2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
57 # include "http-ntlm.h"
70 extern char *version_string;
73 static char *create_authorization_line (const char *, const char *,
74 const char *, const char *,
75 const char *, bool *);
76 static char *basic_authentication_encode (const char *, const char *);
77 static bool known_authentication_scheme_p (const char *, const char *);
78 static void load_cookies (void);
81 # define MIN(x, y) ((x) > (y) ? (y) : (x))
85 static bool cookies_loaded_p;
86 static struct cookie_jar *wget_cookie_jar;
88 #define TEXTHTML_S "text/html"
89 #define TEXTXHTML_S "application/xhtml+xml"
91 /* Some status code validation macros: */
92 #define H_20X(x) (((x) >= 200) && ((x) < 300))
93 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
94 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
95 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
96 || (x) == HTTP_STATUS_SEE_OTHER \
97 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
99 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
100 /* Successful 2xx. */
101 #define HTTP_STATUS_OK 200
102 #define HTTP_STATUS_CREATED 201
103 #define HTTP_STATUS_ACCEPTED 202
104 #define HTTP_STATUS_NO_CONTENT 204
105 #define HTTP_STATUS_PARTIAL_CONTENTS 206
107 /* Redirection 3xx. */
108 #define HTTP_STATUS_MULTIPLE_CHOICES 300
109 #define HTTP_STATUS_MOVED_PERMANENTLY 301
110 #define HTTP_STATUS_MOVED_TEMPORARILY 302
111 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
112 #define HTTP_STATUS_NOT_MODIFIED 304
113 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
115 /* Client error 4xx. */
116 #define HTTP_STATUS_BAD_REQUEST 400
117 #define HTTP_STATUS_UNAUTHORIZED 401
118 #define HTTP_STATUS_FORBIDDEN 403
119 #define HTTP_STATUS_NOT_FOUND 404
120 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
122 /* Server errors 5xx. */
123 #define HTTP_STATUS_INTERNAL 500
124 #define HTTP_STATUS_NOT_IMPLEMENTED 501
125 #define HTTP_STATUS_BAD_GATEWAY 502
126 #define HTTP_STATUS_UNAVAILABLE 503
129 rel_none, rel_name, rel_value, rel_both
136 struct request_header {
138 enum rp release_policy;
140 int hcount, hcapacity;
143 /* Create a new, empty request. At least request_set_method must be
144 called before the request can be used. */
146 static struct request *
149 struct request *req = xnew0 (struct request);
151 req->headers = xnew_array (struct request_header, req->hcapacity);
155 /* Set the request's method and its arguments. METH should be a
156 literal string (or it should outlive the request) because it will
157 not be freed. ARG will be freed by request_free. */
160 request_set_method (struct request *req, const char *meth, char *arg)
166 /* Return the method string passed with the last call to
167 request_set_method. */
170 request_method (const struct request *req)
175 /* Free one header according to the release policy specified with
176 request_set_header. */
179 release_header (struct request_header *hdr)
181 switch (hdr->release_policy)
198 /* Set the request named NAME to VALUE. Specifically, this means that
199 a "NAME: VALUE\r\n" header line will be used in the request. If a
200 header with the same name previously existed in the request, its
201 value will be replaced by this one. A NULL value means do nothing.
203 RELEASE_POLICY determines whether NAME and VALUE should be released
204 (freed) with request_free. Allowed values are:
206 - rel_none - don't free NAME or VALUE
207 - rel_name - free NAME when done
208 - rel_value - free VALUE when done
209 - rel_both - free both NAME and VALUE when done
211 Setting release policy is useful when arguments come from different
212 sources. For example:
214 // Don't free literal strings!
215 request_set_header (req, "Pragma", "no-cache", rel_none);
217 // Don't free a global variable, we'll need it later.
218 request_set_header (req, "Referer", opt.referer, rel_none);
220 // Value freshly allocated, free it when done.
