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 string (typically an HTTP header) at
859 **SOURCE and advance SOURCE to the next parameter. Return false
860 when there are no more parameters to extract. The name of the
861 parameter is returned in NAME, and the value in VALUE. If the
862 parameter has no value, the 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,
874 const char *p = *source;
876 while (ISSPACE (*p)) ++p;
880 return false; /* no error; nothing more to extract */
885 while (*p && !ISSPACE (*p) && *p != '=' && *p != separator) ++p;
887 if (name->b == name->e)
888 return false; /* empty name: error */
889 while (ISSPACE (*p)) ++p;
890 if (*p == separator || !*p) /* no value */
893 if (*p == separator) ++p;
898 return false; /* error */
900 /* *p is '=', extract value */
902 while (ISSPACE (*p)) ++p;
903 if (*p == '"') /* quoted */
906 while (*p && *p != '"') ++p;
910 /* Currently at closing quote; find the end of param. */
911 while (ISSPACE (*p)) ++p;
912 while (*p && *p != separator) ++p;
916 /* garbage after closed quote, e.g. foo="bar"baz */
922 while (*p && *p != separator) ++p;
924 while (value->e != value->b && ISSPACE (value->e[-1]))
926 if (*p == separator) ++p;
933 #define MAX(p, q) ((p) > (q) ? (p) : (q))
936 parse_content_disposition (const char *hdr, char **filename)
938 param_token name, value;
939 while (extract_param (&hdr, &name, &value, ';'))
940 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
942 /* Make the file name begin at the last slash or backslash. */
943 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
944 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
945 if (last_slash && last_bs)
946 value.b = 1 + MAX (last_slash, last_bs);
947 else if (last_slash || last_bs)
948 value.b = 1 + (last_slash ? last_slash : last_bs);
949 if (value.b == value.e)
951 *filename = strdupdelim (value.b, value.e);
957 /* Persistent connections. Currently, we cache the most recently used
958 connection as persistent, provided that the HTTP server agrees to
959 make it such. The persistence data is stored in the variables
960 below. Ideally, it should be possible to cache an arbitrary fixed
961 number of these connections. */
963 /* Whether a persistent connection is active. */
964 static bool pconn_active;
967 /* The socket of the connection. */
970 /* Host and port of the currently active persistent connection. */
974 /* Whether a ssl handshake has occoured on this connection. */
977 /* Whether the connection was authorized. This is only done by
978 NTLM, which authorizes *connections* rather than individual
979 requests. (That practice is peculiar for HTTP, but it is a
980 useful optimization.) */
984 /* NTLM data of the current connection. */
985 struct ntlmdata ntlm;
989 /* Mark the persistent connection as invalid and free the resources it
990 uses. This is used by the CLOSE_* macros after they forcefully
991 close a registered persistent connection. */
994 invalidate_persistent (void)
996 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
997 pconn_active = false;
998 fd_close (pconn.socket);
1003 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1004 persistent. This will enable someone to use the same connection
1005 later. In the context of HTTP, this must be called only AFTER the
1006 response has been received and the server has promised that the
1007 connection will remain alive.
1009 If a previous connection was persistent, it is closed. */
1012 register_persistent (const char *host, int port, int fd, bool ssl)
1016 if (pconn.socket == fd)
1018 /* The connection FD is already registered. */
1023 /* The old persistent connection is still active; close it
1024 first. This situation arises whenever a persistent
1025 connection exists, but we then connect to a different
1026 host, and try to register a persistent connection to that
1028 invalidate_persistent ();
1032 pconn_active = true;
1034 pconn.host = xstrdup (host);
1037 pconn.authorized = false;
1039 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1042 /* Return true if a persistent connection is available for connecting
1046 persistent_available_p (const char *host, int port, bool ssl,
1047 bool *host_lookup_failed)
1049 /* First, check whether a persistent connection is active at all. */
1053 /* If we want SSL and the last connection wasn't or vice versa,
1054 don't use it. Checking for host and port is not enough because
1055 HTTP and HTTPS can apparently coexist on the same port. */
1056 if (ssl != pconn.ssl)
1059 /* If we're not connecting to the same port, we're not interested. */
1060 if (port != pconn.port)
1063 /* If the host is the same, we're in business. If not, there is
1064 still hope -- read below. */
1065 if (0 != strcasecmp (host, pconn.host))
1067 /* Check if pconn.socket is talking to HOST under another name.
1068 This happens often when both sites are virtual hosts
1069 distinguished only by name and served by the same network
1070 interface, and hence the same web server (possibly set up by
1071 the ISP and serving many different web sites). This
1072 admittedly unconventional optimization does not contradict
1073 HTTP and works well with popular server software. */
1077 struct address_list *al;
1080 /* Don't try to talk to two different SSL sites over the same
1081 secure connection! (Besides, it's not clear that
1082 name-based virtual hosting is even possible with SSL.) */
1085 /* If pconn.socket's peer is one of the IP addresses HOST
1086 resolves to, pconn.socket is for all intents and purposes
1087 already talking to HOST. */
1089 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1091 /* Can't get the peer's address -- something must be very
1092 wrong with the connection. */
1093 invalidate_persistent ();
1096 al = lookup_host (host, 0);
1099 *host_lookup_failed = true;
1103 found = address_list_contains (al, &ip);
1104 address_list_release (al);
1109 /* The persistent connection's peer address was found among the
1110 addresses HOST resolved to; therefore, pconn.sock is in fact
1111 already talking to HOST -- no need to reconnect. */
1114 /* Finally, check whether the connection is still open. This is
1115 important because most servers implement liberal (short) timeout
1116 on persistent connections. Wget can of course always reconnect
1117 if the connection doesn't work out, but it's nicer to know in
1118 advance. This test is a logical followup of the first test, but
1119 is "expensive" and therefore placed at the end of the list.
1121 (Current implementation of test_socket_open has a nice side
1122 effect that it treats sockets with pending data as "closed".
