2 Copyright (C) 1996-2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
55 # include "http-ntlm.h"
67 extern char *version_string;
70 # define MIN(x, y) ((x) > (y) ? (y) : (x))
74 static bool cookies_loaded_p;
75 static struct cookie_jar *wget_cookie_jar;
77 #define TEXTHTML_S "text/html"
78 #define TEXTXHTML_S "application/xhtml+xml"
80 /* Some status code validation macros: */
81 #define H_20X(x) (((x) >= 200) && ((x) < 300))
82 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
83 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
84 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
85 || (x) == HTTP_STATUS_SEE_OTHER \
86 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
88 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
90 #define HTTP_STATUS_OK 200
91 #define HTTP_STATUS_CREATED 201
92 #define HTTP_STATUS_ACCEPTED 202
93 #define HTTP_STATUS_NO_CONTENT 204
94 #define HTTP_STATUS_PARTIAL_CONTENTS 206
96 /* Redirection 3xx. */
97 #define HTTP_STATUS_MULTIPLE_CHOICES 300
98 #define HTTP_STATUS_MOVED_PERMANENTLY 301
99 #define HTTP_STATUS_MOVED_TEMPORARILY 302
100 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
101 #define HTTP_STATUS_NOT_MODIFIED 304
102 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
104 /* Client error 4xx. */
105 #define HTTP_STATUS_BAD_REQUEST 400
106 #define HTTP_STATUS_UNAUTHORIZED 401
107 #define HTTP_STATUS_FORBIDDEN 403
108 #define HTTP_STATUS_NOT_FOUND 404
109 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
111 /* Server errors 5xx. */
112 #define HTTP_STATUS_INTERNAL 500
113 #define HTTP_STATUS_NOT_IMPLEMENTED 501
114 #define HTTP_STATUS_BAD_GATEWAY 502
115 #define HTTP_STATUS_UNAVAILABLE 503
118 rel_none, rel_name, rel_value, rel_both
125 struct request_header {
127 enum rp release_policy;
129 int hcount, hcapacity;
132 /* Create a new, empty request. At least request_set_method must be
133 called before the request can be used. */
135 static struct request *
138 struct request *req = xnew0 (struct request);
140 req->headers = xnew_array (struct request_header, req->hcapacity);
144 /* Set the request's method and its arguments. METH should be a
145 literal string (or it should outlive the request) because it will
146 not be freed. ARG will be freed by request_free. */
149 request_set_method (struct request *req, const char *meth, char *arg)
155 /* Return the method string passed with the last call to
156 request_set_method. */
159 request_method (const struct request *req)
164 /* Free one header according to the release policy specified with
165 request_set_header. */
168 release_header (struct request_header *hdr)
170 switch (hdr->release_policy)
187 /* Set the request named NAME to VALUE. Specifically, this means that
188 a "NAME: VALUE\r\n" header line will be used in the request. If a
189 header with the same name previously existed in the request, its
190 value will be replaced by this one. A NULL value means do nothing.
192 RELEASE_POLICY determines whether NAME and VALUE should be released
193 (freed) with request_free. Allowed values are:
195 - rel_none - don't free NAME or VALUE
196 - rel_name - free NAME when done
197 - rel_value - free VALUE when done
198 - rel_both - free both NAME and VALUE when done
200 Setting release policy is useful when arguments come from different
201 sources. For example:
203 // Don't free literal strings!
204 request_set_header (req, "Pragma", "no-cache", rel_none);
206 // Don't free a global variable, we'll need it later.
207 request_set_header (req, "Referer", opt.referer, rel_none);
209 // Value freshly allocated, free it when done.
210 request_set_header (req, "Range",
211 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
216 request_set_header (struct request *req, char *name, char *value,
217 enum rp release_policy)
219 struct request_header *hdr;
224 /* A NULL value is a no-op; if freeing the name is requested,
225 free it now to avoid leaks. */
226 if (release_policy == rel_name || release_policy == rel_both)
231 for (i = 0; i < req->hcount; i++)
233 hdr = &req->headers[i];
234 if (0 == strcasecmp (name, hdr->name))
236 /* Replace existing header. */
237 release_header (hdr);
240 hdr->release_policy = release_policy;
245 /* Install new header. */
247 if (req->hcount >= req->hcapacity)
249 req->hcapacity <<= 1;
250 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
252 hdr = &req->headers[req->hcount++];
255 hdr->release_policy = release_policy;
258 /* Like request_set_header, but sets the whole header line, as
259 provided by the user using the `--header' option. For example,
260 request_set_user_header (req, "Foo: bar") works just like
261 request_set_header (req, "Foo", "bar"). */
264 request_set_user_header (struct request *req, const char *header)
267 const char *p = strchr (header, ':');
270 BOUNDED_TO_ALLOCA (header, p, name);
274 request_set_header (req, xstrdup (name), (char *) p, rel_name);
277 /* Remove the header with specified name from REQ. Returns true if
278 the header was actually removed, false otherwise. */
281 request_remove_header (struct request *req, char *name)
284 for (i = 0; i < req->hcount; i++)
286 struct request_header *hdr = &req->headers[i];
287 if (0 == strcasecmp (name, hdr->name))
289 release_header (hdr);
290 /* Move the remaining headers by one. */
291 if (i < req->hcount - 1)
292 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
300 #define APPEND(p, str) do { \
301 int A_len = strlen (str); \
302 memcpy (p, str, A_len); \
306 /* Construct the request and write it to FD using fd_write. */
309 request_send (const struct request *req, int fd)
311 char *request_string, *p;
312 int i, size, write_error;
314 /* Count the request size. */
317 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
318 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
320 for (i = 0; i < req->hcount; i++)
322 struct request_header *hdr = &req->headers[i];
323 /* NAME ": " VALUE "\r\n" */
324 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
330 p = request_string = alloca_array (char, size);
332 /* Generate the request. */
334 APPEND (p, req->method); *p++ = ' ';
335 APPEND (p, req->arg); *p++ = ' ';
336 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
338 for (i = 0; i < req->hcount; i++)
340 struct request_header *hdr = &req->headers[i];
341 APPEND (p, hdr->name);
342 *p++ = ':', *p++ = ' ';
343 APPEND (p, hdr->value);
344 *p++ = '\r', *p++ = '\n';
347 *p++ = '\r', *p++ = '\n', *p++ = '\0';
348 assert (p - request_string == size);
352 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
354 /* Send the request to the server. */
356 write_error = fd_write (fd, request_string, size - 1, -1);
358 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
363 /* Release the resources used by REQ. */
366 request_free (struct request *req)
369 xfree_null (req->arg);
370 for (i = 0; i < req->hcount; i++)
371 release_header (&req->headers[i]);
372 xfree_null (req->headers);
376 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
377 PROMISED_SIZE bytes are sent over the wire -- if the file is
378 longer, read only that much; if the file is shorter, report an error. */
381 post_file (int sock, const char *file_name, wgint promised_size)
383 static char chunk[8192];
388 DEBUGP (("[writing POST file %s ... ", file_name));
390 fp = fopen (file_name, "rb");
393 while (!feof (fp) && written < promised_size)
396 int length = fread (chunk, 1, sizeof (chunk), fp);
399 towrite = MIN (promised_size - written, length);
400 write_error = fd_write (sock, chunk, towrite, -1);
410 /* If we've written less than was promised, report a (probably
411 nonsensical) error rather than break the promise. */
412 if (written < promised_size)
418 assert (written == promised_size);
419 DEBUGP (("done]\n"));
423 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
424 If so, return the pointer to the position after the line, otherwise
425 return NULL. This is used as callback to fd_read_hunk. The data
426 between START and PEEKED has been read and cannot be "unread"; the
427 data after PEEKED has only been peeked. */
430 response_head_terminator (const char *start, const char *peeked, int peeklen)
434 /* If at first peek, verify whether HUNK starts with "HTTP". If
435 not, this is a HTTP/0.9 request and we must bail out without
437 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
440 /* Look for "\n[\r]\n", and return the following position if found.