221 request_set_header (req, "Range",
222 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
227 request_set_header (struct request *req, char *name, char *value,
228 enum rp release_policy)
230 struct request_header *hdr;
235 /* A NULL value is a no-op; if freeing the name is requested,
236 free it now to avoid leaks. */
237 if (release_policy == rel_name || release_policy == rel_both)
242 for (i = 0; i < req->hcount; i++)
244 hdr = &req->headers[i];
245 if (0 == strcasecmp (name, hdr->name))
247 /* Replace existing header. */
248 release_header (hdr);
251 hdr->release_policy = release_policy;
256 /* Install new header. */
258 if (req->hcount >= req->hcapacity)
260 req->hcapacity <<= 1;
261 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
263 hdr = &req->headers[req->hcount++];
266 hdr->release_policy = release_policy;
269 /* Like request_set_header, but sets the whole header line, as
270 provided by the user using the `--header' option. For example,
271 request_set_user_header (req, "Foo: bar") works just like
272 request_set_header (req, "Foo", "bar"). */
275 request_set_user_header (struct request *req, const char *header)
278 const char *p = strchr (header, ':');
281 BOUNDED_TO_ALLOCA (header, p, name);
285 request_set_header (req, xstrdup (name), (char *) p, rel_name);
288 /* Remove the header with specified name from REQ. Returns true if
289 the header was actually removed, false otherwise. */
292 request_remove_header (struct request *req, char *name)
295 for (i = 0; i < req->hcount; i++)
297 struct request_header *hdr = &req->headers[i];
298 if (0 == strcasecmp (name, hdr->name))
300 release_header (hdr);
301 /* Move the remaining headers by one. */
302 if (i < req->hcount - 1)
303 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
311 #define APPEND(p, str) do { \
312 int A_len = strlen (str); \
313 memcpy (p, str, A_len); \
317 /* Construct the request and write it to FD using fd_write. */
320 request_send (const struct request *req, int fd)
322 char *request_string, *p;
323 int i, size, write_error;
325 /* Count the request size. */
328 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
329 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
331 for (i = 0; i < req->hcount; i++)
333 struct request_header *hdr = &req->headers[i];
334 /* NAME ": " VALUE "\r\n" */
335 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
341 p = request_string = alloca_array (char, size);
343 /* Generate the request. */
345 APPEND (p, req->method); *p++ = ' ';
346 APPEND (p, req->arg); *p++ = ' ';
347 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
349 for (i = 0; i < req->hcount; i++)
351 struct request_header *hdr = &req->headers[i];
352 APPEND (p, hdr->name);
353 *p++ = ':', *p++ = ' ';
354 APPEND (p, hdr->value);
355 *p++ = '\r', *p++ = '\n';
358 *p++ = '\r', *p++ = '\n', *p++ = '\0';
359 assert (p - request_string == size);
363 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
365 /* Send the request to the server. */
367 write_error = fd_write (fd, request_string, size - 1, -1);
369 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
374 /* Release the resources used by REQ. */
377 request_free (struct request *req)
380 xfree_null (req->arg);
381 for (i = 0; i < req->hcount; i++)
382 release_header (&req->headers[i]);
383 xfree_null (req->headers);
387 static struct hash_table *basic_authed_hosts;
389 /* Find out if this host has issued a Basic challenge yet; if so, give
390 * it the username, password. A temporary measure until we can get
391 * proper authentication in place. */
394 maybe_send_basic_creds (const char *hostname, const char *user,
395 const char *passwd, struct request *req)
397 int did_challenge = 0;
399 if (basic_authed_hosts
400 && hash_table_contains(basic_authed_hosts, hostname))
402 DEBUGP(("Found `%s' in basic_authed_hosts.\n", hostname));
403 request_set_header (req, "Authorization",
404 basic_authentication_encode (user, passwd),
410 DEBUGP(("Host `%s' has not issued a general basic challenge.\n",
413 return did_challenge;
417 register_basic_auth_host (const char *hostname)
419 if (!basic_authed_hosts)
421 basic_authed_hosts = make_nocase_string_hash_table (1);
423 if (!hash_table_contains(basic_authed_hosts, hostname))
425 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
426 DEBUGP(("Inserted `%s' into basic_authed_hosts\n", hostname));
431 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
432 PROMISED_SIZE bytes are sent over the wire -- if the file is
433 longer, read only that much; if the file is shorter, report an error. */
436 post_file (int sock, const char *file_name, wgint promised_size)
438 static char chunk[8192];
443 DEBUGP (("[writing POST file %s ... ", file_name));
445 fp = fopen (file_name, "rb");
448 while (!feof (fp) && written < promised_size)
451 int length = fread (chunk, 1, sizeof (chunk), fp);
454 towrite = MIN (promised_size - written, length);
455 write_error = fd_write (sock, chunk, towrite, -1);
465 /* If we've written less than was promised, report a (probably
466 nonsensical) error rather than break the promise. */
467 if (written < promised_size)
473 assert (written == promised_size);
474 DEBUGP (("done]\n"));
478 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
479 If so, return the pointer to the position after the line, otherwise
480 return NULL. This is used as callback to fd_read_hunk. The data
481 between START and PEEKED has been read and cannot be "unread"; the
482 data after PEEKED has only been peeked. */
485 response_head_terminator (const char *start, const char *peeked, int peeklen)
489 /* If at first peek, verify whether HUNK starts with "HTTP". If
490 not, this is a HTTP/0.9 request and we must bail out without
492 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
495 /* Look for "\n[\r]\n", and return the following position if found.
