opt.proxy_user && opt.proxy_passwd)
1308 proxy_user = opt.proxy_user;
1309 proxy_passwd = opt.proxy_passwd;
1313 proxy_user = proxy->user;
1314 proxy_passwd = proxy->passwd;
1316 /* #### This does not appear right. Can't the proxy request,
1317 say, `Digest' authentication? */
1318 if (proxy_user && proxy_passwd)
1319 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1321 /* If we're using a proxy, we will be connecting to the proxy
1325 /* Proxy authorization over SSL is handled below. */
1327 if (u->scheme != SCHEME_HTTPS)
1329 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1333 /* Whether we need to print the host header with braces around
1334 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1335 usual "Host: symbolic-name:1234". */
1336 int squares = strchr (u->host, ':') != NULL;
1337 if (u->port == scheme_default_port (u->scheme))
1338 request_set_header (req, "Host",
1339 aprintf (squares ? "[%s]" : "%s", u->host),
1342 request_set_header (req, "Host",
1343 aprintf (squares ? "[%s]:%d" : "%s:%d",
1348 if (!inhibit_keep_alive)
1349 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1352 request_set_header (req, "Cookie",
1353 cookie_header (wget_cookie_jar,
1354 u->host, u->port, u->path,
1356 u->scheme == SCHEME_HTTPS
1363 if (opt.post_data || opt.post_file_name)
1365 request_set_header (req, "Content-Type",
1366 "application/x-www-form-urlencoded", rel_none);
1368 post_data_size = strlen (opt.post_data);
1371 post_data_size = file_size (opt.post_file_name);
1372 if (post_data_size == -1)
1374 logprintf (LOG_NOTQUIET, _("POST data file missing: %s (%s)\n"),
1375 opt.post_file_name, strerror (errno));
1379 request_set_header (req, "Content-Length",
1380 xstrdup (number_to_static_string (post_data_size)),
1384 /* Add the user headers. */
1385 if (opt.user_headers)
1388 for (i = 0; opt.user_headers[i]; i++)
1389 request_set_user_header (req, opt.user_headers[i]);
1393 /* We need to come back here when the initial attempt to retrieve
1394 without authorization header fails. (Expected to happen at least
1395 for the Digest authorization scheme.) */
1399 /* Establish the connection. */
1401 if (!inhibit_keep_alive)
1403 /* Look for a persistent connection to target host, unless a
1404 proxy is used. The exception is when SSL is in use, in which
1405 case the proxy is nothing but a passthrough to the target
1406 host, registered as a connection to the latter. */
1407 struct url *relevant = conn;
1409 if (u->scheme == SCHEME_HTTPS)
1413 if (persistent_available_p (relevant->host, relevant->port,
1415 relevant->scheme == SCHEME_HTTPS,
1419 &host_lookup_failed))
1421 sock = pconn.socket;
1422 using_ssl = pconn.ssl;
1423 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1424 escnonprint (pconn.host), pconn.port);
1425 DEBUGP (("Reusing fd %d.\n", sock));
1426 if (pconn.authorized)
1427 /* If the connection is already authorized, the "Basic"
1428 authorization added by code above is unnecessary and
1430 request_remove_header (req, "Authorization");
1436 /* In its current implementation, persistent_available_p will
1437 look up conn->host in some cases. If that lookup failed, we
1438 don't need to bother with connect_to_host. */
1439 if (host_lookup_failed)
1445 sock = connect_to_host (conn->host, conn->port);
1454 return (retryable_socket_connect_error (errno)
1455 ? CONERROR : CONIMPOSSIBLE);
1459 if (proxy && u->scheme == SCHEME_HTTPS)
1461 /* When requesting SSL URLs through proxies, use the
1462 CONNECT method to request passthrough. */
1463 struct request *connreq = request_new ();
1464 request_set_method (connreq, "CONNECT",
1465 aprintf ("%s:%d", u->host, u->port));
1466 SET_USER_AGENT (connreq);
1469 request_set_header (connreq, "Proxy-Authorization",
1470 proxyauth, rel_value);
1471 /* Now that PROXYAUTH is part of the CONNECT request,
1472 zero it out so we don't send proxy authorization with
1473 the regular request below. */
1476 /* Examples in rfc2817 use the Host header in CONNECT
1477 requests. I don't see how that gains anything, given
1478 that the contents of Host would be exactly the same as
1479 the contents of CONNECT. */
1481 write_error = request_send (connreq, sock);
1482 request_free (connreq);
1483 if (write_error < 0)
1485 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1487 CLOSE_INVALIDATE (sock);
1491 head = read_http_response_head (sock);
1494 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1496 CLOSE_INVALIDATE (sock);
1505 DEBUGP (("proxy responded with: [%s]\n", head));
1507 resp = resp_new (head);
1508 statcode = resp_status (resp, &message);
1511 if (statcode != 200)
1514 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1515 message ? escnonprint (message) : "?");
1516 xfree_null (message);
1519 xfree_null (message);
1521 /* SOCK is now *really* connected to u->host, so update CONN
