2 Copyright (C) 1996-2006 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, (int) (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))
935 /* Parse the contents of the `Content-Disposition' header, extracting
936 the information useful to Wget. Content-Disposition is a header
937 borrowed from MIME; when used in HTTP, it typically serves for
938 specifying the desired file name of the resource. For example:
940 Content-Disposition: attachment; filename="flora.jpg"
942 Wget will skip the tokens it doesn't care about, such as
943 "attachment" in the previous example; it will also skip other
944 unrecognized params. If the header is syntactically correct and
945 contains a file name, a copy of the file name is stored in
946 *filename and true is returned. Otherwise, the function returns
949 The file name is stripped of directory components and must not be
953 parse_content_disposition (const char *hdr, char **filename)
955 param_token name, value;
956 while (extract_param (&hdr, &name, &value, ';'))
957 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
959 /* Make the file name begin at the last slash or backslash. */
960 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
961 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
962 if (last_slash && last_bs)
963 value.b = 1 + MAX (last_slash, last_bs);
964 else if (last_slash || last_bs)
965 value.b = 1 + (last_slash ? last_slash : last_bs);
966 if (value.b == value.e)
968 *filename = strdupdelim (value.b, value.e);
974 /* Persistent connections. Currently, we cache the most recently used
975 connection as persistent, provided that the HTTP server agrees to
976 make it such. The persistence data is stored in the variables
977 below. Ideally, it should be possible to cache an arbitrary fixed
978 number of these connections. */
980 /* Whether a persistent connection is active. */
981 static bool pconn_active;
984 /* The socket of the connection. */
987 /* Host and port of the currently active persistent connection. */
991 /* Whether a ssl handshake has occoured on this connection. */
994 /* Whether the connection was authorized. This is only done by
995 NTLM, which authorizes *connections* rather than individual
996 requests. (That practice is peculiar for HTTP, but it is a
997 useful optimization.) */
1001 /* NTLM data of the current connection. */
1002 struct ntlmdata ntlm;
1006 /* Mark the persistent connection as invalid and free the resources it
1007 uses. This is used by the CLOSE_* macros after they forcefully
1008 close a registered persistent connection. */
1011 invalidate_persistent (void)
1013 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1014 pconn_active = false;
1015 fd_close (pconn.socket);
1020 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1021 persistent. This will enable someone to use the same connection
1022 later. In the context of HTTP, this must be called only AFTER the
1023 response has been received and the server has promised that the
1024 connection will remain alive.
1026 If a previous connection was persistent, it is closed. */
1029 register_persistent (const char *host, int port, int fd, bool ssl)
1033 if (pconn.socket == fd)
1035 /* The connection FD is already registered. */
1040 /* The old persistent connection is still active; close it
1041 first. This situation arises whenever a persistent
1042 connection exists, but we then connect to a different
1043 host, and try to register a persistent connection to that
1045 invalidate_persistent ();
1049 pconn_active = true;
1051 pconn.host = xstrdup (host);
1054 pconn.authorized = false;
1056 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1059 /* Return true if a persistent connection is available for connecting
1063 persistent_available_p (const char *host, int port, bool ssl,
1064 bool *host_lookup_failed)
1066 /* First, check whether a persistent connection is active at all. */
1070 /* If we want SSL and the last connection wasn't or vice versa,
1071 don't use it. Checking for host and port is not enough because
1072 HTTP and HTTPS can apparently coexist on the same port. */
1073 if (ssl != pconn.ssl)
1076 /* If we're not connecting to the same port, we're not interested. */
1077 if (port != pconn.port)
1080 /* If the host is the same, we're in business. If not, there is
1081 still hope -- read below. */
1082 if (0 != strcasecmp (host, pconn.host))
1084 /* Check if pconn.socket is talking to HOST under another name.
1085 This happens often when both sites are virtual hosts
1086 distinguished only by name and served by the same network
1087 interface, and hence the same web server (possibly set up by
1088 the ISP and serving many different web sites). This
1089 admittedly unconventional optimization does not contradict
1090 HTTP and works well with popular server software. */
1094 struct address_list *al;
1097 /* Don't try to talk to two different SSL sites over the same
1098 secure connection! (Besides, it's not clear that
1099 name-based virtual hosting is even possible with SSL.) */
1102 /* If pconn.socket's peer is one of the IP addresses HOST
1103 resolves to, pconn.socket is for all intents and purposes
1104 already talking to HOST. */
1106 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1108 /* Can't get the peer's address -- something must be very
1109 wrong with the connection. */
1110 invalidate_persistent ();
1113 al = lookup_host (host, 0);
1116 *host_lookup_failed = true;
1120 found = address_list_contains (al, &ip);
1121 address_list_release (al);
1126 /* The persistent connection's peer address was found among the
1127 addresses HOST resolved to; therefore, pconn.sock is in fact
1128 already talking to HOST -- no need to reconnect. */
1131 /* Finally, check whether the connection is still open. This is
1132 important because most servers implement liberal (short) timeout
1133 on persistent connections. Wget can of course always reconnect
1134 if the connection doesn't work out, but it's nicer to know in
1135 advance. This test is a logical followup of the first test, but
1136 is "expensive" and therefore placed at the end of the list.
1138 (Current implementation of test_socket_open has a nice side
1139 effect that it treats sockets with pending data as "closed".
1140 This is exactly what we want: if a broken server sends message
1141 body in response to HEAD, or if it sends more than conent-length
1142 data, we won't reuse the corrupted connection.) */
1144 if (!test_socket_open (pconn.socket))
1146 /* Oops, the socket is no longer open. Now that we know that,
1147 let's invalidate the persistent connection before returning
1149 invalidate_persistent ();
1156 /* The idea behind these two CLOSE macros is to distinguish between
1157 two cases: one when the job we've been doing is finished, and we
1158 want to close the connection and leave, and two when something is
1159 seriously wrong and we're closing the connection as part of
1162 In case of keep_alive, CLOSE_FINISH should leave the connection
1163 open, while CLOSE_INVALIDATE should still close it.
