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"));
858 /* Extract a parameter from the HTTP header at *SOURCE and advance
859 *SOURCE to the next parameter. Return false when there are no more
860 parameters to extract. The name of the parameter is returned in
861 NAME, and the value in VALUE. If the parameter has no value, the
862 token's value is zeroed out.
864 For example, if *SOURCE points to the string "attachment;
865 filename=\"foo bar\"", the first call to this function will return
866 the token named "attachment" and no value, and the second call will
867 return the token named "filename" and value "foo bar". The third
868 call will return false, indicating no more valid tokens. */
871 extract_param (const char **source, param_token *name, param_token *value)
873 const char *p = *source;
875 while (ISSPACE (*p)) ++p;
879 return false; /* no error; nothing more to extract */
884 while (*p && !ISSPACE (*p) && *p != '=' && *p != ';') ++p;
886 if (name->b == name->e)
887 return false; /* empty name: error */
888 while (ISSPACE (*p)) ++p;
889 if (*p == ';' || !*p) /* no value */
897 return false; /* error */
899 /* *p is '=', extract value */
901 while (ISSPACE (*p)) ++p;
902 if (*p == '"') /* quoted */
905 while (*p && *p != '"') ++p;
909 /* Currently at closing quote; find the end of param. */
910 while (ISSPACE (*p)) ++p;
911 while (*p && *p != ';') ++p;
915 /* garbage after closed quote, e.g. foo="bar"baz */
921 while (*p && *p != ';') ++p;
923 while (value->e != value->b && ISSPACE (value->e[-1]))
932 #define MAX(p, q) ((p) > (q) ? (p) : (q))
935 parse_content_disposition (const char *hdr, char **filename)
937 param_token name, value;
938 while (extract_param (&hdr, &name, &value))
939 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
941 /* Make the file name begin at the last slash or backslash. */
942 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
943 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
944 if (last_slash && last_bs)
945 value.b = 1 + MAX (last_slash, last_bs);
946 else if (last_slash || last_bs)
947 value.b = 1 + (last_slash ? last_slash : last_bs);
948 if (value.b == value.e)
950 *filename = strdupdelim (value.b, value.e);
956 /* Persistent connections. Currently, we cache the most recently used
957 connection as persistent, provided that the HTTP server agrees to
958 make it such. The persistence data is stored in the variables
959 below. Ideally, it should be possible to cache an arbitrary fixed
960 number of these connections. */
962 /* Whether a persistent connection is active. */
963 static bool pconn_active;
966 /* The socket of the connection. */
969 /* Host and port of the currently active persistent connection. */
973 /* Whether a ssl handshake has occoured on this connection. */
976 /* Whether the connection was authorized. This is only done by
977 NTLM, which authorizes *connections* rather than individual
978 requests. (That practice is peculiar for HTTP, but it is a
979 useful optimization.) */
983 /* NTLM data of the current connection. */
984 struct ntlmdata ntlm;
988 /* Mark the persistent connection as invalid and free the resources it
989 uses. This is used by the CLOSE_* macros after they forcefully
990 close a registered persistent connection. */
993 invalidate_persistent (void)
995 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
996 pconn_active = false;
997 fd_close (pconn.socket);
1002 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1003 persistent. This will enable someone to use the same connection
1004 later. In the context of HTTP, this must be called only AFTER the
1005 response has been received and the server has promised that the
1006 connection will remain alive.
1008 If a previous connection was persistent, it is closed. */
1011 register_persistent (const char *host, int port, int fd, bool ssl)
1015 if (pconn.socket == fd)
1017 /* The connection FD is already registered. */
1022 /* The old persistent connection is still active; close it
1023 first. This situation arises whenever a persistent
1024 connection exists, but we then connect to a different
1025 host, and try to register a persistent connection to that
1027 invalidate_persistent ();
1031 pconn_active = true;
1033 pconn.host = xstrdup (host);
1036 pconn.authorized = false;
1038 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1041 /* Return true if a persistent connection is available for connecting
1045 persistent_available_p (const char *host, int port, bool ssl,
1046 bool *host_lookup_failed)
1048 /* First, check whether a persistent connection is active at all. */
1052 /* If we want SSL and the last connection wasn't or vice versa,
1053 don't use it. Checking for host and port is not enough because
1054 HTTP and HTTPS can apparently coexist on the same port. */
1055 if (ssl != pconn.ssl)
1058 /* If we're not connecting to the same port, we're not interested. */
1059 if (port != pconn.port)
1062 /* If the host is the same, we're in business. If not, there is
1063 still hope -- read below. */
1064 if (0 != strcasecmp (host, pconn.host))
1066 /* Check if pconn.socket is talking to HOST under another name.
1067 This happens often when both sites are virtual hosts
1068 distinguished only by name and served by the same network
1069 interface, and hence the same web server (possibly set up by
1070 the ISP and serving many different web sites). This
1071 admittedly unconventional optimization does not contradict
1072 HTTP and works well with popular server software. */
1076 struct address_list *al;
1079 /* Don't try to talk to two different SSL sites over the same
1080 secure connection! (Besides, it's not clear that
1081 name-based virtual hosting is even possible with SSL.) */
1084 /* If pconn.socket's peer is one of the IP addresses HOST
1085 resolves to, pconn.socket is for all intents and purposes
1086 already talking to HOST. */
1088 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1090 /* Can't get the peer's address -- something must be very
1091 wrong with the connection. */
1092 invalidate_persistent ();
1095 al = lookup_host (host, 0);
1098 *host_lookup_failed = true;
1102 found = address_list_contains (al, &ip);
1103 address_list_release (al);
1108 /* The persistent connection's peer address was found among the
1109 addresses HOST resolved to; therefore, pconn.sock is in fact
1110 already talking to HOST -- no need to reconnect. */
1113 /* Finally, check whether the connection is still open. This is
1114 important because most servers implement liberal (short) timeout
1115 on persistent connections. Wget can of course always reconnect
1116 if the connection doesn't work out, but it's nicer to know in
1117 advance. This test is a logical followup of the first test, but
1118 is "expensive" and therefore placed at the end of the list.
1120 (Current implementation of test_socket_open has a nice side
1121 effect that it treats sockets with pending data as "closed".