1522 to reflect this. That way register_persistent will
1523 register SOCK as being connected to u->host:u->port. */
1527 if (conn->scheme == SCHEME_HTTPS)
1529 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1536 #endif /* HAVE_SSL */
1539 /* Send the request to server. */
1540 write_error = request_send (req, sock);
1542 if (write_error >= 0)
1546 DEBUGP (("[POST data: %s]\n", opt.post_data));
1547 write_error = fd_write (sock, opt.post_data, post_data_size, -1.0);
1549 else if (opt.post_file_name && post_data_size != 0)
1550 write_error = post_file (sock, opt.post_file_name, post_data_size);
1553 if (write_error < 0)
1555 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1557 CLOSE_INVALIDATE (sock);
1561 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1562 proxy ? "Proxy" : "HTTP");
1567 head = read_http_response_head (sock);
1572 logputs (LOG_NOTQUIET, _("No data received.\n"));
1573 CLOSE_INVALIDATE (sock);
1579 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1581 CLOSE_INVALIDATE (sock);
1586 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1588 resp = resp_new (head);
1590 /* Check for status line. */
1592 statcode = resp_status (resp, &message);
1593 if (!opt.server_response)
1594 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1595 message ? escnonprint (message) : "");
1598 logprintf (LOG_VERBOSE, "\n");
1599 print_server_response (resp, " ");
1602 if (!opt.ignore_length
1603 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1607 parsed = str_to_wgint (hdrval, NULL, 10);
1608 if (parsed == WGINT_MAX && errno == ERANGE)
1610 #### If Content-Length is out of range, it most likely
1611 means that the file is larger than 2G and that we're
1612 compiled without LFS. In that case we should probably
1613 refuse to even attempt to download the file. */
1619 /* Check for keep-alive related responses. */
1620 if (!inhibit_keep_alive && contlen != -1)
1622 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1624 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1626 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1631 /* The server has promised that it will not close the connection
1632 when we're done. This means that we can register it. */
1633 register_persistent (conn->host, conn->port, sock, using_ssl);
1635 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1637 /* Authorization is required. */
1638 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1639 CLOSE_FINISH (sock);
1641 CLOSE_INVALIDATE (sock);
1642 pconn.authorized = 0;
1643 if (!auth_finished && (user && passwd))
1645 /* IIS sends multiple copies of WWW-Authenticate, one with
1646 the value "negotiate", and other(s) with data. Loop over
1647 all the occurrences and pick the one we recognize. */
1649 const char *wabeg, *waend;
1650 char *www_authenticate = NULL;
1652 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1653 &wabeg, &waend)) != -1;
1655 if (known_authentication_scheme_p (wabeg, waend))
1657 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1661 if (!www_authenticate)
1662 /* If the authentication header is missing or
1663 unrecognized, there's no sense in retrying. */
1664 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1665 else if (BEGINS_WITH (www_authenticate, "Basic"))
1666 /* If the authentication scheme is "Basic", which we send
1667 by default, there's no sense in retrying either. (This
1668 should be changed when we stop sending "Basic" data by
1674 pth = url_full_path (u);
1675 request_set_header (req, "Authorization",
1676 create_authorization_line (www_authenticate,
1678 request_method (req),
1682 if (BEGINS_WITH (www_authenticate, "NTLM"))
1685 goto retry_with_auth;
1688 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1692 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1694 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1696 pconn.authorized = 1;
1700 hs->statcode = statcode;
1702 hs->error = xstrdup (_("Malformed status line"));
1704 hs->error = xstrdup (_("(no description)"));
1706 hs->error = xstrdup (message);
1709 type = resp_header_strdup (resp, "Content-Type");
1712 char *tmp = strchr (type, ';');
1715 while (tmp > type && ISSPACE (tmp[-1]))
1720 hs->newloc = resp_header_strdup (resp, "Location");
1721 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1723 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1727 const char *scbeg, *scend;
1728 /* The jar should have been created by now. */
1729 assert (wget_cookie_jar != NULL);
1731 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1732 &scbeg, &scend)) != -1;
1735 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1738 /* u->path doesn't begin with /, which cookies.c expects. */
1739 pth = (char *) alloca (1 + strlen (u->path) + 1);
1741 strcpy (pth + 1, u->path);
1743 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, pth,