1123 This is exactly what we want: if a broken server sends message
1124 body in response to HEAD, or if it sends more than conent-length
1125 data, we won't reuse the corrupted connection.) */
1127 if (!test_socket_open (pconn.socket))
1129 /* Oops, the socket is no longer open. Now that we know that,
1130 let's invalidate the persistent connection before returning
1132 invalidate_persistent ();
1139 /* The idea behind these two CLOSE macros is to distinguish between
1140 two cases: one when the job we've been doing is finished, and we
1141 want to close the connection and leave, and two when something is
1142 seriously wrong and we're closing the connection as part of
1145 In case of keep_alive, CLOSE_FINISH should leave the connection
1146 open, while CLOSE_INVALIDATE should still close it.
1148 Note that the semantics of the flag `keep_alive' is "this
1149 connection *will* be reused (the server has promised not to close
1150 the connection once we're done)", while the semantics of
1151 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1152 active, registered connection". */
1154 #define CLOSE_FINISH(fd) do { \
1157 if (pconn_active && (fd) == pconn.socket) \
1158 invalidate_persistent (); \
1167 #define CLOSE_INVALIDATE(fd) do { \
1168 if (pconn_active && (fd) == pconn.socket) \
1169 invalidate_persistent (); \
1177 wgint len; /* received length */
1178 wgint contlen; /* expected length */
1179 wgint restval; /* the restart value */
1180 int res; /* the result of last read */
1181 char *rderrmsg; /* error message from read error */
1182 char *newloc; /* new location (redirection) */
1183 char *remote_time; /* remote time-stamp string */
1184 char *error; /* textual HTTP error */
1185 int statcode; /* status code */
1186 wgint rd_size; /* amount of data read from socket */
1187 double dltime; /* time it took to download the data */
1188 const char *referer; /* value of the referer header. */
1189 char *local_file; /* local file name. */
1190 bool timestamp_checked; /* true if pre-download time-stamping checks
1191 * have already been performed */
1192 char *orig_file_name; /* name of file to compare for time-stamping
1193 * (might be != local_file if -K is set) */
1194 wgint orig_file_size; /* size of file to compare for time-stamping */
1195 time_t orig_file_tstamp; /* time-stamp of file to compare for
1200 free_hstat (struct http_stat *hs)
1202 xfree_null (hs->newloc);
1203 xfree_null (hs->remote_time);
1204 xfree_null (hs->error);
1205 xfree_null (hs->rderrmsg);
1206 xfree_null (hs->local_file);
1207 xfree_null (hs->orig_file_name);
1209 /* Guard against being called twice. */
1211 hs->remote_time = NULL;
1215 static char *create_authorization_line (const char *, const char *,
1216 const char *, const char *,
1217 const char *, bool *);
1218 static char *basic_authentication_encode (const char *, const char *);
1219 static bool known_authentication_scheme_p (const char *, const char *);
1220 static void load_cookies (void);
1222 #define BEGINS_WITH(line, string_constant) \
1223 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1224 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1225 || !line[sizeof (string_constant) - 1]))
1227 #define SET_USER_AGENT(req) do { \
1228 if (!opt.useragent) \
1229 request_set_header (req, "User-Agent", \
1230 aprintf ("Wget/%s", version_string), rel_value); \
1231 else if (*opt.useragent) \
1232 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1235 /* The flags that allow clobbering the file (opening with "wb").
1236 Defined here to avoid repetition later. #### This will require
1238 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1239 || opt.dirstruct || opt.output_document)
1241 /* Retrieve a document through HTTP protocol. It recognizes status
1242 code, and correctly handles redirections. It closes the network
1243 socket. If it receives an error from the functions below it, it
1244 will print it if there is enough information to do so (almost
1245 always), returning the error to the caller (i.e. http_loop).
1247 Various HTTP parameters are stored to hs.
1249 If PROXY is non-NULL, the connection will be made to the proxy
1250 server, and u->url will be requested. */
1252 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1254 struct request *req;
1257 char *user, *passwd;
1261 wgint contlen, contrange;
1268 /* Set to 1 when the authorization has failed permanently and should
1269 not be tried again. */
1270 bool auth_finished = false;
1272 /* Whether NTLM authentication is used for this request. */
1273 bool ntlm_seen = false;
1275 /* Whether our connection to the remote host is through SSL. */
1276 bool using_ssl = false;
1278 /* Whether a HEAD request will be issued (as opposed to GET or
1280 bool head_only = !!(*dt & HEAD_ONLY);
1283 struct response *resp;
1287 /* Whether this connection will be kept alive after the HTTP request
1291 /* Whether keep-alive should be inhibited.
1293 RFC 2068 requests that 1.0 clients not send keep-alive requests
1294 to proxies. This is because many 1.0 proxies do not interpret
1295 the Connection header and transfer it to the remote server,
1296 causing it to not close the connection and leave both the proxy
1297 and the client hanging. */
1298 bool inhibit_keep_alive =
1299 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1301 /* Headers sent when using POST. */
1302 wgint post_data_size = 0;
1304 bool host_lookup_failed = false;
1307 if (u->scheme == SCHEME_HTTPS)
1309 /* Initialize the SSL context. After this has once been done,
1310 it becomes a no-op. */
1313 scheme_disable (SCHEME_HTTPS);
1314 logprintf (LOG_NOTQUIET,
1315 _("Disabling SSL due to encountered errors.\n"));
1316 return SSLINITFAILED;
1319 #endif /* HAVE_SSL */
1321 /* Initialize certain elements of struct http_stat. */
1325 hs->rderrmsg = NULL;
1327 hs->remote_time = NULL;
1332 /* Prepare the request to send. */
1334 req = request_new ();
1337 const char *meth = "GET";
1340 else if (opt.post_file_name || opt.post_data)
1342 /* Use the full path, i.e. one that includes the leading slash and
1343 the query string. E.g. if u->path is "foo/bar" and u->query is
1344 "param=value", full_path will be "/foo/bar?param=value". */
1347 /* When using SSL over proxy, CONNECT establishes a direct
1348 connection to the HTTPS server. Therefore use the same
1349 argument as when talking to the server directly. */
1350 && u->scheme != SCHEME_HTTPS
1353 meth_arg = xstrdup (u->url);
1355 meth_arg = url_full_path (u);
1356 request_set_method (req, meth, meth_arg);
1359 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1360 if (*dt & SEND_NOCACHE)
1361 request_set_header (req, "Pragma", "no-cache", rel_none);
1363 request_set_header (req, "Range",
1364 aprintf ("bytes=%s-",
1365 number_to_static_string (hs->restval)),
1367 SET_USER_AGENT (req);
1368 request_set_header (req, "Accept", "*/*", rel_none);
1370 /* Find the username and password for authentication. */
1373 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1374 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1375 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1379 /* We have the username and the password, but haven't tried
1380 any authorization yet. Let's see if the "Basic" method
1381 works. If not, we'll come back here and construct a
1382 proper authorization method with the right challenges.
