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);
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);
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);
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 /* If pconn.socket is already talking to HOST, we needn't
973 reconnect. This happens often when both sites are virtual
974 hosts distinguished only by name and served by the same
975 network interface, and hence the same web server (possibly
976 set up by the ISP and serving many different web sites).
977 This admittedly non-standard 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 if name-based
987 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) \
1117 if (opt.useragent) \
1118 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1120 request_set_header (req, "User-Agent", \
1121 aprintf ("Wget/%s", version_string), rel_value);
1124 /* Retrieve a document through HTTP protocol. It recognizes status
1125 code, and correctly handles redirections. It closes the network
1126 socket. If it receives an error from the functions below it, it
1127 will print it if there is enough information to do so (almost
1128 always), returning the error to the caller (i.e. http_loop).
1130 Various HTTP parameters are stored to hs.
1132 If PROXY is non-NULL, the connection will be made to the proxy
1133 server, and u->url will be requested. */
1135 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1137 struct request *req;
1140 char *user, *passwd;
1144 wgint contlen, contrange;
1151 /* Set to 1 when the authorization has failed permanently and should
1152 not be tried again. */
1153 int auth_finished = 0;
1155 /* Whether NTLM authentication is used for this request. */
1158 /* Whether our connection to the remote host is through SSL. */
1162 struct response *resp;
1166 /* Whether this connection will be kept alive after the HTTP request
1170 /* Whether keep-alive should be inhibited.
1172 RFC 2068 requests that 1.0 clients not send keep-alive requests
1173 to proxies. This is because many 1.0 proxies do not interpret
1174 the Connection header and transfer it to the remote server,
1175 causing it to not close the connection and leave both the proxy
1176 and the client hanging. */
1177 int inhibit_keep_alive =
1178 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1180 /* Headers sent when using POST. */
1181 wgint post_data_size = 0;
1183 int host_lookup_failed = 0;
1186 if (u->scheme == SCHEME_HTTPS)
1188 /* Initialize the SSL context. After this has once been done,
1189 it becomes a no-op. */
1190 switch (ssl_init ())
1192 case SSLERRCTXCREATE:
1194 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
1195 return SSLERRCTXCREATE;
1196 case SSLERRCERTFILE:
1197 /* try without certfile */
1198 logprintf (LOG_NOTQUIET,
1199 _("Failed to load certificates from %s\n"),
1201 logprintf (LOG_NOTQUIET,
1202 _("Trying without the specified certificate\n"));
1205 logprintf (LOG_NOTQUIET,
1206 _("Failed to get certificate key from %s\n"),
1208 logprintf (LOG_NOTQUIET,
1209 _("Trying without the specified certificate\n"));
1215 #endif /* HAVE_SSL */
1217 if (!(*dt & HEAD_ONLY))
1218 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1219 know the local filename so we can save to it. */
1220 assert (*hs->local_file != NULL);
1222 /* Initialize certain elements of struct http_stat. */
1227 hs->remote_time = NULL;
1232 /* Prepare the request to send. */
1234 req = request_new ();
1236 const char *meth = "GET";
1237 if (*dt & HEAD_ONLY)
1239 else if (opt.post_file_name || opt.post_data)
1241 /* Use the full path, i.e. one that includes the leading slash and
1242 the query string. E.g. if u->path is "foo/bar" and u->query is
1243 "param=value", full_path will be "/foo/bar?param=value". */
1244 request_set_method (req, meth,
1245 proxy ? xstrdup (u->url) : url_full_path (u));
1248 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1249 if (*dt & SEND_NOCACHE)
1250 request_set_header (req, "Pragma", "no-cache", rel_none);
1252 request_set_header (req, "Range",
1253 aprintf ("bytes=%s-",
1254 number_to_static_string (hs->restval)),
1256 SET_USER_AGENT (req);
1257 request_set_header (req, "Accept", "*/*", rel_none);
1259 /* Find the username and password for authentication. */
1262 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1263 user = user ? user : opt.http_user;
1264 passwd = passwd ? passwd : opt.http_passwd;
1268 /* We have the username and the password, but haven't tried
1269 any authorization yet. Let's see if the "Basic" method
1270 works. If not, we'll come back here and construct a
1271 proper authorization method with the right challenges.
1273 If we didn't employ this kind of logic, every URL that
1274 requires authorization would have to be processed twice,
1275 which is very suboptimal and generates a bunch of false
1276 "unauthorized" errors in the server log.
1278 #### But this logic also has a serious problem when used
1279 with stronger authentications: we *first* transmit the
1280 username and the password in clear text, and *then* attempt a
1281 stronger authentication scheme. That cannot be right! We
1282 are only fortunate that almost everyone still uses the
1283 `Basic' scheme anyway.
