2 Copyright (C) 1996-2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
55 # include "http-ntlm.h"
67 extern char *version_string;
70 # define MIN(x, y) ((x) > (y) ? (y) : (x))
74 static bool cookies_loaded_p;
75 static struct cookie_jar *wget_cookie_jar;
77 #define TEXTHTML_S "text/html"
78 #define TEXTXHTML_S "application/xhtml+xml"
80 /* Some status code validation macros: */
81 #define H_20X(x) (((x) >= 200) && ((x) < 300))
82 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
83 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
84 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
85 || (x) == HTTP_STATUS_SEE_OTHER \
86 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
88 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
90 #define HTTP_STATUS_OK 200
91 #define HTTP_STATUS_CREATED 201
92 #define HTTP_STATUS_ACCEPTED 202
93 #define HTTP_STATUS_NO_CONTENT 204
94 #define HTTP_STATUS_PARTIAL_CONTENTS 206
96 /* Redirection 3xx. */
97 #define HTTP_STATUS_MULTIPLE_CHOICES 300
98 #define HTTP_STATUS_MOVED_PERMANENTLY 301
99 #define HTTP_STATUS_MOVED_TEMPORARILY 302
100 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
101 #define HTTP_STATUS_NOT_MODIFIED 304
102 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
109 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
111 /* Server errors 5xx. */
112 #define HTTP_STATUS_INTERNAL 500
113 #define HTTP_STATUS_NOT_IMPLEMENTED 501
114 #define HTTP_STATUS_BAD_GATEWAY 502
115 #define HTTP_STATUS_UNAVAILABLE 503
118 rel_none, rel_name, rel_value, rel_both
125 struct request_header {
127 enum rp release_policy;
129 int hcount, hcapacity;
132 /* Create a new, empty request. At least request_set_method must be
133 called before the request can be used. */
135 static struct request *
138 struct request *req = xnew0 (struct request);
140 req->headers = xnew_array (struct request_header, req->hcapacity);
144 /* Set the request's method and its arguments. METH should be a
145 literal string (or it should outlive the request) because it will
146 not be freed. ARG will be freed by request_free. */
149 request_set_method (struct request *req, const char *meth, char *arg)
155 /* Return the method string passed with the last call to
156 request_set_method. */
159 request_method (const struct request *req)
164 /* Free one header according to the release policy specified with
165 request_set_header. */
168 release_header (struct request_header *hdr)
170 switch (hdr->release_policy)
187 /* Set the request named NAME to VALUE. Specifically, this means that
188 a "NAME: VALUE\r\n" header line will be used in the request. If a
189 header with the same name previously existed in the request, its
190 value will be replaced by this one. A NULL value means do nothing.
192 RELEASE_POLICY determines whether NAME and VALUE should be released
193 (freed) with request_free. Allowed values are:
195 - rel_none - don't free NAME or VALUE
196 - rel_name - free NAME when done
197 - rel_value - free VALUE when done
198 - rel_both - free both NAME and VALUE when done
200 Setting release policy is useful when arguments come from different
201 sources. For example:
203 // Don't free literal strings!
204 request_set_header (req, "Pragma", "no-cache", rel_none);
206 // Don't free a global variable, we'll need it later.
207 request_set_header (req, "Referer", opt.referer, rel_none);
209 // Value freshly allocated, free it when done.
210 request_set_header (req, "Range",
211 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
216 request_set_header (struct request *req, char *name, char *value,
217 enum rp release_policy)
219 struct request_header *hdr;
224 /* A NULL value is a no-op; if freeing the name is requested,
225 free it now to avoid leaks. */
226 if (release_policy == rel_name || release_policy == rel_both)
231 for (i = 0; i < req->hcount; i++)
233 hdr = &req->headers[i];
234 if (0 == strcasecmp (name, hdr->name))
236 /* Replace existing header. */
237 release_header (hdr);
240 hdr->release_policy = release_policy;
245 /* Install new header. */
247 if (req->hcount >= req->hcapacity)
249 req->hcapacity <<= 1;
250 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
252 hdr = &req->headers[req->hcount++];
255 hdr->release_policy = release_policy;
258 /* Like request_set_header, but sets the whole header line, as
259 provided by the user using the `--header' option. For example,
260 request_set_user_header (req, "Foo: bar") works just like
261 request_set_header (req, "Foo", "bar"). */
264 request_set_user_header (struct request *req, const char *header)
267 const char *p = strchr (header, ':');
270 BOUNDED_TO_ALLOCA (header, p, name);
274 request_set_header (req, xstrdup (name), (char *) p, rel_name);
277 /* Remove the header with specified name from REQ. Returns true if
278 the header was actually removed, false otherwise. */
281 request_remove_header (struct request *req, char *name)
284 for (i = 0; i < req->hcount; i++)
286 struct request_header *hdr = &req->headers[i];
287 if (0 == strcasecmp (name, hdr->name))
289 release_header (hdr);
290 /* Move the remaining headers by one. */
291 if (i < req->hcount - 1)
292 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
300 #define APPEND(p, str) do { \
301 int A_len = strlen (str); \
302 memcpy (p, str, A_len); \
306 /* Construct the request and write it to FD using fd_write. */
309 request_send (const struct request *req, int fd)
311 char *request_string, *p;
312 int i, size, write_error;
314 /* Count the request size. */
317 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
318 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
320 for (i = 0; i < req->hcount; i++)
322 struct request_header *hdr = &req->headers[i];
323 /* NAME ": " VALUE "\r\n" */
324 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
330 p = request_string = alloca_array (char, size);
332 /* Generate the request. */
334 APPEND (p, req->method); *p++ = ' ';
335 APPEND (p, req->arg); *p++ = ' ';
336 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
338 for (i = 0; i < req->hcount; i++)
340 struct request_header *hdr = &req->headers[i];
341 APPEND (p, hdr->name);
342 *p++ = ':', *p++ = ' ';
343 APPEND (p, hdr->value);
344 *p++ = '\r', *p++ = '\n';
347 *p++ = '\r', *p++ = '\n', *p++ = '\0';
348 assert (p - request_string == size);
352 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
354 /* Send the request to the server. */
356 write_error = fd_write (fd, request_string, size - 1, -1);
358 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
363 /* Release the resources used by REQ. */
366 request_free (struct request *req)
369 xfree_null (req->arg);
370 for (i = 0; i < req->hcount; i++)
371 release_header (&req->headers[i]);
372 xfree_null (req->headers);
376 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
377 PROMISED_SIZE bytes are sent over the wire -- if the file is
378 longer, read only that much; if the file is shorter, report an error. */
381 post_file (int sock, const char *file_name, wgint promised_size)
383 static char chunk[8192];
388 DEBUGP (("[writing POST file %s ... ", file_name));
390 fp = fopen (file_name, "rb");
393 while (!feof (fp) && written < promised_size)
396 int length = fread (chunk, 1, sizeof (chunk), fp);
399 towrite = MIN (promised_size - written, length);
400 write_error = fd_write (sock, chunk, towrite, -1);
410 /* If we've written less than was promised, report a (probably
411 nonsensical) error rather than break the promise. */
412 if (written < promised_size)
418 assert (written == promised_size);
419 DEBUGP (("done]\n"));
423 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
424 If so, return the pointer to the position after the line, otherwise
425 return NULL. This is used as callback to fd_read_hunk. The data
426 between START and PEEKED has been read and cannot be "unread"; the
427 data after PEEKED has only been peeked. */
430 response_head_terminator (const char *start, const char *peeked, int peeklen)
434 /* If at first peek, verify whether HUNK starts with "HTTP". If
435 not, this is a HTTP/0.9 request and we must bail out without
437 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
440 /* Look for "\n[\r]\n", and return the following position if found.
441 Start two chars before the current to cover the possibility that
442 part of the terminator (e.g. "\n\r") arrived in the previous
444 p = peeked - start < 2 ? start : peeked - 2;
445 end = peeked + peeklen;
447 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
448 for (; p < end - 2; p++)
451 if (p[1] == '\r' && p[2] == '\n')
453 else if (p[1] == '\n')
456 /* p==end-2: check for \n\n directly preceding END. */
457 if (p[0] == '\n' && p[1] == '\n')
463 /* The maximum size of a single HTTP response we care to read. Rather
464 than being a limit of the reader implementation, this limit
465 prevents Wget from slurping all available memory upon encountering
466 malicious or buggy server output, thus protecting the user. Define
467 it to 0 to remove the limit. */
469 #define HTTP_RESPONSE_MAX_SIZE 65536
471 /* Read the HTTP request head from FD and return it. The error
472 conditions are the same as with fd_read_hunk.