1384 If we didn't employ this kind of logic, every URL that
1385 requires authorization would have to be processed twice,
1386 which is very suboptimal and generates a bunch of false
1387 "unauthorized" errors in the server log.
1389 #### But this logic also has a serious problem when used
1390 with stronger authentications: we *first* transmit the
1391 username and the password in clear text, and *then* attempt a
1392 stronger authentication scheme. That cannot be right! We
1393 are only fortunate that almost everyone still uses the
1394 `Basic' scheme anyway.
1396 There should be an option to prevent this from happening, for
1397 those who use strong authentication schemes and value their
1399 request_set_header (req, "Authorization",
1400 basic_authentication_encode (user, passwd),
1407 char *proxy_user, *proxy_passwd;
1408 /* For normal username and password, URL components override
1409 command-line/wgetrc parameters. With proxy
1410 authentication, it's the reverse, because proxy URLs are
1411 normally the "permanent" ones, so command-line args
1412 should take precedence. */
1413 if (opt.proxy_user && opt.proxy_passwd)
1415 proxy_user = opt.proxy_user;
1416 proxy_passwd = opt.proxy_passwd;
1420 proxy_user = proxy->user;
1421 proxy_passwd = proxy->passwd;
1423 /* #### This does not appear right. Can't the proxy request,
1424 say, `Digest' authentication? */
1425 if (proxy_user && proxy_passwd)
1426 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1428 /* If we're using a proxy, we will be connecting to the proxy
1432 /* Proxy authorization over SSL is handled below. */
1434 if (u->scheme != SCHEME_HTTPS)
1436 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1439 /* Generate the Host header, HOST:PORT. Take into account that:
1441 - Broken server-side software often doesn't recognize the PORT
1442 argument, so we must generate "Host: www.server.com" instead of
1443 "Host: www.server.com:80" (and likewise for https port).
1445 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1446 becomes ambiguous and needs to be rewritten as "Host:
1447 [3ffe:8100:200:2::2]:1234". */
1449 /* Formats arranged for hfmt[add_port][add_squares]. */
1450 static const char *hfmt[][2] = {
1451 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1453 int add_port = u->port != scheme_default_port (u->scheme);
1454 int add_squares = strchr (u->host, ':') != NULL;
1455 request_set_header (req, "Host",
1456 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1460 if (!inhibit_keep_alive)
1461 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1464 request_set_header (req, "Cookie",
1465 cookie_header (wget_cookie_jar,
1466 u->host, u->port, u->path,
1468 u->scheme == SCHEME_HTTPS
1475 if (opt.post_data || opt.post_file_name)
1477 request_set_header (req, "Content-Type",
1478 "application/x-www-form-urlencoded", rel_none);
1480 post_data_size = strlen (opt.post_data);
1483 post_data_size = file_size (opt.post_file_name);
1484 if (post_data_size == -1)
1486 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1487 opt.post_file_name, strerror (errno));
1491 request_set_header (req, "Content-Length",
1492 xstrdup (number_to_static_string (post_data_size)),
1496 /* Add the user headers. */
1497 if (opt.user_headers)
1500 for (i = 0; opt.user_headers[i]; i++)
1501 request_set_user_header (req, opt.user_headers[i]);
1505 /* We need to come back here when the initial attempt to retrieve
1506 without authorization header fails. (Expected to happen at least
1507 for the Digest authorization scheme.) */
1511 /* Establish the connection. */
1513 if (!inhibit_keep_alive)
1515 /* Look for a persistent connection to target host, unless a
1516 proxy is used. The exception is when SSL is in use, in which
1517 case the proxy is nothing but a passthrough to the target
1518 host, registered as a connection to the latter. */
1519 struct url *relevant = conn;
1521 if (u->scheme == SCHEME_HTTPS)
1525 if (persistent_available_p (relevant->host, relevant->port,
1527 relevant->scheme == SCHEME_HTTPS,
1531 &host_lookup_failed))
1533 sock = pconn.socket;
1534 using_ssl = pconn.ssl;
1535 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1536 escnonprint (pconn.host), pconn.port);
1537 DEBUGP (("Reusing fd %d.\n", sock));
1538 if (pconn.authorized)
1539 /* If the connection is already authorized, the "Basic"
1540 authorization added by code above is unnecessary and
1542 request_remove_header (req, "Authorization");
1548 /* In its current implementation, persistent_available_p will
1549 look up conn->host in some cases. If that lookup failed, we
1550 don't need to bother with connect_to_host. */
1551 if (host_lookup_failed)
1557 sock = connect_to_host (conn->host, conn->port);
1566 return (retryable_socket_connect_error (errno)
1567 ? CONERROR : CONIMPOSSIBLE);
1571 if (proxy && u->scheme == SCHEME_HTTPS)