1285 There should be an option to prevent this from happening, for
1286 those who use strong authentication schemes and value their
1288 request_set_header (req, "Authorization",
1289 basic_authentication_encode (user, passwd),
1296 char *proxy_user, *proxy_passwd;
1297 /* For normal username and password, URL components override
1298 command-line/wgetrc parameters. With proxy
1299 authentication, it's the reverse, because proxy URLs are
1300 normally the "permanent" ones, so command-line args
1301 should take precedence. */
1302 if (opt.proxy_user && opt.proxy_passwd)
1304 proxy_user = opt.proxy_user;
1305 proxy_passwd = opt.proxy_passwd;
1309 proxy_user = proxy->user;
1310 proxy_passwd = proxy->passwd;
1312 /* #### This does not appear right. Can't the proxy request,
1313 say, `Digest' authentication? */
1314 if (proxy_user && proxy_passwd)
1315 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1317 /* If we're using a proxy, we will be connecting to the proxy
1321 /* Proxy authorization over SSL is handled below. */
1323 if (u->scheme != SCHEME_HTTPS)
1325 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1329 /* Whether we need to print the host header with braces around
1330 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1331 usual "Host: symbolic-name:1234". */
1332 int squares = strchr (u->host, ':') != NULL;
1333 if (u->port == scheme_default_port (u->scheme))
1334 request_set_header (req, "Host",
1335 aprintf (squares ? "[%s]" : "%s", u->host),
1338 request_set_header (req, "Host",
1339 aprintf (squares ? "[%s]:%d" : "%s:%d",
1344 if (!inhibit_keep_alive)
1345 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1348 request_set_header (req, "Cookie",
1349 cookie_header (wget_cookie_jar,
1350 u->host, u->port, u->path,
1352 u->scheme == SCHEME_HTTPS
1359 if (opt.post_data || opt.post_file_name)
1361 request_set_header (req, "Content-Type",
1362 "application/x-www-form-urlencoded", rel_none);
1364 post_data_size = strlen (opt.post_data);
1367 post_data_size = file_size (opt.post_file_name);
1368 if (post_data_size == -1)
1370 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1371 opt.post_file_name);
1375 request_set_header (req, "Content-Length",
1376 xstrdup (number_to_static_string (post_data_size)),
1380 /* Add the user headers. */
1381 if (opt.user_headers)
1384 for (i = 0; opt.user_headers[i]; i++)
1385 request_set_user_header (req, opt.user_headers[i]);
1389 /* We need to come back here when the initial attempt to retrieve
1390 without authorization header fails. (Expected to happen at least
1391 for the Digest authorization scheme.) */
1395 /* Establish the connection. */
1397 if (!inhibit_keep_alive)
1399 /* Look for a persistent connection to target host, unless a
1400 proxy is used. The exception is when SSL is in use, in which
1401 case the proxy is nothing but a passthrough to the target
1402 host, registered as a connection to the latter. */
1403 struct url *relevant = conn;
1405 if (u->scheme == SCHEME_HTTPS)
1409 if (persistent_available_p (relevant->host, relevant->port,
1411 relevant->scheme == SCHEME_HTTPS,
1415 &host_lookup_failed))
1417 sock = pconn.socket;
1418 using_ssl = pconn.ssl;
1419 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1420 escnonprint (pconn.host), pconn.port);
1421 DEBUGP (("Reusing fd %d.\n", sock));
1422 if (pconn.authorized)
1423 /* If the connection is already authorized, the "Basic"
1424 authorization added by code above is unnecessary and
1426 request_remove_header (req, "Authorization");
1432 /* In its current implementation, persistent_available_p will
1433 look up conn->host in some cases. If that lookup failed, we
1434 don't need to bother with connect_to_host. */
1435 if (host_lookup_failed)
1441 sock = connect_to_host (conn->host, conn->port);
1450 return (retryable_socket_connect_error (errno)
1451 ? CONERROR : CONIMPOSSIBLE);
1455 if (proxy && u->scheme == SCHEME_HTTPS)
1457 /* When requesting SSL URLs through proxies, use the
1458 CONNECT method to request passthrough. */
1459 struct request *connreq = request_new ();
1460 request_set_method (connreq, "CONNECT",
1461 aprintf ("%s:%d", u->host, u->port));
1462 SET_USER_AGENT (req);
1465 request_set_header (connreq, "Proxy-Authorization",
1466 proxyauth, rel_value);
1467 /* Now that PROXYAUTH is part of the CONNECT request,
1468 zero it out so we don't send proxy authorization with
1469 the regular request below. */
1472 /* Examples in rfc2817 use the Host header in CONNECT
1473 requests. I don't see how that gains anything, given
1474 that the contents of Host would be exactly the same as
1475 the contents of CONNECT. */
1477 write_error = request_send (connreq, sock);
1478 request_free (connreq);
1479 if (write_error < 0)
1481 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1483 CLOSE_INVALIDATE (sock);
1487 head = read_http_response_head (sock);
1490 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1492 CLOSE_INVALIDATE (sock);
1501 DEBUGP (("proxy responded with: [%s]\n", head));
1503 resp = resp_new (head);
1504 statcode = resp_status (resp, &message);
1507 if (statcode != 200)
1510 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1511 message ? escnonprint (message) : "?");
1512 xfree_null (message);
1515 xfree_null (message);
1517 /* SOCK is now *really* connected to u->host, so update CONN
1518 to reflect this. That way register_persistent will
1519 register SOCK as being connected to u->host:u->port. */
1523 if (conn->scheme == SCHEME_HTTPS)
1525 if (!ssl_connect (sock))
1532 #endif /* HAVE_SSL */
1535 /* Send the request to server. */
1536 write_error = request_send (req, sock);
1538 if (write_error >= 0)
1542 DEBUGP (("[POST data: %s]\n", opt.post_data));
1543 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1545 else if (opt.post_file_name && post_data_size != 0)