1165 Note that the semantics of the flag `keep_alive' is "this
1166 connection *will* be reused (the server has promised not to close
1167 the connection once we're done)", while the semantics of
1168 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1169 active, registered connection". */
1171 #define CLOSE_FINISH(fd) do { \
1174 if (pconn_active && (fd) == pconn.socket) \
1175 invalidate_persistent (); \
1184 #define CLOSE_INVALIDATE(fd) do { \
1185 if (pconn_active && (fd) == pconn.socket) \
1186 invalidate_persistent (); \
1194 wgint len; /* received length */
1195 wgint contlen; /* expected length */
1196 wgint restval; /* the restart value */
1197 int res; /* the result of last read */
1198 char *rderrmsg; /* error message from read error */
1199 char *newloc; /* new location (redirection) */
1200 char *remote_time; /* remote time-stamp string */
1201 char *error; /* textual HTTP error */
1202 int statcode; /* status code */
1203 wgint rd_size; /* amount of data read from socket */
1204 double dltime; /* time it took to download the data */
1205 const char *referer; /* value of the referer header. */
1206 char *local_file; /* local file name. */
1207 bool timestamp_checked; /* true if pre-download time-stamping checks
1208 * have already been performed */
1209 char *orig_file_name; /* name of file to compare for time-stamping
1210 * (might be != local_file if -K is set) */
1211 wgint orig_file_size; /* size of file to compare for time-stamping */
1212 time_t orig_file_tstamp; /* time-stamp of file to compare for
1217 free_hstat (struct http_stat *hs)
1219 xfree_null (hs->newloc);
1220 xfree_null (hs->remote_time);
1221 xfree_null (hs->error);
1222 xfree_null (hs->rderrmsg);
1223 xfree_null (hs->local_file);
1224 xfree_null (hs->orig_file_name);
1226 /* Guard against being called twice. */
1228 hs->remote_time = NULL;
1232 static char *create_authorization_line (const char *, const char *,
1233 const char *, const char *,
1234 const char *, bool *);
1235 static char *basic_authentication_encode (const char *, const char *);
1236 static bool known_authentication_scheme_p (const char *, const char *);
1237 static void load_cookies (void);
1239 #define BEGINS_WITH(line, string_constant) \
1240 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1241 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1242 || !line[sizeof (string_constant) - 1]))
1244 #define SET_USER_AGENT(req) do { \
1245 if (!opt.useragent) \
1246 request_set_header (req, "User-Agent", \
1247 aprintf ("Wget/%s", version_string), rel_value); \
1248 else if (*opt.useragent) \
1249 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1252 /* The flags that allow clobbering the file (opening with "wb").
1253 Defined here to avoid repetition later. #### This will require
1255 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1256 || opt.dirstruct || opt.output_document)
1258 /* Retrieve a document through HTTP protocol. It recognizes status
1259 code, and correctly handles redirections. It closes the network
1260 socket. If it receives an error from the functions below it, it
1261 will print it if there is enough information to do so (almost
1262 always), returning the error to the caller (i.e. http_loop).
1264 Various HTTP parameters are stored to hs.
1266 If PROXY is non-NULL, the connection will be made to the proxy
1267 server, and u->url will be requested. */
1269 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1271 struct request *req;
1274 char *user, *passwd;
1278 wgint contlen, contrange;
1285 /* Set to 1 when the authorization has failed permanently and should
1286 not be tried again. */
1287 bool auth_finished = false;
1289 /* Whether NTLM authentication is used for this request. */
1290 bool ntlm_seen = false;
1292 /* Whether our connection to the remote host is through SSL. */
1293 bool using_ssl = false;
1295 /* Whether a HEAD request will be issued (as opposed to GET or
1297 bool head_only = !!(*dt & HEAD_ONLY);
1300 struct response *resp;
1304 /* Whether this connection will be kept alive after the HTTP request
1308 /* Whether keep-alive should be inhibited.
1310 RFC 2068 requests that 1.0 clients not send keep-alive requests
1311 to proxies. This is because many 1.0 proxies do not interpret
1312 the Connection header and transfer it to the remote server,
1313 causing it to not close the connection and leave both the proxy
1314 and the client hanging. */
1315 bool inhibit_keep_alive =
1316 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1318 /* Headers sent when using POST. */
1319 wgint post_data_size = 0;
1321 bool host_lookup_failed = false;
1324 if (u->scheme == SCHEME_HTTPS)
1326 /* Initialize the SSL context. After this has once been done,
1327 it becomes a no-op. */
1330 scheme_disable (SCHEME_HTTPS);
1331 logprintf (LOG_NOTQUIET,
1332 _("Disabling SSL due to encountered errors.\n"));
1333 return SSLINITFAILED;
1336 #endif /* HAVE_SSL */
1338 /* Initialize certain elements of struct http_stat. */
1342 hs->rderrmsg = NULL;
1344 hs->remote_time = NULL;
1349 /* Prepare the request to send. */
1351 req = request_new ();
1354 const char *meth = "GET";
1357 else if (opt.post_file_name || opt.post_data)
1359 /* Use the full path, i.e. one that includes the leading slash and
1360 the query string. E.g. if u->path is "foo/bar" and u->query is
1361 "param=value", full_path will be "/foo/bar?param=value". */
1364 /* When using SSL over proxy, CONNECT establishes a direct
1365 connection to the HTTPS server. Therefore use the same
1366 argument as when talking to the server directly. */
1367 && u->scheme != SCHEME_HTTPS
1370 meth_arg = xstrdup (u->url);
1372 meth_arg = url_full_path (u);
1373 request_set_method (req, meth, meth_arg);
1376 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1377 if (*dt & SEND_NOCACHE)
1378 request_set_header (req, "Pragma", "no-cache", rel_none);
1380 request_set_header (req, "Range",
1381 aprintf ("bytes=%s-",
1382 number_to_static_string (hs->restval)),
1384 SET_USER_AGENT (req);
1385 request_set_header (req, "Accept", "*/*", rel_none);
1387 /* Find the username and password for authentication. */
1390 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1391 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1392 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1396 /* We have the username and the password, but haven't tried
1397 any authorization yet. Let's see if the "Basic" method
1398 works. If not, we'll come back here and construct a
1399 proper authorization method with the right challenges.
1401 If we didn't employ this kind of logic, every URL that
1402 requires authorization would have to be processed twice,
1403 which is very suboptimal and generates a bunch of false
1404 "unauthorized" errors in the server log.
1406 #### But this logic also has a serious problem when used
1407 with stronger authentications: we *first* transmit the
1408 username and the password in clear text, and *then* attempt a
1409 stronger authentication scheme. That cannot be right! We
1410 are only fortunate that almost everyone still uses the
1411 `Basic' scheme anyway.