1573 /* When requesting SSL URLs through proxies, use the
1574 CONNECT method to request passthrough. */
1575 struct request *connreq = request_new ();
1576 request_set_method (connreq, "CONNECT",
1577 aprintf ("%s:%d", u->host, u->port));
1578 SET_USER_AGENT (connreq);
1581 request_set_header (connreq, "Proxy-Authorization",
1582 proxyauth, rel_value);
1583 /* Now that PROXYAUTH is part of the CONNECT request,
1584 zero it out so we don't send proxy authorization with
1585 the regular request below. */
1588 /* Examples in rfc2817 use the Host header in CONNECT
1589 requests. I don't see how that gains anything, given
1590 that the contents of Host would be exactly the same as
1591 the contents of CONNECT. */
1593 write_error = request_send (connreq, sock);
1594 request_free (connreq);
1595 if (write_error < 0)
1597 CLOSE_INVALIDATE (sock);
1601 head = read_http_response_head (sock);
1604 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1606 CLOSE_INVALIDATE (sock);
1615 DEBUGP (("proxy responded with: [%s]\n", head));
1617 resp = resp_new (head);
1618 statcode = resp_status (resp, &message);
1621 if (statcode != 200)
1624 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1625 message ? escnonprint (message) : "?");
1626 xfree_null (message);
1629 xfree_null (message);
1631 /* SOCK is now *really* connected to u->host, so update CONN
1632 to reflect this. That way register_persistent will
1633 register SOCK as being connected to u->host:u->port. */
1637 if (conn->scheme == SCHEME_HTTPS)
1639 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1646 #endif /* HAVE_SSL */
1649 /* Send the request to server. */
1650 write_error = request_send (req, sock);
1652 if (write_error >= 0)
1656 DEBUGP (("[POST data: %s]\n", opt.post_data));
1657 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1659 else if (opt.post_file_name && post_data_size != 0)
1660 write_error = post_file (sock, opt.post_file_name, post_data_size);
1663 if (write_error < 0)
1665 CLOSE_INVALIDATE (sock);
1669 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1670 proxy ? "Proxy" : "HTTP");
1675 head = read_http_response_head (sock);
1680 logputs (LOG_NOTQUIET, _("No data received.\n"));
1681 CLOSE_INVALIDATE (sock);
1687 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1689 CLOSE_INVALIDATE (sock);
1694 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1696 resp = resp_new (head);
1698 /* Check for status line. */
1700 statcode = resp_status (resp, &message);
1701 if (!opt.server_response)
1702 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1703 message ? escnonprint (message) : "");
1706 logprintf (LOG_VERBOSE, "\n");
1707 print_server_response (resp, " ");
1710 /* Determine the local filename if needed. Notice that if -O is used
1711 * hstat.local_file is set by http_loop to the argument of -O. */
1712 if (!hs->local_file)
1714 /* Honor Content-Disposition whether possible. */
1715 if (!resp_header_copy (resp, "Content-Disposition", hdrval, sizeof (hdrval))
1716 || !parse_content_disposition (hdrval, &hs->local_file))
1718 /* Choose filename according to URL name. */
1719 hs->local_file = url_file_name (u);
1723 /* TODO: perform this check only once. */
1724 if (opt.noclobber && file_exists_p (hs->local_file))
1726 /* If opt.noclobber is turned on and file already exists, do not
1727 retrieve the file */
1728 logprintf (LOG_VERBOSE, _("\
1729 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1730 /* If the file is there, we suppose it's retrieved OK. */
1733 /* #### Bogusness alert. */
1734 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1735 if (has_html_suffix_p (hs->local_file))
1741 /* Support timestamping */
1742 /* TODO: move this code out of gethttp. */
1743 if (opt.timestamping && !hs->timestamp_checked)
1745 size_t filename_len = strlen (hs->local_file);
1746 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1747 bool local_dot_orig_file_exists = false;
1748 char *local_filename = NULL;
1751 if (opt.backup_converted)
1752 /* If -K is specified, we'll act on the assumption that it was specified
1753 last time these files were downloaded as well, and instead of just
1754 comparing local file X against server file X, we'll compare local
1755 file X.orig (if extant, else X) against server file X. If -K
1756 _wasn't_ specified last time, or the server contains files called
1757 *.orig, -N will be back to not operating correctly with -k. */
1759 /* Would a single s[n]printf() call be faster? --dan
1761 Definitely not. sprintf() is horribly slow. It's a
1762 different question whether the difference between the two
1763 affects a program. Usually I'd say "no", but at one
1764 point I profiled Wget, and found that a measurable and
1765 non-negligible amount of time was lost calling sprintf()
1766 in url.c. Replacing sprintf with inline calls to
1767 strcpy() and number_to_string() made a difference.