1748 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1750 wgint first_byte_pos, last_byte_pos, entity_length;
1751 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1753 contrange = first_byte_pos;
1758 /* 20x responses are counted among successful by default. */
1759 if (H_20X (statcode))
1762 /* Return if redirected. */
1763 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1765 /* RFC2068 says that in case of the 300 (multiple choices)
1766 response, the server can output a preferred URL through
1767 `Location' header; otherwise, the request should be treated
1768 like GET. So, if the location is set, it will be a
1769 redirection; otherwise, just proceed normally. */
1770 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1774 logprintf (LOG_VERBOSE,
1775 _("Location: %s%s\n"),
1776 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1777 hs->newloc ? _(" [following]") : "");
1778 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1779 CLOSE_FINISH (sock);
1781 CLOSE_INVALIDATE (sock);
1787 /* If content-type is not given, assume text/html. This is because
1788 of the multitude of broken CGI's that "forget" to generate the
1791 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1792 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1797 if (opt.html_extension && (*dt & TEXTHTML))
1798 /* -E / --html-extension / html_extension = on was specified, and this is a
1799 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1800 already the file's suffix, tack on ".html". */
1802 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1804 if (last_period_in_local_filename == NULL
1805 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1806 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1808 int local_filename_len = strlen (*hs->local_file);
1809 /* Resize the local file, allowing for ".html" preceded by
1810 optional ".NUMBER". */
1811 *hs->local_file = xrealloc (*hs->local_file,
1812 local_filename_len + 24 + sizeof (".html"));
1813 strcpy(*hs->local_file + local_filename_len, ".html");
1814 /* If clobbering is not allowed and the file, as named,
1815 exists, tack on ".NUMBER.html" instead. */
1820 sprintf (*hs->local_file + local_filename_len,
1821 ".%d.html", ext_num++);
1822 while (file_exists_p (*hs->local_file));
1824 *dt |= ADDED_HTML_EXTENSION;
1828 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1830 /* If `-c' is in use and the file has been fully downloaded (or
1831 the remote file has shrunk), Wget effectively requests bytes
1832 after the end of file and the server response with 416. */
1833 logputs (LOG_VERBOSE, _("\
1834 \n The file is already fully retrieved; nothing to do.\n\n"));
1835 /* In case the caller inspects. */
1838 /* Mark as successfully retrieved. */
1841 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1842 might be more bytes in the body. */
1843 return RETRUNNEEDED;
1845 if ((contrange != 0 && contrange != hs->restval)
1846 || (H_PARTIAL (statcode) && !contrange))
1848 /* The Range request was somehow misunderstood by the server.
1851 CLOSE_INVALIDATE (sock);
1854 hs->contlen = contlen + contrange;
1860 /* No need to print this output if the body won't be
1861 downloaded at all, or if the original server response is
1863 logputs (LOG_VERBOSE, _("Length: "));
1866 logputs (LOG_VERBOSE, with_thousand_seps (contlen + contrange));
1867 if (contlen + contrange >= 1024)
1868 logprintf (LOG_VERBOSE, " (%s)",
1869 human_readable (contlen + contrange));
1872 if (contlen >= 1024)
1873 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1874 with_thousand_seps (contlen),
1875 human_readable (contlen));
1877 logprintf (LOG_VERBOSE, _(", %s remaining"),
1878 with_thousand_seps (contlen));
1882 logputs (LOG_VERBOSE,
1883 opt.ignore_length ? _("ignored") : _("unspecified"));
1885 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1887 logputs (LOG_VERBOSE, "\n");
1891 type = NULL; /* We don't need it any more. */
1893 /* Return if we have no intention of further downloading. */
1894 if (!(*dt & RETROKF) || head_only)
1896 /* In case the caller cares to look... */
1900 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1901 servers not to send body in response to a HEAD request. If
1902 you encounter such a server (more likely a broken CGI), use
1903 `--no-http-keep-alive'. */
1904 CLOSE_FINISH (sock);
1905 return RETRFINISHED;
1908 /* Open the local file. */
1911 mkalldirs (*hs->local_file);
1913 rotate_backups (*hs->local_file);
1915 fp = fopen (*hs->local_file, "ab");
1916 else if (ALLOW_CLOBBER)
1917 fp = fopen (*hs->local_file, "wb");
1920 fp = fopen_excl (*hs->local_file, 1);
1921 if (!fp && errno == EEXIST)
1923 /* We cannot just invent a new name and use it (which is
1924 what functions like unique_create typically do)
1925 because we told the user we'd use this name.