1413 There should be an option to prevent this from happening, for
1414 those who use strong authentication schemes and value their
1416 request_set_header (req, "Authorization",
1417 basic_authentication_encode (user, passwd),
1424 char *proxy_user, *proxy_passwd;
1425 /* For normal username and password, URL components override
1426 command-line/wgetrc parameters. With proxy
1427 authentication, it's the reverse, because proxy URLs are
1428 normally the "permanent" ones, so command-line args
1429 should take precedence. */
1430 if (opt.proxy_user && opt.proxy_passwd)
1432 proxy_user = opt.proxy_user;
1433 proxy_passwd = opt.proxy_passwd;
1437 proxy_user = proxy->user;
1438 proxy_passwd = proxy->passwd;
1440 /* #### This does not appear right. Can't the proxy request,
1441 say, `Digest' authentication? */
1442 if (proxy_user && proxy_passwd)
1443 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1445 /* If we're using a proxy, we will be connecting to the proxy
1449 /* Proxy authorization over SSL is handled below. */
1451 if (u->scheme != SCHEME_HTTPS)
1453 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1456 /* Generate the Host header, HOST:PORT. Take into account that:
1458 - Broken server-side software often doesn't recognize the PORT
1459 argument, so we must generate "Host: www.server.com" instead of
1460 "Host: www.server.com:80" (and likewise for https port).
1462 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1463 becomes ambiguous and needs to be rewritten as "Host:
1464 [3ffe:8100:200:2::2]:1234". */
1466 /* Formats arranged for hfmt[add_port][add_squares]. */
1467 static const char *hfmt[][2] = {
1468 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1470 int add_port = u->port != scheme_default_port (u->scheme);
1471 int add_squares = strchr (u->host, ':') != NULL;
1472 request_set_header (req, "Host",
1473 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1477 if (!inhibit_keep_alive)
1478 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1481 request_set_header (req, "Cookie",
1482 cookie_header (wget_cookie_jar,
1483 u->host, u->port, u->path,
1485 u->scheme == SCHEME_HTTPS
1492 if (opt.post_data || opt.post_file_name)
1494 request_set_header (req, "Content-Type",
1495 "application/x-www-form-urlencoded", rel_none);
1497 post_data_size = strlen (opt.post_data);
1500 post_data_size = file_size (opt.post_file_name);
1501 if (post_data_size == -1)
1503 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1504 opt.post_file_name, strerror (errno));
1508 request_set_header (req, "Content-Length",
1509 xstrdup (number_to_static_string (post_data_size)),
1513 /* Add the user headers. */
1514 if (opt.user_headers)
1517 for (i = 0; opt.user_headers[i]; i++)
1518 request_set_user_header (req, opt.user_headers[i]);
1522 /* We need to come back here when the initial attempt to retrieve
1523 without authorization header fails. (Expected to happen at least
1524 for the Digest authorization scheme.) */
1528 /* Establish the connection. */
1530 if (!inhibit_keep_alive)
1532 /* Look for a persistent connection to target host, unless a
1533 proxy is used. The exception is when SSL is in use, in which
1534 case the proxy is nothing but a passthrough to the target
1535 host, registered as a connection to the latter. */
1536 struct url *relevant = conn;
1538 if (u->scheme == SCHEME_HTTPS)
1542 if (persistent_available_p (relevant->host, relevant->port,
1544 relevant->scheme == SCHEME_HTTPS,
1548 &host_lookup_failed))
1550 sock = pconn.socket;
1551 using_ssl = pconn.ssl;
1552 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1553 escnonprint (pconn.host), pconn.port);
1554 DEBUGP (("Reusing fd %d.\n", sock));
1555 if (pconn.authorized)
1556 /* If the connection is already authorized, the "Basic"
1557 authorization added by code above is unnecessary and
1559 request_remove_header (req, "Authorization");
1565 /* In its current implementation, persistent_available_p will
1566 look up conn->host in some cases. If that lookup failed, we
1567 don't need to bother with connect_to_host. */
1568 if (host_lookup_failed)
1574 sock = connect_to_host (conn->host, conn->port);
1583 return (retryable_socket_connect_error (errno)
1584 ? CONERROR : CONIMPOSSIBLE);
1588 if (proxy && u->scheme == SCHEME_HTTPS)
1590 /* When requesting SSL URLs through proxies, use the
1591 CONNECT method to request passthrough. */
1592 struct request *connreq = request_new ();
1593 request_set_method (connreq, "CONNECT",
1594 aprintf ("%s:%d", u->host, u->port));
1595 SET_USER_AGENT (connreq);
1598 request_set_header (connreq, "Proxy-Authorization",
1599 proxyauth, rel_value);
1600 /* Now that PROXYAUTH is part of the CONNECT request,
1601 zero it out so we don't send proxy authorization with
1602 the regular request below. */
1605 /* Examples in rfc2817 use the Host header in CONNECT
1606 requests. I don't see how that gains anything, given
1607 that the contents of Host would be exactly the same as
1608 the contents of CONNECT. */
1610 write_error = request_send (connreq, sock);
1611 request_free (connreq);
1612 if (write_error < 0)
1614 CLOSE_INVALIDATE (sock);
1618 head = read_http_response_head (sock);
1621 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1623 CLOSE_INVALIDATE (sock);
1632 DEBUGP (("proxy responded with: [%s]\n", head));
1634 resp = resp_new (head);
1635 statcode = resp_status (resp, &message);
1638 if (statcode != 200)
1641 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1642 message ? escnonprint (message) : "?");
1643 xfree_null (message);
1646 xfree_null (message);
1648 /* SOCK is now *really* connected to u->host, so update CONN
1649 to reflect this. That way register_persistent will
1650 register SOCK as being connected to u->host:u->port. */
1654 if (conn->scheme == SCHEME_HTTPS)
1656 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1663 #endif /* HAVE_SSL */
1666 /* Send the request to server. */
1667 write_error = request_send (req, sock);
1669 if (write_error >= 0)
1673 DEBUGP (("[POST data: %s]\n", opt.post_data));
1674 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1676 else if (opt.post_file_name && post_data_size != 0)
1677 write_error = post_file (sock, opt.post_file_name, post_data_size);
1680 if (write_error < 0)
1682 CLOSE_INVALIDATE (sock);
1686 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1687 proxy ? "Proxy" : "HTTP");
1692 head = read_http_response_head (sock);
1697 logputs (LOG_NOTQUIET, _("No data received.\n"));
1698 CLOSE_INVALIDATE (sock);
1704 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1706 CLOSE_INVALIDATE (sock);
1711 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1713 resp = resp_new (head);
1715 /* Check for status line. */
1717 statcode = resp_status (resp, &message);
1718 if (!opt.server_response)
1719 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1720 message ? escnonprint (message) : "");
1723 logprintf (LOG_VERBOSE, "\n");
1724 print_server_response (resp, " ");
1727 /* Determine the local filename if needed. Notice that if -O is used
1728 * hstat.local_file is set by http_loop to the argument of -O. */
1729 if (!hs->local_file)
1731 /* Honor Content-Disposition whether possible. */
1732 if (!opt.content_disposition
1733 || !resp_header_copy (resp, "Content-Disposition",
1734 hdrval, sizeof (hdrval))
1735 || !parse_content_disposition (hdrval, &hs->local_file))
1737 /* The Content-Disposition header is missing or broken.