1769 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1770 memcpy (filename_plus_orig_suffix + filename_len,
1771 ".orig", sizeof (".orig"));
1773 /* Try to stat() the .orig file. */
1774 if (stat (filename_plus_orig_suffix, &st) == 0)
1776 local_dot_orig_file_exists = 1;
1777 local_filename = filename_plus_orig_suffix;
1781 if (!local_dot_orig_file_exists)
1782 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1783 if (stat (hs->local_file, &st) == 0)
1784 local_filename = hs->local_file;
1786 if (local_filename != NULL)
1787 /* There was a local file, so we'll check later to see if the version
1788 the server has is the same version we already have, allowing us to
1791 hs->orig_file_name = xstrdup (local_filename);
1792 hs->orig_file_size = st.st_size;
1793 hs->orig_file_tstamp = st.st_mtime;
1795 /* Modification time granularity is 2 seconds for Windows, so
1796 increase local time by 1 second for later comparison. */
1797 ++hs->orig_file_tstamp;
1802 if (!opt.ignore_length
1803 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1807 parsed = str_to_wgint (hdrval, NULL, 10);
1808 if (parsed == WGINT_MAX && errno == ERANGE)
1810 #### If Content-Length is out of range, it most likely
1811 means that the file is larger than 2G and that we're
1812 compiled without LFS. In that case we should probably
1813 refuse to even attempt to download the file. */
1819 /* Check for keep-alive related responses. */
1820 if (!inhibit_keep_alive && contlen != -1)
1822 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1824 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1826 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1831 /* The server has promised that it will not close the connection
1832 when we're done. This means that we can register it. */
1833 register_persistent (conn->host, conn->port, sock, using_ssl);
1835 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1837 /* Authorization is required. */
1838 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1839 CLOSE_FINISH (sock);
1841 CLOSE_INVALIDATE (sock);
1842 pconn.authorized = false;
1843 if (!auth_finished && (user && passwd))
1845 /* IIS sends multiple copies of WWW-Authenticate, one with
1846 the value "negotiate", and other(s) with data. Loop over
1847 all the occurrences and pick the one we recognize. */
1849 const char *wabeg, *waend;
1850 char *www_authenticate = NULL;
1852 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1853 &wabeg, &waend)) != -1;
1855 if (known_authentication_scheme_p (wabeg, waend))
1857 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1861 if (!www_authenticate)
1862 /* If the authentication header is missing or
1863 unrecognized, there's no sense in retrying. */
1864 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1865 else if (BEGINS_WITH (www_authenticate, "Basic"))
1866 /* If the authentication scheme is "Basic", which we send
1867 by default, there's no sense in retrying either. (This
1868 should be changed when we stop sending "Basic" data by
1874 pth = url_full_path (u);
1875 request_set_header (req, "Authorization",
1876 create_authorization_line (www_authenticate,
1878 request_method (req),
1882 if (BEGINS_WITH (www_authenticate, "NTLM"))
1885 goto retry_with_auth;
1888 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1892 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1894 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1896 pconn.authorized = true;
1900 hs->statcode = statcode;
1902 hs->error = xstrdup (_("Malformed status line"));
1904 hs->error = xstrdup (_("(no description)"));
1906 hs->error = xstrdup (message);
1907 xfree_null (message);
1909 type = resp_header_strdup (resp, "Content-Type");
1912 char *tmp = strchr (type, ';');
1915 while (tmp > type && ISSPACE (tmp[-1]))
1920 hs->newloc = resp_header_strdup (resp, "Location");
1921 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1923 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1927 const char *scbeg, *scend;
1928 /* The jar should have been created by now. */
1929 assert (wget_cookie_jar != NULL);
1931 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1932 &scbeg, &scend)) != -1;
1935 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1936 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1937 u->path, set_cookie);
1941 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1943 wgint first_byte_pos, last_byte_pos, entity_length;
1944 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1946 contrange = first_byte_pos;
1950 /* 20x responses are counted among successful by default. */
1951 if (H_20X (statcode))
1954 /* Return if redirected. */
1955 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1957 /* RFC2068 says that in case of the 300 (multiple choices)
1958 response, the server can output a preferred URL through
1959 `Location' header; otherwise, the request should be treated
1960 like GET. So, if the location is set, it will be a
1961 redirection; otherwise, just proceed normally. */
1962 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1966 logprintf (LOG_VERBOSE,
1967 _("Location: %s%s\n"),
1968 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1969 hs->newloc ? _(" [following]") : "");
1970 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1971 CLOSE_FINISH (sock);
1973 CLOSE_INVALIDATE (sock);
1979 /* If content-type is not given, assume text/html. This is because
1980 of the multitude of broken CGI's that "forget" to generate the
1983 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1984 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1989 if (opt.html_extension && (*dt & TEXTHTML))
1990 /* -E / --html-extension / html_extension = on was specified, and this is a
1991 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1992 already the file's suffix, tack on ".html". */
1994 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
1996 if (last_period_in_local_filename == NULL
1997 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1998 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2000 int local_filename_len = strlen (hs->local_file);
2001 /* Resize the local file, allowing for ".html" preceded by
2002 optional ".NUMBER". */
2003 hs->local_file = xrealloc (hs->local_file,
2004 local_filename_len + 24 + sizeof (".html"));
2005 strcpy(hs->local_file + local_filename_len, ".html");
2006 /* If clobbering is not allowed and the file, as named,
2007 exists, tack on ".NUMBER.html" instead. */
2008 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2012 sprintf (hs->local_file + local_filename_len,
2013 ".%d.html", ext_num++);
2014 while (file_exists_p (hs->local_file));
2016 *dt |= ADDED_HTML_EXTENSION;
2020 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2022 /* If `-c' is in use and the file has been fully downloaded (or
2023 the remote file has shrunk), Wget effectively requests bytes
2024 after the end of file and the server response with 416. */
2025 logputs (LOG_VERBOSE, _("\
2026 \n The file is already fully retrieved; nothing to do.\n\n"));
2027 /* In case the caller inspects. */
2030 /* Mark as successfully retrieved. */
2033 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2034 might be more bytes in the body. */
2035 return RETRUNNEEDED;
2037 if ((contrange != 0 && contrange != hs->restval)
2038 || (H_PARTIAL (statcode) && !contrange))
2040 /* The Range request was somehow misunderstood by the server.
2043 CLOSE_INVALIDATE (sock);
2046 hs->contlen = contlen + contrange;
2052 /* No need to print this output if the body won't be
2053 downloaded at all, or if the original server response is
2055 logputs (LOG_VERBOSE, _("Length: "));
2058 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2059 if (contlen + contrange >= 1024)
2060 logprintf (LOG_VERBOSE, " (%s)",
2061 human_readable (contlen + contrange));
2064 if (contlen >= 1024)
2065 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2066 number_to_static_string (contlen),
2067 human_readable (contlen));
2069 logprintf (LOG_VERBOSE, _(", %s remaining"),
2070 number_to_static_string (contlen));
2074 logputs (LOG_VERBOSE,
2075 opt.ignore_length ? _("ignored") : _("unspecified"));
2077 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2079 logputs (LOG_VERBOSE, "\n");
2083 type = NULL; /* We don't need it any more. */
2085 /* Return if we have no intention of further downloading. */
2086 if (!(*dt & RETROKF) || head_only)
2088 /* In case the caller cares to look... */
2093 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2094 servers not to send body in response to a HEAD request, and
2095 those that do will likely be caught by test_socket_open.