1122 This is exactly what we want: if a broken server sends message
1123 body in response to HEAD, or if it sends more than conent-length
1124 data, we won't reuse the corrupted connection.) */
1126 if (!test_socket_open (pconn.socket))
1128 /* Oops, the socket is no longer open. Now that we know that,
1129 let's invalidate the persistent connection before returning
1131 invalidate_persistent ();
1138 /* The idea behind these two CLOSE macros is to distinguish between
1139 two cases: one when the job we've been doing is finished, and we
1140 want to close the connection and leave, and two when something is
1141 seriously wrong and we're closing the connection as part of
1144 In case of keep_alive, CLOSE_FINISH should leave the connection
1145 open, while CLOSE_INVALIDATE should still close it.
1147 Note that the semantics of the flag `keep_alive' is "this
1148 connection *will* be reused (the server has promised not to close
1149 the connection once we're done)", while the semantics of
1150 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1151 active, registered connection". */
1153 #define CLOSE_FINISH(fd) do { \
1156 if (pconn_active && (fd) == pconn.socket) \
1157 invalidate_persistent (); \
1166 #define CLOSE_INVALIDATE(fd) do { \
1167 if (pconn_active && (fd) == pconn.socket) \
1168 invalidate_persistent (); \
1176 wgint len; /* received length */
1177 wgint contlen; /* expected length */
1178 wgint restval; /* the restart value */
1179 int res; /* the result of last read */
1180 char *rderrmsg; /* error message from read error */
1181 char *newloc; /* new location (redirection) */
1182 char *remote_time; /* remote time-stamp string */
1183 char *error; /* textual HTTP error */
1184 int statcode; /* status code */
1185 wgint rd_size; /* amount of data read from socket */
1186 double dltime; /* time it took to download the data */
1187 const char *referer; /* value of the referer header. */
1188 char *local_file; /* local file name. */
1189 bool timestamp_checked; /* true if pre-download time-stamping checks
1190 * have already been performed */
1191 char *orig_file_name; /* name of file to compare for time-stamping
1192 * (might be != local_file if -K is set) */
1193 wgint orig_file_size; /* size of file to compare for time-stamping */
1194 time_t orig_file_tstamp; /* time-stamp of file to compare for
1199 free_hstat (struct http_stat *hs)
1201 xfree_null (hs->newloc);
1202 xfree_null (hs->remote_time);
1203 xfree_null (hs->error);
1204 xfree_null (hs->rderrmsg);
1205 xfree_null (hs->local_file);
1206 xfree_null (hs->orig_file_name);
1208 /* Guard against being called twice. */
1210 hs->remote_time = NULL;
1214 static char *create_authorization_line (const char *, const char *,
1215 const char *, const char *,
1216 const char *, bool *);
1217 static char *basic_authentication_encode (const char *, const char *);
1218 static bool known_authentication_scheme_p (const char *, const char *);
1219 static void load_cookies (void);
1221 #define BEGINS_WITH(line, string_constant) \
1222 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1223 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1224 || !line[sizeof (string_constant) - 1]))
1226 #define SET_USER_AGENT(req) do { \
1227 if (!opt.useragent) \
1228 request_set_header (req, "User-Agent", \
1229 aprintf ("Wget/%s", version_string), rel_value); \
1230 else if (*opt.useragent) \
1231 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1234 /* The flags that allow clobbering the file (opening with "wb").
1235 Defined here to avoid repetition later. #### This will require
1237 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1238 || opt.dirstruct || opt.output_document)
1240 /* Retrieve a document through HTTP protocol. It recognizes status
1241 code, and correctly handles redirections. It closes the network
1242 socket. If it receives an error from the functions below it, it
1243 will print it if there is enough information to do so (almost
1244 always), returning the error to the caller (i.e. http_loop).
1246 Various HTTP parameters are stored to hs.
1248 If PROXY is non-NULL, the connection will be made to the proxy
1249 server, and u->url will be requested. */
1251 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1253 struct request *req;
1256 char *user, *passwd;
1260 wgint contlen, contrange;
1267 /* Set to 1 when the authorization has failed permanently and should
1268 not be tried again. */
1269 bool auth_finished = false;
1271 /* Whether NTLM authentication is used for this request. */
1272 bool ntlm_seen = false;
1274 /* Whether our connection to the remote host is through SSL. */
1275 bool using_ssl = false;
1277 /* Whether a HEAD request will be issued (as opposed to GET or
1279 bool head_only = !!(*dt & HEAD_ONLY);
1282 struct response *resp;
1286 /* Whether this connection will be kept alive after the HTTP request
1290 /* Whether keep-alive should be inhibited.
1292 RFC 2068 requests that 1.0 clients not send keep-alive requests
1293 to proxies. This is because many 1.0 proxies do not interpret
1294 the Connection header and transfer it to the remote server,
1295 causing it to not close the connection and leave both the proxy
1296 and the client hanging. */
1297 bool inhibit_keep_alive =
1298 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1300 /* Headers sent when using POST. */
1301 wgint post_data_size = 0;
1303 bool host_lookup_failed = false;
1306 if (u->scheme == SCHEME_HTTPS)
1308 /* Initialize the SSL context. After this has once been done,
1309 it becomes a no-op. */
1312 scheme_disable (SCHEME_HTTPS);
1313 logprintf (LOG_NOTQUIET,
1314 _("Disabling SSL due to encountered errors.\n"));
1315 return SSLINITFAILED;
1318 #endif /* HAVE_SSL */
1320 /* Initialize certain elements of struct http_stat. */
1324 hs->rderrmsg = NULL;
1326 hs->remote_time = NULL;
1331 /* Prepare the request to send. */
1333 req = request_new ();
1336 const char *meth = "GET";
1339 else if (opt.post_file_name || opt.post_data)
1341 /* Use the full path, i.e. one that includes the leading slash and
1342 the query string. E.g. if u->path is "foo/bar" and u->query is
1343 "param=value", full_path will be "/foo/bar?param=value". */
1346 /* When using SSL over proxy, CONNECT establishes a direct
1347 connection to the HTTPS server. Therefore use the same
1348 argument as when talking to the server directly. */
1349 && u->scheme != SCHEME_HTTPS
1352 meth_arg = xstrdup (u->url);
1354 meth_arg = url_full_path (u);
1355 request_set_method (req, meth, meth_arg);
1358 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1359 if (*dt & SEND_NOCACHE)
1360 request_set_header (req, "Pragma", "no-cache", rel_none);
1362 request_set_header (req, "Range",
1363 aprintf ("bytes=%s-",
1364 number_to_static_string (hs->restval)),
1366 SET_USER_AGENT (req);
1367 request_set_header (req, "Accept", "*/*", rel_none);
1369 /* Find the username and password for authentication. */
1372 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1373 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1374 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1378 /* We have the username and the password, but haven't tried
1379 any authorization yet. Let's see if the "Basic" method
1380 works. If not, we'll come back here and construct a
1381 proper authorization method with the right challenges.