1546 write_error = post_file (sock, opt.post_file_name, post_data_size);
1549 if (write_error < 0)
1551 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1553 CLOSE_INVALIDATE (sock);
1557 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1558 proxy ? "Proxy" : "HTTP");
1563 head = read_http_response_head (sock);
1568 logputs (LOG_NOTQUIET, _("No data received.\n"));
1569 CLOSE_INVALIDATE (sock);
1575 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1577 CLOSE_INVALIDATE (sock);
1582 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1584 resp = resp_new (head);
1586 /* Check for status line. */
1588 statcode = resp_status (resp, &message);
1589 if (!opt.server_response)
1590 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1591 message ? escnonprint (message) : "");
1594 logprintf (LOG_VERBOSE, "\n");
1595 print_server_response (resp, " ");
1598 if (!opt.ignore_length
1599 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1603 parsed = str_to_wgint (hdrval, NULL, 10);
1604 if (parsed == WGINT_MAX && errno == ERANGE)
1606 #### If Content-Length is out of range, it most likely
1607 means that the file is larger than 2G and that we're
1608 compiled without LFS. In that case we should probably
1609 refuse to even attempt to download the file. */
1615 /* Check for keep-alive related responses. */
1616 if (!inhibit_keep_alive && contlen != -1)
1618 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1620 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1622 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1627 /* The server has promised that it will not close the connection
1628 when we're done. This means that we can register it. */
1629 register_persistent (conn->host, conn->port, sock, using_ssl);
1631 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1633 /* Authorization is required. */
1636 if (skip_short_body (sock, contlen))
1637 CLOSE_FINISH (sock);
1639 CLOSE_INVALIDATE (sock);
1641 pconn.authorized = 0;
1642 if (auth_finished || !(user && passwd))
1644 /* If we have tried it already, then there is not point
1646 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1650 /* IIS sometimes sends two instances of WWW-Authenticate
1651 header, one with the keyword "negotiate", and other with
1652 useful data. Loop over all occurrences of this header
1653 and use the one we recognize. */
1655 const char *wabeg, *waend;
1656 char *www_authenticate = NULL;
1658 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1659 &wabeg, &waend)) != -1;
1661 if (known_authentication_scheme_p (wabeg, waend))
1663 www_authenticate = strdupdelim (wabeg, waend);
1666 /* If the authentication header is missing or recognized, or
1667 if the authentication scheme is "Basic" (which we send by
1668 default), there's no sense in retrying. */
1669 if (!www_authenticate
1670 || BEGINS_WITH (www_authenticate, "Basic"))
1672 xfree_null (www_authenticate);
1673 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1678 pth = url_full_path (u);
1679 request_set_header (req, "Authorization",
1680 create_authorization_line (www_authenticate,
1682 request_method (req),
1686 if (BEGINS_WITH (www_authenticate, "NTLM"))
1689 xfree (www_authenticate);
1690 goto retry_with_auth;
1696 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1698 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1700 pconn.authorized = 1;
1704 hs->statcode = statcode;
1706 hs->error = xstrdup (_("Malformed status line"));
1708 hs->error = xstrdup (_("(no description)"));
1710 hs->error = xstrdup (message);
1713 type = resp_header_strdup (resp, "Content-Type");
1716 char *tmp = strchr (type, ';');
1719 while (tmp > type && ISSPACE (tmp[-1]))
1724 hs->newloc = resp_header_strdup (resp, "Location");
1725 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1727 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1731 const char *scbeg, *scend;
1732 /* The jar should have been created by now. */
1733 assert (wget_cookie_jar != NULL);
1735 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1736 &scbeg, &scend)) != -1;
1739 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1742 /* u->path doesn't begin with /, which cookies.c expects. */
1743 pth = (char *) alloca (1 + strlen (u->path) + 1);
1745 strcpy (pth + 1, u->path);
1747 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, pth,
1752 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1754 wgint first_byte_pos, last_byte_pos, entity_length;
1755 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1757 contrange = first_byte_pos;
1762 /* 20x responses are counted among successful by default. */
1763 if (H_20X (statcode))
1766 /* Return if redirected. */
1767 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1769 /* RFC2068 says that in case of the 300 (multiple choices)
1770 response, the server can output a preferred URL through
1771 `Location' header; otherwise, the request should be treated
1772 like GET. So, if the location is set, it will be a
1773 redirection; otherwise, just proceed normally. */
1774 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1778 logprintf (LOG_VERBOSE,
1779 _("Location: %s%s\n"),
1780 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1781 hs->newloc ? _(" [following]") : "");
1784 if (skip_short_body (sock, contlen))
1785 CLOSE_FINISH (sock);
1787 CLOSE_INVALIDATE (sock);
1794 /* If content-type is not given, assume text/html. This is because
1795 of the multitude of broken CGI's that "forget" to generate the
1798 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1799 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1804 if (opt.html_extension && (*dt & TEXTHTML))
1805 /* -E / --html-extension / html_extension = on was specified, and this is a
1806 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1807 already the file's suffix, tack on ".html". */