2096 If not, they can be worked around using
2097 `--no-http-keep-alive'. */
2098 CLOSE_FINISH (sock);
2099 else if (keep_alive && skip_short_body (sock, contlen))
2100 /* Successfully skipped the body; also keep using the socket. */
2101 CLOSE_FINISH (sock);
2103 CLOSE_INVALIDATE (sock);
2104 return RETRFINISHED;
2107 /* Print fetch message, if opt.verbose. */
2110 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2111 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2114 /* Open the local file. */
2117 mkalldirs (hs->local_file);
2119 rotate_backups (hs->local_file);
2121 fp = fopen (hs->local_file, "ab");
2122 else if (ALLOW_CLOBBER)
2123 fp = fopen (hs->local_file, "wb");
2126 fp = fopen_excl (hs->local_file, true);
2127 if (!fp && errno == EEXIST)
2129 /* We cannot just invent a new name and use it (which is
2130 what functions like unique_create typically do)
2131 because we told the user we'd use this name.
2132 Instead, return and retry the download. */
2133 logprintf (LOG_NOTQUIET,
2134 _("%s has sprung into existence.\n"),
2136 CLOSE_INVALIDATE (sock);
2137 return FOPEN_EXCL_ERR;
2142 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2143 CLOSE_INVALIDATE (sock);
2150 /* This confuses the timestamping code that checks for file size.
2151 #### The timestamping code should be smarter about file size. */
2152 if (opt.save_headers && hs->restval == 0)
2153 fwrite (head, 1, strlen (head), fp);
2155 /* Now we no longer need to store the response header. */
2158 /* Download the request body. */
2161 /* If content-length is present, read that much; otherwise, read
2162 until EOF. The HTTP spec doesn't require the server to
2163 actually close the connection when it's done sending data. */
2164 flags |= rb_read_exactly;
2165 if (hs->restval > 0 && contrange == 0)
2166 /* If the server ignored our range request, instruct fd_read_body
2167 to skip the first RESTVAL bytes of body. */
2168 flags |= rb_skip_startpos;
2169 hs->len = hs->restval;
2171 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2172 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2176 CLOSE_FINISH (sock);
2180 hs->rderrmsg = xstrdup (fd_errstr (sock));
2181 CLOSE_INVALIDATE (sock);
2188 return RETRFINISHED;
2191 /* The genuine HTTP loop! This is the part where the retrieval is
2192 retried, and retried, and retried, and... */
2194 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2195 int *dt, struct url *proxy)
2198 bool got_head = false; /* used for time-stamping */
2201 uerr_t err, ret = TRYLIMEXC;
2202 time_t tmr = -1; /* remote time-stamp */
2203 wgint local_size = 0; /* the size of the local file */
2204 struct http_stat hstat; /* HTTP status */
2207 /* Assert that no value for *LOCAL_FILE was passed. */
2208 assert (local_file == NULL || *local_file == NULL);
2210 /* Set LOCAL_FILE parameter. */
2211 if (local_file && opt.output_document)
2212 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2214 /* Reset NEWLOC parameter. */
2217 /* This used to be done in main(), but it's a better idea to do it
2218 here so that we don't go through the hoops if we're just using
2223 /* Warn on (likely bogus) wildcard usage in HTTP. */
2224 if (opt.ftp_glob && has_wildcards_p (u->path))
2225 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2227 /* Setup hstat struct. */
2229 hstat.referer = referer;
2231 if (opt.output_document)
2232 hstat.local_file = xstrdup (opt.output_document);
2234 /* Reset the counter. */
2237 /* Reset the document type. */
2243 /* Increment the pass counter. */
2245 sleep_between_retrievals (count);
2247 /* Get the current time string. */
2248 tms = time_str (NULL);
2250 /* Print fetch message, if opt.verbose. */
2253 char *hurl = url_string (u, true);
2258 sprintf (tmp, _("(try:%2d)"), count);
2259 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2264 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2269 ws_changetitle (hurl);
2274 /* Default document type is empty. However, if spider mode is
2275 on or time-stamping is employed, HEAD_ONLY commands is
2276 encoded within *dt. */
2277 if (opt.spider || (opt.timestamping && !got_head))
2282 /* Decide whether or not to restart. */
2284 && stat (hstat.local_file, &st) == 0
2285 && S_ISREG (st.st_mode))
2286 /* When -c is used, continue from on-disk size. (Can't use
2287 hstat.len even if count>1 because we don't want a failed
2288 first attempt to clobber existing data.) */
2289 hstat.restval = st.st_size;
2291 /* otherwise, continue where the previous try left off */
2292 hstat.restval = hstat.len;
2296 /* Decide whether to send the no-cache directive. We send it in
2298 a) we're using a proxy, and we're past our first retrieval.
2299 Some proxies are notorious for caching incomplete data, so
2300 we require a fresh get.