474 To support HTTP/0.9 responses, this function tries to make sure
475 that the data begins with "HTTP". If this is not the case, no data
476 is read and an empty request is returned, so that the remaining
477 data can be treated as body. */
480 read_http_response_head (int fd)
482 return fd_read_hunk (fd, response_head_terminator, 512,
483 HTTP_RESPONSE_MAX_SIZE);
487 /* The response data. */
490 /* The array of pointers that indicate where each header starts.
491 For example, given this HTTP response:
498 The headers are located like this:
500 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
502 headers[0] headers[1] headers[2] headers[3]
504 I.e. headers[0] points to the beginning of the request,
505 headers[1] points to the end of the first header and the
506 beginning of the second one, etc. */
508 const char **headers;
511 /* Create a new response object from the text of the HTTP response,
512 available in HEAD. That text is automatically split into
513 constituent header lines for fast retrieval using
516 static struct response *
517 resp_new (const char *head)
522 struct response *resp = xnew0 (struct response);
527 /* Empty head means that we're dealing with a headerless
528 (HTTP/0.9) response. In that case, don't set HEADERS at
533 /* Split HEAD into header lines, so that resp_header_* functions
534 don't need to do this over and over again. */
540 DO_REALLOC (resp->headers, size, count + 1, const char *);
541 resp->headers[count++] = hdr;
543 /* Break upon encountering an empty line. */
544 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
547 /* Find the end of HDR, including continuations. */
550 const char *end = strchr (hdr, '\n');
556 while (*hdr == ' ' || *hdr == '\t');
558 DO_REALLOC (resp->headers, size, count + 1, const char *);
559 resp->headers[count] = NULL;
564 /* Locate the header named NAME in the request data, starting with
565 position START. This allows the code to loop through the request
566 data, filtering for all requests of a given name. Returns the
567 found position, or -1 for failure. The code that uses this
568 function typically looks like this:
570 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
571 ... do something with header ...
573 If you only care about one header, use resp_header_get instead of
577 resp_header_locate (const struct response *resp, const char *name, int start,
578 const char **begptr, const char **endptr)
581 const char **headers = resp->headers;
584 if (!headers || !headers[1])
587 name_len = strlen (name);
593 for (; headers[i + 1]; i++)
595 const char *b = headers[i];
596 const char *e = headers[i + 1];
598 && b[name_len] == ':'
599 && 0 == strncasecmp (b, name, name_len))
602 while (b < e && ISSPACE (*b))
604 while (b < e && ISSPACE (e[-1]))
614 /* Find and retrieve the header named NAME in the request data. If
615 found, set *BEGPTR to its starting, and *ENDPTR to its ending
616 position, and return true. Otherwise return false.
618 This function is used as a building block for resp_header_copy
619 and resp_header_strdup. */
622 resp_header_get (const struct response *resp, const char *name,
623 const char **begptr, const char **endptr)
625 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
629 /* Copy the response header named NAME to buffer BUF, no longer than
630 BUFSIZE (BUFSIZE includes the terminating 0). If the header
631 exists, true is returned, false otherwise. If there should be no
632 limit on the size of the header, use resp_header_strdup instead.
634 If BUFSIZE is 0, no data is copied, but the boolean indication of
635 whether the header is present is still returned. */
638 resp_header_copy (const struct response *resp, const char *name,
639 char *buf, int bufsize)
642 if (!resp_header_get (resp, name, &b, &e))
646 int len = MIN (e - b, bufsize - 1);
647 memcpy (buf, b, len);
653 /* Return the value of header named NAME in RESP, allocated with
654 malloc. If such a header does not exist in RESP, return NULL. */
657 resp_header_strdup (const struct response *resp, const char *name)
660 if (!resp_header_get (resp, name, &b, &e))
662 return strdupdelim (b, e);
665 /* Parse the HTTP status line, which is of format:
667 HTTP-Version SP Status-Code SP Reason-Phrase
669 The function returns the status-code, or -1 if the status line
670 appears malformed. The pointer to "reason-phrase" message is
671 returned in *MESSAGE. */
674 resp_status (const struct response *resp, char **message)
681 /* For a HTTP/0.9 response, assume status 200. */
683 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
687 p = resp->headers[0];
688 end = resp->headers[1];
694 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
698 /* Match the HTTP version. This is optional because Gnutella
699 servers have been reported to not specify HTTP version. */
700 if (p < end && *p == '/')
703 while (p < end && ISDIGIT (*p))
705 if (p < end && *p == '.')
707 while (p < end && ISDIGIT (*p))
711 while (p < end && ISSPACE (*p))
713 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
716 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
721 while (p < end && ISSPACE (*p))
723 while (p < end && ISSPACE (end[-1]))
725 *message = strdupdelim (p, end);
731 /* Release the resources used by RESP. */
734 resp_free (struct response *resp)
736 xfree_null (resp->headers);
740 /* Print the server response, line by line, omitting the trailing CRLF
741 from individual header lines, and prefixed with PREFIX. */
744 print_server_response (const struct response *resp, const char *prefix)
749 for (i = 0; resp->headers[i + 1]; i++)
751 const char *b = resp->headers[i];
752 const char *e = resp->headers[i + 1];
754 if (b < e && e[-1] == '\n')
756 if (b < e && e[-1] == '\r')
758 /* This is safe even on printfs with broken handling of "%.<n>s"
759 because resp->headers ends with \0. */
760 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
764 /* Parse the `Content-Range' header and extract the information it
765 contains. Returns true if successful, false otherwise. */
767 parse_content_range (const char *hdr, wgint *first_byte_ptr,
768 wgint *last_byte_ptr, wgint *entity_length_ptr)
772 /* Ancient versions of Netscape proxy server, presumably predating
773 rfc2068, sent out `Content-Range' without the "bytes"
775 if (0 == strncasecmp (hdr, "bytes", 5))
778 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
782 while (ISSPACE (*hdr))
789 for (num = 0; ISDIGIT (*hdr); hdr++)
790 num = 10 * num + (*hdr - '0');
791 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
793 *first_byte_ptr = num;
795 for (num = 0; ISDIGIT (*hdr); hdr++)
796 num = 10 * num + (*hdr - '0');
797 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
799 *last_byte_ptr = num;
801 for (num = 0; ISDIGIT (*hdr); hdr++)
802 num = 10 * num + (*hdr - '0');
803 *entity_length_ptr = num;
807 /* Read the body of the request, but don't store it anywhere and don't
808 display a progress gauge. This is useful for reading the bodies of
809 administrative responses to which we will soon issue another
810 request. The response is not useful to the user, but reading it
811 allows us to continue using the same connection to the server.
813 If reading fails, false is returned, true otherwise. In debug
814 mode, the body is displayed for debugging purposes. */
817 skip_short_body (int fd, wgint contlen)
820 SKIP_SIZE = 512, /* size of the download buffer */
821 SKIP_THRESHOLD = 4096 /* the largest size we read */
823 char dlbuf[SKIP_SIZE + 1];
824 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
826 /* We shouldn't get here with unknown contlen. (This will change
827 with HTTP/1.1, which supports "chunked" transfer.) */
828 assert (contlen != -1);
830 /* If the body is too large, it makes more sense to simply close the
831 connection than to try to read the body. */
832 if (contlen > SKIP_THRESHOLD)
835 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
839 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
842 /* Don't normally report the error since this is an
843 optimization that should be invisible to the user. */
844 DEBUGP (("] aborting (%s).\n",
845 ret < 0 ? fd_errstr (fd) : "EOF received"));
849 /* Safe even if %.*s bogusly expects terminating \0 because
850 we've zero-terminated dlbuf above. */
851 DEBUGP (("%.*s", ret, dlbuf));
854 DEBUGP (("] done.\n"));
859 extract_param_value_delim (const char *begin, const char *end,
860 const char *param_name, char **param_value)
868 assert (param_value);
870 len = strlen (param_name);
872 /* skip initial whitespaces */
874 while (*p && ISSPACE (*p) && p < end) ++p;
877 && 0 == strncasecmp (p, param_name, len))
881 /* skip white spaces, equal sign and inital quote */
883 while (*p && (ISSPACE (*p) || *p == '\"' || *p == '=') && p < end) ++p;
885 /* find last quote */
887 while (*e && *e != '\"' && e < end) ++e;
889 *param_value = strdupdelim (p, e);
898 /* A token consists of characters in the [b, e) range. */
902 /* Extract a parameter from the HTTP header at *SOURCE and advance
903 *SOURCE to the next parameter. Return false when there are no more
904 parameters to extract. The name of the parameter is returned in
905 NAME, and the value in VALUE. If the parameter has no value, the
906 token's value is zeroed out.