1926 Instead, return and retry the download. */
1927 logprintf (LOG_NOTQUIET,
1928 _("%s has sprung into existence.\n"),
1930 CLOSE_INVALIDATE (sock);
1931 return FOPEN_EXCL_ERR;
1936 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1937 CLOSE_INVALIDATE (sock);
1944 /* #### This confuses the timestamping code that checks for file
1945 size. Maybe we should save some additional information? */
1946 if (opt.save_headers)
1947 fwrite (head, 1, strlen (head), fp);
1949 /* Download the request body. */
1952 flags |= rb_read_exactly;
1953 if (hs->restval > 0 && contrange == 0)
1954 /* If the server ignored our range request, instruct fd_read_body
1955 to skip the first RESTVAL bytes of body. */
1956 flags |= rb_skip_startpos;
1957 hs->len = hs->restval;
1959 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1960 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1964 CLOSE_FINISH (sock);
1966 CLOSE_INVALIDATE (sock);
1969 /* Close or flush the file. We have to be careful to check for
1970 error here. Checking the result of fwrite() is not enough --
1971 errors could go unnoticed! */
1974 flush_res = fclose (fp);
1976 flush_res = fflush (fp);
1977 if (flush_res == EOF)
1982 return RETRFINISHED;
1985 /* The genuine HTTP loop! This is the part where the retrieval is
1986 retried, and retried, and retried, and... */
1988 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1989 int *dt, struct url *proxy)
1992 int use_ts, got_head = 0; /* time-stamping info */
1993 char *filename_plus_orig_suffix;
1994 char *local_filename = NULL;
1995 char *tms, *locf, *tmrate;
1997 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1998 wgint local_size = 0; /* the size of the local file */
1999 size_t filename_len;
2000 struct http_stat hstat; /* HTTP status */
2004 /* This used to be done in main(), but it's a better idea to do it
2005 here so that we don't go through the hoops if we're just using
2009 if (!wget_cookie_jar)
2010 wget_cookie_jar = cookie_jar_new ();
2011 if (opt.cookies_input && !cookies_loaded_p)
2013 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2014 cookies_loaded_p = 1;
2020 /* Warn on (likely bogus) wildcard usage in HTTP. */
2021 if (has_wildcards_p (u->path))
2022 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2026 /* Determine the local filename. */
2027 if (local_file && *local_file)
2028 hstat.local_file = local_file;
2029 else if (local_file && !opt.output_document)
2031 *local_file = url_file_name (u);
2032 hstat.local_file = local_file;
2036 dummy = url_file_name (u);
2037 hstat.local_file = &dummy;
2038 /* be honest about where we will save the file */
2039 if (local_file && opt.output_document)
2040 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2043 if (!opt.output_document)
2044 locf = *hstat.local_file;
2046 locf = opt.output_document;
2048 hstat.referer = referer;
2050 filename_len = strlen (*hstat.local_file);
2051 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2053 if (opt.noclobber && file_exists_p (*hstat.local_file))
2055 /* If opt.noclobber is turned on and file already exists, do not
2056 retrieve the file */
2057 logprintf (LOG_VERBOSE, _("\
2058 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2059 /* If the file is there, we suppose it's retrieved OK. */
2062 /* #### Bogusness alert. */
2063 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2064 if (has_html_suffix_p (*hstat.local_file))
2072 if (opt.timestamping)
2074 int local_dot_orig_file_exists = 0;
2076 if (opt.backup_converted)
2077 /* If -K is specified, we'll act on the assumption that it was specified
2078 last time these files were downloaded as well, and instead of just
2079 comparing local file X against server file X, we'll compare local
2080 file X.orig (if extant, else X) against server file X. If -K
2081 _wasn't_ specified last time, or the server contains files called
2082 *.orig, -N will be back to not operating correctly with -k. */
2084 /* Would a single s[n]printf() call be faster? --dan
2086 Definitely not. sprintf() is horribly slow. It's a
2087 different question whether the difference between the two
2088 affects a program. Usually I'd say "no", but at one
2089 point I profiled Wget, and found that a measurable and
2090 non-negligible amount of time was lost calling sprintf()
2091 in url.c. Replacing sprintf with inline calls to
2092 strcpy() and number_to_string() made a difference.
2094 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2095 memcpy (filename_plus_orig_suffix + filename_len,
2096 ".orig", sizeof (".orig"));
2098 /* Try to stat() the .orig file. */
2099 if (stat (filename_plus_orig_suffix, &st) == 0)
2101 local_dot_orig_file_exists = 1;
2102 local_filename = filename_plus_orig_suffix;
2106 if (!local_dot_orig_file_exists)
2107 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2108 if (stat (*hstat.local_file, &st) == 0)
2109 local_filename = *hstat.local_file;
2111 if (local_filename != NULL)
2112 /* There was a local file, so we'll check later to see if the version
2113 the server has is the same version we already have, allowing us to
2119 /* Modification time granularity is 2 seconds for Windows, so
2120 increase local time by 1 second for later comparison. */
2123 local_size = st.st_size;
2127 /* Reset the counter. */
2133 /* Increment the pass counter. */
2135 sleep_between_retrievals (count);
2136 /* Get the current time string. */
2137 tms = time_str (NULL);
2138 /* Print fetch message, if opt.verbose. */
2141 char *hurl = url_string (u, 1);
2145 sprintf (tmp, _("(try:%2d)"), count);
2146 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2147 tms, hurl, tmp, locf);
2149 ws_changetitle (hurl);
2154 /* Default document type is empty. However, if spider mode is
2155 on or time-stamping is employed, HEAD_ONLY commands is
2156 encoded within *dt. */
2157 if (opt.spider || (use_ts && !got_head))
2162 /* Decide whether or not to restart. */
2164 && stat (locf, &st) == 0
2165 && S_ISREG (st.st_mode))
2166 /* When -c is used, continue from on-disk size. (Can't use
2167 hstat.len even if count>1 because we don't want a failed
2168 first attempt to clobber existing data.) */
2169 hstat.restval = st.st_size;
2171 /* otherwise, continue where the previous try left off */
2172 hstat.restval = hstat.len;
2176 /* Decide whether to send the no-cache directive. We send it in
2178 a) we're using a proxy, and we're past our first retrieval.