1383 If we didn't employ this kind of logic, every URL that
1384 requires authorization would have to be processed twice,
1385 which is very suboptimal and generates a bunch of false
1386 "unauthorized" errors in the server log.
1388 #### But this logic also has a serious problem when used
1389 with stronger authentications: we *first* transmit the
1390 username and the password in clear text, and *then* attempt a
1391 stronger authentication scheme. That cannot be right! We
1392 are only fortunate that almost everyone still uses the
1393 `Basic' scheme anyway.
1395 There should be an option to prevent this from happening, for
1396 those who use strong authentication schemes and value their
1398 request_set_header (req, "Authorization",
1399 basic_authentication_encode (user, passwd),
1406 char *proxy_user, *proxy_passwd;
1407 /* For normal username and password, URL components override
1408 command-line/wgetrc parameters. With proxy
1409 authentication, it's the reverse, because proxy URLs are
1410 normally the "permanent" ones, so command-line args
1411 should take precedence. */
1412 if (opt.proxy_user && opt.proxy_passwd)
1414 proxy_user = opt.proxy_user;
1415 proxy_passwd = opt.proxy_passwd;
1419 proxy_user = proxy->user;
1420 proxy_passwd = proxy->passwd;
1422 /* #### This does not appear right. Can't the proxy request,
1423 say, `Digest' authentication? */
1424 if (proxy_user && proxy_passwd)
1425 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1427 /* If we're using a proxy, we will be connecting to the proxy
1431 /* Proxy authorization over SSL is handled below. */
1433 if (u->scheme != SCHEME_HTTPS)
1435 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1438 /* Generate the Host header, HOST:PORT. Take into account that:
1440 - Broken server-side software often doesn't recognize the PORT
1441 argument, so we must generate "Host: www.server.com" instead of
1442 "Host: www.server.com:80" (and likewise for https port).
1444 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1445 becomes ambiguous and needs to be rewritten as "Host:
1446 [3ffe:8100:200:2::2]:1234". */
1448 /* Formats arranged for hfmt[add_port][add_squares]. */
1449 static const char *hfmt[][2] = {
1450 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1452 int add_port = u->port != scheme_default_port (u->scheme);
1453 int add_squares = strchr (u->host, ':') != NULL;
1454 request_set_header (req, "Host",
1455 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1459 if (!inhibit_keep_alive)
1460 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1463 request_set_header (req, "Cookie",
1464 cookie_header (wget_cookie_jar,
1465 u->host, u->port, u->path,
1467 u->scheme == SCHEME_HTTPS
1474 if (opt.post_data || opt.post_file_name)
1476 request_set_header (req, "Content-Type",
1477 "application/x-www-form-urlencoded", rel_none);
1479 post_data_size = strlen (opt.post_data);
1482 post_data_size = file_size (opt.post_file_name);
1483 if (post_data_size == -1)
1485 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1486 opt.post_file_name, strerror (errno));
1490 request_set_header (req, "Content-Length",
1491 xstrdup (number_to_static_string (post_data_size)),
1495 /* Add the user headers. */
1496 if (opt.user_headers)
1499 for (i = 0; opt.user_headers[i]; i++)
1500 request_set_user_header (req, opt.user_headers[i]);
1504 /* We need to come back here when the initial attempt to retrieve
1505 without authorization header fails. (Expected to happen at least
1506 for the Digest authorization scheme.) */
1510 /* Establish the connection. */
1512 if (!inhibit_keep_alive)
1514 /* Look for a persistent connection to target host, unless a
1515 proxy is used. The exception is when SSL is in use, in which
1516 case the proxy is nothing but a passthrough to the target
1517 host, registered as a connection to the latter. */
1518 struct url *relevant = conn;
1520 if (u->scheme == SCHEME_HTTPS)
1524 if (persistent_available_p (relevant->host, relevant->port,
1526 relevant->scheme == SCHEME_HTTPS,
1530 &host_lookup_failed))
1532 sock = pconn.socket;
1533 using_ssl = pconn.ssl;
1534 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1535 escnonprint (pconn.host), pconn.port);
1536 DEBUGP (("Reusing fd %d.\n", sock));
1537 if (pconn.authorized)
1538 /* If the connection is already authorized, the "Basic"
1539 authorization added by code above is unnecessary and
1541 request_remove_header (req, "Authorization");
1547 /* In its current implementation, persistent_available_p will
1548 look up conn->host in some cases. If that lookup failed, we
1549 don't need to bother with connect_to_host. */
1550 if (host_lookup_failed)
1556 sock = connect_to_host (conn->host, conn->port);
1565 return (retryable_socket_connect_error (errno)
1566 ? CONERROR : CONIMPOSSIBLE);
1570 if (proxy && u->scheme == SCHEME_HTTPS)
1572 /* When requesting SSL URLs through proxies, use the
1573 CONNECT method to request passthrough. */
1574 struct request *connreq = request_new ();
1575 request_set_method (connreq, "CONNECT",
1576 aprintf ("%s:%d", u->host, u->port));
1577 SET_USER_AGENT (connreq);
1580 request_set_header (connreq, "Proxy-Authorization",
1581 proxyauth, rel_value);
1582 /* Now that PROXYAUTH is part of the CONNECT request,
1583 zero it out so we don't send proxy authorization with
1584 the regular request below. */
1587 /* Examples in rfc2817 use the Host header in CONNECT
1588 requests. I don't see how that gains anything, given
1589 that the contents of Host would be exactly the same as
1590 the contents of CONNECT. */
1592 write_error = request_send (connreq, sock);
1593 request_free (connreq);
1594 if (write_error < 0)
1596 CLOSE_INVALIDATE (sock);
1600 head = read_http_response_head (sock);
1603 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1605 CLOSE_INVALIDATE (sock);
1614 DEBUGP (("proxy responded with: [%s]\n", head));
1616 resp = resp_new (head);
1617 statcode = resp_status (resp, &message);
1620 if (statcode != 200)
1623 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1624 message ? escnonprint (message) : "?");
1625 xfree_null (message);
1628 xfree_null (message);
1630 /* SOCK is now *really* connected to u->host, so update CONN
1631 to reflect this. That way register_persistent will
1632 register SOCK as being connected to u->host:u->port. */
1636 if (conn->scheme == SCHEME_HTTPS)
1638 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1645 #endif /* HAVE_SSL */
1648 /* Send the request to server. */
1649 write_error = request_send (req, sock);