1809 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1811 if (last_period_in_local_filename == NULL
1812 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1813 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1815 size_t local_filename_len = strlen(*hs->local_file);
1817 *hs->local_file = xrealloc(*hs->local_file,
1818 local_filename_len + sizeof(".html"));
1819 strcpy(*hs->local_file + local_filename_len, ".html");
1821 *dt |= ADDED_HTML_EXTENSION;
1825 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1827 /* If `-c' is in use and the file has been fully downloaded (or
1828 the remote file has shrunk), Wget effectively requests bytes
1829 after the end of file and the server response with 416. */
1830 logputs (LOG_VERBOSE, _("\
1831 \n The file is already fully retrieved; nothing to do.\n\n"));
1832 /* In case the caller inspects. */
1835 /* Mark as successfully retrieved. */
1838 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1839 might be more bytes in the body. */
1840 return RETRUNNEEDED;
1842 if ((contrange != 0 && contrange != hs->restval)
1843 || (H_PARTIAL (statcode) && !contrange))
1845 /* The Range request was somehow misunderstood by the server.
1848 CLOSE_INVALIDATE (sock);
1851 hs->contlen = contlen + contrange;
1857 /* No need to print this output if the body won't be
1858 downloaded at all, or if the original server response is
1860 logputs (LOG_VERBOSE, _("Length: "));
1863 logputs (LOG_VERBOSE, with_thousand_seps (contlen + contrange));
1864 if (contlen + contrange >= 1024)
1865 logprintf (LOG_VERBOSE, " (%s)",
1866 human_readable (contlen + contrange));
1869 if (contlen >= 1024)
1870 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1871 with_thousand_seps (contlen),
1872 human_readable (contlen));
1874 logprintf (LOG_VERBOSE, _(", %s remaining"),
1875 with_thousand_seps (contlen));
1879 logputs (LOG_VERBOSE,
1880 opt.ignore_length ? _("ignored") : _("unspecified"));
1882 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1884 logputs (LOG_VERBOSE, "\n");
1888 type = NULL; /* We don't need it any more. */
1890 /* Return if we have no intention of further downloading. */
1891 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1893 /* In case the caller cares to look... */
1897 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1898 servers not to send body in response to a HEAD request. If
1899 you encounter such a server (more likely a broken CGI), use
1900 `--no-http-keep-alive'. */
1901 CLOSE_FINISH (sock);
1902 return RETRFINISHED;
1905 /* Open the local file. */
1908 mkalldirs (*hs->local_file);
1910 rotate_backups (*hs->local_file);
1912 fp = fopen (*hs->local_file, "ab");
1913 else if (opt.noclobber || opt.always_rest || opt.timestamping || opt.dirstruct
1914 || opt.output_document)
1915 fp = fopen (*hs->local_file, "wb");
1918 fp = fopen_excl (*hs->local_file, 1);
1919 if (!fp && errno == EEXIST)
1921 /* We cannot just invent a new name and use it (which is
1922 what functions like unique_create typically do)
1923 because we told the user we'd use this name.
1924 Instead, return and retry the download. */
1925 logprintf (LOG_NOTQUIET,
1926 _("%s has sprung into existence.\n"),
1928 CLOSE_INVALIDATE (sock);
1929 return FOPEN_EXCL_ERR;
1934 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1935 CLOSE_INVALIDATE (sock);
1942 /* #### This confuses the timestamping code that checks for file
1943 size. Maybe we should save some additional information? */
1944 if (opt.save_headers)
1945 fwrite (head, 1, strlen (head), fp);
1947 /* Download the request body. */
1950 flags |= rb_read_exactly;
1951 if (hs->restval > 0 && contrange == 0)
1952 /* If the server ignored our range request, instruct fd_read_body
1953 to skip the first RESTVAL bytes of body. */
1954 flags |= rb_skip_startpos;
1955 hs->len = hs->restval;
1957 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1958 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1962 CLOSE_FINISH (sock);
1964 CLOSE_INVALIDATE (sock);
1967 /* Close or flush the file. We have to be careful to check for
1968 error here. Checking the result of fwrite() is not enough --
1969 errors could go unnoticed! */
1972 flush_res = fclose (fp);
1974 flush_res = fflush (fp);
1975 if (flush_res == EOF)
1980 return RETRFINISHED;
1983 /* The genuine HTTP loop! This is the part where the retrieval is
1984 retried, and retried, and retried, and... */
1986 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1987 int *dt, struct url *proxy)
1990 int use_ts, got_head = 0; /* time-stamping info */
1991 char *filename_plus_orig_suffix;
1992 char *local_filename = NULL;
1993 char *tms, *locf, *tmrate;
1995 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1996 wgint local_size = 0; /* the size of the local file */
1997 size_t filename_len;
1998 struct http_stat hstat; /* HTTP status */
2002 /* This used to be done in main(), but it's a better idea to do it
2003 here so that we don't go through the hoops if we're just using
2007 if (!wget_cookie_jar)
2008 wget_cookie_jar = cookie_jar_new ();
2009 if (opt.cookies_input && !cookies_loaded_p)
2011 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2012 cookies_loaded_p = 1;
2018 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
2019 has_wildcards_p because it would also warn on `?', and we know that
2020 shows up in CGI paths a *lot*. */
2021 if (strchr (u->url, '*'))
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, will not retrieve.\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. */
2165 hstat.restval = hstat.len; /* continue where we left off */
2166 else if (opt.always_rest
2167 && stat (locf, &st) == 0
2168 && S_ISREG (st.st_mode))
2169 hstat.restval = st.st_size;
2171 /* Decide whether to send the no-cache directive. We send it in
2173 a) we're using a proxy, and we're past our first retrieval.