2179 Some proxies are notorious for caching incomplete data, so
2180 we require a fresh get.
2181 b) caching is explicitly inhibited. */
2182 if ((proxy && count > 1) /* a */
2183 || !opt.allow_cache /* b */
2185 *dt |= SEND_NOCACHE;
2187 *dt &= ~SEND_NOCACHE;
2189 /* Try fetching the document, or at least its head. */
2190 err = gethttp (u, &hstat, dt, proxy);
2192 /* It's unfortunate that wget determines the local filename before finding
2193 out the Content-Type of the file. Barring a major restructuring of the
2194 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2195 *hstat.local_file to tack on ".html". */
2196 if (!opt.output_document)
2197 locf = *hstat.local_file;
2200 tms = time_str (NULL);
2201 /* Get the new location (with or without the redirection). */
2203 *newloc = xstrdup (hstat.newloc);
2206 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2207 case CONERROR: case READERR: case WRITEFAILED:
2208 case RANGEERR: case FOPEN_EXCL_ERR:
2209 /* Non-fatal errors continue executing the loop, which will
2210 bring them to "while" statement at the end, to judge
2211 whether the number of tries was exceeded. */
2212 free_hstat (&hstat);
2213 printwhat (count, opt.ntry);
2214 if (err == FOPEN_EXCL_ERR)
2216 /* Re-determine the file name. */
2217 if (local_file && *local_file)
2219 xfree (*local_file);
2220 *local_file = url_file_name (u);
2221 hstat.local_file = local_file;
2226 dummy = url_file_name (u);
2227 hstat.local_file = &dummy;
2229 /* be honest about where we will save the file */
2230 if (local_file && opt.output_document)
2231 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2232 if (!opt.output_document)
2233 locf = *hstat.local_file;
2235 locf = opt.output_document;
2238 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2239 case SSLINITFAILED: case CONTNOTSUPPORTED:
2240 /* Fatal errors just return from the function. */
2241 free_hstat (&hstat);
2244 case FWRITEERR: case FOPENERR:
2245 /* Another fatal error. */
2246 logputs (LOG_VERBOSE, "\n");
2247 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2248 *hstat.local_file, strerror (errno));
2249 free_hstat (&hstat);
2253 /* Another fatal error. */
2254 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2255 free_hstat (&hstat);
2259 /* Return the new location to the caller. */
2262 logprintf (LOG_NOTQUIET,
2263 _("ERROR: Redirection (%d) without location.\n"),
2265 free_hstat (&hstat);
2269 free_hstat (&hstat);
2273 /* The file was already fully retrieved. */
2274 free_hstat (&hstat);
2278 /* Deal with you later. */
2281 /* All possibilities should have been exhausted. */
2284 if (!(*dt & RETROKF))
2288 /* #### Ugly ugly ugly! */
2289 char *hurl = url_string (u, 1);
2290 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2293 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2294 tms, hstat.statcode, escnonprint (hstat.error));
2295 logputs (LOG_VERBOSE, "\n");
2296 free_hstat (&hstat);
2301 /* Did we get the time-stamp? */
2304 if (opt.timestamping && !hstat.remote_time)
2306 logputs (LOG_NOTQUIET, _("\
2307 Last-modified header missing -- time-stamps turned off.\n"));
2309 else if (hstat.remote_time)
2311 /* Convert the date-string into struct tm. */
2312 tmr = http_atotm (hstat.remote_time);
2313 if (tmr == (time_t) (-1))
2314 logputs (LOG_VERBOSE, _("\
2315 Last-modified header invalid -- time-stamp ignored.\n"));
2319 /* The time-stamping section. */
2324 use_ts = 0; /* no more time-stamping */
2325 count = 0; /* the retrieve count for HEAD is
2327 if (hstat.remote_time && tmr != (time_t) (-1))
2329 /* Now time-stamping can be used validly. Time-stamping
2330 means that if the sizes of the local and remote file
2331 match, and local file is newer than the remote file,
2332 it will not be retrieved. Otherwise, the normal
2333 download procedure is resumed. */
2335 (hstat.contlen == -1 || local_size == hstat.contlen))
2337 logprintf (LOG_VERBOSE, _("\
2338 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2340 free_hstat (&hstat);
2344 else if (tml >= tmr)
2345 logprintf (LOG_VERBOSE, _("\
2346 The sizes do not match (local %s) -- retrieving.\n"),
2347 number_to_static_string (local_size));
2349 logputs (LOG_VERBOSE,
2350 _("Remote file is newer, retrieving.\n"));
2352 free_hstat (&hstat);
2355 if ((tmr != (time_t) (-1))
2357 && ((hstat.len == hstat.contlen) ||
2358 ((hstat.res == 0) &&
2359 ((hstat.contlen == -1) ||
2360 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2362 /* #### This code repeats in http.c and ftp.c. Move it to a
2364 const char *fl = NULL;
2365 if (opt.output_document)