441 Start two chars before the current to cover the possibility that
442 part of the terminator (e.g. "\n\r") arrived in the previous
444 p = peeked - start < 2 ? start : peeked - 2;
445 end = peeked + peeklen;
447 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
448 for (; p < end - 2; p++)
451 if (p[1] == '\r' && p[2] == '\n')
453 else if (p[1] == '\n')
456 /* p==end-2: check for \n\n directly preceding END. */
457 if (p[0] == '\n' && p[1] == '\n')
463 /* The maximum size of a single HTTP response we care to read. Rather
464 than being a limit of the reader implementation, this limit
465 prevents Wget from slurping all available memory upon encountering
466 malicious or buggy server output, thus protecting the user. Define
467 it to 0 to remove the limit. */
469 #define HTTP_RESPONSE_MAX_SIZE 65536
471 /* Read the HTTP request head from FD and return it. The error
472 conditions are the same as with fd_read_hunk.
474 To support HTTP/0.9 responses, this function tries to make sure
475 that the data begins with "HTTP". If this is not the case, no data
476 is read and an empty request is returned, so that the remaining
477 data can be treated as body. */
480 read_http_response_head (int fd)
482 return fd_read_hunk (fd, response_head_terminator, 512,
483 HTTP_RESPONSE_MAX_SIZE);
487 /* The response data. */
490 /* The array of pointers that indicate where each header starts.
491 For example, given this HTTP response:
498 The headers are located like this:
500 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
502 headers[0] headers[1] headers[2] headers[3]
504 I.e. headers[0] points to the beginning of the request,
505 headers[1] points to the end of the first header and the
506 beginning of the second one, etc. */
508 const char **headers;
511 /* Create a new response object from the text of the HTTP response,
512 available in HEAD. That text is automatically split into
513 constituent header lines for fast retrieval using
516 static struct response *
517 resp_new (const char *head)
522 struct response *resp = xnew0 (struct response);
527 /* Empty head means that we're dealing with a headerless
528 (HTTP/0.9) response. In that case, don't set HEADERS at
533 /* Split HEAD into header lines, so that resp_header_* functions
534 don't need to do this over and over again. */
540 DO_REALLOC (resp->headers, size, count + 1, const char *);
541 resp->headers[count++] = hdr;
543 /* Break upon encountering an empty line. */
544 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
547 /* Find the end of HDR, including continuations. */
550 const char *end = strchr (hdr, '\n');
556 while (*hdr == ' ' || *hdr == '\t');
558 DO_REALLOC (resp->headers, size, count + 1, const char *);
559 resp->headers[count] = NULL;
564 /* Locate the header named NAME in the request data, starting with
565 position START. This allows the code to loop through the request
566 data, filtering for all requests of a given name. Returns the
567 found position, or -1 for failure. The code that uses this
568 function typically looks like this:
570 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
571 ... do something with header ...
573 If you only care about one header, use resp_header_get instead of
577 resp_header_locate (const struct response *resp, const char *name, int start,
578 const char **begptr, const char **endptr)
581 const char **headers = resp->headers;
584 if (!headers || !headers[1])
587 name_len = strlen (name);
593 for (; headers[i + 1]; i++)
595 const char *b = headers[i];
596 const char *e = headers[i + 1];
598 && b[name_len] == ':'
599 && 0 == strncasecmp (b, name, name_len))
602 while (b < e && ISSPACE (*b))
604 while (b < e && ISSPACE (e[-1]))
614 /* Find and retrieve the header named NAME in the request data. If
615 found, set *BEGPTR to its starting, and *ENDPTR to its ending
616 position, and return true. Otherwise return false.
618 This function is used as a building block for resp_header_copy
619 and resp_header_strdup. */
622 resp_header_get (const struct response *resp, const char *name,
623 const char **begptr, const char **endptr)
625 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
629 /* Copy the response header named NAME to buffer BUF, no longer than
630 BUFSIZE (BUFSIZE includes the terminating 0). If the header
631 exists, true is returned, false otherwise. If there should be no
632 limit on the size of the header, use resp_header_strdup instead.
634 If BUFSIZE is 0, no data is copied, but the boolean indication of
635 whether the header is present is still returned. */
638 resp_header_copy (const struct response *resp, const char *name,
639 char *buf, int bufsize)
642 if (!resp_header_get (resp, name, &b, &e))
646 int len = MIN (e - b, bufsize - 1);
647 memcpy (buf, b, len);
653 /* Return the value of header named NAME in RESP, allocated with
654 malloc. If such a header does not exist in RESP, return NULL. */
657 resp_header_strdup (const struct response *resp, const char *name)
660 if (!resp_header_get (resp, name, &b, &e))
662 return strdupdelim (b, e);
665 /* Parse the HTTP status line, which is of format:
667 HTTP-Version SP Status-Code SP Reason-Phrase
669 The function returns the status-code, or -1 if the status line
670 appears malformed. The pointer to "reason-phrase" message is
671 returned in *MESSAGE. */
674 resp_status (const struct response *resp, char **message)
681 /* For a HTTP/0.9 response, assume status 200. */
683 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
687 p = resp->headers[0];
688 end = resp->headers[1];
694 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
698 /* Match the HTTP version. This is optional because Gnutella
699 servers have been reported to not specify HTTP version. */
700 if (p < end && *p == '/')
703 while (p < end && ISDIGIT (*p))
705 if (p < end && *p == '.')
707 while (p < end && ISDIGIT (*p))
711 while (p < end && ISSPACE (*p))
713 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
716 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
721 while (p < end && ISSPACE (*p))
723 while (p < end && ISSPACE (end[-1]))
725 *message = strdupdelim (p, end);
731 /* Release the resources used by RESP. */
734 resp_free (struct response *resp)
736 xfree_null (resp->headers);
740 /* Print the server response, line by line, omitting the trailing CRLF
741 from individual header lines, and prefixed with PREFIX. */
744 print_server_response (const struct response *resp, const char *prefix)
749 for (i = 0; resp->headers[i + 1]; i++)
751 const char *b = resp->headers[i];
752 const char *e = resp->headers[i + 1];
754 if (b < e && e[-1] == '\n')
756 if (b < e && e[-1] == '\r')
758 /* This is safe even on printfs with broken handling of "%.<n>s"
759 because resp->headers ends with \0. */
760 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
764 /* Parse the `Content-Range' header and extract the information it
765 contains. Returns true if successful, false otherwise. */
767 parse_content_range (const char *hdr, wgint *first_byte_ptr,
768 wgint *last_byte_ptr, wgint *entity_length_ptr)
772 /* Ancient versions of Netscape proxy server, presumably predating
773 rfc2068, sent out `Content-Range' without the "bytes"
775 if (0 == strncasecmp (hdr, "bytes", 5))
778 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
782 while (ISSPACE (*hdr))
789 for (num = 0; ISDIGIT (*hdr); hdr++)
790 num = 10 * num + (*hdr - '0');
791 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
793 *first_byte_ptr = num;
795 for (num = 0; ISDIGIT (*hdr); hdr++)
796 num = 10 * num + (*hdr - '0');
797 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
799 *last_byte_ptr = num;
801 for (num = 0; ISDIGIT (*hdr); hdr++)
802 num = 10 * num + (*hdr - '0');
803 *entity_length_ptr = num;
807 /* Read the body of the request, but don't store it anywhere and don't
808 display a progress gauge. This is useful for reading the bodies of
809 administrative responses to which we will soon issue another
810 request. The response is not useful to the user, but reading it
811 allows us to continue using the same connection to the server.