908 For example, if *SOURCE points to the string "attachment;
909 filename=\"foo bar\"", the first call to this function will return
910 the token named "attachment" and no value, and the second call will
911 return the token named "filename" and value "foo bar". The third
912 call will return false, indicating no more valid tokens. */
915 extract_param (const char **source, param_token *name, param_token *value)
917 const char *p = *source;
919 while (ISSPACE (*p)) ++p;
921 return false; /* nothing more to extract */
925 while (*p && !ISSPACE (*p) && *p != '=' && *p != ';') ++p;
927 while (ISSPACE (*p)) ++p;
928 if (*p == ';' || !*p) /* no value */
936 return false; /* error */
938 /* *p is '=', extract value */
940 while (ISSPACE (*p)) ++p;
941 if (*p == '"') /* quoted */
944 while (*p && *p != '"') ++p;
948 /* Currently at closing quote; find the end of param. */
949 while (ISSPACE (*p)) ++p;
950 while (*p && *p != ';') ++p;
954 /* garbage after closed quote, e.g. foo="bar"baz */
960 while (*p && *p != ';') ++p;
962 while (value->e != value->b && ISSPACE (value->e[-1]))
971 #define MAX(p, q) ((p) > (q) ? (p) : (q))
974 parse_content_disposition (const char *hdr, char **filename)
976 param_token name, value;
977 while (extract_param (&hdr, &name, &value))
978 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
980 /* Make the file name begin at the last slash or backslash. */
981 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
982 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
983 if (last_slash && last_bs)
984 value.b = 1 + MAX (last_slash, last_bs);
985 else if (last_slash || last_bs)
986 value.b = 1 + (last_slash ? last_slash : last_bs);
987 if (value.b == value.e)
989 *filename = strdupdelim (value.b, value.e);
995 /* Persistent connections. Currently, we cache the most recently used
996 connection as persistent, provided that the HTTP server agrees to
997 make it such. The persistence data is stored in the variables
998 below. Ideally, it should be possible to cache an arbitrary fixed
999 number of these connections. */
1001 /* Whether a persistent connection is active. */
1002 static bool pconn_active;
1005 /* The socket of the connection. */
1008 /* Host and port of the currently active persistent connection. */
1012 /* Whether a ssl handshake has occoured on this connection. */
1015 /* Whether the connection was authorized. This is only done by
1016 NTLM, which authorizes *connections* rather than individual
1017 requests. (That practice is peculiar for HTTP, but it is a
1018 useful optimization.) */
1022 /* NTLM data of the current connection. */
1023 struct ntlmdata ntlm;
1027 /* Mark the persistent connection as invalid and free the resources it
1028 uses. This is used by the CLOSE_* macros after they forcefully
1029 close a registered persistent connection. */
1032 invalidate_persistent (void)
1034 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1035 pconn_active = false;
1036 fd_close (pconn.socket);
1041 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1042 persistent. This will enable someone to use the same connection
1043 later. In the context of HTTP, this must be called only AFTER the
1044 response has been received and the server has promised that the
1045 connection will remain alive.
1047 If a previous connection was persistent, it is closed. */
1050 register_persistent (const char *host, int port, int fd, bool ssl)
1054 if (pconn.socket == fd)
1056 /* The connection FD is already registered. */
1061 /* The old persistent connection is still active; close it
1062 first. This situation arises whenever a persistent
1063 connection exists, but we then connect to a different
1064 host, and try to register a persistent connection to that
1066 invalidate_persistent ();
1070 pconn_active = true;
1072 pconn.host = xstrdup (host);
1075 pconn.authorized = false;
1077 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1080 /* Return true if a persistent connection is available for connecting
1084 persistent_available_p (const char *host, int port, bool ssl,
1085 bool *host_lookup_failed)
1087 /* First, check whether a persistent connection is active at all. */
1091 /* If we want SSL and the last connection wasn't or vice versa,
1092 don't use it. Checking for host and port is not enough because
1093 HTTP and HTTPS can apparently coexist on the same port. */
1094 if (ssl != pconn.ssl)
1097 /* If we're not connecting to the same port, we're not interested. */
1098 if (port != pconn.port)
1101 /* If the host is the same, we're in business. If not, there is
1102 still hope -- read below. */
1103 if (0 != strcasecmp (host, pconn.host))
1105 /* Check if pconn.socket is talking to HOST under another name.
1106 This happens often when both sites are virtual hosts
1107 distinguished only by name and served by the same network
1108 interface, and hence the same web server (possibly set up by
1109 the ISP and serving many different web sites). This
1110 admittedly unconventional optimization does not contradict
1111 HTTP and works well with popular server software. */
1115 struct address_list *al;
1118 /* Don't try to talk to two different SSL sites over the same
1119 secure connection! (Besides, it's not clear that
1120 name-based virtual hosting is even possible with SSL.) */
1123 /* If pconn.socket's peer is one of the IP addresses HOST
1124 resolves to, pconn.socket is for all intents and purposes
1125 already talking to HOST. */
1127 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1129 /* Can't get the peer's address -- something must be very
1130 wrong with the connection. */
1131 invalidate_persistent ();
1134 al = lookup_host (host, 0);
1137 *host_lookup_failed = true;
1141 found = address_list_contains (al, &ip);
1142 address_list_release (al);
1147 /* The persistent connection's peer address was found among the
1148 addresses HOST resolved to; therefore, pconn.sock is in fact
1149 already talking to HOST -- no need to reconnect. */
1152 /* Finally, check whether the connection is still open. This is
1153 important because most servers implement liberal (short) timeout
1154 on persistent connections. Wget can of course always reconnect
1155 if the connection doesn't work out, but it's nicer to know in
1156 advance. This test is a logical followup of the first test, but
1157 is "expensive" and therefore placed at the end of the list.
1159 (Current implementation of test_socket_open has a nice side
1160 effect that it treats sockets with pending data as "closed".
1161 This is exactly what we want: if a broken server sends message
1162 body in response to HEAD, or if it sends more than conent-length
1163 data, we won't reuse the corrupted connection.) */
1165 if (!test_socket_open (pconn.socket))
1167 /* Oops, the socket is no longer open. Now that we know that,
1168 let's invalidate the persistent connection before returning
1170 invalidate_persistent ();
1177 /* The idea behind these two CLOSE macros is to distinguish between
1178 two cases: one when the job we've been doing is finished, and we
1179 want to close the connection and leave, and two when something is
1180 seriously wrong and we're closing the connection as part of
1183 In case of keep_alive, CLOSE_FINISH should leave the connection
1184 open, while CLOSE_INVALIDATE should still close it.
1186 Note that the semantics of the flag `keep_alive' is "this
1187 connection *will* be reused (the server has promised not to close
1188 the connection once we're done)", while the semantics of
1189 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1190 active, registered connection". */
1192 #define CLOSE_FINISH(fd) do { \
1195 if (pconn_active && (fd) == pconn.socket) \
1196 invalidate_persistent (); \
1205 #define CLOSE_INVALIDATE(fd) do { \
1206 if (pconn_active && (fd) == pconn.socket) \
1207 invalidate_persistent (); \
1215 wgint len; /* received length */
1216 wgint contlen; /* expected length */
1217 wgint restval; /* the restart value */
1218 int res; /* the result of last read */
1219 char *rderrmsg; /* error message from read error */
1220 char *newloc; /* new location (redirection) */
1221 char *remote_time; /* remote time-stamp string */
1222 char *error; /* textual HTTP error */
1223 int statcode; /* status code */
1224 wgint rd_size; /* amount of data read from socket */
1225 double dltime; /* time it took to download the data */
1226 const char *referer; /* value of the referer header. */
1227 char *local_file; /* local file name. */
1228 bool timestamp_checked; /* true if pre-download time-stamping checks
1229 * have already been performed */
1230 char *orig_file_name; /* name of file to compare for time-stamping
1231 * (might be != local_file if -K is set) */
1232 wgint orig_file_size; /* size of file to compare for time-stamping */
1233 time_t orig_file_tstamp; /* time-stamp of file to compare for
1238 free_hstat (struct http_stat *hs)
1240 xfree_null (hs->newloc);
1241 xfree_null (hs->remote_time);
1242 xfree_null (hs->error);
1243 xfree_null (hs->rderrmsg);
1244 xfree_null (hs->local_file);
1245 xfree_null (hs->orig_file_name);
1247 /* Guard against being called twice. */
1249 hs->remote_time = NULL;
1253 static char *create_authorization_line (const char *, const char *,
1254 const char *, const char *,
1255 const char *, bool *);
1256 static char *basic_authentication_encode (const char *, const char *);
1257 static bool known_authentication_scheme_p (const char *, const char *);
1258 static void load_cookies (void);
1260 #define BEGINS_WITH(line, string_constant) \
1261 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1262 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1263 || !line[sizeof (string_constant) - 1]))
1265 #define SET_USER_AGENT(req) do { \
1266 if (!opt.useragent) \
1267 request_set_header (req, "User-Agent", \
1268 aprintf ("Wget/%s", version_string), rel_value); \
1269 else if (*opt.useragent) \
1270 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1273 /* The flags that allow clobbering the file (opening with "wb").