1738 * Choose unique file name according to given URL. */
1739 hs->local_file = url_file_name (u);
1743 DEBUGP (("hs->local_file is: %s %s\n", hs->local_file,
1744 file_exists_p (hs->local_file) ? "(existing)" : "(not existing)"));
1746 /* TODO: perform this check only once. */
1747 if (file_exists_p (hs->local_file))
1751 /* If opt.noclobber is turned on and file already exists, do not
1752 retrieve the file */
1753 logprintf (LOG_VERBOSE, _("\
1754 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1755 /* If the file is there, we suppose it's retrieved OK. */
1758 /* #### Bogusness alert. */
1759 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1760 if (has_html_suffix_p (hs->local_file))
1767 char *unique = unique_name (hs->local_file, true);
1768 if (unique != hs->local_file)
1769 xfree (hs->local_file);
1770 hs->local_file = unique;
1774 /* Support timestamping */
1775 /* TODO: move this code out of gethttp. */
1776 if (opt.timestamping && !hs->timestamp_checked)
1778 size_t filename_len = strlen (hs->local_file);
1779 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1780 bool local_dot_orig_file_exists = false;
1781 char *local_filename = NULL;
1784 if (opt.backup_converted)
1785 /* If -K is specified, we'll act on the assumption that it was specified
1786 last time these files were downloaded as well, and instead of just
1787 comparing local file X against server file X, we'll compare local
1788 file X.orig (if extant, else X) against server file X. If -K
1789 _wasn't_ specified last time, or the server contains files called
1790 *.orig, -N will be back to not operating correctly with -k. */
1792 /* Would a single s[n]printf() call be faster? --dan
1794 Definitely not. sprintf() is horribly slow. It's a
1795 different question whether the difference between the two
1796 affects a program. Usually I'd say "no", but at one
1797 point I profiled Wget, and found that a measurable and
1798 non-negligible amount of time was lost calling sprintf()
1799 in url.c. Replacing sprintf with inline calls to
1800 strcpy() and number_to_string() made a difference.
1802 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1803 memcpy (filename_plus_orig_suffix + filename_len,
1804 ".orig", sizeof (".orig"));
1806 /* Try to stat() the .orig file. */
1807 if (stat (filename_plus_orig_suffix, &st) == 0)
1809 local_dot_orig_file_exists = 1;
1810 local_filename = filename_plus_orig_suffix;
1814 if (!local_dot_orig_file_exists)
1815 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1816 if (stat (hs->local_file, &st) == 0)
1817 local_filename = hs->local_file;
1819 if (local_filename != NULL)
1820 /* There was a local file, so we'll check later to see if the version
1821 the server has is the same version we already have, allowing us to
1824 hs->orig_file_name = xstrdup (local_filename);
1825 hs->orig_file_size = st.st_size;
1826 hs->orig_file_tstamp = st.st_mtime;
1828 /* Modification time granularity is 2 seconds for Windows, so
1829 increase local time by 1 second for later comparison. */
1830 ++hs->orig_file_tstamp;
1835 if (!opt.ignore_length
1836 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1840 parsed = str_to_wgint (hdrval, NULL, 10);
1841 if (parsed == WGINT_MAX && errno == ERANGE)
1843 #### If Content-Length is out of range, it most likely
1844 means that the file is larger than 2G and that we're
1845 compiled without LFS. In that case we should probably
1846 refuse to even attempt to download the file. */
1852 /* Check for keep-alive related responses. */
1853 if (!inhibit_keep_alive && contlen != -1)
1855 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1857 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1859 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1864 /* The server has promised that it will not close the connection
1865 when we're done. This means that we can register it. */
1866 register_persistent (conn->host, conn->port, sock, using_ssl);
1868 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1870 /* Authorization is required. */
1871 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1872 CLOSE_FINISH (sock);
1874 CLOSE_INVALIDATE (sock);
1875 pconn.authorized = false;
1876 if (!auth_finished && (user && passwd))
1878 /* IIS sends multiple copies of WWW-Authenticate, one with
1879 the value "negotiate", and other(s) with data. Loop over
1880 all the occurrences and pick the one we recognize. */
1882 const char *wabeg, *waend;
1883 char *www_authenticate = NULL;
1885 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1886 &wabeg, &waend)) != -1;
1888 if (known_authentication_scheme_p (wabeg, waend))
1890 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1894 if (!www_authenticate)
1895 /* If the authentication header is missing or
1896 unrecognized, there's no sense in retrying. */
1897 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1898 else if (BEGINS_WITH (www_authenticate, "Basic"))
1899 /* If the authentication scheme is "Basic", which we send
1900 by default, there's no sense in retrying either. (This
1901 should be changed when we stop sending "Basic" data by
1907 pth = url_full_path (u);
1908 request_set_header (req, "Authorization",
1909 create_authorization_line (www_authenticate,
1911 request_method (req),
1915 if (BEGINS_WITH (www_authenticate, "NTLM"))
1918 goto retry_with_auth;
1921 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1925 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1927 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1929 pconn.authorized = true;
1933 hs->statcode = statcode;
1935 hs->error = xstrdup (_("Malformed status line"));
1937 hs->error = xstrdup (_("(no description)"));
1939 hs->error = xstrdup (message);
1940 xfree_null (message);
1942 type = resp_header_strdup (resp, "Content-Type");
1945 char *tmp = strchr (type, ';');
1948 while (tmp > type && ISSPACE (tmp[-1]))
1953 hs->newloc = resp_header_strdup (resp, "Location");
1954 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1956 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1960 const char *scbeg, *scend;