2367 if (output_stream_regular)
2368 fl = opt.output_document;
2371 fl = *hstat.local_file;
2375 /* End of time-stamping section. */
2379 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2380 escnonprint (hstat.error));
2385 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2387 if (hstat.len == hstat.contlen)
2391 logprintf (LOG_VERBOSE,
2392 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2394 number_to_static_string (hstat.len),
2395 number_to_static_string (hstat.contlen));
2396 logprintf (LOG_NONVERBOSE,
2397 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2399 number_to_static_string (hstat.len),
2400 number_to_static_string (hstat.contlen),
2404 total_downloaded_bytes += hstat.len;
2406 /* Remember that we downloaded the file for later ".orig" code. */
2407 if (*dt & ADDED_HTML_EXTENSION)
2408 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2410 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2412 free_hstat (&hstat);
2416 else if (hstat.res == 0) /* No read error */
2418 if (hstat.contlen == -1) /* We don't know how much we were supposed
2419 to get, so assume we succeeded. */
2423 logprintf (LOG_VERBOSE,
2424 _("%s (%s) - `%s' saved [%s]\n\n"),
2426 number_to_static_string (hstat.len));
2427 logprintf (LOG_NONVERBOSE,
2428 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2429 tms, u->url, number_to_static_string (hstat.len),
2433 total_downloaded_bytes += hstat.len;
2435 /* Remember that we downloaded the file for later ".orig" code. */
2436 if (*dt & ADDED_HTML_EXTENSION)
2437 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2439 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2441 free_hstat (&hstat);
2445 else if (hstat.len < hstat.contlen) /* meaning we lost the
2446 connection too soon */
2448 logprintf (LOG_VERBOSE,
2449 _("%s (%s) - Connection closed at byte %s. "),
2450 tms, tmrate, number_to_static_string (hstat.len));
2451 printwhat (count, opt.ntry);
2452 free_hstat (&hstat);
2455 else if (!opt.kill_longer) /* meaning we got more than expected */
2457 logprintf (LOG_VERBOSE,
2458 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2460 number_to_static_string (hstat.len),
2461 number_to_static_string (hstat.contlen));
2462 logprintf (LOG_NONVERBOSE,
2463 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2465 number_to_static_string (hstat.len),
2466 number_to_static_string (hstat.contlen),
2469 total_downloaded_bytes += hstat.len;
2471 /* Remember that we downloaded the file for later ".orig" code. */
2472 if (*dt & ADDED_HTML_EXTENSION)
2473 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2475 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2477 free_hstat (&hstat);
2481 else /* the same, but not accepted */
2483 logprintf (LOG_VERBOSE,
2484 _("%s (%s) - Connection closed at byte %s/%s. "),
2486 number_to_static_string (hstat.len),
2487 number_to_static_string (hstat.contlen));
2488 printwhat (count, opt.ntry);
2489 free_hstat (&hstat);
2493 else /* now hstat.res can only be -1 */
2495 if (hstat.contlen == -1)
2497 logprintf (LOG_VERBOSE,
2498 _("%s (%s) - Read error at byte %s (%s)."),
2499 tms, tmrate, number_to_static_string (hstat.len),
2501 printwhat (count, opt.ntry);
2502 free_hstat (&hstat);
2505 else /* hstat.res == -1 and contlen is given */
2507 logprintf (LOG_VERBOSE,
2508 _("%s (%s) - Read error at byte %s/%s (%s). "),
2510 number_to_static_string (hstat.len),
2511 number_to_static_string (hstat.contlen),
2513 printwhat (count, opt.ntry);
2514 free_hstat (&hstat);
2520 while (!opt.ntry || (count < opt.ntry));
2524 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2525 than local timezone.
2527 mktime is similar but assumes struct tm, also known as the
2528 "broken-down" form of time, is in local time zone. mktime_from_utc
2529 uses mktime to make the conversion understanding that an offset
2530 will be introduced by the local time assumption.
2532 mktime_from_utc then measures the introduced offset by applying
2533 gmtime to the initial result and applying mktime to the resulting
2534 "broken-down" form. The difference between the two mktime results
2535 is the measured offset which is then subtracted from the initial
2536 mktime result to yield a calendar time which is the value returned.
2538 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2539 consistent offset (the non DST offset) since tm and tm+o might be
2540 on opposite sides of a DST change.
2542 Some implementations of mktime return -1 for the nonexistent
2543 localtime hour at the beginning of DST. In this event, use
2544 mktime(tm - 1hr) + 3600.