1651 if (write_error >= 0)
1655 DEBUGP (("[POST data: %s]\n", opt.post_data));
1656 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1658 else if (opt.post_file_name && post_data_size != 0)
1659 write_error = post_file (sock, opt.post_file_name, post_data_size);
1662 if (write_error < 0)
1664 CLOSE_INVALIDATE (sock);
1668 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1669 proxy ? "Proxy" : "HTTP");
1674 head = read_http_response_head (sock);
1679 logputs (LOG_NOTQUIET, _("No data received.\n"));
1680 CLOSE_INVALIDATE (sock);
1686 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1688 CLOSE_INVALIDATE (sock);
1693 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1695 resp = resp_new (head);
1697 /* Check for status line. */
1699 statcode = resp_status (resp, &message);
1700 if (!opt.server_response)
1701 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1702 message ? escnonprint (message) : "");
1705 logprintf (LOG_VERBOSE, "\n");
1706 print_server_response (resp, " ");
1709 /* Determine the local filename if needed. Notice that if -O is used
1710 * hstat.local_file is set by http_loop to the argument of -O. */
1711 if (!hs->local_file)
1713 /* Honor Content-Disposition whether possible. */
1714 if (!resp_header_copy (resp, "Content-Disposition", hdrval, sizeof (hdrval))
1715 || !parse_content_disposition (hdrval, &hs->local_file))
1717 /* Choose filename according to URL name. */
1718 hs->local_file = url_file_name (u);
1722 /* TODO: perform this check only once. */
1723 if (opt.noclobber && file_exists_p (hs->local_file))
1725 /* If opt.noclobber is turned on and file already exists, do not
1726 retrieve the file */
1727 logprintf (LOG_VERBOSE, _("\
1728 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1729 /* If the file is there, we suppose it's retrieved OK. */
1732 /* #### Bogusness alert. */
1733 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1734 if (has_html_suffix_p (hs->local_file))
1740 /* Support timestamping */
1741 /* TODO: move this code out of gethttp. */
1742 if (opt.timestamping && !hs->timestamp_checked)
1744 size_t filename_len = strlen (hs->local_file);
1745 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1746 bool local_dot_orig_file_exists = false;
1747 char *local_filename = NULL;
1750 if (opt.backup_converted)
1751 /* If -K is specified, we'll act on the assumption that it was specified
1752 last time these files were downloaded as well, and instead of just
1753 comparing local file X against server file X, we'll compare local
1754 file X.orig (if extant, else X) against server file X. If -K
1755 _wasn't_ specified last time, or the server contains files called
1756 *.orig, -N will be back to not operating correctly with -k. */
1758 /* Would a single s[n]printf() call be faster? --dan
1760 Definitely not. sprintf() is horribly slow. It's a
1761 different question whether the difference between the two
1762 affects a program. Usually I'd say "no", but at one
1763 point I profiled Wget, and found that a measurable and
1764 non-negligible amount of time was lost calling sprintf()
1765 in url.c. Replacing sprintf with inline calls to
1766 strcpy() and number_to_string() made a difference.
1768 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1769 memcpy (filename_plus_orig_suffix + filename_len,
1770 ".orig", sizeof (".orig"));
1772 /* Try to stat() the .orig file. */
1773 if (stat (filename_plus_orig_suffix, &st) == 0)
1775 local_dot_orig_file_exists = 1;
1776 local_filename = filename_plus_orig_suffix;
1780 if (!local_dot_orig_file_exists)
1781 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1782 if (stat (hs->local_file, &st) == 0)
1783 local_filename = hs->local_file;
1785 if (local_filename != NULL)
1786 /* There was a local file, so we'll check later to see if the version
1787 the server has is the same version we already have, allowing us to
1790 hs->orig_file_name = xstrdup (local_filename);
1791 hs->orig_file_size = st.st_size;
1792 hs->orig_file_tstamp = st.st_mtime;
1794 /* Modification time granularity is 2 seconds for Windows, so
1795 increase local time by 1 second for later comparison. */
1796 ++hs->orig_file_tstamp;
1801 if (!opt.ignore_length
1802 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1806 parsed = str_to_wgint (hdrval, NULL, 10);
1807 if (parsed == WGINT_MAX && errno == ERANGE)
1809 #### If Content-Length is out of range, it most likely
1810 means that the file is larger than 2G and that we're
1811 compiled without LFS. In that case we should probably
1812 refuse to even attempt to download the file. */
1818 /* Check for keep-alive related responses. */
1819 if (!inhibit_keep_alive && contlen != -1)
1821 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1823 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1825 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1830 /* The server has promised that it will not close the connection
1831 when we're done. This means that we can register it. */
1832 register_persistent (conn->host, conn->port, sock, using_ssl);
1834 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1836 /* Authorization is required. */
1837 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1838 CLOSE_FINISH (sock);
1840 CLOSE_INVALIDATE (sock);
1841 pconn.authorized = false;
1842 if (!auth_finished && (user && passwd))
1844 /* IIS sends multiple copies of WWW-Authenticate, one with
1845 the value "negotiate", and other(s) with data. Loop over
1846 all the occurrences and pick the one we recognize. */
1848 const char *wabeg, *waend;
1849 char *www_authenticate = NULL;
1851 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1852 &wabeg, &waend)) != -1;
1854 if (known_authentication_scheme_p (wabeg, waend))
1856 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1860 if (!www_authenticate)
1861 /* If the authentication header is missing or
1862 unrecognized, there's no sense in retrying. */
1863 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1864 else if (BEGINS_WITH (www_authenticate, "Basic"))
1865 /* If the authentication scheme is "Basic", which we send
1866 by default, there's no sense in retrying either. (This
1867 should be changed when we stop sending "Basic" data by
1873 pth = url_full_path (u);
1874 request_set_header (req, "Authorization",
1875 create_authorization_line (www_authenticate,
1877 request_method (req),
1881 if (BEGINS_WITH (www_authenticate, "NTLM"))
1884 goto retry_with_auth;
1887 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1891 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1893 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1895 pconn.authorized = true;