1961 /* The jar should have been created by now. */
1962 assert (wget_cookie_jar != NULL);
1964 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1965 &scbeg, &scend)) != -1;
1968 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1969 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1970 u->path, set_cookie);
1974 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1976 wgint first_byte_pos, last_byte_pos, entity_length;
1977 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1979 contrange = first_byte_pos;
1983 /* 20x responses are counted among successful by default. */
1984 if (H_20X (statcode))
1987 /* Return if redirected. */
1988 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1990 /* RFC2068 says that in case of the 300 (multiple choices)
1991 response, the server can output a preferred URL through
1992 `Location' header; otherwise, the request should be treated
1993 like GET. So, if the location is set, it will be a
1994 redirection; otherwise, just proceed normally. */
1995 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1999 logprintf (LOG_VERBOSE,
2000 _("Location: %s%s\n"),
2001 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2002 hs->newloc ? _(" [following]") : "");
2003 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2004 CLOSE_FINISH (sock);
2006 CLOSE_INVALIDATE (sock);
2012 /* If content-type is not given, assume text/html. This is because
2013 of the multitude of broken CGI's that "forget" to generate the
2016 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2017 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2022 DEBUGP (("TEXTHTML is %s.\n", *dt | TEXTHTML ? "on": "off"));
2024 if (opt.html_extension && (*dt & TEXTHTML))
2025 /* -E / --html-extension / html_extension = on was specified, and this is a
2026 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2027 already the file's suffix, tack on ".html". */
2029 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2031 if (last_period_in_local_filename == NULL
2032 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2033 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2035 int local_filename_len = strlen (hs->local_file);
2036 /* Resize the local file, allowing for ".html" preceded by
2037 optional ".NUMBER". */
2038 hs->local_file = xrealloc (hs->local_file,
2039 local_filename_len + 24 + sizeof (".html"));
2040 strcpy(hs->local_file + local_filename_len, ".html");
2041 /* If clobbering is not allowed and the file, as named,
2042 exists, tack on ".NUMBER.html" instead. */
2043 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2047 sprintf (hs->local_file + local_filename_len,
2048 ".%d.html", ext_num++);
2049 while (file_exists_p (hs->local_file));
2051 *dt |= ADDED_HTML_EXTENSION;
2055 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2057 /* If `-c' is in use and the file has been fully downloaded (or
2058 the remote file has shrunk), Wget effectively requests bytes
2059 after the end of file and the server response with 416. */
2060 logputs (LOG_VERBOSE, _("\
2061 \n The file is already fully retrieved; nothing to do.\n\n"));
2062 /* In case the caller inspects. */
2065 /* Mark as successfully retrieved. */
2068 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2069 might be more bytes in the body. */
2070 return RETRUNNEEDED;
2072 if ((contrange != 0 && contrange != hs->restval)
2073 || (H_PARTIAL (statcode) && !contrange))
2075 /* The Range request was somehow misunderstood by the server.
2078 CLOSE_INVALIDATE (sock);
2081 hs->contlen = contlen + contrange;
2087 /* No need to print this output if the body won't be
2088 downloaded at all, or if the original server response is
2090 logputs (LOG_VERBOSE, _("Length: "));
2093 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2094 if (contlen + contrange >= 1024)
2095 logprintf (LOG_VERBOSE, " (%s)",
2096 human_readable (contlen + contrange));
2099 if (contlen >= 1024)
2100 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2101 number_to_static_string (contlen),
2102 human_readable (contlen));
2104 logprintf (LOG_VERBOSE, _(", %s remaining"),
2105 number_to_static_string (contlen));
2109 logputs (LOG_VERBOSE,
2110 opt.ignore_length ? _("ignored") : _("unspecified"));
2112 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2114 logputs (LOG_VERBOSE, "\n");
2118 type = NULL; /* We don't need it any more. */
2120 /* Return if we have no intention of further downloading. */
2121 if (!(*dt & RETROKF) || head_only)
2123 /* In case the caller cares to look... */
2128 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2129 servers not to send body in response to a HEAD request, and
2130 those that do will likely be caught by test_socket_open.
2131 If not, they can be worked around using
2132 `--no-http-keep-alive'. */
2133 CLOSE_FINISH (sock);
2134 else if (keep_alive && skip_short_body (sock, contlen))
2135 /* Successfully skipped the body; also keep using the socket. */
2136 CLOSE_FINISH (sock);
2138 CLOSE_INVALIDATE (sock);
2139 return RETRFINISHED;
2142 /* Open the local file. */
2145 mkalldirs (hs->local_file);
2147 rotate_backups (hs->local_file);
2149 fp = fopen (hs->local_file, "ab");
2150 else if (ALLOW_CLOBBER)
2151 fp = fopen (hs->local_file, "wb");
2154 fp = fopen_excl (hs->local_file, true);
2155 if (!fp && errno == EEXIST)
2157 /* We cannot just invent a new name and use it (which is
2158 what functions like unique_create typically do)
2159 because we told the user we'd use this name.
2160 Instead, return and retry the download. */
2161 logprintf (LOG_NOTQUIET,
2162 _("%s has sprung into existence.\n"),
2164 CLOSE_INVALIDATE (sock);
2165 return FOPEN_EXCL_ERR;
2170 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2171 CLOSE_INVALIDATE (sock);
2178 /* Print fetch message, if opt.verbose. */
2181 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2182 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2185 /* This confuses the timestamping code that checks for file size.