2548 gmtime(t+o) --> tm+o
2549 mktime(tm+o) --> t+2o
2550 t+o - (t+2o - t+o) = t
2552 Note that glibc contains a function of the same purpose named
2553 `timegm' (reverse of gmtime). But obviously, it is not universally
2554 available, and unfortunately it is not straightforwardly
2555 extractable for use here. Perhaps configure should detect timegm
2556 and use it where available.
2558 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2559 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2560 Further improved by Roger with assistance from Edward J. Sabol
2561 based on input by Jamie Zawinski. */
2564 mktime_from_utc (struct tm *t)
2575 return -1; /* can't deal with output from strptime */
2586 return -1; /* can't deal with output from gmtime */
2589 return (tl - (tb - tl));
2592 /* Check whether the result of strptime() indicates success.
2593 strptime() returns the pointer to how far it got to in the string.
2594 The processing has been successful if the string is at `GMT' or
2595 `+X', or at the end of the string.
2597 In extended regexp parlance, the function returns 1 if P matches
2598 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2599 can return) is considered a failure and 0 is returned. */
2601 check_end (const char *p)
2605 while (ISSPACE (*p))
2608 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2609 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2615 /* Convert the textual specification of time in TIME_STRING to the
2616 number of seconds since the Epoch.
2618 TIME_STRING can be in any of the three formats RFC2616 allows the
2619 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2620 as well as the time format used in the Set-Cookie header.
2621 Timezones are ignored, and should be GMT.
2623 Return the computed time_t representation, or -1 if the conversion
2626 This function uses strptime with various string formats for parsing
2627 TIME_STRING. This results in a parser that is not as lenient in
2628 interpreting TIME_STRING as I would like it to be. Being based on
2629 strptime, it always allows shortened months, one-digit days, etc.,
2630 but due to the multitude of formats in which time can be
2631 represented, an ideal HTTP time parser would be even more
2632 forgiving. It should completely ignore things like week days and
2633 concentrate only on the various forms of representing years,
2634 months, days, hours, minutes, and seconds. For example, it would
2635 be nice if it accepted ISO 8601 out of the box.
2637 I've investigated free and PD code for this purpose, but none was
2638 usable. getdate was big and unwieldy, and had potential copyright
2639 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2640 distributed with phttpd, is excellent, but we cannot use it because
2641 it is not assigned to the FSF. So I stuck it with strptime. */
2644 http_atotm (const char *time_string)
2646 /* NOTE: Solaris strptime man page claims that %n and %t match white
2647 space, but that's not universally available. Instead, we simply
2648 use ` ' to mean "skip all WS", which works under all strptime
2649 implementations I've tested. */
2651 static const char *time_formats[] = {
2652 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2653 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2654 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2655 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2656 (used in Set-Cookie, defined in the
2657 Netscape cookie specification.) */
2661 for (i = 0; i < countof (time_formats); i++)
2665 /* Some versions of strptime use the existing contents of struct
2666 tm to recalculate the date according to format. Zero it out
2667 to prevent garbage from the stack influencing strptime. */
2670 /* Solaris strptime fails to recognize English month names in
2671 non-English locales, which we work around by not setting the
2672 LC_TIME category. Another way would be to temporarily set
2673 locale to C before invoking strptime, but that's slow and
2674 messy. GNU strptime does not have this problem because it
2675 recognizes English month names along with the local ones. */
2677 if (check_end (strptime (time_string, time_formats[i], &t)))
2678 return mktime_from_utc (&t);
2681 /* All formats have failed. */
2685 /* Authorization support: We support three authorization schemes:
2687 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2689 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2690 consisting of answering to the server's challenge with the proper
2693 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2694 Stenberg for libcurl. Like digest, NTLM is based on a
2695 challenge-response mechanism, but unlike digest, it is non-standard
2696 (authenticates TCP connections rather than requests), undocumented
2697 and Microsoft-specific. */
2699 /* Create the authentication header contents for the `Basic' scheme.