1899 hs->statcode = statcode;
1901 hs->error = xstrdup (_("Malformed status line"));
1903 hs->error = xstrdup (_("(no description)"));
1905 hs->error = xstrdup (message);
1906 xfree_null (message);
1908 type = resp_header_strdup (resp, "Content-Type");
1911 char *tmp = strchr (type, ';');
1914 while (tmp > type && ISSPACE (tmp[-1]))
1919 hs->newloc = resp_header_strdup (resp, "Location");
1920 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1922 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1926 const char *scbeg, *scend;
1927 /* The jar should have been created by now. */
1928 assert (wget_cookie_jar != NULL);
1930 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1931 &scbeg, &scend)) != -1;
1934 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1935 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1936 u->path, set_cookie);
1940 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1942 wgint first_byte_pos, last_byte_pos, entity_length;
1943 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1945 contrange = first_byte_pos;
1949 /* 20x responses are counted among successful by default. */
1950 if (H_20X (statcode))
1953 /* Return if redirected. */
1954 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1956 /* RFC2068 says that in case of the 300 (multiple choices)
1957 response, the server can output a preferred URL through
1958 `Location' header; otherwise, the request should be treated
1959 like GET. So, if the location is set, it will be a
1960 redirection; otherwise, just proceed normally. */
1961 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1965 logprintf (LOG_VERBOSE,
1966 _("Location: %s%s\n"),
1967 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1968 hs->newloc ? _(" [following]") : "");
1969 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1970 CLOSE_FINISH (sock);
1972 CLOSE_INVALIDATE (sock);
1978 /* If content-type is not given, assume text/html. This is because
1979 of the multitude of broken CGI's that "forget" to generate the
1982 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1983 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1988 if (opt.html_extension && (*dt & TEXTHTML))
1989 /* -E / --html-extension / html_extension = on was specified, and this is a
1990 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1991 already the file's suffix, tack on ".html". */
1993 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
1995 if (last_period_in_local_filename == NULL
1996 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1997 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1999 int local_filename_len = strlen (hs->local_file);
2000 /* Resize the local file, allowing for ".html" preceded by
2001 optional ".NUMBER". */
2002 hs->local_file = xrealloc (hs->local_file,
2003 local_filename_len + 24 + sizeof (".html"));
2004 strcpy(hs->local_file + local_filename_len, ".html");
2005 /* If clobbering is not allowed and the file, as named,
2006 exists, tack on ".NUMBER.html" instead. */
2007 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2011 sprintf (hs->local_file + local_filename_len,
2012 ".%d.html", ext_num++);
2013 while (file_exists_p (hs->local_file));
2015 *dt |= ADDED_HTML_EXTENSION;
2019 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2021 /* If `-c' is in use and the file has been fully downloaded (or
2022 the remote file has shrunk), Wget effectively requests bytes
2023 after the end of file and the server response with 416. */
2024 logputs (LOG_VERBOSE, _("\
2025 \n The file is already fully retrieved; nothing to do.\n\n"));
2026 /* In case the caller inspects. */
2029 /* Mark as successfully retrieved. */
2032 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2033 might be more bytes in the body. */
2034 return RETRUNNEEDED;
2036 if ((contrange != 0 && contrange != hs->restval)
2037 || (H_PARTIAL (statcode) && !contrange))
2039 /* The Range request was somehow misunderstood by the server.
2042 CLOSE_INVALIDATE (sock);
2045 hs->contlen = contlen + contrange;
2051 /* No need to print this output if the body won't be
2052 downloaded at all, or if the original server response is
2054 logputs (LOG_VERBOSE, _("Length: "));
2057 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2058 if (contlen + contrange >= 1024)
2059 logprintf (LOG_VERBOSE, " (%s)",
2060 human_readable (contlen + contrange));
2063 if (contlen >= 1024)
2064 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2065 number_to_static_string (contlen),
2066 human_readable (contlen));
2068 logprintf (LOG_VERBOSE, _(", %s remaining"),
2069 number_to_static_string (contlen));
2073 logputs (LOG_VERBOSE,
2074 opt.ignore_length ? _("ignored") : _("unspecified"));
2076 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2078 logputs (LOG_VERBOSE, "\n");
2082 type = NULL; /* We don't need it any more. */
2084 /* Return if we have no intention of further downloading. */
2085 if (!(*dt & RETROKF) || head_only)
2087 /* In case the caller cares to look... */
2092 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2093 servers not to send body in response to a HEAD request, and
2094 those that do will likely be caught by test_socket_open.
2095 If not, they can be worked around using
2096 `--no-http-keep-alive'. */
2097 CLOSE_FINISH (sock);
2098 else if (keep_alive && skip_short_body (sock, contlen))
2099 /* Successfully skipped the body; also keep using the socket. */
2100 CLOSE_FINISH (sock);
2102 CLOSE_INVALIDATE (sock);
2103 return RETRFINISHED;
2106 /* Print fetch message, if opt.verbose. */
2109 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2110 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2113 /* Open the local file. */
2116 mkalldirs (hs->local_file);
2118 rotate_backups (hs->local_file);
2120 fp = fopen (hs->local_file, "ab");
2121 else if (ALLOW_CLOBBER)
2122 fp = fopen (hs->local_file, "wb");
2125 fp = fopen_excl (hs->local_file, true);
2126 if (!fp && errno == EEXIST)
2128 /* We cannot just invent a new name and use it (which is
2129 what functions like unique_create typically do)
2130 because we told the user we'd use this name.
2131 Instead, return and retry the download. */
2132 logprintf (LOG_NOTQUIET,
2133 _("%s has sprung into existence.\n"),
2135 CLOSE_INVALIDATE (sock);
2136 return FOPEN_EXCL_ERR;
2141 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2142 CLOSE_INVALIDATE (sock);
2149 /* This confuses the timestamping code that checks for file size.