813 If reading fails, false is returned, true otherwise. In debug
814 mode, the body is displayed for debugging purposes. */
817 skip_short_body (int fd, wgint contlen)
820 SKIP_SIZE = 512, /* size of the download buffer */
821 SKIP_THRESHOLD = 4096 /* the largest size we read */
823 char dlbuf[SKIP_SIZE + 1];
824 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
826 /* We shouldn't get here with unknown contlen. (This will change
827 with HTTP/1.1, which supports "chunked" transfer.) */
828 assert (contlen != -1);
830 /* If the body is too large, it makes more sense to simply close the
831 connection than to try to read the body. */
832 if (contlen > SKIP_THRESHOLD)
835 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
839 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
842 /* Don't normally report the error since this is an
843 optimization that should be invisible to the user. */
844 DEBUGP (("] aborting (%s).\n",
845 ret < 0 ? fd_errstr (fd) : "EOF received"));
849 /* Safe even if %.*s bogusly expects terminating \0 because
850 we've zero-terminated dlbuf above. */
851 DEBUGP (("%.*s", ret, dlbuf));
854 DEBUGP (("] done.\n"));
859 extract_param_value_delim (const char *begin, const char *end,
860 const char *param_name, char **param_value)
868 assert (param_value);
870 len = strlen (param_name);
872 /* skip initial whitespaces */
874 while (*p && ISSPACE (*p) && p < end) ++p;
877 && 0 == strncasecmp (p, param_name, len))
881 /* skip white spaces, equal sign and inital quote */
883 while (*p && (ISSPACE (*p) || *p == '\"' || *p == '=') && p < end) ++p;
885 /* find last quote */
887 while (*e && *e != '\"' && e < end) ++e;
889 *param_value = strdupdelim (p, e);
897 /* Parse the `Content-Disposition' header and extract the information it
898 contains. Returns true if successful, false otherwise. */
900 parse_content_disposition (const char *hdrval, char **filename)
902 const char *b = hdrval; /* b - begin */
903 const char *e = hdrval; /* e - end */
913 /* process chars b->e-1 */
914 if (true == extract_param_value_delim (b, e - 1, "filename", filename))
923 /* process chars b->e */
924 if (true == extract_param_value_delim (b, e, "filename", filename))
931 /* Persistent connections. Currently, we cache the most recently used
932 connection as persistent, provided that the HTTP server agrees to
933 make it such. The persistence data is stored in the variables
934 below. Ideally, it should be possible to cache an arbitrary fixed
935 number of these connections. */
937 /* Whether a persistent connection is active. */
938 static bool pconn_active;
941 /* The socket of the connection. */
944 /* Host and port of the currently active persistent connection. */
948 /* Whether a ssl handshake has occoured on this connection. */
951 /* Whether the connection was authorized. This is only done by
952 NTLM, which authorizes *connections* rather than individual
953 requests. (That practice is peculiar for HTTP, but it is a
954 useful optimization.) */
958 /* NTLM data of the current connection. */
959 struct ntlmdata ntlm;
963 /* Mark the persistent connection as invalid and free the resources it
964 uses. This is used by the CLOSE_* macros after they forcefully
965 close a registered persistent connection. */
968 invalidate_persistent (void)
970 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
971 pconn_active = false;
972 fd_close (pconn.socket);
977 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
978 persistent. This will enable someone to use the same connection
979 later. In the context of HTTP, this must be called only AFTER the
980 response has been received and the server has promised that the
981 connection will remain alive.
983 If a previous connection was persistent, it is closed. */
986 register_persistent (const char *host, int port, int fd, bool ssl)
990 if (pconn.socket == fd)
992 /* The connection FD is already registered. */
997 /* The old persistent connection is still active; close it
998 first. This situation arises whenever a persistent
999 connection exists, but we then connect to a different
1000 host, and try to register a persistent connection to that
1002 invalidate_persistent ();
1006 pconn_active = true;
1008 pconn.host = xstrdup (host);
1011 pconn.authorized = false;
1013 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1016 /* Return true if a persistent connection is available for connecting
1020 persistent_available_p (const char *host, int port, bool ssl,
1021 bool *host_lookup_failed)
1023 /* First, check whether a persistent connection is active at all. */
1027 /* If we want SSL and the last connection wasn't or vice versa,
1028 don't use it. Checking for host and port is not enough because
1029 HTTP and HTTPS can apparently coexist on the same port. */
1030 if (ssl != pconn.ssl)
1033 /* If we're not connecting to the same port, we're not interested. */
1034 if (port != pconn.port)
1037 /* If the host is the same, we're in business. If not, there is
1038 still hope -- read below. */
1039 if (0 != strcasecmp (host, pconn.host))
1041 /* Check if pconn.socket is talking to HOST under another name.
1042 This happens often when both sites are virtual hosts
1043 distinguished only by name and served by the same network
1044 interface, and hence the same web server (possibly set up by
1045 the ISP and serving many different web sites). This
1046 admittedly unconventional optimization does not contradict
1047 HTTP and works well with popular server software. */
1051 struct address_list *al;
1054 /* Don't try to talk to two different SSL sites over the same
1055 secure connection! (Besides, it's not clear that
1056 name-based virtual hosting is even possible with SSL.) */
1059 /* If pconn.socket's peer is one of the IP addresses HOST
1060 resolves to, pconn.socket is for all intents and purposes
1061 already talking to HOST. */
1063 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1065 /* Can't get the peer's address -- something must be very
1066 wrong with the connection. */
1067 invalidate_persistent ();
1070 al = lookup_host (host, 0);
1073 *host_lookup_failed = true;
1077 found = address_list_contains (al, &ip);
1078 address_list_release (al);
1083 /* The persistent connection's peer address was found among the
1084 addresses HOST resolved to; therefore, pconn.sock is in fact
1085 already talking to HOST -- no need to reconnect. */
1088 /* Finally, check whether the connection is still open. This is
1089 important because most servers implement liberal (short) timeout
1090 on persistent connections. Wget can of course always reconnect
1091 if the connection doesn't work out, but it's nicer to know in
1092 advance. This test is a logical followup of the first test, but
1093 is "expensive" and therefore placed at the end of the list.
1095 (Current implementation of test_socket_open has a nice side
1096 effect that it treats sockets with pending data as "closed".
1097 This is exactly what we want: if a broken server sends message
1098 body in response to HEAD, or if it sends more than conent-length
1099 data, we won't reuse the corrupted connection.) */
1101 if (!test_socket_open (pconn.socket))
1103 /* Oops, the socket is no longer open. Now that we know that,
1104 let's invalidate the persistent connection before returning
1106 invalidate_persistent ();
1113 /* The idea behind these two CLOSE macros is to distinguish between
1114 two cases: one when the job we've been doing is finished, and we
1115 want to close the connection and leave, and two when something is
1116 seriously wrong and we're closing the connection as part of
1119 In case of keep_alive, CLOSE_FINISH should leave the connection
1120 open, while CLOSE_INVALIDATE should still close it.
1122 Note that the semantics of the flag `keep_alive' is "this
1123 connection *will* be reused (the server has promised not to close
1124 the connection once we're done)", while the semantics of
1125 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1126 active, registered connection". */
1128 #define CLOSE_FINISH(fd) do { \
1131 if (pconn_active && (fd) == pconn.socket) \
1132 invalidate_persistent (); \
1141 #define CLOSE_INVALIDATE(fd) do { \
1142 if (pconn_active && (fd) == pconn.socket) \
1143 invalidate_persistent (); \
1151 wgint len; /* received length */
1152 wgint contlen; /* expected length */
1153 wgint restval; /* the restart value */
1154 int res; /* the result of last read */
1155 char *rderrmsg; /* error message from read error */
1156 char *newloc; /* new location (redirection) */
1157 char *remote_time; /* remote time-stamp string */
1158 char *error; /* textual HTTP error */
1159 int statcode; /* status code */
1160 wgint rd_size; /* amount of data read from socket */
1161 double dltime; /* time it took to download the data */
1162 const char *referer; /* value of the referer header. */
1163 char *local_file; /* local file name. */
1164 bool timestamp_checked; /* true if pre-download time-stamping checks
1165 * have already been performed */
1166 char *orig_file_name; /* name of file to compare for time-stamping
1167 * (might be != local_file if -K is set) */
1168 wgint orig_file_size; /* size of file to compare for time-stamping */
1169 time_t orig_file_tstamp; /* time-stamp of file to compare for
1174 free_hstat (struct http_stat *hs)
1176 xfree_null (hs->newloc);
1177 xfree_null (hs->remote_time);
1178 xfree_null (hs->error);
1179 xfree_null (hs->rderrmsg);
1180 xfree_null (hs->local_file);
1181 xfree_null (hs->orig_file_name);
1183 /* Guard against being called twice. */
1185 hs->remote_time = NULL;
1189 static char *create_authorization_line (const char *, const char *,
1190 const char *, const char *,
1191 const char *, bool *);
1192 static char *basic_authentication_encode (const char *, const char *);
1193 static bool known_authentication_scheme_p (const char *, const char *);
1194 static void load_cookies (void);
1196 #define BEGINS_WITH(line, string_constant) \
1197 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1198 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1199 || !line[sizeof (string_constant) - 1]))
1201 #define SET_USER_AGENT(req) do { \
1202 if (!opt.useragent) \
1203 request_set_header (req, "User-Agent", \
1204 aprintf ("Wget/%s", version_string), rel_value); \
1205 else if (*opt.useragent) \
1206 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1209 /* The flags that allow clobbering the file (opening with "wb").