2174 Some proxies are notorious for caching incomplete data, so
2175 we require a fresh get.
2176 b) caching is explicitly inhibited. */
2177 if ((proxy && count > 1) /* a */
2178 || !opt.allow_cache /* b */
2180 *dt |= SEND_NOCACHE;
2182 *dt &= ~SEND_NOCACHE;
2184 /* Try fetching the document, or at least its head. */
2185 err = gethttp (u, &hstat, dt, proxy);
2187 /* It's unfortunate that wget determines the local filename before finding
2188 out the Content-Type of the file. Barring a major restructuring of the
2189 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2190 *hstat.local_file to tack on ".html". */
2191 if (!opt.output_document)
2192 locf = *hstat.local_file;
2195 tms = time_str (NULL);
2196 /* Get the new location (with or without the redirection). */
2198 *newloc = xstrdup (hstat.newloc);
2201 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2202 case CONERROR: case READERR: case WRITEFAILED:
2203 case RANGEERR: case FOPEN_EXCL_ERR:
2204 /* Non-fatal errors continue executing the loop, which will
2205 bring them to "while" statement at the end, to judge
2206 whether the number of tries was exceeded. */
2207 free_hstat (&hstat);
2208 printwhat (count, opt.ntry);
2209 if (err == FOPEN_EXCL_ERR)
2211 /* Re-determine the file name. */
2212 if (local_file && *local_file)
2214 xfree (*local_file);
2215 *local_file = url_file_name (u);
2216 hstat.local_file = local_file;
2221 dummy = url_file_name (u);
2222 hstat.local_file = &dummy;
2224 /* be honest about where we will save the file */
2225 if (local_file && opt.output_document)
2226 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2227 if (!opt.output_document)
2228 locf = *hstat.local_file;
2230 locf = opt.output_document;
2234 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2235 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
2236 /* Fatal errors just return from the function. */
2237 free_hstat (&hstat);
2241 case FWRITEERR: case FOPENERR:
2242 /* Another fatal error. */
2243 logputs (LOG_VERBOSE, "\n");
2244 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2245 *hstat.local_file, strerror (errno));
2246 free_hstat (&hstat);
2251 /* Another fatal error. */
2252 logputs (LOG_VERBOSE, "\n");
2253 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2254 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);
2274 /* The file was already fully retrieved. */
2275 free_hstat (&hstat);
2280 /* Deal with you later. */
2283 /* All possibilities should have been exhausted. */
2286 if (!(*dt & RETROKF))
2290 /* #### Ugly ugly ugly! */
2291 char *hurl = url_string (u, 1);
2292 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2295 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2296 tms, hstat.statcode, escnonprint (hstat.error));
2297 logputs (LOG_VERBOSE, "\n");
2298 free_hstat (&hstat);
2303 /* Did we get the time-stamp? */
2306 if (opt.timestamping && !hstat.remote_time)
2308 logputs (LOG_NOTQUIET, _("\
2309 Last-modified header missing -- time-stamps turned off.\n"));
2311 else if (hstat.remote_time)
2313 /* Convert the date-string into struct tm. */
2314 tmr = http_atotm (hstat.remote_time);
2315 if (tmr == (time_t) (-1))
2316 logputs (LOG_VERBOSE, _("\
2317 Last-modified header invalid -- time-stamp ignored.\n"));
2321 /* The time-stamping section. */
2326 use_ts = 0; /* no more time-stamping */
2327 count = 0; /* the retrieve count for HEAD is
2329 if (hstat.remote_time && tmr != (time_t) (-1))
2331 /* Now time-stamping can be used validly. Time-stamping
2332 means that if the sizes of the local and remote file
2333 match, and local file is newer than the remote file,
2334 it will not be retrieved. Otherwise, the normal
2335 download procedure is resumed. */
2337 (hstat.contlen == -1 || local_size == hstat.contlen))
2339 logprintf (LOG_VERBOSE, _("\
2340 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2342 free_hstat (&hstat);
2346 else if (tml >= tmr)
2347 logprintf (LOG_VERBOSE, _("\
2348 The sizes do not match (local %s) -- retrieving.\n"),
2349 number_to_static_string (local_size));
2351 logputs (LOG_VERBOSE,
2352 _("Remote file is newer, retrieving.\n"));
2354 free_hstat (&hstat);
2357 if ((tmr != (time_t) (-1))
2359 && ((hstat.len == hstat.contlen) ||
2360 ((hstat.res == 0) &&
2361 ((hstat.contlen == -1) ||
2362 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2364 /* #### This code repeats in http.c and ftp.c. Move it to a
2366 const char *fl = NULL;