2150 #### The timestamping code should be smarter about file size. */
2151 if (opt.save_headers && hs->restval == 0)
2152 fwrite (head, 1, strlen (head), fp);
2154 /* Now we no longer need to store the response header. */
2157 /* Download the request body. */
2160 /* If content-length is present, read that much; otherwise, read
2161 until EOF. The HTTP spec doesn't require the server to
2162 actually close the connection when it's done sending data. */
2163 flags |= rb_read_exactly;
2164 if (hs->restval > 0 && contrange == 0)
2165 /* If the server ignored our range request, instruct fd_read_body
2166 to skip the first RESTVAL bytes of body. */
2167 flags |= rb_skip_startpos;
2168 hs->len = hs->restval;
2170 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2171 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2175 CLOSE_FINISH (sock);
2179 hs->rderrmsg = xstrdup (fd_errstr (sock));
2180 CLOSE_INVALIDATE (sock);
2187 return RETRFINISHED;
2190 /* The genuine HTTP loop! This is the part where the retrieval is
2191 retried, and retried, and retried, and... */
2193 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2194 int *dt, struct url *proxy)
2197 bool got_head = false; /* used for time-stamping */
2201 time_t tmr = -1; /* remote time-stamp */
2202 wgint local_size = 0; /* the size of the local file */
2203 struct http_stat hstat; /* HTTP status */
2206 /* Assert that no value for *LOCAL_FILE was passed. */
2207 assert (local_file == NULL || *local_file == NULL);
2209 /* Set LOCAL_FILE parameter. */
2210 if (local_file && opt.output_document)
2211 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2213 /* Reset NEWLOC parameter. */
2216 /* This used to be done in main(), but it's a better idea to do it
2217 here so that we don't go through the hoops if we're just using
2222 /* Warn on (likely bogus) wildcard usage in HTTP. */
2223 if (opt.ftp_glob && has_wildcards_p (u->path))
2224 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2226 /* Setup hstat struct. */
2228 hstat.referer = referer;
2230 if (opt.output_document)
2231 hstat.local_file = xstrdup (opt.output_document);
2233 /* Reset the counter. */
2236 /* Reset the document type. */
2242 /* Increment the pass counter. */
2244 sleep_between_retrievals (count);
2246 /* Get the current time string. */
2247 tms = time_str (NULL);
2249 /* Print fetch message, if opt.verbose. */
2252 char *hurl = url_string (u, true);
2257 sprintf (tmp, _("(try:%2d)"), count);
2258 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2263 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2268 ws_changetitle (hurl);
2273 /* Default document type is empty. However, if spider mode is
2274 on or time-stamping is employed, HEAD_ONLY commands is
2275 encoded within *dt. */
2276 if (opt.spider || (opt.timestamping && !got_head))
2281 /* Decide whether or not to restart. */
2283 && stat (hstat.local_file, &st) == 0
2284 && S_ISREG (st.st_mode))
2285 /* When -c is used, continue from on-disk size. (Can't use
2286 hstat.len even if count>1 because we don't want a failed
2287 first attempt to clobber existing data.) */
2288 hstat.restval = st.st_size;
2290 /* otherwise, continue where the previous try left off */
2291 hstat.restval = hstat.len;
2295 /* Decide whether to send the no-cache directive. We send it in
2297 a) we're using a proxy, and we're past our first retrieval.
2298 Some proxies are notorious for caching incomplete data, so
2299 we require a fresh get.
2300 b) caching is explicitly inhibited. */
2301 if ((proxy && count > 1) /* a */
2302 || !opt.allow_cache /* b */
2304 *dt |= SEND_NOCACHE;
2306 *dt &= ~SEND_NOCACHE;
2308 /* Try fetching the document, or at least its head. */
2309 err = gethttp (u, &hstat, dt, proxy);
2312 tms = time_str (NULL);
2314 /* Get the new location (with or without the redirection). */
2316 *newloc = xstrdup (hstat.newloc);
2320 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2321 case CONERROR: case READERR: case WRITEFAILED:
2322 case RANGEERR: case FOPEN_EXCL_ERR:
2323 /* Non-fatal errors continue executing the loop, which will
2324 bring them to "while" statement at the end, to judge
2325 whether the number of tries was exceeded. */
2326 /* free_hstat (&hstat); */
2327 printwhat (count, opt.ntry);
2329 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2330 case SSLINITFAILED: case CONTNOTSUPPORTED:
2331 /* Fatal errors just return from the function. */
2332 free_hstat (&hstat);
2334 case FWRITEERR: case FOPENERR:
2335 /* Another fatal error. */
2336 logputs (LOG_VERBOSE, "\n");
2337 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2338 hstat.local_file, strerror (errno));
2339 free_hstat (&hstat);
2342 /* Another fatal error. */
2343 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2344 free_hstat (&hstat);
2347 /* Return the new location to the caller. */
2350 logprintf (LOG_NOTQUIET,
2351 _("ERROR: Redirection (%d) without location.\n"),
2353 free_hstat (&hstat);
2356 free_hstat (&hstat);
2359 /* The file was already fully retrieved. */
2360 free_hstat (&hstat);
2363 /* Deal with you later. */
2366 /* All possibilities should have been exhausted. */
2370 if (!(*dt & RETROKF))
2374 /* #### Ugly ugly ugly! */
2375 char *hurl = url_string (u, true);
2376 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2379 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2380 tms, hstat.statcode, escnonprint (hstat.error));
2381 logputs (LOG_VERBOSE, "\n");
2382 free_hstat (&hstat);
2386 /* Did we get the time-stamp? */
2389 if (opt.timestamping && !hstat.remote_time)
2391 logputs (LOG_NOTQUIET, _("\
2392 Last-modified header missing -- time-stamps turned off.\n"));
2394 else if (hstat.remote_time)
2396 /* Convert the date-string into struct tm. */
2397 tmr = http_atotm (hstat.remote_time);
2398 if (tmr == (time_t) (-1))
2399 logputs (LOG_VERBOSE, _("\
2400 Last-modified header invalid -- time-stamp ignored.\n"));
2404 /* The time-stamping section. */
2405 if (opt.timestamping && !got_head)
2407 got_head = true; /* no more time-stamping */
2409 count = 0; /* the retrieve count for HEAD is reset */
2411 if (hstat.remote_time && tmr != (time_t) (-1))
2413 /* Now time-stamping can be used validly. Time-stamping
2414 means that if the sizes of the local and remote file
2415 match, and local file is newer than the remote file,
2416 it will not be retrieved. Otherwise, the normal
2417 download procedure is resumed. */
2418 if (hstat.orig_file_tstamp >= tmr)
2420 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2422 logprintf (LOG_VERBOSE, _("\
2423 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2424 hstat.orig_file_name);
2425 free_hstat (&hstat);
2430 logprintf (LOG_VERBOSE, _("\
2431 The sizes do not match (local %s) -- retrieving.\n"),
2432 number_to_static_string (local_size));
2436 logputs (LOG_VERBOSE,
2437 _("Remote file is newer, retrieving.\n"));
2439 logputs (LOG_VERBOSE, "\n");
2442 /* free_hstat (&hstat); */
2443 hstat.timestamp_checked = true;
2447 if ((tmr != (time_t) (-1))
2449 && ((hstat.len == hstat.contlen) ||