2301 b) caching is explicitly inhibited. */
2302 if ((proxy && count > 1) /* a */
2303 || !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 printwhat (count, opt.ntry);
2328 case FWRITEERR: case FOPENERR:
2329 /* Another fatal error. */
2330 logputs (LOG_VERBOSE, "\n");
2331 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2332 hstat.local_file, strerror (errno));
2333 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2334 case SSLINITFAILED: case CONTNOTSUPPORTED:
2335 /* Fatal errors just return from the function. */
2339 /* Another fatal error. */
2340 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2344 /* Return the new location to the caller. */
2347 logprintf (LOG_NOTQUIET,
2348 _("ERROR: Redirection (%d) without location.\n"),
2358 /* The file was already fully retrieved. */
2362 /* Deal with you later. */
2365 /* All possibilities should have been exhausted. */
2369 if (!(*dt & RETROKF))
2373 /* #### Ugly ugly ugly! */
2374 char *hurl = url_string (u, true);
2375 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2378 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2379 tms, hstat.statcode, escnonprint (hstat.error));
2380 logputs (LOG_VERBOSE, "\n");
2385 /* Did we get the time-stamp? */
2388 if (opt.timestamping && !hstat.remote_time)
2390 logputs (LOG_NOTQUIET, _("\
2391 Last-modified header missing -- time-stamps turned off.\n"));
2393 else if (hstat.remote_time)
2395 /* Convert the date-string into struct tm. */
2396 tmr = http_atotm (hstat.remote_time);
2397 if (tmr == (time_t) (-1))
2398 logputs (LOG_VERBOSE, _("\
2399 Last-modified header invalid -- time-stamp ignored.\n"));
2403 /* The time-stamping section. */
2404 if (opt.timestamping && !got_head)
2406 got_head = true; /* no more time-stamping */
2408 count = 0; /* the retrieve count for HEAD is reset */
2410 if (hstat.remote_time && tmr != (time_t) (-1))
2412 /* Now time-stamping can be used validly. Time-stamping
2413 means that if the sizes of the local and remote file
2414 match, and local file is newer than the remote file,
2415 it will not be retrieved. Otherwise, the normal
2416 download procedure is resumed. */
2417 if (hstat.orig_file_tstamp >= tmr)
2419 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2421 logprintf (LOG_VERBOSE, _("\
2422 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2423 hstat.orig_file_name);
2429 logprintf (LOG_VERBOSE, _("\
2430 The sizes do not match (local %s) -- retrieving.\n"),
2431 number_to_static_string (local_size));
2435 logputs (LOG_VERBOSE,
2436 _("Remote file is newer, retrieving.\n"));
2438 logputs (LOG_VERBOSE, "\n");
2441 /* free_hstat (&hstat); */
2442 hstat.timestamp_checked = true;
2446 if ((tmr != (time_t) (-1))
2448 && ((hstat.len == hstat.contlen) ||
2449 ((hstat.res == 0) && (hstat.contlen == -1))))
2451 /* #### This code repeats in http.c and ftp.c. Move it to a
2453 const char *fl = NULL;
2454 if (opt.output_document)
2456 if (output_stream_regular)
2457 fl = opt.output_document;
2460 fl = hstat.local_file;
2464 /* End of time-stamping section. */
2468 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2469 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);
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);
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);
2543 /* Getting here would mean reading more data than
2544 requested with content-length, which we never do. */
2547 else /* from now on hstat.res can only be -1 */
2549 if (hstat.contlen == -1)
2551 logprintf (LOG_VERBOSE,
2552 _("%s (%s) - Read error at byte %s (%s)."),
2553 tms, tmrate, number_to_static_string (hstat.len),
2555 printwhat (count, opt.ntry);
2558 else /* hstat.res == -1 and contlen is given */
2560 logprintf (LOG_VERBOSE,
2561 _("%s (%s) - Read error at byte %s/%s (%s). "),
2563 number_to_static_string (hstat.len),
2564 number_to_static_string (hstat.contlen),
2566 printwhat (count, opt.ntry);
2572 while (!opt.ntry || (count < opt.ntry));
2576 *local_file = xstrdup (hstat.local_file);
2577 free_hstat (&hstat);
2582 /* Check whether the result of strptime() indicates success.
2583 strptime() returns the pointer to how far it got to in the string.
2584 The processing has been successful if the string is at `GMT' or
2585 `+X', or at the end of the string.
2587 In extended regexp parlance, the function returns 1 if P matches
2588 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2589 can return) is considered a failure and 0 is returned. */
2591 check_end (const char *p)
2595 while (ISSPACE (*p))
2598 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2599 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2605 /* Convert the textual specification of time in TIME_STRING to the
2606 number of seconds since the Epoch.
2608 TIME_STRING can be in any of the three formats RFC2616 allows the
2609 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2610 as well as the time format used in the Set-Cookie header.
2611 Timezones are ignored, and should be GMT.
2613 Return the computed time_t representation, or -1 if the conversion
2616 This function uses strptime with various string formats for parsing
2617 TIME_STRING. This results in a parser that is not as lenient in
2618 interpreting TIME_STRING as I would like it to be. Being based on
2619 strptime, it always allows shortened months, one-digit days, etc.,
2620 but due to the multitude of formats in which time can be
2621 represented, an ideal HTTP time parser would be even more
2622 forgiving. It should completely ignore things like week days and
2623 concentrate only on the various forms of representing years,
2624 months, days, hours, minutes, and seconds. For example, it would
2625 be nice if it accepted ISO 8601 out of the box.
2627 I've investigated free and PD code for this purpose, but none was
2628 usable. getdate was big and unwieldy, and had potential copyright
2629 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2630 distributed with phttpd, is excellent, but we cannot use it because
2631 it is not assigned to the FSF. So I stuck it with strptime. */
2634 http_atotm (const char *time_string)
2636 /* NOTE: Solaris strptime man page claims that %n and %t match white
2637 space, but that's not universally available. Instead, we simply
2638 use ` ' to mean "skip all WS", which works under all strptime
2639 implementations I've tested. */
2641 static const char *time_formats[] = {
2642 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2643 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2644 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2645 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2646 (used in Set-Cookie, defined in the
2647 Netscape cookie specification.) */
2649 const char *oldlocale;
2651 time_t ret = (time_t) -1;
2653 /* Solaris strptime fails to recognize English month names in
2654 non-English locales, which we work around by temporarily setting
2655 locale to C before invoking strptime. */
2656 oldlocale = setlocale (LC_TIME, NULL);
2657 setlocale (LC_TIME, "C");
2659 for (i = 0; i < countof (time_formats); i++)
2663 /* Some versions of strptime use the existing contents of struct
2664 tm to recalculate the date according to format. Zero it out
2665 to prevent stack garbage from influencing strptime. */
2668 if (check_end (strptime (time_string, time_formats[i], &t)))
2675 /* Restore the previous locale. */
2676 setlocale (LC_TIME, oldlocale);
2681 /* Authorization support: We support three authorization schemes:
2683 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2685 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2686 consisting of answering to the server's challenge with the proper
2689 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2690 Stenberg for libcurl. Like digest, NTLM is based on a
2691 challenge-response mechanism, but unlike digest, it is non-standard
2692 (authenticates TCP connections rather than requests), undocumented
2693 and Microsoft-specific. */
2695 /* Create the authentication header contents for the `Basic' scheme.