2367 if (opt.output_document)
2369 if (output_stream_regular)
2370 fl = opt.output_document;
2373 fl = *hstat.local_file;
2377 /* End of time-stamping section. */
2381 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2382 escnonprint (hstat.error));
2387 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2389 if (hstat.len == hstat.contlen)
2393 logprintf (LOG_VERBOSE,
2394 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2396 number_to_static_string (hstat.len),
2397 number_to_static_string (hstat.contlen));
2398 logprintf (LOG_NONVERBOSE,
2399 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2401 number_to_static_string (hstat.len),
2402 number_to_static_string (hstat.contlen),
2406 total_downloaded_bytes += hstat.len;
2408 /* Remember that we downloaded the file for later ".orig" code. */
2409 if (*dt & ADDED_HTML_EXTENSION)
2410 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2412 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2414 free_hstat (&hstat);
2418 else if (hstat.res == 0) /* No read error */
2420 if (hstat.contlen == -1) /* We don't know how much we were supposed
2421 to get, so assume we succeeded. */
2425 logprintf (LOG_VERBOSE,
2426 _("%s (%s) - `%s' saved [%s]\n\n"),
2428 number_to_static_string (hstat.len));
2429 logprintf (LOG_NONVERBOSE,
2430 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2431 tms, u->url, number_to_static_string (hstat.len),
2435 total_downloaded_bytes += hstat.len;
2437 /* Remember that we downloaded the file for later ".orig" code. */
2438 if (*dt & ADDED_HTML_EXTENSION)
2439 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2441 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2443 free_hstat (&hstat);
2447 else if (hstat.len < hstat.contlen) /* meaning we lost the
2448 connection too soon */
2450 logprintf (LOG_VERBOSE,
2451 _("%s (%s) - Connection closed at byte %s. "),
2452 tms, tmrate, number_to_static_string (hstat.len));
2453 printwhat (count, opt.ntry);
2454 free_hstat (&hstat);
2457 else if (!opt.kill_longer) /* meaning we got more than expected */
2459 logprintf (LOG_VERBOSE,
2460 _("%s (%s) - `%s' saved [%s/%s])\n\n"),
2462 number_to_static_string (hstat.len),
2463 number_to_static_string (hstat.contlen));
2464 logprintf (LOG_NONVERBOSE,
2465 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2467 number_to_static_string (hstat.len),
2468 number_to_static_string (hstat.contlen),
2471 total_downloaded_bytes += hstat.len;
2473 /* Remember that we downloaded the file for later ".orig" code. */
2474 if (*dt & ADDED_HTML_EXTENSION)
2475 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2477 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2479 free_hstat (&hstat);
2483 else /* the same, but not accepted */
2485 logprintf (LOG_VERBOSE,
2486 _("%s (%s) - Connection closed at byte %s/%s. "),
2488 number_to_static_string (hstat.len),
2489 number_to_static_string (hstat.contlen));
2490 printwhat (count, opt.ntry);
2491 free_hstat (&hstat);
2495 else /* now hstat.res can only be -1 */
2497 if (hstat.contlen == -1)
2499 logprintf (LOG_VERBOSE,
2500 _("%s (%s) - Read error at byte %s (%s)."),
2501 tms, tmrate, number_to_static_string (hstat.len),
2503 printwhat (count, opt.ntry);
2504 free_hstat (&hstat);
2507 else /* hstat.res == -1 and contlen is given */
2509 logprintf (LOG_VERBOSE,
2510 _("%s (%s) - Read error at byte %s/%s (%s). "),
2512 number_to_static_string (hstat.len),
2513 number_to_static_string (hstat.contlen),
2515 printwhat (count, opt.ntry);
2516 free_hstat (&hstat);
2523 while (!opt.ntry || (count < opt.ntry));
2527 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2528 than local timezone.
2530 mktime is similar but assumes struct tm, also known as the
2531 "broken-down" form of time, is in local time zone. mktime_from_utc
2532 uses mktime to make the conversion understanding that an offset
2533 will be introduced by the local time assumption.
2535 mktime_from_utc then measures the introduced offset by applying
2536 gmtime to the initial result and applying mktime to the resulting
2537 "broken-down" form. The difference between the two mktime results
2538 is the measured offset which is then subtracted from the initial
2539 mktime result to yield a calendar time which is the value returned.
2541 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2542 consistent offset (the non DST offset) since tm and tm+o might be
2543 on opposite sides of a DST change.