2450 ((hstat.res == 0) && (hstat.contlen == -1))))
2452 /* #### This code repeats in http.c and ftp.c. Move it to a
2454 const char *fl = NULL;
2455 if (opt.output_document)
2457 if (output_stream_regular)
2458 fl = opt.output_document;
2461 fl = hstat.local_file;
2465 /* End of time-stamping section. */
2469 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2470 escnonprint (hstat.error));
2474 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2475 total_download_time += hstat.dltime;
2477 if (hstat.len == hstat.contlen)
2481 logprintf (LOG_VERBOSE,
2482 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2483 tms, tmrate, hstat.local_file,
2484 number_to_static_string (hstat.len),
2485 number_to_static_string (hstat.contlen));
2486 logprintf (LOG_NONVERBOSE,
2487 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2489 number_to_static_string (hstat.len),
2490 number_to_static_string (hstat.contlen),
2491 hstat.local_file, count);
2494 total_downloaded_bytes += hstat.len;
2496 /* Remember that we downloaded the file for later ".orig" code. */
2497 if (*dt & ADDED_HTML_EXTENSION)
2498 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2500 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2502 free_hstat (&hstat);
2505 else if (hstat.res == 0) /* No read error */
2507 if (hstat.contlen == -1) /* We don't know how much we were supposed
2508 to get, so assume we succeeded. */
2512 logprintf (LOG_VERBOSE,
2513 _("%s (%s) - `%s' saved [%s]\n\n"),
2514 tms, tmrate, hstat.local_file,
2515 number_to_static_string (hstat.len));
2516 logprintf (LOG_NONVERBOSE,
2517 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2518 tms, u->url, number_to_static_string (hstat.len),
2519 hstat.local_file, count);
2522 total_downloaded_bytes += hstat.len;
2524 /* Remember that we downloaded the file for later ".orig" code. */
2525 if (*dt & ADDED_HTML_EXTENSION)
2526 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2528 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2530 free_hstat (&hstat);
2533 else if (hstat.len < hstat.contlen) /* meaning we lost the
2534 connection too soon */
2536 logprintf (LOG_VERBOSE,
2537 _("%s (%s) - Connection closed at byte %s. "),
2538 tms, tmrate, number_to_static_string (hstat.len));
2539 printwhat (count, opt.ntry);
2540 /* free_hstat (&hstat); */
2544 /* Getting here would mean reading more data than
2545 requested with content-length, which we never do. */
2548 else /* from now on hstat.res can only be -1 */
2550 if (hstat.contlen == -1)
2552 logprintf (LOG_VERBOSE,
2553 _("%s (%s) - Read error at byte %s (%s)."),
2554 tms, tmrate, number_to_static_string (hstat.len),
2556 printwhat (count, opt.ntry);
2557 /* free_hstat (&hstat); */
2560 else /* hstat.res == -1 and contlen is given */
2562 logprintf (LOG_VERBOSE,
2563 _("%s (%s) - Read error at byte %s/%s (%s). "),
2565 number_to_static_string (hstat.len),
2566 number_to_static_string (hstat.contlen),
2568 printwhat (count, opt.ntry);
2569 /* free_hstat (&hstat); */
2575 while (!opt.ntry || (count < opt.ntry));
2580 /* Check whether the result of strptime() indicates success.
2581 strptime() returns the pointer to how far it got to in the string.
2582 The processing has been successful if the string is at `GMT' or
2583 `+X', or at the end of the string.
2585 In extended regexp parlance, the function returns 1 if P matches
2586 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2587 can return) is considered a failure and 0 is returned. */
2589 check_end (const char *p)
2593 while (ISSPACE (*p))
2596 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2597 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2603 /* Convert the textual specification of time in TIME_STRING to the
2604 number of seconds since the Epoch.
2606 TIME_STRING can be in any of the three formats RFC2616 allows the
2607 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2608 as well as the time format used in the Set-Cookie header.
2609 Timezones are ignored, and should be GMT.
2611 Return the computed time_t representation, or -1 if the conversion
2614 This function uses strptime with various string formats for parsing
2615 TIME_STRING. This results in a parser that is not as lenient in
2616 interpreting TIME_STRING as I would like it to be. Being based on
2617 strptime, it always allows shortened months, one-digit days, etc.,
2618 but due to the multitude of formats in which time can be
2619 represented, an ideal HTTP time parser would be even more
2620 forgiving. It should completely ignore things like week days and
2621 concentrate only on the various forms of representing years,
2622 months, days, hours, minutes, and seconds. For example, it would
2623 be nice if it accepted ISO 8601 out of the box.
2625 I've investigated free and PD code for this purpose, but none was
2626 usable. getdate was big and unwieldy, and had potential copyright
2627 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2628 distributed with phttpd, is excellent, but we cannot use it because
2629 it is not assigned to the FSF. So I stuck it with strptime. */
2632 http_atotm (const char *time_string)
2634 /* NOTE: Solaris strptime man page claims that %n and %t match white
2635 space, but that's not universally available. Instead, we simply
2636 use ` ' to mean "skip all WS", which works under all strptime
2637 implementations I've tested. */
2639 static const char *time_formats[] = {
2640 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2641 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2642 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2643 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2644 (used in Set-Cookie, defined in the
2645 Netscape cookie specification.) */
2647 const char *oldlocale;
2649 time_t ret = (time_t) -1;
2651 /* Solaris strptime fails to recognize English month names in
2652 non-English locales, which we work around by temporarily setting
2653 locale to C before invoking strptime. */
2654 oldlocale = setlocale (LC_TIME, NULL);
2655 setlocale (LC_TIME, "C");
2657 for (i = 0; i < countof (time_formats); i++)
2661 /* Some versions of strptime use the existing contents of struct
2662 tm to recalculate the date according to format. Zero it out
2663 to prevent stack garbage from influencing strptime. */
2666 if (check_end (strptime (time_string, time_formats[i], &t)))
2673 /* Restore the previous locale. */
2674 setlocale (LC_TIME, oldlocale);
2679 /* Authorization support: We support three authorization schemes:
2681 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2683 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2684 consisting of answering to the server's challenge with the proper
2687 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2688 Stenberg for libcurl. Like digest, NTLM is based on a
2689 challenge-response mechanism, but unlike digest, it is non-standard
2690 (authenticates TCP connections rather than requests), undocumented
2691 and Microsoft-specific. */
2693 /* Create the authentication header contents for the `Basic' scheme.