2696 This is done by encoding the string "USER:PASS" to base64 and
2697 prepending the string "Basic " in front of it. */
2700 basic_authentication_encode (const char *user, const char *passwd)
2703 int len1 = strlen (user) + 1 + strlen (passwd);
2705 t1 = (char *)alloca (len1 + 1);
2706 sprintf (t1, "%s:%s", user, passwd);
2708 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2709 base64_encode (t1, len1, t2);
2711 return concat_strings ("Basic ", t2, (char *) 0);
2714 #define SKIP_WS(x) do { \
2715 while (ISSPACE (*(x))) \
2719 #ifdef ENABLE_DIGEST
2720 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2721 an array of 16 bytes containing the hash keys, and BUF should be a
2722 buffer of 33 writable characters (32 for hex digits plus one for
2723 zero termination). */
2725 dump_hash (char *buf, const unsigned char *hash)
2729 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2731 *buf++ = XNUM_TO_digit (*hash >> 4);
2732 *buf++ = XNUM_TO_digit (*hash & 0xf);
2737 /* Take the line apart to find the challenge, and compose a digest
2738 authorization header. See RFC2069 section 2.1.2. */
2740 digest_authentication_encode (const char *au, const char *user,
2741 const char *passwd, const char *method,
2744 static char *realm, *opaque, *nonce;
2749 { "realm", &realm },
2750 { "opaque", &opaque },
2754 param_token name, value;
2756 realm = opaque = nonce = NULL;
2758 au += 6; /* skip over `Digest' */
2759 while (extract_param (&au, &name, &value, ','))
2762 for (i = 0; i < countof (options); i++)
2763 if (name.e - name.b == strlen (options[i].name)
2764 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2766 *options[i].variable = strdupdelim (value.b, value.e);
2770 if (!realm || !nonce || !user || !passwd || !path || !method)
2773 xfree_null (opaque);
2778 /* Calculate the digest value. */
2780 ALLOCA_MD5_CONTEXT (ctx);
2781 unsigned char hash[MD5_HASHLEN];
2782 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2783 char response_digest[MD5_HASHLEN * 2 + 1];
2785 /* A1BUF = H(user ":" realm ":" password) */
2787 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2788 gen_md5_update ((unsigned char *)":", 1, ctx);
2789 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2790 gen_md5_update ((unsigned char *)":", 1, ctx);
2791 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2792 gen_md5_finish (ctx, hash);
2793 dump_hash (a1buf, hash);
2795 /* A2BUF = H(method ":" path) */
2797 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2798 gen_md5_update ((unsigned char *)":", 1, ctx);
2799 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2800 gen_md5_finish (ctx, hash);
2801 dump_hash (a2buf, hash);
2803 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2805 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2806 gen_md5_update ((unsigned char *)":", 1, ctx);
2807 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2808 gen_md5_update ((unsigned char *)":", 1, ctx);
2809 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2810 gen_md5_finish (ctx, hash);
2811 dump_hash (response_digest, hash);
2813 res = xmalloc (strlen (user)
2818 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2819 + (opaque ? strlen (opaque) : 0)
2821 sprintf (res, "Digest \
2822 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2823 user, realm, nonce, path, response_digest);
2826 char *p = res + strlen (res);
2827 strcat (p, ", opaque=\"");
2834 #endif /* ENABLE_DIGEST */
2836 /* Computing the size of a string literal must take into account that
2837 value returned by sizeof includes the terminating \0. */
2838 #define STRSIZE(literal) (sizeof (literal) - 1)
2840 /* Whether chars in [b, e) begin with the literal string provided as
2841 first argument and are followed by whitespace or terminating \0.
2842 The comparison is case-insensitive. */
2843 #define STARTS(literal, b, e) \
2844 ((e) - (b) >= STRSIZE (literal) \
2845 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2846 && ((e) - (b) == STRSIZE (literal) \
2847 || ISSPACE (b[STRSIZE (literal)])))
2850 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2852 return STARTS ("Basic", hdrbeg, hdrend)
2853 #ifdef ENABLE_DIGEST
2854 || STARTS ("Digest", hdrbeg, hdrend)
2857 || STARTS ("NTLM", hdrbeg, hdrend)
2864 /* Create the HTTP authorization request header. When the
2865 `WWW-Authenticate' response header is seen, according to the
2866 authorization scheme specified in that header (`Basic' and `Digest'
2867 are supported by the current implementation), produce an
2868 appropriate HTTP authorization request header. */
2870 create_authorization_line (const char *au, const char *user,
2871 const char *passwd, const char *method,
2872 const char *path, bool *finished)
2874 /* We are called only with known schemes, so we can dispatch on the
2876 switch (TOUPPER (*au))
2878 case 'B': /* Basic */
2880 return basic_authentication_encode (user, passwd);
2881 #ifdef ENABLE_DIGEST
2882 case 'D': /* Digest */
2884 return digest_authentication_encode (au, user, passwd, method, path);
2887 case 'N': /* NTLM */
2888 if (!ntlm_input (&pconn.ntlm, au))
2893 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2896 /* We shouldn't get here -- this function should be only called
2897 with values approved by known_authentication_scheme_p. */
2905 if (!wget_cookie_jar)
2906 wget_cookie_jar = cookie_jar_new ();
2907 if (opt.cookies_input && !cookies_loaded_p)
2909 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2910 cookies_loaded_p = true;
2917 if (wget_cookie_jar)
2918 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2924 xfree_null (pconn.host);
2925 if (wget_cookie_jar)
2926 cookie_jar_delete (wget_cookie_jar);
2933 test_parse_content_disposition()
2941 { "filename=\"file.ext\"", "file.ext", true },
2942 { "attachment; filename=\"file.ext\"", "file.ext", true },
2943 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
2944 { "attachment", NULL, false },
2947 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
2950 bool res = parse_content_disposition (test_array[i].hdrval, &filename);
2952 mu_assert ("test_parse_content_disposition: wrong result",
2953 res == test_array[i].result
2955 || 0 == strcmp (test_array[i].filename, filename)));
2957 /* printf ("test %d: %s\n", i, res == false ? "false" : filename); */
2963 #endif /* TESTING */