2186 #### The timestamping code should be smarter about file size. */
2187 if (opt.save_headers && hs->restval == 0)
2188 fwrite (head, 1, strlen (head), fp);
2190 /* Now we no longer need to store the response header. */
2193 /* Download the request body. */
2196 /* If content-length is present, read that much; otherwise, read
2197 until EOF. The HTTP spec doesn't require the server to
2198 actually close the connection when it's done sending data. */
2199 flags |= rb_read_exactly;
2200 if (hs->restval > 0 && contrange == 0)
2201 /* If the server ignored our range request, instruct fd_read_body
2202 to skip the first RESTVAL bytes of body. */
2203 flags |= rb_skip_startpos;
2204 hs->len = hs->restval;
2206 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2207 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2211 CLOSE_FINISH (sock);
2215 hs->rderrmsg = xstrdup (fd_errstr (sock));
2216 CLOSE_INVALIDATE (sock);
2223 return RETRFINISHED;
2226 /* The genuine HTTP loop! This is the part where the retrieval is
2227 retried, and retried, and retried, and... */
2229 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2230 int *dt, struct url *proxy)
2233 bool got_head = false; /* used for time-stamping */
2236 uerr_t err, ret = TRYLIMEXC;
2237 time_t tmr = -1; /* remote time-stamp */
2238 wgint local_size = 0; /* the size of the local file */
2239 struct http_stat hstat; /* HTTP status */
2242 /* Assert that no value for *LOCAL_FILE was passed. */
2243 assert (local_file == NULL || *local_file == NULL);
2245 /* Set LOCAL_FILE parameter. */
2246 if (local_file && opt.output_document)
2247 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2249 /* Reset NEWLOC parameter. */
2252 /* This used to be done in main(), but it's a better idea to do it
2253 here so that we don't go through the hoops if we're just using
2258 /* Warn on (likely bogus) wildcard usage in HTTP. */
2259 if (opt.ftp_glob && has_wildcards_p (u->path))
2260 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2262 /* Setup hstat struct. */
2264 hstat.referer = referer;
2266 if (opt.output_document)
2267 hstat.local_file = xstrdup (opt.output_document);
2269 /* Reset the counter. */
2272 /* Reset the document type. */
2278 /* Increment the pass counter. */
2280 sleep_between_retrievals (count);
2282 /* Get the current time string. */
2283 tms = time_str (time (NULL));
2285 /* Print fetch message, if opt.verbose. */
2288 char *hurl = url_string (u, true);
2293 sprintf (tmp, _("(try:%2d)"), count);
2294 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2299 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2304 ws_changetitle (hurl);
2309 /* Default document type is empty. However, if spider mode is
2310 on or time-stamping is employed, HEAD_ONLY commands is
2311 encoded within *dt. */
2312 if ((opt.spider && !opt.recursive) || (opt.timestamping && !got_head))
2317 /* Decide whether or not to restart. */
2319 && stat (hstat.local_file, &st) == 0
2320 && S_ISREG (st.st_mode))
2321 /* When -c is used, continue from on-disk size. (Can't use
2322 hstat.len even if count>1 because we don't want a failed
2323 first attempt to clobber existing data.) */
2324 hstat.restval = st.st_size;
2326 /* otherwise, continue where the previous try left off */
2327 hstat.restval = hstat.len;
2331 /* Decide whether to send the no-cache directive. We send it in
2333 a) we're using a proxy, and we're past our first retrieval.
2334 Some proxies are notorious for caching incomplete data, so
2335 we require a fresh get.
2336 b) caching is explicitly inhibited. */
2337 if ((proxy && count > 1) /* a */
2338 || !opt.allow_cache) /* b */
2339 *dt |= SEND_NOCACHE;
2341 *dt &= ~SEND_NOCACHE;
2343 /* Try fetching the document, or at least its head. */
2344 err = gethttp (u, &hstat, dt, proxy);
2347 tms = time_str (time (NULL));
2349 /* Get the new location (with or without the redirection). */
2351 *newloc = xstrdup (hstat.newloc);
2355 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2356 case CONERROR: case READERR: case WRITEFAILED:
2357 case RANGEERR: case FOPEN_EXCL_ERR:
2358 /* Non-fatal errors continue executing the loop, which will
2359 bring them to "while" statement at the end, to judge
2360 whether the number of tries was exceeded. */
2361 printwhat (count, opt.ntry);
2363 case FWRITEERR: case FOPENERR:
2364 /* Another fatal error. */
2365 logputs (LOG_VERBOSE, "\n");
2366 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2367 hstat.local_file, strerror (errno));
2368 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2369 case SSLINITFAILED: case CONTNOTSUPPORTED:
2370 /* Fatal errors just return from the function. */
2374 /* Another fatal error. */
2375 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2379 /* Return the new location to the caller. */
2382 logprintf (LOG_NOTQUIET,
2383 _("ERROR: Redirection (%d) without location.\n"),
2393 /* The file was already fully retrieved. */
2397 /* Deal with you later. */
2400 /* All possibilities should have been exhausted. */
2404 if (!(*dt & RETROKF))
2409 /* #### Ugly ugly ugly! */
2410 hurl = url_string (u, true);
2411 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2413 if (opt.spider && opt.recursive)
2415 if (!hurl) hurl = url_string (u, true);
2416 nonexisting_url (hurl, referer);
2418 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2419 tms, hstat.statcode, escnonprint (hstat.error));
2420 logputs (LOG_VERBOSE, "\n");
2426 /* Did we get the time-stamp? */
2429 if (opt.timestamping && !hstat.remote_time)
2431 logputs (LOG_NOTQUIET, _("\
2432 Last-modified header missing -- time-stamps turned off.\n"));
2434 else if (hstat.remote_time)
2436 /* Convert the date-string into struct tm. */
2437 tmr = http_atotm (hstat.remote_time);
2438 if (tmr == (time_t) (-1))
2439 logputs (LOG_VERBOSE, _("\
2440 Last-modified header invalid -- time-stamp ignored.\n"));
2444 /* The time-stamping section. */
2445 if (opt.timestamping && !got_head)
2447 got_head = true; /* no more time-stamping */
2449 count = 0; /* the retrieve count for HEAD is reset */
2451 if (hstat.remote_time && tmr != (time_t) (-1))
2453 /* Now time-stamping can be used validly. Time-stamping
2454 means that if the sizes of the local and remote file
2455 match, and local file is newer than the remote file,
2456 it will not be retrieved. Otherwise, the normal
2457 download procedure is resumed. */
2458 if (hstat.orig_file_tstamp >= tmr)
2460 if (hstat.contlen == -1 || hstat.orig_file_size == hstat.contlen)
2462 logprintf (LOG_VERBOSE, _("\
2463 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2464 hstat.orig_file_name);
2470 logprintf (LOG_VERBOSE, _("\
2471 The sizes do not match (local %s) -- retrieving.\n"),
2472 number_to_static_string (local_size));
2476 logputs (LOG_VERBOSE,
2477 _("Remote file is newer, retrieving.\n"));
2479 logputs (LOG_VERBOSE, "\n");
2482 /* free_hstat (&hstat); */
2483 hstat.timestamp_checked = true;
2487 if ((tmr != (time_t) (-1))
2488 && (!opt.spider || opt.recursive)
2489 && ((hstat.len == hstat.contlen) ||