1274 Defined here to avoid repetition later. #### This will require
1276 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1277 || opt.dirstruct || opt.output_document)
1279 /* Retrieve a document through HTTP protocol. It recognizes status
1280 code, and correctly handles redirections. It closes the network
1281 socket. If it receives an error from the functions below it, it
1282 will print it if there is enough information to do so (almost
1283 always), returning the error to the caller (i.e. http_loop).
1285 Various HTTP parameters are stored to hs.
1287 If PROXY is non-NULL, the connection will be made to the proxy
1288 server, and u->url will be requested. */
1290 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1292 struct request *req;
1295 char *user, *passwd;
1299 wgint contlen, contrange;
1306 /* Set to 1 when the authorization has failed permanently and should
1307 not be tried again. */
1308 bool auth_finished = false;
1310 /* Whether NTLM authentication is used for this request. */
1311 bool ntlm_seen = false;
1313 /* Whether our connection to the remote host is through SSL. */
1314 bool using_ssl = false;
1316 /* Whether a HEAD request will be issued (as opposed to GET or
1318 bool head_only = !!(*dt & HEAD_ONLY);
1321 struct response *resp;
1325 /* Whether this connection will be kept alive after the HTTP request
1329 /* Whether keep-alive should be inhibited.
1331 RFC 2068 requests that 1.0 clients not send keep-alive requests
1332 to proxies. This is because many 1.0 proxies do not interpret
1333 the Connection header and transfer it to the remote server,
1334 causing it to not close the connection and leave both the proxy
1335 and the client hanging. */
1336 bool inhibit_keep_alive =
1337 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1339 /* Headers sent when using POST. */
1340 wgint post_data_size = 0;
1342 bool host_lookup_failed = false;
1345 if (u->scheme == SCHEME_HTTPS)
1347 /* Initialize the SSL context. After this has once been done,
1348 it becomes a no-op. */
1351 scheme_disable (SCHEME_HTTPS);
1352 logprintf (LOG_NOTQUIET,
1353 _("Disabling SSL due to encountered errors.\n"));
1354 return SSLINITFAILED;
1357 #endif /* HAVE_SSL */
1359 /* Initialize certain elements of struct http_stat. */
1363 hs->rderrmsg = NULL;
1365 hs->remote_time = NULL;
1370 /* Prepare the request to send. */
1372 req = request_new ();
1375 const char *meth = "GET";
1378 else if (opt.post_file_name || opt.post_data)
1380 /* Use the full path, i.e. one that includes the leading slash and
1381 the query string. E.g. if u->path is "foo/bar" and u->query is
1382 "param=value", full_path will be "/foo/bar?param=value". */
1385 /* When using SSL over proxy, CONNECT establishes a direct
1386 connection to the HTTPS server. Therefore use the same
1387 argument as when talking to the server directly. */
1388 && u->scheme != SCHEME_HTTPS
1391 meth_arg = xstrdup (u->url);
1393 meth_arg = url_full_path (u);
1394 request_set_method (req, meth, meth_arg);
1397 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1398 if (*dt & SEND_NOCACHE)
1399 request_set_header (req, "Pragma", "no-cache", rel_none);
1401 request_set_header (req, "Range",
1402 aprintf ("bytes=%s-",
1403 number_to_static_string (hs->restval)),
1405 SET_USER_AGENT (req);
1406 request_set_header (req, "Accept", "*/*", rel_none);
1408 /* Find the username and password for authentication. */
1411 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1412 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1413 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1417 /* We have the username and the password, but haven't tried
1418 any authorization yet. Let's see if the "Basic" method
1419 works. If not, we'll come back here and construct a
1420 proper authorization method with the right challenges.
1422 If we didn't employ this kind of logic, every URL that
1423 requires authorization would have to be processed twice,
1424 which is very suboptimal and generates a bunch of false
1425 "unauthorized" errors in the server log.
1427 #### But this logic also has a serious problem when used
1428 with stronger authentications: we *first* transmit the
1429 username and the password in clear text, and *then* attempt a
1430 stronger authentication scheme. That cannot be right! We
1431 are only fortunate that almost everyone still uses the
1432 `Basic' scheme anyway.
1434 There should be an option to prevent this from happening, for
1435 those who use strong authentication schemes and value their
1437 request_set_header (req, "Authorization",
1438 basic_authentication_encode (user, passwd),
1445 char *proxy_user, *proxy_passwd;
1446 /* For normal username and password, URL components override
1447 command-line/wgetrc parameters. With proxy
1448 authentication, it's the reverse, because proxy URLs are
1449 normally the "permanent" ones, so command-line args
1450 should take precedence. */
1451 if (opt.proxy_user && opt.proxy_passwd)
1453 proxy_user = opt.proxy_user;
1454 proxy_passwd = opt.proxy_passwd;
1458 proxy_user = proxy->user;
1459 proxy_passwd = proxy->passwd;
1461 /* #### This does not appear right. Can't the proxy request,
1462 say, `Digest' authentication? */
1463 if (proxy_user && proxy_passwd)
1464 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1466 /* If we're using a proxy, we will be connecting to the proxy
1470 /* Proxy authorization over SSL is handled below. */
1472 if (u->scheme != SCHEME_HTTPS)
1474 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1477 /* Generate the Host header, HOST:PORT. Take into account that:
1479 - Broken server-side software often doesn't recognize the PORT
1480 argument, so we must generate "Host: www.server.com" instead of
1481 "Host: www.server.com:80" (and likewise for https port).