1210 Defined here to avoid repetition later. #### This will require
1212 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1213 || opt.dirstruct || opt.output_document)
1215 /* Retrieve a document through HTTP protocol. It recognizes status
1216 code, and correctly handles redirections. It closes the network
1217 socket. If it receives an error from the functions below it, it
1218 will print it if there is enough information to do so (almost
1219 always), returning the error to the caller (i.e. http_loop).
1221 Various HTTP parameters are stored to hs.
1223 If PROXY is non-NULL, the connection will be made to the proxy
1224 server, and u->url will be requested. */
1226 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1228 struct request *req;
1231 char *user, *passwd;
1235 wgint contlen, contrange;
1242 /* Set to 1 when the authorization has failed permanently and should
1243 not be tried again. */
1244 bool auth_finished = false;
1246 /* Whether NTLM authentication is used for this request. */
1247 bool ntlm_seen = false;
1249 /* Whether our connection to the remote host is through SSL. */
1250 bool using_ssl = false;
1252 /* Whether a HEAD request will be issued (as opposed to GET or
1254 bool head_only = !!(*dt & HEAD_ONLY);
1257 struct response *resp;
1261 /* Whether this connection will be kept alive after the HTTP request
1265 /* Whether keep-alive should be inhibited.
1267 RFC 2068 requests that 1.0 clients not send keep-alive requests
1268 to proxies. This is because many 1.0 proxies do not interpret
1269 the Connection header and transfer it to the remote server,
1270 causing it to not close the connection and leave both the proxy
1271 and the client hanging. */
1272 bool inhibit_keep_alive =
1273 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1275 /* Headers sent when using POST. */
1276 wgint post_data_size = 0;
1278 bool host_lookup_failed = false;
1281 if (u->scheme == SCHEME_HTTPS)
1283 /* Initialize the SSL context. After this has once been done,
1284 it becomes a no-op. */
1287 scheme_disable (SCHEME_HTTPS);
1288 logprintf (LOG_NOTQUIET,
1289 _("Disabling SSL due to encountered errors.\n"));
1290 return SSLINITFAILED;
1293 #endif /* HAVE_SSL */
1295 /* Initialize certain elements of struct http_stat. */
1299 hs->rderrmsg = NULL;
1301 hs->remote_time = NULL;
1306 /* Prepare the request to send. */
1308 req = request_new ();
1311 const char *meth = "GET";
1314 else if (opt.post_file_name || opt.post_data)
1316 /* Use the full path, i.e. one that includes the leading slash and
1317 the query string. E.g. if u->path is "foo/bar" and u->query is
1318 "param=value", full_path will be "/foo/bar?param=value". */
1321 /* When using SSL over proxy, CONNECT establishes a direct
1322 connection to the HTTPS server. Therefore use the same
1323 argument as when talking to the server directly. */
1324 && u->scheme != SCHEME_HTTPS
1327 meth_arg = xstrdup (u->url);
1329 meth_arg = url_full_path (u);
1330 request_set_method (req, meth, meth_arg);
1333 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1334 if (*dt & SEND_NOCACHE)
1335 request_set_header (req, "Pragma", "no-cache", rel_none);
1337 request_set_header (req, "Range",
1338 aprintf ("bytes=%s-",
1339 number_to_static_string (hs->restval)),
1341 SET_USER_AGENT (req);
1342 request_set_header (req, "Accept", "*/*", rel_none);
1344 /* Find the username and password for authentication. */
1347 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1348 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1349 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1353 /* We have the username and the password, but haven't tried
1354 any authorization yet. Let's see if the "Basic" method
1355 works. If not, we'll come back here and construct a
1356 proper authorization method with the right challenges.
1358 If we didn't employ this kind of logic, every URL that
1359 requires authorization would have to be processed twice,
1360 which is very suboptimal and generates a bunch of false
1361 "unauthorized" errors in the server log.
1363 #### But this logic also has a serious problem when used
1364 with stronger authentications: we *first* transmit the
1365 username and the password in clear text, and *then* attempt a
1366 stronger authentication scheme. That cannot be right! We
1367 are only fortunate that almost everyone still uses the
1368 `Basic' scheme anyway.
1370 There should be an option to prevent this from happening, for
1371 those who use strong authentication schemes and value their
1373 request_set_header (req, "Authorization",
1374 basic_authentication_encode (user, passwd),
1381 char *proxy_user, *proxy_passwd;
1382 /* For normal username and password, URL components override
1383 command-line/wgetrc parameters. With proxy
1384 authentication, it's the reverse, because proxy URLs are
1385 normally the "permanent" ones, so command-line args
1386 should take precedence. */
1387 if (opt.proxy_user && opt.proxy_passwd)
1389 proxy_user = opt.proxy_user;
1390 proxy_passwd = opt.proxy_passwd;
1394 proxy_user = proxy->user;
1395 proxy_passwd = proxy->passwd;
1397 /* #### This does not appear right. Can't the proxy request,
1398 say, `Digest' authentication? */
1399 if (proxy_user && proxy_passwd)
1400 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1402 /* If we're using a proxy, we will be connecting to the proxy
1406 /* Proxy authorization over SSL is handled below. */
1408 if (u->scheme != SCHEME_HTTPS)
1410 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1413 /* Generate the Host header, HOST:PORT. Take into account that:
1415 - Broken server-side software often doesn't recognize the PORT
1416 argument, so we must generate "Host: www.server.com" instead of
1417 "Host: www.server.com:80" (and likewise for https port).
1419 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1420 becomes ambiguous and needs to be rewritten as "Host:
1421 [3ffe:8100:200:2::2]:1234". */
1423 /* Formats arranged for hfmt[add_port][add_squares]. */
1424 static const char *hfmt[][2] = {
1425 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1427 int add_port = u->port != scheme_default_port (u->scheme);
1428 int add_squares = strchr (u->host, ':') != NULL;
1429 request_set_header (req, "Host",
1430 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1434 if (!inhibit_keep_alive)
1435 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1438 request_set_header (req, "Cookie",
1439 cookie_header (wget_cookie_jar,
1440 u->host, u->port, u->path,
1442 u->scheme == SCHEME_HTTPS
1449 if (opt.post_data || opt.post_file_name)
1451 request_set_header (req, "Content-Type",
1452 "application/x-www-form-urlencoded", rel_none);
1454 post_data_size = strlen (opt.post_data);
1457 post_data_size = file_size (opt.post_file_name);
1458 if (post_data_size == -1)
1460 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1461 opt.post_file_name, strerror (errno));
1465 request_set_header (req, "Content-Length",
1466 xstrdup (number_to_static_string (post_data_size)),
1470 /* Add the user headers. */
1471 if (opt.user_headers)
1474 for (i = 0; opt.user_headers[i]; i++)
1475 request_set_user_header (req, opt.user_headers[i]);
1479 /* We need to come back here when the initial attempt to retrieve
1480 without authorization header fails. (Expected to happen at least
1481 for the Digest authorization scheme.) */
1485 /* Establish the connection. */
1487 if (!inhibit_keep_alive)
1489 /* Look for a persistent connection to target host, unless a
1490 proxy is used. The exception is when SSL is in use, in which
1491 case the proxy is nothing but a passthrough to the target
1492 host, registered as a connection to the latter. */
1493 struct url *relevant = conn;
1495 if (u->scheme == SCHEME_HTTPS)
1499 if (persistent_available_p (relevant->host, relevant->port,
1501 relevant->scheme == SCHEME_HTTPS,
1505 &host_lookup_failed))
1507 sock = pconn.socket;
1508 using_ssl = pconn.ssl;
1509 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1510 escnonprint (pconn.host), pconn.port);
1511 DEBUGP (("Reusing fd %d.\n", sock));
1512 if (pconn.authorized)
1513 /* If the connection is already authorized, the "Basic"
1514 authorization added by code above is unnecessary and
1516 request_remove_header (req, "Authorization");
1522 /* In its current implementation, persistent_available_p will
1523 look up conn->host in some cases. If that lookup failed, we
1524 don't need to bother with connect_to_host. */
1525 if (host_lookup_failed)
1531 sock = connect_to_host (conn->host, conn->port);
1540 return (retryable_socket_connect_error (errno)
1541 ? CONERROR : CONIMPOSSIBLE);
1545 if (proxy && u->scheme == SCHEME_HTTPS)
1547 /* When requesting SSL URLs through proxies, use the
1548 CONNECT method to request passthrough. */
1549 struct request *connreq = request_new ();
1550 request_set_method (connreq, "CONNECT",
1551 aprintf ("%s:%d", u->host, u->port));
1552 SET_USER_AGENT (connreq);
1555 request_set_header (connreq, "Proxy-Authorization",
1556 proxyauth, rel_value);
1557 /* Now that PROXYAUTH is part of the CONNECT request,
1558 zero it out so we don't send proxy authorization with
1559 the regular request below. */
1562 /* Examples in rfc2817 use the Host header in CONNECT
1563 requests. I don't see how that gains anything, given
1564 that the contents of Host would be exactly the same as
1565 the contents of CONNECT. */
1567 write_error = request_send (connreq, sock);
1568 request_free (connreq);
1569 if (write_error < 0)
1571 CLOSE_INVALIDATE (sock);
1575 head = read_http_response_head (sock);
1578 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1580 CLOSE_INVALIDATE (sock);
1589 DEBUGP (("proxy responded with: [%s]\n", head));
1591 resp = resp_new (head);
1592 statcode = resp_status (resp, &message);
1595 if (statcode != 200)
1598 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1599 message ? escnonprint (message) : "?");
1600 xfree_null (message);
1603 xfree_null (message);
1605 /* SOCK is now *really* connected to u->host, so update CONN
1606 to reflect this. That way register_persistent will
1607 register SOCK as being connected to u->host:u->port. */
1611 if (conn->scheme == SCHEME_HTTPS)
1613 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1620 #endif /* HAVE_SSL */
1623 /* Send the request to server. */
1624 write_error = request_send (req, sock);
1626 if (write_error >= 0)
1630 DEBUGP (("[POST data: %s]\n", opt.post_data));