2694 This is done by encoding the string "USER:PASS" to base64 and
2695 prepending the string "Basic " in front of it. */
2698 basic_authentication_encode (const char *user, const char *passwd)
2701 int len1 = strlen (user) + 1 + strlen (passwd);
2703 t1 = (char *)alloca (len1 + 1);
2704 sprintf (t1, "%s:%s", user, passwd);
2706 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2707 base64_encode (t1, len1, t2);
2709 return concat_strings ("Basic ", t2, (char *) 0);
2712 #define SKIP_WS(x) do { \
2713 while (ISSPACE (*(x))) \
2717 #ifdef ENABLE_DIGEST
2718 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2719 of a field in such a header. If the field is the one specified by
2720 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2721 digest authorization code), extract its value in the (char*)
2722 variable pointed by RET. Returns negative on a malformed header,
2723 or number of bytes that have been parsed by this call. */
2725 extract_header_attr (const char *au, const char *attr_name, char **ret)
2728 const char *cp = au;
2730 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2732 cp += strlen (attr_name);
2745 for (ep = cp; *ep && *ep != '\"'; ep++)
2750 *ret = strdupdelim (cp, ep);
2757 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2758 an array of 16 bytes containing the hash keys, and BUF should be a
2759 buffer of 33 writable characters (32 for hex digits plus one for
2760 zero termination). */
2762 dump_hash (char *buf, const unsigned char *hash)
2766 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2768 *buf++ = XNUM_TO_digit (*hash >> 4);
2769 *buf++ = XNUM_TO_digit (*hash & 0xf);
2774 /* Take the line apart to find the challenge, and compose a digest
2775 authorization header. See RFC2069 section 2.1.2. */
2777 digest_authentication_encode (const char *au, const char *user,
2778 const char *passwd, const char *method,
2781 static char *realm, *opaque, *nonce;
2786 { "realm", &realm },
2787 { "opaque", &opaque },
2792 realm = opaque = nonce = NULL;
2794 au += 6; /* skip over `Digest' */
2800 for (i = 0; i < countof (options); i++)
2802 int skip = extract_header_attr (au, options[i].name,
2803 options[i].variable);
2807 xfree_null (opaque);
2817 if (i == countof (options))
2819 while (*au && *au != '=')
2827 while (*au && *au != '\"')
2834 while (*au && *au != ',')
2839 if (!realm || !nonce || !user || !passwd || !path || !method)
2842 xfree_null (opaque);
2847 /* Calculate the digest value. */
2849 ALLOCA_MD5_CONTEXT (ctx);
2850 unsigned char hash[MD5_HASHLEN];
2851 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2852 char response_digest[MD5_HASHLEN * 2 + 1];
2854 /* A1BUF = H(user ":" realm ":" password) */
2856 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2857 gen_md5_update ((unsigned char *)":", 1, ctx);
2858 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2859 gen_md5_update ((unsigned char *)":", 1, ctx);
2860 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2861 gen_md5_finish (ctx, hash);
2862 dump_hash (a1buf, hash);
2864 /* A2BUF = H(method ":" path) */
2866 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2867 gen_md5_update ((unsigned char *)":", 1, ctx);
2868 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2869 gen_md5_finish (ctx, hash);
2870 dump_hash (a2buf, hash);
2872 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2874 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2875 gen_md5_update ((unsigned char *)":", 1, ctx);
2876 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2877 gen_md5_update ((unsigned char *)":", 1, ctx);
2878 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2879 gen_md5_finish (ctx, hash);
2880 dump_hash (response_digest, hash);
2882 res = xmalloc (strlen (user)
2887 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2888 + (opaque ? strlen (opaque) : 0)
2890 sprintf (res, "Digest \
2891 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2892 user, realm, nonce, path, response_digest);
2895 char *p = res + strlen (res);
2896 strcat (p, ", opaque=\"");
2903 #endif /* ENABLE_DIGEST */
2905 /* Computing the size of a string literal must take into account that
2906 value returned by sizeof includes the terminating \0. */
2907 #define STRSIZE(literal) (sizeof (literal) - 1)
2909 /* Whether chars in [b, e) begin with the literal string provided as
2910 first argument and are followed by whitespace or terminating \0.
2911 The comparison is case-insensitive. */
2912 #define STARTS(literal, b, e) \
2913 ((e) - (b) >= STRSIZE (literal) \
2914 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2915 && ((e) - (b) == STRSIZE (literal) \
2916 || ISSPACE (b[STRSIZE (literal)])))
2919 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2921 return STARTS ("Basic", hdrbeg, hdrend)
2922 #ifdef ENABLE_DIGEST
2923 || STARTS ("Digest", hdrbeg, hdrend)
2926 || STARTS ("NTLM", hdrbeg, hdrend)
2933 /* Create the HTTP authorization request header. When the
2934 `WWW-Authenticate' response header is seen, according to the
2935 authorization scheme specified in that header (`Basic' and `Digest'
2936 are supported by the current implementation), produce an
2937 appropriate HTTP authorization request header. */
2939 create_authorization_line (const char *au, const char *user,
2940 const char *passwd, const char *method,
2941 const char *path, bool *finished)
2943 /* We are called only with known schemes, so we can dispatch on the
2945 switch (TOUPPER (*au))
2947 case 'B': /* Basic */
2949 return basic_authentication_encode (user, passwd);
2950 #ifdef ENABLE_DIGEST
2951 case 'D': /* Digest */
2953 return digest_authentication_encode (au, user, passwd, method, path);
2956 case 'N': /* NTLM */
2957 if (!ntlm_input (&pconn.ntlm, au))
2962 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2965 /* We shouldn't get here -- this function should be only called
2966 with values approved by known_authentication_scheme_p. */
2974 if (!wget_cookie_jar)
2975 wget_cookie_jar = cookie_jar_new ();
2976 if (opt.cookies_input && !cookies_loaded_p)
2978 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2979 cookies_loaded_p = true;
2986 if (wget_cookie_jar)
2987 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2993 xfree_null (pconn.host);
2994 if (wget_cookie_jar)
2995 cookie_jar_delete (wget_cookie_jar);
3002 test_parse_content_disposition()
3010 { "filename=\"file.ext\"", "file.ext", true },
3011 { "attachment; filename=\"file.ext\"", "file.ext", true },
3012 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
3013 { "attachment", NULL, false },
3016 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3019 bool res = parse_content_disposition (test_array[i].hdrval, &filename);
3021 mu_assert ("test_parse_content_disposition: wrong result",
3022 res == test_array[i].result
3024 || 0 == strcmp (test_array[i].filename, filename)));
3026 /* printf ("test %d: %s\n", i, res == false ? "false" : filename); */
3032 #endif /* TESTING */