1483 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1484 becomes ambiguous and needs to be rewritten as "Host:
1485 [3ffe:8100:200:2::2]:1234". */
1487 /* Formats arranged for hfmt[add_port][add_squares]. */
1488 static const char *hfmt[][2] = {
1489 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1491 int add_port = u->port != scheme_default_port (u->scheme);
1492 int add_squares = strchr (u->host, ':') != NULL;
1493 request_set_header (req, "Host",
1494 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1498 if (!inhibit_keep_alive)
1499 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1502 request_set_header (req, "Cookie",
1503 cookie_header (wget_cookie_jar,
1504 u->host, u->port, u->path,
1506 u->scheme == SCHEME_HTTPS
1513 if (opt.post_data || opt.post_file_name)
1515 request_set_header (req, "Content-Type",
1516 "application/x-www-form-urlencoded", rel_none);
1518 post_data_size = strlen (opt.post_data);
1521 post_data_size = file_size (opt.post_file_name);
1522 if (post_data_size == -1)
1524 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1525 opt.post_file_name, strerror (errno));
1529 request_set_header (req, "Content-Length",
1530 xstrdup (number_to_static_string (post_data_size)),
1534 /* Add the user headers. */
1535 if (opt.user_headers)
1538 for (i = 0; opt.user_headers[i]; i++)
1539 request_set_user_header (req, opt.user_headers[i]);
1543 /* We need to come back here when the initial attempt to retrieve
1544 without authorization header fails. (Expected to happen at least
1545 for the Digest authorization scheme.) */
1549 /* Establish the connection. */
1551 if (!inhibit_keep_alive)
1553 /* Look for a persistent connection to target host, unless a
1554 proxy is used. The exception is when SSL is in use, in which
1555 case the proxy is nothing but a passthrough to the target
1556 host, registered as a connection to the latter. */
1557 struct url *relevant = conn;
1559 if (u->scheme == SCHEME_HTTPS)
1563 if (persistent_available_p (relevant->host, relevant->port,
1565 relevant->scheme == SCHEME_HTTPS,
1569 &host_lookup_failed))
1571 sock = pconn.socket;
1572 using_ssl = pconn.ssl;
1573 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1574 escnonprint (pconn.host), pconn.port);
1575 DEBUGP (("Reusing fd %d.\n", sock));
1576 if (pconn.authorized)
1577 /* If the connection is already authorized, the "Basic"
1578 authorization added by code above is unnecessary and
1580 request_remove_header (req, "Authorization");
1586 /* In its current implementation, persistent_available_p will
1587 look up conn->host in some cases. If that lookup failed, we
1588 don't need to bother with connect_to_host. */
1589 if (host_lookup_failed)
1595 sock = connect_to_host (conn->host, conn->port);
1604 return (retryable_socket_connect_error (errno)
1605 ? CONERROR : CONIMPOSSIBLE);
1609 if (proxy && u->scheme == SCHEME_HTTPS)
1611 /* When requesting SSL URLs through proxies, use the
1612 CONNECT method to request passthrough. */
1613 struct request *connreq = request_new ();
1614 request_set_method (connreq, "CONNECT",
1615 aprintf ("%s:%d", u->host, u->port));
1616 SET_USER_AGENT (connreq);
1619 request_set_header (connreq, "Proxy-Authorization",
1620 proxyauth, rel_value);
1621 /* Now that PROXYAUTH is part of the CONNECT request,
1622 zero it out so we don't send proxy authorization with
1623 the regular request below. */
1626 /* Examples in rfc2817 use the Host header in CONNECT
1627 requests. I don't see how that gains anything, given
1628 that the contents of Host would be exactly the same as
1629 the contents of CONNECT. */
1631 write_error = request_send (connreq, sock);
1632 request_free (connreq);
1633 if (write_error < 0)
1635 CLOSE_INVALIDATE (sock);
1639 head = read_http_response_head (sock);
1642 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1644 CLOSE_INVALIDATE (sock);
1653 DEBUGP (("proxy responded with: [%s]\n", head));
1655 resp = resp_new (head);
1656 statcode = resp_status (resp, &message);
1659 if (statcode != 200)
1662 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1663 message ? escnonprint (message) : "?");
1664 xfree_null (message);
1667 xfree_null (message);
1669 /* SOCK is now *really* connected to u->host, so update CONN
1670 to reflect this. That way register_persistent will
1671 register SOCK as being connected to u->host:u->port. */
1675 if (conn->scheme == SCHEME_HTTPS)
1677 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1684 #endif /* HAVE_SSL */
1687 /* Send the request to server. */
1688 write_error = request_send (req, sock);
1690 if (write_error >= 0)
1694 DEBUGP (("[POST data: %s]\n", opt.post_data));
1695 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1697 else if (opt.post_file_name && post_data_size != 0)
1698 write_error = post_file (sock, opt.post_file_name, post_data_size);
1701 if (write_error < 0)
1703 CLOSE_INVALIDATE (sock);
1707 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1708 proxy ? "Proxy" : "HTTP");
1713 head = read_http_response_head (sock);
1718 logputs (LOG_NOTQUIET, _("No data received.\n"));
1719 CLOSE_INVALIDATE (sock);
1725 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1727 CLOSE_INVALIDATE (sock);
1732 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1734 resp = resp_new (head);
1736 /* Check for status line. */
1738 statcode = resp_status (resp, &message);
1739 if (!opt.server_response)
1740 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1741 message ? escnonprint (message) : "");
1744 logprintf (LOG_VERBOSE, "\n");
1745 print_server_response (resp, " ");
1748 /* Determine the local filename if needed. Notice that if -O is used
1749 * hstat.local_file is set by http_loop to the argument of -O. */
1750 if (!hs->local_file)
1752 /* Honor Content-Disposition whether possible. */
1753 if (!resp_header_copy (resp, "Content-Disposition", hdrval, sizeof (hdrval))
1754 || !parse_content_disposition (hdrval, &hs->local_file))
1756 /* Choose filename according to URL name. */
1757 hs->local_file = url_file_name (u);
1761 /* TODO: perform this check only once. */
1762 if (opt.noclobber && file_exists_p (hs->local_file))
1764 /* If opt.noclobber is turned on and file already exists, do not
1765 retrieve the file */
1766 logprintf (LOG_VERBOSE, _("\
1767 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1768 /* If the file is there, we suppose it's retrieved OK. */
1771 /* #### Bogusness alert. */
1772 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1773 if (has_html_suffix_p (hs->local_file))
1779 /* Support timestamping */
1780 /* TODO: move this code out of gethttp. */
1781 if (opt.timestamping && !hs->timestamp_checked)
1783 size_t filename_len = strlen (hs->local_file);
1784 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1785 bool local_dot_orig_file_exists = false;
1786 char *local_filename = NULL;
1789 if (opt.backup_converted)
1790 /* If -K is specified, we'll act on the assumption that it was specified
1791 last time these files were downloaded as well, and instead of just
1792 comparing local file X against server file X, we'll compare local
1793 file X.orig (if extant, else X) against server file X. If -K
1794 _wasn't_ specified last time, or the server contains files called
1795 *.orig, -N will be back to not operating correctly with -k. */
1797 /* Would a single s[n]printf() call be faster? --dan
1799 Definitely not. sprintf() is horribly slow. It's a
1800 different question whether the difference between the two
1801 affects a program. Usually I'd say "no", but at one
1802 point I profiled Wget, and found that a measurable and
1803 non-negligible amount of time was lost calling sprintf()
1804 in url.c. Replacing sprintf with inline calls to
1805 strcpy() and number_to_string() made a difference.
1807 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1808 memcpy (filename_plus_orig_suffix + filename_len,
1809 ".orig", sizeof (".orig"));
1811 /* Try to stat() the .orig file. */
1812 if (stat (filename_plus_orig_suffix, &st) == 0)
1814 local_dot_orig_file_exists = 1;
1815 local_filename = filename_plus_orig_suffix;
1819 if (!local_dot_orig_file_exists)
1820 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1821 if (stat (hs->local_file, &st) == 0)
1822 local_filename = hs->local_file;
1824 if (local_filename != NULL)
1825 /* There was a local file, so we'll check later to see if the version
1826 the server has is the same version we already have, allowing us to
1829 hs->orig_file_name = xstrdup (local_filename);
1830 hs->orig_file_size = st.st_size;
1831 hs->orig_file_tstamp = st.st_mtime;
1833 /* Modification time granularity is 2 seconds for Windows, so
1834 increase local time by 1 second for later comparison. */
1835 ++hs->orig_file_tstamp;
1840 if (!opt.ignore_length
1841 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1845 parsed = str_to_wgint (hdrval, NULL, 10);
1846 if (parsed == WGINT_MAX && errno == ERANGE)
1848 #### If Content-Length is out of range, it most likely
1849 means that the file is larger than 2G and that we're
1850 compiled without LFS. In that case we should probably
1851 refuse to even attempt to download the file. */
1857 /* Check for keep-alive related responses. */
1858 if (!inhibit_keep_alive && contlen != -1)
1860 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1862 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1864 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1869 /* The server has promised that it will not close the connection
1870 when we're done. This means that we can register it. */
1871 register_persistent (conn->host, conn->port, sock, using_ssl);
1873 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1875 /* Authorization is required. */
1876 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1877 CLOSE_FINISH (sock);
1879 CLOSE_INVALIDATE (sock);
1880 pconn.authorized = false;
1881 if (!auth_finished && (user && passwd))
1883 /* IIS sends multiple copies of WWW-Authenticate, one with
1884 the value "negotiate", and other(s) with data. Loop over
1885 all the occurrences and pick the one we recognize. */
1887 const char *wabeg, *waend;
1888 char *www_authenticate = NULL;
1890 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1891 &wabeg, &waend)) != -1;
1893 if (known_authentication_scheme_p (wabeg, waend))
1895 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1899 if (!www_authenticate)
1900 /* If the authentication header is missing or
1901 unrecognized, there's no sense in retrying. */