2545 Some implementations of mktime return -1 for the nonexistent
2546 localtime hour at the beginning of DST. In this event, use
2547 mktime(tm - 1hr) + 3600.
2551 gmtime(t+o) --> tm+o
2552 mktime(tm+o) --> t+2o
2553 t+o - (t+2o - t+o) = t
2555 Note that glibc contains a function of the same purpose named
2556 `timegm' (reverse of gmtime). But obviously, it is not universally
2557 available, and unfortunately it is not straightforwardly
2558 extractable for use here. Perhaps configure should detect timegm
2559 and use it where available.
2561 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2562 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2563 Further improved by Roger with assistance from Edward J. Sabol
2564 based on input by Jamie Zawinski. */
2567 mktime_from_utc (struct tm *t)
2578 return -1; /* can't deal with output from strptime */
2589 return -1; /* can't deal with output from gmtime */
2592 return (tl - (tb - tl));
2595 /* Check whether the result of strptime() indicates success.
2596 strptime() returns the pointer to how far it got to in the string.
2597 The processing has been successful if the string is at `GMT' or
2598 `+X', or at the end of the string.
2600 In extended regexp parlance, the function returns 1 if P matches
2601 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2602 can return) is considered a failure and 0 is returned. */
2604 check_end (const char *p)
2608 while (ISSPACE (*p))
2611 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2612 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2618 /* Convert the textual specification of time in TIME_STRING to the
2619 number of seconds since the Epoch.
2621 TIME_STRING can be in any of the three formats RFC2068 allows the
2622 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2623 Timezones are ignored, and should be GMT.
2625 Return the computed time_t representation, or -1 if the conversion
2628 This function uses strptime with various string formats for parsing
2629 TIME_STRING. This results in a parser that is not as lenient in
2630 interpreting TIME_STRING as I would like it to be. Being based on
2631 strptime, it always allows shortened months, one-digit days, etc.,
2632 but due to the multitude of formats in which time can be
2633 represented, an ideal HTTP time parser would be even more
2634 forgiving. It should completely ignore things like week days and
2635 concentrate only on the various forms of representing years,
2636 months, days, hours, minutes, and seconds. For example, it would
2637 be nice if it accepted ISO 8601 out of the box.
2639 I've investigated free and PD code for this purpose, but none was
2640 usable. getdate was big and unwieldy, and had potential copyright
2641 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2642 distributed with phttpd, is excellent, but we cannot use it because
2643 it is not assigned to the FSF. So I stuck it with strptime. */
2646 http_atotm (const char *time_string)
2648 /* NOTE: Solaris strptime man page claims that %n and %t match white
2649 space, but that's not universally available. Instead, we simply
2650 use ` ' to mean "skip all WS", which works under all strptime
2651 implementations I've tested. */
2653 static const char *time_formats[] = {
2654 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2655 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2656 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2657 (google.com uses this for their cookies.) */
2658 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2664 /* According to Roger Beeman, we need to initialize tm_isdst, since
2665 strptime won't do it. */
2668 /* Note that under foreign locales Solaris strptime() fails to
2669 recognize English dates, which renders this function useless. We
2670 solve this by being careful not to affect LC_TIME when
2671 initializing locale.
2673 Another solution would be to temporarily set locale to C, invoke
2674 strptime(), and restore it back. This is slow and dirty,
2675 however, and locale support other than LC_MESSAGES can mess other
2676 things, so I rather chose to stick with just setting LC_MESSAGES.
2678 GNU strptime does not have this problem because it recognizes
2679 both international and local dates. */
2681 for (i = 0; i < countof (time_formats); i++)
2682 if (check_end (strptime (time_string, time_formats[i], &t)))
2683 return mktime_from_utc (&t);
2685 /* All formats have failed. */
2689 /* Authorization support: We support three authorization schemes:
2691 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2693 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2694 consisting of answering to the server's challenge with the proper
2697 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2698 Stenberg for libcurl. Like digest, NTLM is based on a
2699 challenge-response mechanism, but unlike digest, it is non-standard
2700 (authenticates TCP connections rather than requests), undocumented
2701 and Microsoft-specific. */
2703 /* Create the authentication header contents for the `Basic' scheme.