1631 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1633 else if (opt.post_file_name && post_data_size != 0)
1634 write_error = post_file (sock, opt.post_file_name, post_data_size);
1637 if (write_error < 0)
1639 CLOSE_INVALIDATE (sock);
1643 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1644 proxy ? "Proxy" : "HTTP");
1649 head = read_http_response_head (sock);
1654 logputs (LOG_NOTQUIET, _("No data received.\n"));
1655 CLOSE_INVALIDATE (sock);
1661 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1663 CLOSE_INVALIDATE (sock);
1668 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1670 resp = resp_new (head);
1672 /* Check for status line. */
1674 statcode = resp_status (resp, &message);
1675 if (!opt.server_response)
1676 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1677 message ? escnonprint (message) : "");
1680 logprintf (LOG_VERBOSE, "\n");
1681 print_server_response (resp, " ");
1684 /* Determine the local filename if needed. Notice that if -O is used
1685 * hstat.local_file is set by http_loop to the argument of -O. */
1686 if (!hs->local_file)
1688 /* Honor Content-Disposition whether possible. */
1689 if (!resp_header_copy (resp, "Content-Disposition", hdrval, sizeof (hdrval))
1690 || false == parse_content_disposition (hdrval, &hs->local_file))
1692 /* Choose filename according to URL name. */
1693 hs->local_file = url_file_name (u);
1697 /* TODO: perform this check only once. */
1698 if (opt.noclobber && file_exists_p (hs->local_file))
1700 /* If opt.noclobber is turned on and file already exists, do not
1701 retrieve the file */
1702 logprintf (LOG_VERBOSE, _("\
1703 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1704 /* If the file is there, we suppose it's retrieved OK. */
1707 /* #### Bogusness alert. */
1708 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1709 if (has_html_suffix_p (hs->local_file))
1715 /* Support timestamping */
1716 /* TODO: move this code out of gethttp. */
1717 if (opt.timestamping && !hs->timestamp_checked)
1719 size_t filename_len = strlen (hs->local_file);
1720 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1721 bool local_dot_orig_file_exists = false;
1722 char *local_filename = NULL;
1725 if (opt.backup_converted)
1726 /* If -K is specified, we'll act on the assumption that it was specified
1727 last time these files were downloaded as well, and instead of just
1728 comparing local file X against server file X, we'll compare local
1729 file X.orig (if extant, else X) against server file X. If -K
1730 _wasn't_ specified last time, or the server contains files called
1731 *.orig, -N will be back to not operating correctly with -k. */
1733 /* Would a single s[n]printf() call be faster? --dan
1735 Definitely not. sprintf() is horribly slow. It's a
1736 different question whether the difference between the two
1737 affects a program. Usually I'd say "no", but at one
1738 point I profiled Wget, and found that a measurable and
1739 non-negligible amount of time was lost calling sprintf()
1740 in url.c. Replacing sprintf with inline calls to
1741 strcpy() and number_to_string() made a difference.
1743 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1744 memcpy (filename_plus_orig_suffix + filename_len,
1745 ".orig", sizeof (".orig"));
1747 /* Try to stat() the .orig file. */
1748 if (stat (filename_plus_orig_suffix, &st) == 0)
1750 local_dot_orig_file_exists = 1;
1751 local_filename = filename_plus_orig_suffix;
1755 if (!local_dot_orig_file_exists)
1756 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1757 if (stat (hs->local_file, &st) == 0)
1758 local_filename = hs->local_file;
1760 if (local_filename != NULL)
1761 /* There was a local file, so we'll check later to see if the version
1762 the server has is the same version we already have, allowing us to
1765 hs->orig_file_name = xstrdup (local_filename);
1766 hs->orig_file_size = st.st_size;
1767 hs->orig_file_tstamp = st.st_mtime;
1769 /* Modification time granularity is 2 seconds for Windows, so
1770 increase local time by 1 second for later comparison. */
1771 ++hs->orig_file_tstamp;
1776 if (!opt.ignore_length
1777 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1781 parsed = str_to_wgint (hdrval, NULL, 10);
1782 if (parsed == WGINT_MAX && errno == ERANGE)
1784 #### If Content-Length is out of range, it most likely
1785 means that the file is larger than 2G and that we're
1786 compiled without LFS. In that case we should probably
1787 refuse to even attempt to download the file. */
1793 /* Check for keep-alive related responses. */
1794 if (!inhibit_keep_alive && contlen != -1)
1796 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1798 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1800 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1805 /* The server has promised that it will not close the connection
1806 when we're done. This means that we can register it. */
1807 register_persistent (conn->host, conn->port, sock, using_ssl);
1809 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1811 /* Authorization is required. */
1812 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1813 CLOSE_FINISH (sock);
1815 CLOSE_INVALIDATE (sock);
1816 pconn.authorized = false;
1817 if (!auth_finished && (user && passwd))
1819 /* IIS sends multiple copies of WWW-Authenticate, one with
1820 the value "negotiate", and other(s) with data. Loop over
1821 all the occurrences and pick the one we recognize. */
1823 const char *wabeg, *waend;
1824 char *www_authenticate = NULL;
1826 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1827 &wabeg, &waend)) != -1;
1829 if (known_authentication_scheme_p (wabeg, waend))
1831 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1835 if (!www_authenticate)
1836 /* If the authentication header is missing or
1837 unrecognized, there's no sense in retrying. */
1838 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1839 else if (BEGINS_WITH (www_authenticate, "Basic"))
1840 /* If the authentication scheme is "Basic", which we send
1841 by default, there's no sense in retrying either. (This
1842 should be changed when we stop sending "Basic" data by
1848 pth = url_full_path (u);
1849 request_set_header (req, "Authorization",
1850 create_authorization_line (www_authenticate,
1852 request_method (req),
1856 if (BEGINS_WITH (www_authenticate, "NTLM"))
1859 goto retry_with_auth;
1862 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1866 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1868 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1870 pconn.authorized = true;
1874 hs->statcode = statcode;
1876 hs->error = xstrdup (_("Malformed status line"));
1878 hs->error = xstrdup (_("(no description)"));
1880 hs->error = xstrdup (message);
1881 xfree_null (message);
1883 type = resp_header_strdup (resp, "Content-Type");
1886 char *tmp = strchr (type, ';');
1889 while (tmp > type && ISSPACE (tmp[-1]))
1894 hs->newloc = resp_header_strdup (resp, "Location");
1895 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1897 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1901 const char *scbeg, *scend;
1902 /* The jar should have been created by now. */
1903 assert (wget_cookie_jar != NULL);
1905 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1906 &scbeg, &scend)) != -1;
1909 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1910 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1911 u->path, set_cookie);
1915 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1917 wgint first_byte_pos, last_byte_pos, entity_length;
1918 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1920 contrange = first_byte_pos;
1924 /* 20x responses are counted among successful by default. */
1925 if (H_20X (statcode))
1928 /* Return if redirected. */
1929 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1931 /* RFC2068 says that in case of the 300 (multiple choices)
1932 response, the server can output a preferred URL through
1933 `Location' header; otherwise, the request should be treated
1934 like GET. So, if the location is set, it will be a
1935 redirection; otherwise, just proceed normally. */
1936 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1940 logprintf (LOG_VERBOSE,
1941 _("Location: %s%s\n"),
1942 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1943 hs->newloc ? _(" [following]") : "");
1944 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1945 CLOSE_FINISH (sock);
1947 CLOSE_INVALIDATE (sock);
1953 /* If content-type is not given, assume text/html. This is because
1954 of the multitude of broken CGI's that "forget" to generate the
1957 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1958 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1963 if (opt.html_extension && (*dt & TEXTHTML))
1964 /* -E / --html-extension / html_extension = on was specified, and this is a
1965 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1966 already the file's suffix, tack on ".html". */
1968 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
1970 if (last_period_in_local_filename == NULL
1971 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1972 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1974 int local_filename_len = strlen (hs->local_file);
1975 /* Resize the local file, allowing for ".html" preceded by
1976 optional ".NUMBER". */
1977 hs->local_file = xrealloc (hs->local_file,
1978 local_filename_len + 24 + sizeof (".html"));
1979 strcpy(hs->local_file + local_filename_len, ".html");
1980 /* If clobbering is not allowed and the file, as named,
1981 exists, tack on ".NUMBER.html" instead. */
1982 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
1986 sprintf (hs->local_file + local_filename_len,
1987 ".%d.html", ext_num++);
1988 while (file_exists_p (hs->local_file));
1990 *dt |= ADDED_HTML_EXTENSION;
1994 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1996 /* If `-c' is in use and the file has been fully downloaded (or
1997 the remote file has shrunk), Wget effectively requests bytes
1998 after the end of file and the server response with 416. */
1999 logputs (LOG_VERBOSE, _("\
2000 \n The file is already fully retrieved; nothing to do.\n\n"));
2001 /* In case the caller inspects. */
2004 /* Mark as successfully retrieved. */
2007 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2008 might be more bytes in the body. */
2009 return RETRUNNEEDED;
2011 if ((contrange != 0 && contrange != hs->restval)
2012 || (H_PARTIAL (statcode) && !contrange))
2014 /* The Range request was somehow misunderstood by the server.