2490 ((hstat.res == 0) && (hstat.contlen == -1))))
2492 /* #### This code repeats in http.c and ftp.c. Move it to a
2494 const char *fl = NULL;
2495 if (opt.output_document)
2497 if (output_stream_regular)
2498 fl = opt.output_document;
2501 fl = hstat.local_file;
2505 /* End of time-stamping section. */
2507 if (opt.spider && !opt.recursive)
2509 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2510 escnonprint (hstat.error));
2515 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2516 total_download_time += hstat.dltime;
2518 if (hstat.len == hstat.contlen)
2522 logprintf (LOG_VERBOSE,
2523 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2524 tms, tmrate, hstat.local_file,
2525 number_to_static_string (hstat.len),
2526 number_to_static_string (hstat.contlen));
2527 logprintf (LOG_NONVERBOSE,
2528 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2530 number_to_static_string (hstat.len),
2531 number_to_static_string (hstat.contlen),
2532 hstat.local_file, count);
2535 total_downloaded_bytes += hstat.len;
2537 /* Remember that we downloaded the file for later ".orig" code. */
2538 if (*dt & ADDED_HTML_EXTENSION)
2539 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2541 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2546 else if (hstat.res == 0) /* No read error */
2548 if (hstat.contlen == -1) /* We don't know how much we were supposed
2549 to get, so assume we succeeded. */
2553 logprintf (LOG_VERBOSE,
2554 _("%s (%s) - `%s' saved [%s]\n\n"),
2555 tms, tmrate, hstat.local_file,
2556 number_to_static_string (hstat.len));
2557 logprintf (LOG_NONVERBOSE,
2558 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2559 tms, u->url, number_to_static_string (hstat.len),
2560 hstat.local_file, count);
2563 total_downloaded_bytes += hstat.len;
2565 /* Remember that we downloaded the file for later ".orig" code. */
2566 if (*dt & ADDED_HTML_EXTENSION)
2567 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2569 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2574 else if (hstat.len < hstat.contlen) /* meaning we lost the
2575 connection too soon */
2577 logprintf (LOG_VERBOSE,
2578 _("%s (%s) - Connection closed at byte %s. "),
2579 tms, tmrate, number_to_static_string (hstat.len));
2580 printwhat (count, opt.ntry);
2584 /* Getting here would mean reading more data than
2585 requested with content-length, which we never do. */
2588 else /* from now on hstat.res can only be -1 */
2590 if (hstat.contlen == -1)
2592 logprintf (LOG_VERBOSE,
2593 _("%s (%s) - Read error at byte %s (%s)."),
2594 tms, tmrate, number_to_static_string (hstat.len),
2596 printwhat (count, opt.ntry);
2599 else /* hstat.res == -1 and contlen is given */
2601 logprintf (LOG_VERBOSE,
2602 _("%s (%s) - Read error at byte %s/%s (%s). "),
2604 number_to_static_string (hstat.len),
2605 number_to_static_string (hstat.contlen),
2607 printwhat (count, opt.ntry);
2613 while (!opt.ntry || (count < opt.ntry));
2617 *local_file = xstrdup (hstat.local_file);
2618 free_hstat (&hstat);
2623 /* Check whether the result of strptime() indicates success.
2624 strptime() returns the pointer to how far it got to in the string.
2625 The processing has been successful if the string is at `GMT' or
2626 `+X', or at the end of the string.
2628 In extended regexp parlance, the function returns 1 if P matches
2629 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2630 can return) is considered a failure and 0 is returned. */
2632 check_end (const char *p)
2636 while (ISSPACE (*p))
2639 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2640 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2646 /* Convert the textual specification of time in TIME_STRING to the
2647 number of seconds since the Epoch.
2649 TIME_STRING can be in any of the three formats RFC2616 allows the
2650 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2651 as well as the time format used in the Set-Cookie header.
2652 Timezones are ignored, and should be GMT.
2654 Return the computed time_t representation, or -1 if the conversion
2657 This function uses strptime with various string formats for parsing
2658 TIME_STRING. This results in a parser that is not as lenient in
2659 interpreting TIME_STRING as I would like it to be. Being based on
2660 strptime, it always allows shortened months, one-digit days, etc.,
2661 but due to the multitude of formats in which time can be
2662 represented, an ideal HTTP time parser would be even more
2663 forgiving. It should completely ignore things like week days and
2664 concentrate only on the various forms of representing years,
2665 months, days, hours, minutes, and seconds. For example, it would
2666 be nice if it accepted ISO 8601 out of the box.
2668 I've investigated free and PD code for this purpose, but none was
2669 usable. getdate was big and unwieldy, and had potential copyright
2670 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2671 distributed with phttpd, is excellent, but we cannot use it because
2672 it is not assigned to the FSF. So I stuck it with strptime. */
2675 http_atotm (const char *time_string)
2677 /* NOTE: Solaris strptime man page claims that %n and %t match white
2678 space, but that's not universally available. Instead, we simply
2679 use ` ' to mean "skip all WS", which works under all strptime
2680 implementations I've tested. */
2682 static const char *time_formats[] = {
2683 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2684 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2685 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2686 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2687 (used in Set-Cookie, defined in the
2688 Netscape cookie specification.) */
2690 const char *oldlocale;
2692 time_t ret = (time_t) -1;
2694 /* Solaris strptime fails to recognize English month names in
2695 non-English locales, which we work around by temporarily setting
2696 locale to C before invoking strptime. */
2697 oldlocale = setlocale (LC_TIME, NULL);
2698 setlocale (LC_TIME, "C");
2700 for (i = 0; i < countof (time_formats); i++)
2704 /* Some versions of strptime use the existing contents of struct
2705 tm to recalculate the date according to format. Zero it out
2706 to prevent stack garbage from influencing strptime. */
2709 if (check_end (strptime (time_string, time_formats[i], &t)))
2716 /* Restore the previous locale. */
2717 setlocale (LC_TIME, oldlocale);
2722 /* Authorization support: We support three authorization schemes:
2724 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2726 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2727 consisting of answering to the server's challenge with the proper
2730 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2731 Stenberg for libcurl. Like digest, NTLM is based on a
2732 challenge-response mechanism, but unlike digest, it is non-standard
2733 (authenticates TCP connections rather than requests), undocumented
2734 and Microsoft-specific. */
2736 /* Create the authentication header contents for the `Basic' scheme.