2704 This is done by encoding the string `USER:PASS' in base64 and
2705 prepending `HEADER: Basic ' to it. */
2708 basic_authentication_encode (const char *user, const char *passwd)
2711 int len1 = strlen (user) + 1 + strlen (passwd);
2713 t1 = (char *)alloca (len1 + 1);
2714 sprintf (t1, "%s:%s", user, passwd);
2716 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2717 base64_encode (t1, len1, t2);
2719 return concat_strings ("Basic ", t2, (char *) 0);
2722 #define SKIP_WS(x) do { \
2723 while (ISSPACE (*(x))) \
2727 #ifdef ENABLE_DIGEST
2728 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2729 of a field in such a header. If the field is the one specified by
2730 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2731 digest authorization code), extract its value in the (char*)
2732 variable pointed by RET. Returns negative on a malformed header,
2733 or number of bytes that have been parsed by this call. */
2735 extract_header_attr (const char *au, const char *attr_name, char **ret)
2738 const char *cp = au;
2740 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2742 cp += strlen (attr_name);
2755 for (ep = cp; *ep && *ep != '\"'; ep++)
2760 *ret = strdupdelim (cp, ep);
2767 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2768 an array of 16 bytes containing the hash keys, and BUF should be a
2769 buffer of 33 writable characters (32 for hex digits plus one for
2770 zero termination). */
2772 dump_hash (unsigned char *buf, const unsigned char *hash)
2776 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2778 *buf++ = XNUM_TO_digit (*hash >> 4);
2779 *buf++ = XNUM_TO_digit (*hash & 0xf);
2784 /* Take the line apart to find the challenge, and compose a digest
2785 authorization header. See RFC2069 section 2.1.2. */
2787 digest_authentication_encode (const char *au, const char *user,
2788 const char *passwd, const char *method,
2791 static char *realm, *opaque, *nonce;
2796 { "realm", &realm },
2797 { "opaque", &opaque },
2802 realm = opaque = nonce = NULL;
2804 au += 6; /* skip over `Digest' */
2810 for (i = 0; i < countof (options); i++)
2812 int skip = extract_header_attr (au, options[i].name,
2813 options[i].variable);
2817 xfree_null (opaque);
2827 if (i == countof (options))
2829 while (*au && *au != '=')
2837 while (*au && *au != '\"')
2844 while (*au && *au != ',')
2849 if (!realm || !nonce || !user || !passwd || !path || !method)
2852 xfree_null (opaque);
2857 /* Calculate the digest value. */
2859 ALLOCA_MD5_CONTEXT (ctx);
2860 unsigned char hash[MD5_HASHLEN];
2861 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2862 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2864 /* A1BUF = H(user ":" realm ":" password) */
2866 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2867 gen_md5_update ((unsigned char *)":", 1, ctx);
2868 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2869 gen_md5_update ((unsigned char *)":", 1, ctx);
2870 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2871 gen_md5_finish (ctx, hash);
2872 dump_hash (a1buf, hash);
2874 /* A2BUF = H(method ":" path) */
2876 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2877 gen_md5_update ((unsigned char *)":", 1, ctx);
2878 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2879 gen_md5_finish (ctx, hash);
2880 dump_hash (a2buf, hash);
2882 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2884 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2885 gen_md5_update ((unsigned char *)":", 1, ctx);
2886 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2887 gen_md5_update ((unsigned char *)":", 1, ctx);
2888 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2889 gen_md5_finish (ctx, hash);
2890 dump_hash (response_digest, hash);
2892 res = (char*) xmalloc (strlen (user)
2897 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2898 + (opaque ? strlen (opaque) : 0)
2900 sprintf (res, "Digest \
2901 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2902 user, realm, nonce, path, response_digest);
2905 char *p = res + strlen (res);
2906 strcat (p, ", opaque=\"");
2913 #endif /* ENABLE_DIGEST */
2915 /* Computing the size of a string literal must take into account that
2916 value returned by sizeof includes the terminating \0. */
2917 #define STRSIZE(literal) (sizeof (literal) - 1)
2919 /* Whether chars in [b, e) begin with the literal string provided as
2920 first argument and are followed by whitespace or terminating \0.
2921 The comparison is case-insensitive. */
2922 #define STARTS(literal, b, e) \
2923 ((e) - (b) >= STRSIZE (literal) \
2924 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2925 && ((e) - (b) == STRSIZE (literal) \
2926 || ISSPACE (b[STRSIZE (literal)])))
2929 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2931 return STARTS ("Basic", hdrbeg, hdrend)
2932 #ifdef ENABLE_DIGEST
2933 || STARTS ("Digest", hdrbeg, hdrend)
2936 || STARTS ("NTLM", hdrbeg, hdrend)
2943 /* Create the HTTP authorization request header. When the
2944 `WWW-Authenticate' response header is seen, according to the
2945 authorization scheme specified in that header (`Basic' and `Digest'
2946 are supported by the current implementation), produce an
2947 appropriate HTTP authorization request header. */
2949 create_authorization_line (const char *au, const char *user,
2950 const char *passwd, const char *method,
2951 const char *path, int *finished)
2953 /* We are called only with known schemes, so we can dispatch on the
2955 switch (TOUPPER (*au))
2957 case 'B': /* Basic */
2959 return basic_authentication_encode (user, passwd);
2960 #ifdef ENABLE_DIGEST
2961 case 'D': /* Digest */
2963 return digest_authentication_encode (au, user, passwd, method, path);
2966 case 'N': /* NTLM */
2967 if (!ntlm_input (&pconn.ntlm, au))
2972 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2975 /* We shouldn't get here -- this function should be only called
2976 with values approved by known_authentication_scheme_p. */
2984 if (wget_cookie_jar)
2985 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2991 xfree_null (pconn.host);
2992 if (wget_cookie_jar)
2993 cookie_jar_delete (wget_cookie_jar);