2017 CLOSE_INVALIDATE (sock);
2020 hs->contlen = contlen + contrange;
2026 /* No need to print this output if the body won't be
2027 downloaded at all, or if the original server response is
2029 logputs (LOG_VERBOSE, _("Length: "));
2032 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2033 if (contlen + contrange >= 1024)
2034 logprintf (LOG_VERBOSE, " (%s)",
2035 human_readable (contlen + contrange));
2038 if (contlen >= 1024)
2039 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2040 number_to_static_string (contlen),
2041 human_readable (contlen));
2043 logprintf (LOG_VERBOSE, _(", %s remaining"),
2044 number_to_static_string (contlen));
2048 logputs (LOG_VERBOSE,
2049 opt.ignore_length ? _("ignored") : _("unspecified"));
2051 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2053 logputs (LOG_VERBOSE, "\n");
2057 type = NULL; /* We don't need it any more. */
2059 /* Return if we have no intention of further downloading. */
2060 if (!(*dt & RETROKF) || head_only)
2062 /* In case the caller cares to look... */
2067 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2068 servers not to send body in response to a HEAD request, and
2069 those that do will likely be caught by test_socket_open.
2070 If not, they can be worked around using
2071 `--no-http-keep-alive'. */
2072 CLOSE_FINISH (sock);
2073 else if (keep_alive && skip_short_body (sock, contlen))
2074 /* Successfully skipped the body; also keep using the socket. */
2075 CLOSE_FINISH (sock);
2077 CLOSE_INVALIDATE (sock);
2078 return RETRFINISHED;
2081 /* Print fetch message, if opt.verbose. */
2084 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2085 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2088 /* Open the local file. */
2091 mkalldirs (hs->local_file);
2093 rotate_backups (hs->local_file);
2095 fp = fopen (hs->local_file, "ab");
2096 else if (ALLOW_CLOBBER)
2097 fp = fopen (hs->local_file, "wb");
2100 fp = fopen_excl (hs->local_file, true);
2101 if (!fp && errno == EEXIST)
2103 /* We cannot just invent a new name and use it (which is
2104 what functions like unique_create typically do)
2105 because we told the user we'd use this name.
2106 Instead, return and retry the download. */
2107 logprintf (LOG_NOTQUIET,
2108 _("%s has sprung into existence.\n"),
2110 CLOSE_INVALIDATE (sock);
2111 return FOPEN_EXCL_ERR;
2116 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2117 CLOSE_INVALIDATE (sock);
2124 /* This confuses the timestamping code that checks for file size.
2125 #### The timestamping code should be smarter about file size. */
2126 if (opt.save_headers && hs->restval == 0)
2127 fwrite (head, 1, strlen (head), fp);
2129 /* Now we no longer need to store the response header. */
2132 /* Download the request body. */
2135 /* If content-length is present, read that much; otherwise, read
2136 until EOF. The HTTP spec doesn't require the server to
2137 actually close the connection when it's done sending data. */
2138 flags |= rb_read_exactly;
2139 if (hs->restval > 0 && contrange == 0)
2140 /* If the server ignored our range request, instruct fd_read_body
2141 to skip the first RESTVAL bytes of body. */
2142 flags |= rb_skip_startpos;
2143 hs->len = hs->restval;
2145 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2146 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2150 CLOSE_FINISH (sock);
2154 hs->rderrmsg = xstrdup (fd_errstr (sock));
2155 CLOSE_INVALIDATE (sock);
2162 return RETRFINISHED;
2165 /* The genuine HTTP loop! This is the part where the retrieval is
2166 retried, and retried, and retried, and... */
2168 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2169 int *dt, struct url *proxy)
2172 bool got_head = false; /* used for time-stamping */
2176 time_t tmr = -1; /* remote time-stamp */
2177 wgint local_size = 0; /* the size of the local file */
2178 struct http_stat hstat; /* HTTP status */
2181 /* Assert that no value for *LOCAL_FILE was passed. */
2182 assert (local_file == NULL || *local_file == NULL);
2184 /* Set LOCAL_FILE parameter. */
2185 if (local_file && opt.output_document)
2186 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2188 /* Reset NEWLOC parameter. */
2191 /* This used to be done in main(), but it's a better idea to do it
2192 here so that we don't go through the hoops if we're just using
2197 /* Warn on (likely bogus) wildcard usage in HTTP. */
2198 if (opt.ftp_glob && has_wildcards_p (u->path))
2199 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2201 /* Setup hstat struct. */
2203 hstat.referer = referer;
2205 if (opt.output_document)
2206 hstat.local_file = xstrdup (opt.output_document);
2208 /* Reset the counter. */
2211 /* Reset the document type. */
2217 /* Increment the pass counter. */
2219 sleep_between_retrievals (count);
2221 /* Get the current time string. */
2222 tms = time_str (NULL);
2224 /* Print fetch message, if opt.verbose. */
2227 char *hurl = url_string (u, true);
2232 sprintf (tmp, _("(try:%2d)"), count);
2233 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2238 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2243 ws_changetitle (hurl);
2248 /* Default document type is empty. However, if spider mode is
2249 on or time-stamping is employed, HEAD_ONLY commands is
2250 encoded within *dt. */
2251 if (opt.spider || (opt.timestamping && !got_head))
2256 /* Decide whether or not to restart. */
2258 && stat (hstat.local_file, &st) == 0
2259 && S_ISREG (st.st_mode))
2260 /* When -c is used, continue from on-disk size. (Can't use
2261 hstat.len even if count>1 because we don't want a failed
2262 first attempt to clobber existing data.) */
2263 hstat.restval = st.st_size;
2265 /* otherwise, continue where the previous try left off */
2266 hstat.restval = hstat.len;
2270 /* Decide whether to send the no-cache directive. We send it in
2272 a) we're using a proxy, and we're past our first retrieval.
2273 Some proxies are notorious for caching incomplete data, so
2274 we require a fresh get.