1902 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1903 else if (BEGINS_WITH (www_authenticate, "Basic"))
1904 /* If the authentication scheme is "Basic", which we send
1905 by default, there's no sense in retrying either. (This
1906 should be changed when we stop sending "Basic" data by
1912 pth = url_full_path (u);
1913 request_set_header (req, "Authorization",
1914 create_authorization_line (www_authenticate,
1916 request_method (req),
1920 if (BEGINS_WITH (www_authenticate, "NTLM"))
1923 goto retry_with_auth;
1926 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1930 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1932 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1934 pconn.authorized = true;
1938 hs->statcode = statcode;
1940 hs->error = xstrdup (_("Malformed status line"));
1942 hs->error = xstrdup (_("(no description)"));
1944 hs->error = xstrdup (message);
1945 xfree_null (message);
1947 type = resp_header_strdup (resp, "Content-Type");
1950 char *tmp = strchr (type, ';');
1953 while (tmp > type && ISSPACE (tmp[-1]))
1958 hs->newloc = resp_header_strdup (resp, "Location");
1959 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1961 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1965 const char *scbeg, *scend;
1966 /* The jar should have been created by now. */
1967 assert (wget_cookie_jar != NULL);
1969 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1970 &scbeg, &scend)) != -1;
1973 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1974 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1975 u->path, set_cookie);
1979 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1981 wgint first_byte_pos, last_byte_pos, entity_length;
1982 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1984 contrange = first_byte_pos;
1988 /* 20x responses are counted among successful by default. */
1989 if (H_20X (statcode))
1992 /* Return if redirected. */
1993 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1995 /* RFC2068 says that in case of the 300 (multiple choices)
1996 response, the server can output a preferred URL through
1997 `Location' header; otherwise, the request should be treated
1998 like GET. So, if the location is set, it will be a
1999 redirection; otherwise, just proceed normally. */
2000 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2004 logprintf (LOG_VERBOSE,
2005 _("Location: %s%s\n"),
2006 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2007 hs->newloc ? _(" [following]") : "");
2008 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2009 CLOSE_FINISH (sock);
2011 CLOSE_INVALIDATE (sock);
2017 /* If content-type is not given, assume text/html. This is because
2018 of the multitude of broken CGI's that "forget" to generate the
2021 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2022 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2027 if (opt.html_extension && (*dt & TEXTHTML))
2028 /* -E / --html-extension / html_extension = on was specified, and this is a
2029 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2030 already the file's suffix, tack on ".html". */
2032 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2034 if (last_period_in_local_filename == NULL
2035 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2036 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2038 int local_filename_len = strlen (hs->local_file);
2039 /* Resize the local file, allowing for ".html" preceded by
2040 optional ".NUMBER". */
2041 hs->local_file = xrealloc (hs->local_file,
2042 local_filename_len + 24 + sizeof (".html"));
2043 strcpy(hs->local_file + local_filename_len, ".html");
2044 /* If clobbering is not allowed and the file, as named,
2045 exists, tack on ".NUMBER.html" instead. */
2046 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2050 sprintf (hs->local_file + local_filename_len,
2051 ".%d.html", ext_num++);
2052 while (file_exists_p (hs->local_file));
2054 *dt |= ADDED_HTML_EXTENSION;
2058 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2060 /* If `-c' is in use and the file has been fully downloaded (or
2061 the remote file has shrunk), Wget effectively requests bytes
2062 after the end of file and the server response with 416. */
2063 logputs (LOG_VERBOSE, _("\
2064 \n The file is already fully retrieved; nothing to do.\n\n"));
2065 /* In case the caller inspects. */
2068 /* Mark as successfully retrieved. */
2071 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2072 might be more bytes in the body. */
2073 return RETRUNNEEDED;
2075 if ((contrange != 0 && contrange != hs->restval)
2076 || (H_PARTIAL (statcode) && !contrange))
2078 /* The Range request was somehow misunderstood by the server.
2081 CLOSE_INVALIDATE (sock);
2084 hs->contlen = contlen + contrange;
2090 /* No need to print this output if the body won't be
2091 downloaded at all, or if the original server response is
2093 logputs (LOG_VERBOSE, _("Length: "));
2096 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2097 if (contlen + contrange >= 1024)
2098 logprintf (LOG_VERBOSE, " (%s)",
2099 human_readable (contlen + contrange));
2102 if (contlen >= 1024)
2103 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2104 number_to_static_string (contlen),
2105 human_readable (contlen));
2107 logprintf (LOG_VERBOSE, _(", %s remaining"),
2108 number_to_static_string (contlen));
2112 logputs (LOG_VERBOSE,
2113 opt.ignore_length ? _("ignored") : _("unspecified"));
2115 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2117 logputs (LOG_VERBOSE, "\n");
2121 type = NULL; /* We don't need it any more. */
2123 /* Return if we have no intention of further downloading. */
2124 if (!(*dt & RETROKF) || head_only)
2126 /* In case the caller cares to look... */
2131 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2132 servers not to send body in response to a HEAD request, and
2133 those that do will likely be caught by test_socket_open.
2134 If not, they can be worked around using
2135 `--no-http-keep-alive'. */
2136 CLOSE_FINISH (sock);
2137 else if (keep_alive && skip_short_body (sock, contlen))
2138 /* Successfully skipped the body; also keep using the socket. */
2139 CLOSE_FINISH (sock);
2141 CLOSE_INVALIDATE (sock);
2142 return RETRFINISHED;
2145 /* Print fetch message, if opt.verbose. */
2148 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2149 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2152 /* Open the local file. */
2155 mkalldirs (hs->local_file);
2157 rotate_backups (hs->local_file);
2159 fp = fopen (hs->local_file, "ab");
2160 else if (ALLOW_CLOBBER)
2161 fp = fopen (hs->local_file, "wb");
2164 fp = fopen_excl (hs->local_file, true);
2165 if (!fp && errno == EEXIST)
2167 /* We cannot just invent a new name and use it (which is
2168 what functions like unique_create typically do)
2169 because we told the user we'd use this name.
2170 Instead, return and retry the download. */
2171 logprintf (LOG_NOTQUIET,
2172 _("%s has sprung into existence.\n"),
2174 CLOSE_INVALIDATE (sock);
2175 return FOPEN_EXCL_ERR;
2180 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2181 CLOSE_INVALIDATE (sock);
2188 /* This confuses the timestamping code that checks for file size.
2189 #### The timestamping code should be smarter about file size. */
2190 if (opt.save_headers && hs->restval == 0)
2191 fwrite (head, 1, strlen (head), fp);
2193 /* Now we no longer need to store the response header. */
2196 /* Download the request body. */
2199 /* If content-length is present, read that much; otherwise, read
2200 until EOF. The HTTP spec doesn't require the server to
2201 actually close the connection when it's done sending data. */
2202 flags |= rb_read_exactly;
2203 if (hs->restval > 0 && contrange == 0)
2204 /* If the server ignored our range request, instruct fd_read_body
2205 to skip the first RESTVAL bytes of body. */
2206 flags |= rb_skip_startpos;
2207 hs->len = hs->restval;
2209 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2210 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2214 CLOSE_FINISH (sock);
2218 hs->rderrmsg = xstrdup (fd_errstr (sock));
2219 CLOSE_INVALIDATE (sock);
2226 return RETRFINISHED;
2229 /* The genuine HTTP loop! This is the part where the retrieval is
2230 retried, and retried, and retried, and... */
2232 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2233 int *dt, struct url *proxy)
2236 bool got_head = false; /* used for time-stamping */
2240 time_t tmr = -1; /* remote time-stamp */
2241 wgint local_size = 0; /* the size of the local file */
2242 struct http_stat hstat; /* HTTP status */
2245 /* Assert that no value for *LOCAL_FILE was passed. */
2246 assert (local_file == NULL || *local_file == NULL);
2248 /* Set LOCAL_FILE parameter. */
2249 if (local_file && opt.output_document)
2250 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2252 /* Reset NEWLOC parameter. */
2255 /* This used to be done in main(), but it's a better idea to do it
2256 here so that we don't go through the hoops if we're just using
2261 /* Warn on (likely bogus) wildcard usage in HTTP. */
2262 if (opt.ftp_glob && has_wildcards_p (u->path))
2263 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2265 /* Setup hstat struct. */
2267 hstat.referer = referer;
2269 if (opt.output_document)
2270 hstat.local_file = xstrdup (opt.output_document);
2272 /* Reset the counter. */
2275 /* Reset the document type. */
2281 /* Increment the pass counter. */
2283 sleep_between_retrievals (count);
2285 /* Get the current time string. */
2286 tms = time_str (NULL);
2288 /* Print fetch message, if opt.verbose. */
2291 char *hurl = url_string (u, true);
2296 sprintf (tmp, _("(try:%2d)"), count);
2297 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2302 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2307 ws_changetitle (hurl);
2312 /* Default document type is empty. However, if spider mode is
2313 on or time-stamping is employed, HEAD_ONLY commands is
2314 encoded within *dt. */
2315 if (opt.spider || (opt.timestamping && !got_head))
2320 /* Decide whether or not to restart. */
2322 && stat (hstat.local_file, &st) == 0
2323 && S_ISREG (st.st_mode))
2324 /* When -c is used, continue from on-disk size. (Can't use
2325 hstat.len even if count>1 because we don't want a failed
2326 first attempt to clobber existing data.) */
2327 hstat.restval = st.st_size;
2329 /* otherwise, continue where the previous try left off */
2330 hstat.restval = hstat.len;
2334 /* Decide whether to send the no-cache directive. We send it in
2336 a) we're using a proxy, and we're past our first retrieval.