2700 This is done by encoding the string "USER:PASS" to base64 and
2701 prepending the string "Basic " in front of it. */
2704 basic_authentication_encode (const char *user, const char *passwd)
2707 int len1 = strlen (user) + 1 + strlen (passwd);
2709 t1 = (char *)alloca (len1 + 1);
2710 sprintf (t1, "%s:%s", user, passwd);
2712 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2713 base64_encode (t1, len1, t2);
2715 return concat_strings ("Basic ", t2, (char *) 0);
2718 #define SKIP_WS(x) do { \
2719 while (ISSPACE (*(x))) \
2723 #ifdef ENABLE_DIGEST
2724 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2725 of a field in such a header. If the field is the one specified by
2726 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2727 digest authorization code), extract its value in the (char*)
2728 variable pointed by RET. Returns negative on a malformed header,
2729 or number of bytes that have been parsed by this call. */
2731 extract_header_attr (const char *au, const char *attr_name, char **ret)
2734 const char *cp = au;
2736 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2738 cp += strlen (attr_name);
2751 for (ep = cp; *ep && *ep != '\"'; ep++)
2756 *ret = strdupdelim (cp, ep);
2763 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2764 an array of 16 bytes containing the hash keys, and BUF should be a
2765 buffer of 33 writable characters (32 for hex digits plus one for
2766 zero termination). */
2768 dump_hash (unsigned char *buf, const unsigned char *hash)
2772 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2774 *buf++ = XNUM_TO_digit (*hash >> 4);
2775 *buf++ = XNUM_TO_digit (*hash & 0xf);
2780 /* Take the line apart to find the challenge, and compose a digest
2781 authorization header. See RFC2069 section 2.1.2. */
2783 digest_authentication_encode (const char *au, const char *user,
2784 const char *passwd, const char *method,
2787 static char *realm, *opaque, *nonce;
2792 { "realm", &realm },
2793 { "opaque", &opaque },
2798 realm = opaque = nonce = NULL;
2800 au += 6; /* skip over `Digest' */
2806 for (i = 0; i < countof (options); i++)
2808 int skip = extract_header_attr (au, options[i].name,
2809 options[i].variable);
2813 xfree_null (opaque);
2823 if (i == countof (options))
2825 while (*au && *au != '=')
2833 while (*au && *au != '\"')
2840 while (*au && *au != ',')
2845 if (!realm || !nonce || !user || !passwd || !path || !method)
2848 xfree_null (opaque);
2853 /* Calculate the digest value. */
2855 ALLOCA_MD5_CONTEXT (ctx);
2856 unsigned char hash[MD5_HASHLEN];
2857 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2858 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2860 /* A1BUF = H(user ":" realm ":" password) */
2862 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2863 gen_md5_update ((unsigned char *)":", 1, ctx);
2864 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2865 gen_md5_update ((unsigned char *)":", 1, ctx);
2866 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2867 gen_md5_finish (ctx, hash);
2868 dump_hash (a1buf, hash);
2870 /* A2BUF = H(method ":" path) */
2872 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2873 gen_md5_update ((unsigned char *)":", 1, ctx);
2874 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2875 gen_md5_finish (ctx, hash);
2876 dump_hash (a2buf, hash);
2878 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2880 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2881 gen_md5_update ((unsigned char *)":", 1, ctx);
2882 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2883 gen_md5_update ((unsigned char *)":", 1, ctx);
2884 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2885 gen_md5_finish (ctx, hash);
2886 dump_hash (response_digest, hash);
2888 res = (char*) xmalloc (strlen (user)
2893 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2894 + (opaque ? strlen (opaque) : 0)
2896 sprintf (res, "Digest \
2897 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2898 user, realm, nonce, path, response_digest);
2901 char *p = res + strlen (res);
2902 strcat (p, ", opaque=\"");
2909 #endif /* ENABLE_DIGEST */
2911 /* Computing the size of a string literal must take into account that
2912 value returned by sizeof includes the terminating \0. */
2913 #define STRSIZE(literal) (sizeof (literal) - 1)
2915 /* Whether chars in [b, e) begin with the literal string provided as
2916 first argument and are followed by whitespace or terminating \0.
2917 The comparison is case-insensitive. */
2918 #define STARTS(literal, b, e) \
2919 ((e) - (b) >= STRSIZE (literal) \
2920 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2921 && ((e) - (b) == STRSIZE (literal) \
2922 || ISSPACE (b[STRSIZE (literal)])))
2925 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2927 return STARTS ("Basic", hdrbeg, hdrend)
2928 #ifdef ENABLE_DIGEST
2929 || STARTS ("Digest", hdrbeg, hdrend)
2932 || STARTS ("NTLM", hdrbeg, hdrend)
2939 /* Create the HTTP authorization request header. When the
2940 `WWW-Authenticate' response header is seen, according to the
2941 authorization scheme specified in that header (`Basic' and `Digest'
2942 are supported by the current implementation), produce an
2943 appropriate HTTP authorization request header. */
2945 create_authorization_line (const char *au, const char *user,
2946 const char *passwd, const char *method,
2947 const char *path, int *finished)
2949 /* We are called only with known schemes, so we can dispatch on the
2951 switch (TOUPPER (*au))
2953 case 'B': /* Basic */
2955 return basic_authentication_encode (user, passwd);
2956 #ifdef ENABLE_DIGEST
2957 case 'D': /* Digest */
2959 return digest_authentication_encode (au, user, passwd, method, path);
2962 case 'N': /* NTLM */
2963 if (!ntlm_input (&pconn.ntlm, au))
2968 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2971 /* We shouldn't get here -- this function should be only called
2972 with values approved by known_authentication_scheme_p. */
2980 if (wget_cookie_jar)
2981 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2987 xfree_null (pconn.host);
2988 if (wget_cookie_jar)
2989 cookie_jar_delete (wget_cookie_jar);