2275 b) caching is explicitly inhibited. */
2276 if ((proxy && count > 1) /* a */
2277 || !opt.allow_cache /* b */
2279 *dt |= SEND_NOCACHE;
2281 *dt &= ~SEND_NOCACHE;
2283 /* Try fetching the document, or at least its head. */
2284 err = gethttp (u, &hstat, dt, proxy);
2287 tms = time_str (NULL);
2289 /* Get the new location (with or without the redirection). */
2291 *newloc = xstrdup (hstat.newloc);
2295 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2296 case CONERROR: case READERR: case WRITEFAILED:
2297 case RANGEERR: case FOPEN_EXCL_ERR:
2298 /* Non-fatal errors continue executing the loop, which will
2299 bring them to "while" statement at the end, to judge
2300 whether the number of tries was exceeded. */
2301 /* free_hstat (&hstat); */
2302 printwhat (count, opt.ntry);
2304 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2305 case SSLINITFAILED: case CONTNOTSUPPORTED:
2306 /* Fatal errors just return from the function. */
2307 free_hstat (&hstat);
2309 case FWRITEERR: case FOPENERR:
2310 /* Another fatal error. */
2311 logputs (LOG_VERBOSE, "\n");
2312 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2313 hstat.local_file, strerror (errno));
2314 free_hstat (&hstat);
2317 /* Another fatal error. */
2318 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2319 free_hstat (&hstat);
2322 /* Return the new location to the caller. */
2325 logprintf (LOG_NOTQUIET,
2326 _("ERROR: Redirection (%d) without location.\n"),
2328 free_hstat (&hstat);
2331 free_hstat (&hstat);
2334 /* The file was already fully retrieved. */
2335 free_hstat (&hstat);
2338 /* Deal with you later. */
2341 /* All possibilities should have been exhausted. */
2345 if (!(*dt & RETROKF))
2349 /* #### Ugly ugly ugly! */
2350 char *hurl = url_string (u, true);
2351 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2354 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2355 tms, hstat.statcode, escnonprint (hstat.error));
2356 logputs (LOG_VERBOSE, "\n");
2357 free_hstat (&hstat);
2361 /* Did we get the time-stamp? */
2364 if (opt.timestamping && !hstat.remote_time)
2366 logputs (LOG_NOTQUIET, _("\
2367 Last-modified header missing -- time-stamps turned off.\n"));
2369 else if (hstat.remote_time)
2371 /* Convert the date-string into struct tm. */
2372 tmr = http_atotm (hstat.remote_time);
2373 if (tmr == (time_t) (-1))
2374 logputs (LOG_VERBOSE, _("\
2375 Last-modified header invalid -- time-stamp ignored.\n"));
2379 /* The time-stamping section. */
2380 if (opt.timestamping && !got_head)
2382 got_head = true; /* no more time-stamping */
2384 count = 0; /* the retrieve count for HEAD is reset */
2386 if (hstat.remote_time && tmr != (time_t) (-1))
2388 /* Now time-stamping can be used validly. Time-stamping
2389 means that if the sizes of the local and remote file
2390 match, and local file is newer than the remote file,
2391 it will not be retrieved. Otherwise, the normal
2392 download procedure is resumed. */
2393 if (hstat.orig_file_tstamp >= tmr)
2395 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2397 logprintf (LOG_VERBOSE, _("\
2398 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2399 hstat.orig_file_name);
2400 free_hstat (&hstat);
2405 logprintf (LOG_VERBOSE, _("\
2406 The sizes do not match (local %s) -- retrieving.\n"),
2407 number_to_static_string (local_size));
2411 logputs (LOG_VERBOSE,
2412 _("Remote file is newer, retrieving.\n"));
2414 logputs (LOG_VERBOSE, "\n");
2417 /* free_hstat (&hstat); */
2418 hstat.timestamp_checked = true;
2422 if ((tmr != (time_t) (-1))
2424 && ((hstat.len == hstat.contlen) ||
2425 ((hstat.res == 0) && (hstat.contlen == -1))))
2427 /* #### This code repeats in http.c and ftp.c. Move it to a
2429 const char *fl = NULL;
2430 if (opt.output_document)
2432 if (output_stream_regular)
2433 fl = opt.output_document;
2436 fl = hstat.local_file;
2440 /* End of time-stamping section. */
2444 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2445 escnonprint (hstat.error));
2449 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2450 total_download_time += hstat.dltime;
2452 if (hstat.len == hstat.contlen)
2456 logprintf (LOG_VERBOSE,
2457 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2458 tms, tmrate, hstat.local_file,
2459 number_to_static_string (hstat.len),
2460 number_to_static_string (hstat.contlen));
2461 logprintf (LOG_NONVERBOSE,
2462 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2464 number_to_static_string (hstat.len),
2465 number_to_static_string (hstat.contlen),
2466 hstat.local_file, count);
2469 total_downloaded_bytes += hstat.len;
2471 /* Remember that we downloaded the file for later ".orig" code. */
2472 if (*dt & ADDED_HTML_EXTENSION)
2473 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2475 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2477 free_hstat (&hstat);
2480 else if (hstat.res == 0) /* No read error */
2482 if (hstat.contlen == -1) /* We don't know how much we were supposed
2483 to get, so assume we succeeded. */
2487 logprintf (LOG_VERBOSE,
2488 _("%s (%s) - `%s' saved [%s]\n\n"),
2489 tms, tmrate, hstat.local_file,
2490 number_to_static_string (hstat.len));
2491 logprintf (LOG_NONVERBOSE,
2492 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2493 tms, u->url, number_to_static_string (hstat.len),
2494 hstat.local_file, count);
2497 total_downloaded_bytes += hstat.len;
2499 /* Remember that we downloaded the file for later ".orig" code. */
2500 if (*dt & ADDED_HTML_EXTENSION)
2501 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2503 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2505 free_hstat (&hstat);
2508 else if (hstat.len < hstat.contlen) /* meaning we lost the
2509 connection too soon */
2511 logprintf (LOG_VERBOSE,
2512 _("%s (%s) - Connection closed at byte %s. "),
2513 tms, tmrate, number_to_static_string (hstat.len));
2514 printwhat (count, opt.ntry);
2515 /* free_hstat (&hstat); */
2519 /* Getting here would mean reading more data than
2520 requested with content-length, which we never do. */
2523 else /* from now on hstat.res can only be -1 */
2525 if (hstat.contlen == -1)
2527 logprintf (LOG_VERBOSE,
2528 _("%s (%s) - Read error at byte %s (%s)."),
2529 tms, tmrate, number_to_static_string (hstat.len),
2531 printwhat (count, opt.ntry);
2532 /* free_hstat (&hstat); */
2535 else /* hstat.res == -1 and contlen is given */
2537 logprintf (LOG_VERBOSE,
2538 _("%s (%s) - Read error at byte %s/%s (%s). "),
2540 number_to_static_string (hstat.len),
2541 number_to_static_string (hstat.contlen),
2543 printwhat (count, opt.ntry);
2544 /* free_hstat (&hstat); */
2550 while (!opt.ntry || (count < opt.ntry));
2555 /* Check whether the result of strptime() indicates success.
2556 strptime() returns the pointer to how far it got to in the string.
2557 The processing has been successful if the string is at `GMT' or
2558 `+X', or at the end of the string.
2560 In extended regexp parlance, the function returns 1 if P matches
2561 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2562 can return) is considered a failure and 0 is returned. */
2564 check_end (const char *p)
2568 while (ISSPACE (*p))
2571 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2572 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2578 /* Convert the textual specification of time in TIME_STRING to the
2579 number of seconds since the Epoch.
2581 TIME_STRING can be in any of the three formats RFC2616 allows the
2582 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2583 as well as the time format used in the Set-Cookie header.
2584 Timezones are ignored, and should be GMT.
2586 Return the computed time_t representation, or -1 if the conversion
2589 This function uses strptime with various string formats for parsing
2590 TIME_STRING. This results in a parser that is not as lenient in
2591 interpreting TIME_STRING as I would like it to be. Being based on
2592 strptime, it always allows shortened months, one-digit days, etc.,
2593 but due to the multitude of formats in which time can be
2594 represented, an ideal HTTP time parser would be even more
2595 forgiving. It should completely ignore things like week days and
2596 concentrate only on the various forms of representing years,
2597 months, days, hours, minutes, and seconds. For example, it would
2598 be nice if it accepted ISO 8601 out of the box.