2737 This is done by encoding the string "USER:PASS" to base64 and
2738 prepending the string "Basic " in front of it. */
2741 basic_authentication_encode (const char *user, const char *passwd)
2744 int len1 = strlen (user) + 1 + strlen (passwd);
2746 t1 = (char *)alloca (len1 + 1);
2747 sprintf (t1, "%s:%s", user, passwd);
2749 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2750 base64_encode (t1, len1, t2);
2752 return concat_strings ("Basic ", t2, (char *) 0);
2755 #define SKIP_WS(x) do { \
2756 while (ISSPACE (*(x))) \
2760 #ifdef ENABLE_DIGEST
2761 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2762 an array of 16 bytes containing the hash keys, and BUF should be a
2763 buffer of 33 writable characters (32 for hex digits plus one for
2764 zero termination). */
2766 dump_hash (char *buf, const unsigned char *hash)
2770 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2772 *buf++ = XNUM_TO_digit (*hash >> 4);
2773 *buf++ = XNUM_TO_digit (*hash & 0xf);
2778 /* Take the line apart to find the challenge, and compose a digest
2779 authorization header. See RFC2069 section 2.1.2. */
2781 digest_authentication_encode (const char *au, const char *user,
2782 const char *passwd, const char *method,
2785 static char *realm, *opaque, *nonce;
2790 { "realm", &realm },
2791 { "opaque", &opaque },
2795 param_token name, value;
2797 realm = opaque = nonce = NULL;
2799 au += 6; /* skip over `Digest' */
2800 while (extract_param (&au, &name, &value, ','))
2803 for (i = 0; i < countof (options); i++)
2804 if (name.e - name.b == strlen (options[i].name)
2805 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2807 *options[i].variable = strdupdelim (value.b, value.e);
2811 if (!realm || !nonce || !user || !passwd || !path || !method)
2814 xfree_null (opaque);
2819 /* Calculate the digest value. */
2821 ALLOCA_MD5_CONTEXT (ctx);
2822 unsigned char hash[MD5_HASHLEN];
2823 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2824 char response_digest[MD5_HASHLEN * 2 + 1];
2826 /* A1BUF = H(user ":" realm ":" password) */
2828 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2829 gen_md5_update ((unsigned char *)":", 1, ctx);
2830 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2831 gen_md5_update ((unsigned char *)":", 1, ctx);
2832 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2833 gen_md5_finish (ctx, hash);
2834 dump_hash (a1buf, hash);
2836 /* A2BUF = H(method ":" path) */
2838 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2839 gen_md5_update ((unsigned char *)":", 1, ctx);
2840 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2841 gen_md5_finish (ctx, hash);
2842 dump_hash (a2buf, hash);
2844 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2846 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2847 gen_md5_update ((unsigned char *)":", 1, ctx);
2848 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2849 gen_md5_update ((unsigned char *)":", 1, ctx);
2850 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2851 gen_md5_finish (ctx, hash);
2852 dump_hash (response_digest, hash);
2854 res = xmalloc (strlen (user)
2859 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2860 + (opaque ? strlen (opaque) : 0)
2862 sprintf (res, "Digest \
2863 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2864 user, realm, nonce, path, response_digest);
2867 char *p = res + strlen (res);
2868 strcat (p, ", opaque=\"");
2875 #endif /* ENABLE_DIGEST */
2877 /* Computing the size of a string literal must take into account that
2878 value returned by sizeof includes the terminating \0. */
2879 #define STRSIZE(literal) (sizeof (literal) - 1)
2881 /* Whether chars in [b, e) begin with the literal string provided as
2882 first argument and are followed by whitespace or terminating \0.
2883 The comparison is case-insensitive. */
2884 #define STARTS(literal, b, e) \
2885 ((e) - (b) >= STRSIZE (literal) \
2886 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2887 && ((e) - (b) == STRSIZE (literal) \
2888 || ISSPACE (b[STRSIZE (literal)])))
2891 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2893 return STARTS ("Basic", hdrbeg, hdrend)
2894 #ifdef ENABLE_DIGEST
2895 || STARTS ("Digest", hdrbeg, hdrend)
2898 || STARTS ("NTLM", hdrbeg, hdrend)
2905 /* Create the HTTP authorization request header. When the
2906 `WWW-Authenticate' response header is seen, according to the
2907 authorization scheme specified in that header (`Basic' and `Digest'
2908 are supported by the current implementation), produce an
2909 appropriate HTTP authorization request header. */
2911 create_authorization_line (const char *au, const char *user,
2912 const char *passwd, const char *method,
2913 const char *path, bool *finished)
2915 /* We are called only with known schemes, so we can dispatch on the
2917 switch (TOUPPER (*au))
2919 case 'B': /* Basic */
2921 return basic_authentication_encode (user, passwd);
2922 #ifdef ENABLE_DIGEST
2923 case 'D': /* Digest */
2925 return digest_authentication_encode (au, user, passwd, method, path);
2928 case 'N': /* NTLM */
2929 if (!ntlm_input (&pconn.ntlm, au))
2934 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2937 /* We shouldn't get here -- this function should be only called
2938 with values approved by known_authentication_scheme_p. */
2946 if (!wget_cookie_jar)
2947 wget_cookie_jar = cookie_jar_new ();
2948 if (opt.cookies_input && !cookies_loaded_p)
2950 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
2951 cookies_loaded_p = true;
2958 if (wget_cookie_jar)
2959 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2965 xfree_null (pconn.host);
2966 if (wget_cookie_jar)
2967 cookie_jar_delete (wget_cookie_jar);
2974 test_parse_content_disposition()
2982 { "filename=\"file.ext\"", "file.ext", true },
2983 { "attachment; filename=\"file.ext\"", "file.ext", true },
2984 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
2985 { "attachment", NULL, false },
2988 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
2991 bool res = parse_content_disposition (test_array[i].hdrval, &filename);
2993 mu_assert ("test_parse_content_disposition: wrong result",
2994 res == test_array[i].result
2996 || 0 == strcmp (test_array[i].filename, filename)));
3002 #endif /* TESTING */