2337 Some proxies are notorious for caching incomplete data, so
2338 we require a fresh get.
2339 b) caching is explicitly inhibited. */
2340 if ((proxy && count > 1) /* a */
2341 || !opt.allow_cache /* b */
2343 *dt |= SEND_NOCACHE;
2345 *dt &= ~SEND_NOCACHE;
2347 /* Try fetching the document, or at least its head. */
2348 err = gethttp (u, &hstat, dt, proxy);
2351 tms = time_str (NULL);
2353 /* Get the new location (with or without the redirection). */
2355 *newloc = xstrdup (hstat.newloc);
2359 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2360 case CONERROR: case READERR: case WRITEFAILED:
2361 case RANGEERR: case FOPEN_EXCL_ERR:
2362 /* Non-fatal errors continue executing the loop, which will
2363 bring them to "while" statement at the end, to judge
2364 whether the number of tries was exceeded. */
2365 /* free_hstat (&hstat); */
2366 printwhat (count, opt.ntry);
2368 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2369 case SSLINITFAILED: case CONTNOTSUPPORTED:
2370 /* Fatal errors just return from the function. */
2371 free_hstat (&hstat);
2373 case FWRITEERR: case FOPENERR:
2374 /* Another fatal error. */
2375 logputs (LOG_VERBOSE, "\n");
2376 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2377 hstat.local_file, strerror (errno));
2378 free_hstat (&hstat);
2381 /* Another fatal error. */
2382 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2383 free_hstat (&hstat);
2386 /* Return the new location to the caller. */
2389 logprintf (LOG_NOTQUIET,
2390 _("ERROR: Redirection (%d) without location.\n"),
2392 free_hstat (&hstat);
2395 free_hstat (&hstat);
2398 /* The file was already fully retrieved. */
2399 free_hstat (&hstat);
2402 /* Deal with you later. */
2405 /* All possibilities should have been exhausted. */
2409 if (!(*dt & RETROKF))
2413 /* #### Ugly ugly ugly! */
2414 char *hurl = url_string (u, true);
2415 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2418 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2419 tms, hstat.statcode, escnonprint (hstat.error));
2420 logputs (LOG_VERBOSE, "\n");
2421 free_hstat (&hstat);
2425 /* Did we get the time-stamp? */
2428 if (opt.timestamping && !hstat.remote_time)
2430 logputs (LOG_NOTQUIET, _("\
2431 Last-modified header missing -- time-stamps turned off.\n"));
2433 else if (hstat.remote_time)
2435 /* Convert the date-string into struct tm. */
2436 tmr = http_atotm (hstat.remote_time);
2437 if (tmr == (time_t) (-1))
2438 logputs (LOG_VERBOSE, _("\
2439 Last-modified header invalid -- time-stamp ignored.\n"));
2443 /* The time-stamping section. */
2444 if (opt.timestamping && !got_head)
2446 got_head = true; /* no more time-stamping */
2448 count = 0; /* the retrieve count for HEAD is reset */
2450 if (hstat.remote_time && tmr != (time_t) (-1))
2452 /* Now time-stamping can be used validly. Time-stamping
2453 means that if the sizes of the local and remote file
2454 match, and local file is newer than the remote file,
2455 it will not be retrieved. Otherwise, the normal
2456 download procedure is resumed. */
2457 if (hstat.orig_file_tstamp >= tmr)
2459 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2461 logprintf (LOG_VERBOSE, _("\
2462 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2463 hstat.orig_file_name);
2464 free_hstat (&hstat);
2469 logprintf (LOG_VERBOSE, _("\
2470 The sizes do not match (local %s) -- retrieving.\n"),
2471 number_to_static_string (local_size));
2475 logputs (LOG_VERBOSE,
2476 _("Remote file is newer, retrieving.\n"));
2478 logputs (LOG_VERBOSE, "\n");
2481 /* free_hstat (&hstat); */
2482 hstat.timestamp_checked = true;
2486 if ((tmr != (time_t) (-1))
2488 && ((hstat.len == hstat.contlen) ||
2489 ((hstat.res == 0) && (hstat.contlen == -1))))
2491 /* #### This code repeats in http.c and ftp.c. Move it to a
2493 const char *fl = NULL;
2494 if (opt.output_document)
2496 if (output_stream_regular)
2497 fl = opt.output_document;
2500 fl = hstat.local_file;
2504 /* End of time-stamping section. */
2508 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2509 escnonprint (hstat.error));
2513 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2514 total_download_time += hstat.dltime;
2516 if (hstat.len == hstat.contlen)
2520 logprintf (LOG_VERBOSE,
2521 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2522 tms, tmrate, hstat.local_file,
2523 number_to_static_string (hstat.len),
2524 number_to_static_string (hstat.contlen));
2525 logprintf (LOG_NONVERBOSE,
2526 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2528 number_to_static_string (hstat.len),
2529 number_to_static_string (hstat.contlen),
2530 hstat.local_file, count);
2533 total_downloaded_bytes += hstat.len;
2535 /* Remember that we downloaded the file for later ".orig" code. */
2536 if (*dt & ADDED_HTML_EXTENSION)
2537 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2539 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2541 free_hstat (&hstat);
2544 else if (hstat.res == 0) /* No read error */
2546 if (hstat.contlen == -1) /* We don't know how much we were supposed
2547 to get, so assume we succeeded. */
2551 logprintf (LOG_VERBOSE,
2552 _("%s (%s) - `%s' saved [%s]\n\n"),
2553 tms, tmrate, hstat.local_file,
2554 number_to_static_string (hstat.len));
2555 logprintf (LOG_NONVERBOSE,
2556 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2557 tms, u->url, number_to_static_string (hstat.len),
2558 hstat.local_file, count);
2561 total_downloaded_bytes += hstat.len;
2563 /* Remember that we downloaded the file for later ".orig" code. */
2564 if (*dt & ADDED_HTML_EXTENSION)
2565 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2567 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2569 free_hstat (&hstat);
2572 else if (hstat.len < hstat.contlen) /* meaning we lost the
2573 connection too soon */
2575 logprintf (LOG_VERBOSE,
2576 _("%s (%s) - Connection closed at byte %s. "),
2577 tms, tmrate, number_to_static_string (hstat.len));
2578 printwhat (count, opt.ntry);
2579 /* free_hstat (&hstat); */
2583 /* Getting here would mean reading more data than
2584 requested with content-length, which we never do. */
2587 else /* from now on hstat.res can only be -1 */
2589 if (hstat.contlen == -1)
2591 logprintf (LOG_VERBOSE,
2592 _("%s (%s) - Read error at byte %s (%s)."),
2593 tms, tmrate, number_to_static_string (hstat.len),
2595 printwhat (count, opt.ntry);
2596 /* free_hstat (&hstat); */
2599 else /* hstat.res == -1 and contlen is given */
2601 logprintf (LOG_VERBOSE,
2602 _("%s (%s) - Read error at byte %s/%s (%s). "),
2604 number_to_static_string (hstat.len),
2605 number_to_static_string (hstat.contlen),
2607 printwhat (count, opt.ntry);
2608 /* free_hstat (&hstat); */
2614 while (!opt.ntry || (count < opt.ntry));
2619 /* Check whether the result of strptime() indicates success.
2620 strptime() returns the pointer to how far it got to in the string.
2621 The processing has been successful if the string is at `GMT' or
2622 `+X', or at the end of the string.
2624 In extended regexp parlance, the function returns 1 if P matches
2625 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2626 can return) is considered a failure and 0 is returned. */
2628 check_end (const char *p)
2632 while (ISSPACE (*p))
2635 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2636 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2642 /* Convert the textual specification of time in TIME_STRING to the
2643 number of seconds since the Epoch.
2645 TIME_STRING can be in any of the three formats RFC2616 allows the
2646 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2647 as well as the time format used in the Set-Cookie header.
2648 Timezones are ignored, and should be GMT.
2650 Return the computed time_t representation, or -1 if the conversion
2653 This function uses strptime with various string formats for parsing
2654 TIME_STRING. This results in a parser that is not as lenient in
2655 interpreting TIME_STRING as I would like it to be. Being based on
2656 strptime, it always allows shortened months, one-digit days, etc.,
2657 but due to the multitude of formats in which time can be
2658 represented, an ideal HTTP time parser would be even more
2659 forgiving. It should completely ignore things like week days and
2660 concentrate only on the various forms of representing years,
2661 months, days, hours, minutes, and seconds. For example, it would
2662 be nice if it accepted ISO 8601 out of the box.