2600 I've investigated free and PD code for this purpose, but none was
2601 usable. getdate was big and unwieldy, and had potential copyright
2602 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2603 distributed with phttpd, is excellent, but we cannot use it because
2604 it is not assigned to the FSF. So I stuck it with strptime. */
2607 http_atotm (const char *time_string)
2609 /* NOTE: Solaris strptime man page claims that %n and %t match white
2610 space, but that's not universally available. Instead, we simply
2611 use ` ' to mean "skip all WS", which works under all strptime
2612 implementations I've tested. */
2614 static const char *time_formats[] = {
2615 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2616 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2617 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2618 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2619 (used in Set-Cookie, defined in the
2620 Netscape cookie specification.) */
2622 const char *oldlocale;
2624 time_t ret = (time_t) -1;
2626 /* Solaris strptime fails to recognize English month names in
2627 non-English locales, which we work around by temporarily setting
2628 locale to C before invoking strptime. */
2629 oldlocale = setlocale (LC_TIME, NULL);
2630 setlocale (LC_TIME, "C");
2632 for (i = 0; i < countof (time_formats); i++)
2636 /* Some versions of strptime use the existing contents of struct
2637 tm to recalculate the date according to format. Zero it out
2638 to prevent stack garbage from influencing strptime. */
2641 if (check_end (strptime (time_string, time_formats[i], &t)))
2648 /* Restore the previous locale. */
2649 setlocale (LC_TIME, oldlocale);
2654 /* Authorization support: We support three authorization schemes:
2656 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2658 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2659 consisting of answering to the server's challenge with the proper
2662 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2663 Stenberg for libcurl. Like digest, NTLM is based on a
2664 challenge-response mechanism, but unlike digest, it is non-standard
2665 (authenticates TCP connections rather than requests), undocumented
2666 and Microsoft-specific. */
2668 /* Create the authentication header contents for the `Basic' scheme.
2669 This is done by encoding the string "USER:PASS" to base64 and
2670 prepending the string "Basic " in front of it. */
2673 basic_authentication_encode (const char *user, const char *passwd)
2676 int len1 = strlen (user) + 1 + strlen (passwd);
2678 t1 = (char *)alloca (len1 + 1);
2679 sprintf (t1, "%s:%s", user, passwd);
2681 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2682 base64_encode (t1, len1, t2);
2684 return concat_strings ("Basic ", t2, (char *) 0);
2687 #define SKIP_WS(x) do { \
2688 while (ISSPACE (*(x))) \
2692 #ifdef ENABLE_DIGEST
2693 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2694 of a field in such a header. If the field is the one specified by
2695 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2696 digest authorization code), extract its value in the (char*)
2697 variable pointed by RET. Returns negative on a malformed header,
2698 or number of bytes that have been parsed by this call. */
2700 extract_header_attr (const char *au, const char *attr_name, char **ret)
2703 const char *cp = au;
2705 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2707 cp += strlen (attr_name);
2720 for (ep = cp; *ep && *ep != '\"'; ep++)
2725 *ret = strdupdelim (cp, ep);
2732 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2733 an array of 16 bytes containing the hash keys, and BUF should be a
2734 buffer of 33 writable characters (32 for hex digits plus one for
2735 zero termination). */
2737 dump_hash (char *buf, const unsigned char *hash)
2741 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2743 *buf++ = XNUM_TO_digit (*hash >> 4);
2744 *buf++ = XNUM_TO_digit (*hash & 0xf);
2749 /* Take the line apart to find the challenge, and compose a digest
2750 authorization header. See RFC2069 section 2.1.2. */
2752 digest_authentication_encode (const char *au, const char *user,
2753 const char *passwd, const char *method,
2756 static char *realm, *opaque, *nonce;
2761 { "realm", &realm },
2762 { "opaque", &opaque },
2767 realm = opaque = nonce = NULL;
2769 au += 6; /* skip over `Digest' */
2775 for (i = 0; i < countof (options); i++)
2777 int skip = extract_header_attr (au, options[i].name,
2778 options[i].variable);
2782 xfree_null (opaque);
2792 if (i == countof (options))
2794 while (*au && *au != '=')
2802 while (*au && *au != '\"')
2809 while (*au && *au != ',')
2814 if (!realm || !nonce || !user || !passwd || !path || !method)
2817 xfree_null (opaque);
2822 /* Calculate the digest value. */
2824 ALLOCA_MD5_CONTEXT (ctx);
2825 unsigned char hash[MD5_HASHLEN];
2826 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2827 char response_digest[MD5_HASHLEN * 2 + 1];
2829 /* A1BUF = H(user ":" realm ":" password) */
2831 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2832 gen_md5_update ((unsigned char *)":", 1, ctx);
2833 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2834 gen_md5_update ((unsigned char *)":", 1, ctx);
2835 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2836 gen_md5_finish (ctx, hash);
2837 dump_hash (a1buf, hash);
2839 /* A2BUF = H(method ":" path) */
2841 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2842 gen_md5_update ((unsigned char *)":", 1, ctx);
2843 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2844 gen_md5_finish (ctx, hash);
2845 dump_hash (a2buf, hash);
2847 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2849 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2850 gen_md5_update ((unsigned char *)":", 1, ctx);
2851 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2852 gen_md5_update ((unsigned char *)":", 1, ctx);
2853 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2854 gen_md5_finish (ctx, hash);
2855 dump_hash (response_digest, hash);
2857 res = xmalloc (strlen (user)
2862 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2863 + (opaque ? strlen (opaque) : 0)
2865 sprintf (res, "Digest \
2866 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2867 user, realm, nonce, path, response_digest);
2870 char *p = res + strlen (res);
2871 strcat (p, ", opaque=\"");
2878 #endif /* ENABLE_DIGEST */
2880 /* Computing the size of a string literal must take into account that
2881 value returned by sizeof includes the terminating \0. */
2882 #define STRSIZE(literal) (sizeof (literal) - 1)
2884 /* Whether chars in [b, e) begin with the literal string provided as
2885 first argument and are followed by whitespace or terminating \0.
2886 The comparison is case-insensitive. */
2887 #define STARTS(literal, b, e) \
2888 ((e) - (b) >= STRSIZE (literal) \
2889 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2890 && ((e) - (b) == STRSIZE (literal) \
2891 || ISSPACE (b[STRSIZE (literal)])))
2894 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2896 return STARTS ("Basic", hdrbeg, hdrend)
2897 #ifdef ENABLE_DIGEST
2898 || STARTS ("Digest", hdrbeg, hdrend)
2901 || STARTS ("NTLM", hdrbeg, hdrend)
2908 /* Create the HTTP authorization request header. When the
2909 `WWW-Authenticate' response header is seen, according to the
2910 authorization scheme specified in that header (`Basic' and `Digest'
2911 are supported by the current implementation), produce an
2912 appropriate HTTP authorization request header. */
2914 create_authorization_line (const char *au, const char *user,
2915 const char *passwd, const char *method,
2916 const char *path, bool *finished)
2918 /* We are called only with known schemes, so we can dispatch on the
2920 switch (TOUPPER (*au))
2922 case 'B': /* Basic */
2924 return basic_authentication_encode (user, passwd);
2925 #ifdef ENABLE_DIGEST
2926 case 'D': /* Digest */
2928 return digest_authentication_encode (au, user, passwd, method, path);
2931 case 'N': /* NTLM */
2932 if (!ntlm_input (&pconn.ntlm, au))
2937 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2940 /* We shouldn't get here -- this function should be only called
2941 with values approved by known_authentication_scheme_p. */
2949 if (!wget_cookie_jar)
2950 wget_cookie_jar = cookie_jar_new ();
2951 if (opt.cookies_input && !cookies_loaded_p)
2953 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2954 cookies_loaded_p = true;
2961 if (wget_cookie_jar)
2962 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2968 xfree_null (pconn.host);
2969 if (wget_cookie_jar)
2970 cookie_jar_delete (wget_cookie_jar);
2977 test_parse_content_disposition()
2985 { "filename=\"file.ext\"", "file.ext", true },
2986 { "attachment; filename=\"file.ext\"", "file.ext", true },
2987 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
2988 { "attachment", NULL, false },
2991 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
2994 bool res = parse_content_disposition (test_array[i].hdrval, &filename);
2996 mu_assert ("test_parse_content_disposition: wrong result",
2997 res == test_array[i].result
2999 || 0 == strcmp (test_array[i].filename, filename)));
3001 /* printf ("test %d: %s\n", i, res == false ? "false" : filename); */
3007 #endif /* TESTING */