2664 I've investigated free and PD code for this purpose, but none was
2665 usable. getdate was big and unwieldy, and had potential copyright
2666 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2667 distributed with phttpd, is excellent, but we cannot use it because
2668 it is not assigned to the FSF. So I stuck it with strptime. */
2671 http_atotm (const char *time_string)
2673 /* NOTE: Solaris strptime man page claims that %n and %t match white
2674 space, but that's not universally available. Instead, we simply
2675 use ` ' to mean "skip all WS", which works under all strptime
2676 implementations I've tested. */
2678 static const char *time_formats[] = {
2679 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2680 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2681 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2682 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2683 (used in Set-Cookie, defined in the
2684 Netscape cookie specification.) */
2686 const char *oldlocale;
2688 time_t ret = (time_t) -1;
2690 /* Solaris strptime fails to recognize English month names in
2691 non-English locales, which we work around by temporarily setting
2692 locale to C before invoking strptime. */
2693 oldlocale = setlocale (LC_TIME, NULL);
2694 setlocale (LC_TIME, "C");
2696 for (i = 0; i < countof (time_formats); i++)
2700 /* Some versions of strptime use the existing contents of struct
2701 tm to recalculate the date according to format. Zero it out
2702 to prevent stack garbage from influencing strptime. */
2705 if (check_end (strptime (time_string, time_formats[i], &t)))
2712 /* Restore the previous locale. */
2713 setlocale (LC_TIME, oldlocale);
2718 /* Authorization support: We support three authorization schemes:
2720 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2722 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2723 consisting of answering to the server's challenge with the proper
2726 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2727 Stenberg for libcurl. Like digest, NTLM is based on a
2728 challenge-response mechanism, but unlike digest, it is non-standard
2729 (authenticates TCP connections rather than requests), undocumented
2730 and Microsoft-specific. */
2732 /* Create the authentication header contents for the `Basic' scheme.
2733 This is done by encoding the string "USER:PASS" to base64 and
2734 prepending the string "Basic " in front of it. */
2737 basic_authentication_encode (const char *user, const char *passwd)
2740 int len1 = strlen (user) + 1 + strlen (passwd);
2742 t1 = (char *)alloca (len1 + 1);
2743 sprintf (t1, "%s:%s", user, passwd);
2745 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2746 base64_encode (t1, len1, t2);
2748 return concat_strings ("Basic ", t2, (char *) 0);
2751 #define SKIP_WS(x) do { \
2752 while (ISSPACE (*(x))) \
2756 #ifdef ENABLE_DIGEST
2757 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2758 of a field in such a header. If the field is the one specified by
2759 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2760 digest authorization code), extract its value in the (char*)
2761 variable pointed by RET. Returns negative on a malformed header,
2762 or number of bytes that have been parsed by this call. */
2764 extract_header_attr (const char *au, const char *attr_name, char **ret)
2767 const char *cp = au;
2769 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2771 cp += strlen (attr_name);
2784 for (ep = cp; *ep && *ep != '\"'; ep++)
2789 *ret = strdupdelim (cp, ep);
2796 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2797 an array of 16 bytes containing the hash keys, and BUF should be a
2798 buffer of 33 writable characters (32 for hex digits plus one for
2799 zero termination). */
2801 dump_hash (char *buf, const unsigned char *hash)
2805 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2807 *buf++ = XNUM_TO_digit (*hash >> 4);
2808 *buf++ = XNUM_TO_digit (*hash & 0xf);
2813 /* Take the line apart to find the challenge, and compose a digest
2814 authorization header. See RFC2069 section 2.1.2. */
2816 digest_authentication_encode (const char *au, const char *user,
2817 const char *passwd, const char *method,
2820 static char *realm, *opaque, *nonce;
2825 { "realm", &realm },
2826 { "opaque", &opaque },
2831 realm = opaque = nonce = NULL;
2833 au += 6; /* skip over `Digest' */
2839 for (i = 0; i < countof (options); i++)
2841 int skip = extract_header_attr (au, options[i].name,
2842 options[i].variable);
2846 xfree_null (opaque);
2856 if (i == countof (options))
2858 while (*au && *au != '=')
2866 while (*au && *au != '\"')
2873 while (*au && *au != ',')
2878 if (!realm || !nonce || !user || !passwd || !path || !method)
2881 xfree_null (opaque);
2886 /* Calculate the digest value. */
2888 ALLOCA_MD5_CONTEXT (ctx);
2889 unsigned char hash[MD5_HASHLEN];
2890 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2891 char response_digest[MD5_HASHLEN * 2 + 1];
2893 /* A1BUF = H(user ":" realm ":" password) */
2895 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2896 gen_md5_update ((unsigned char *)":", 1, ctx);
2897 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2898 gen_md5_update ((unsigned char *)":", 1, ctx);
2899 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2900 gen_md5_finish (ctx, hash);
2901 dump_hash (a1buf, hash);
2903 /* A2BUF = H(method ":" path) */
2905 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2906 gen_md5_update ((unsigned char *)":", 1, ctx);
2907 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2908 gen_md5_finish (ctx, hash);
2909 dump_hash (a2buf, hash);
2911 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2913 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2914 gen_md5_update ((unsigned char *)":", 1, ctx);
2915 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2916 gen_md5_update ((unsigned char *)":", 1, ctx);
2917 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2918 gen_md5_finish (ctx, hash);
2919 dump_hash (response_digest, hash);
2921 res = xmalloc (strlen (user)
2926 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2927 + (opaque ? strlen (opaque) : 0)
2929 sprintf (res, "Digest \
2930 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2931 user, realm, nonce, path, response_digest);
2934 char *p = res + strlen (res);
2935 strcat (p, ", opaque=\"");
2942 #endif /* ENABLE_DIGEST */
2944 /* Computing the size of a string literal must take into account that
2945 value returned by sizeof includes the terminating \0. */
2946 #define STRSIZE(literal) (sizeof (literal) - 1)
2948 /* Whether chars in [b, e) begin with the literal string provided as
2949 first argument and are followed by whitespace or terminating \0.
2950 The comparison is case-insensitive. */
2951 #define STARTS(literal, b, e) \
2952 ((e) - (b) >= STRSIZE (literal) \
2953 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2954 && ((e) - (b) == STRSIZE (literal) \
2955 || ISSPACE (b[STRSIZE (literal)])))
2958 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2960 return STARTS ("Basic", hdrbeg, hdrend)
2961 #ifdef ENABLE_DIGEST
2962 || STARTS ("Digest", hdrbeg, hdrend)
2965 || STARTS ("NTLM", hdrbeg, hdrend)
2972 /* Create the HTTP authorization request header. When the
2973 `WWW-Authenticate' response header is seen, according to the
2974 authorization scheme specified in that header (`Basic' and `Digest'
2975 are supported by the current implementation), produce an
2976 appropriate HTTP authorization request header. */
2978 create_authorization_line (const char *au, const char *user,
2979 const char *passwd, const char *method,
2980 const char *path, bool *finished)
2982 /* We are called only with known schemes, so we can dispatch on the
2984 switch (TOUPPER (*au))
2986 case 'B': /* Basic */
2988 return basic_authentication_encode (user, passwd);
2989 #ifdef ENABLE_DIGEST
2990 case 'D': /* Digest */
2992 return digest_authentication_encode (au, user, passwd, method, path);
2995 case 'N': /* NTLM */
2996 if (!ntlm_input (&pconn.ntlm, au))
3001 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3004 /* We shouldn't get here -- this function should be only called
3005 with values approved by known_authentication_scheme_p. */
3013 if (!wget_cookie_jar)
3014 wget_cookie_jar = cookie_jar_new ();
3015 if (opt.cookies_input && !cookies_loaded_p)
3017 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3018 cookies_loaded_p = true;
3025 if (wget_cookie_jar)
3026 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3032 xfree_null (pconn.host);
3033 if (wget_cookie_jar)
3034 cookie_jar_delete (wget_cookie_jar);
3041 test_parse_content_disposition()
3049 { "filename=\"file.ext\"", "file.ext", true },
3050 { "attachment; filename=\"file.ext\"", "file.ext", true },
3051 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
3052 { "attachment", NULL, false },
3055 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3058 bool res = parse_content_disposition (test_array[i].hdrval, &filename);
3060 mu_assert ("test_parse_content_disposition: wrong result",
3061 res == test_array[i].result
3063 || 0 == strcmp (test_array[i].filename, filename)));
3065 /* printf ("test %d: %s\n", i, res == false ? "false" : filename); */
3071 #endif /* TESTING */