496 Start two chars before the current to cover the possibility that
497 part of the terminator (e.g. "\n\r") arrived in the previous
499 p = peeked - start < 2 ? start : peeked - 2;
500 end = peeked + peeklen;
502 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
503 for (; p < end - 2; p++)
506 if (p[1] == '\r' && p[2] == '\n')
508 else if (p[1] == '\n')
511 /* p==end-2: check for \n\n directly preceding END. */
512 if (p[0] == '\n' && p[1] == '\n')
518 /* The maximum size of a single HTTP response we care to read. Rather
519 than being a limit of the reader implementation, this limit
520 prevents Wget from slurping all available memory upon encountering
521 malicious or buggy server output, thus protecting the user. Define
522 it to 0 to remove the limit. */
524 #define HTTP_RESPONSE_MAX_SIZE 65536
526 /* Read the HTTP request head from FD and return it. The error
527 conditions are the same as with fd_read_hunk.
529 To support HTTP/0.9 responses, this function tries to make sure
530 that the data begins with "HTTP". If this is not the case, no data
531 is read and an empty request is returned, so that the remaining
532 data can be treated as body. */
535 read_http_response_head (int fd)
537 return fd_read_hunk (fd, response_head_terminator, 512,
538 HTTP_RESPONSE_MAX_SIZE);
542 /* The response data. */
545 /* The array of pointers that indicate where each header starts.
546 For example, given this HTTP response:
553 The headers are located like this:
555 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
557 headers[0] headers[1] headers[2] headers[3]
559 I.e. headers[0] points to the beginning of the request,
560 headers[1] points to the end of the first header and the
561 beginning of the second one, etc. */
563 const char **headers;
566 /* Create a new response object from the text of the HTTP response,
567 available in HEAD. That text is automatically split into
568 constituent header lines for fast retrieval using
571 static struct response *
572 resp_new (const char *head)
577 struct response *resp = xnew0 (struct response);
582 /* Empty head means that we're dealing with a headerless
583 (HTTP/0.9) response. In that case, don't set HEADERS at
588 /* Split HEAD into header lines, so that resp_header_* functions
589 don't need to do this over and over again. */
595 DO_REALLOC (resp->headers, size, count + 1, const char *);
596 resp->headers[count++] = hdr;
598 /* Break upon encountering an empty line. */
599 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
602 /* Find the end of HDR, including continuations. */
605 const char *end = strchr (hdr, '\n');
611 while (*hdr == ' ' || *hdr == '\t');
613 DO_REALLOC (resp->headers, size, count + 1, const char *);
614 resp->headers[count] = NULL;
619 /* Locate the header named NAME in the request data, starting with
620 position START. This allows the code to loop through the request
621 data, filtering for all requests of a given name. Returns the
622 found position, or -1 for failure. The code that uses this
623 function typically looks like this:
625 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
626 ... do something with header ...
628 If you only care about one header, use resp_header_get instead of
632 resp_header_locate (const struct response *resp, const char *name, int start,
633 const char **begptr, const char **endptr)
636 const char **headers = resp->headers;
639 if (!headers || !headers[1])
642 name_len = strlen (name);
648 for (; headers[i + 1]; i++)
650 const char *b = headers[i];
651 const char *e = headers[i + 1];
653 && b[name_len] == ':'
654 && 0 == strncasecmp (b, name, name_len))
657 while (b < e && ISSPACE (*b))
659 while (b < e && ISSPACE (e[-1]))
669 /* Find and retrieve the header named NAME in the request data. If
670 found, set *BEGPTR to its starting, and *ENDPTR to its ending
671 position, and return true. Otherwise return false.
673 This function is used as a building block for resp_header_copy
674 and resp_header_strdup. */
677 resp_header_get (const struct response *resp, const char *name,
678 const char **begptr, const char **endptr)
680 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
684 /* Copy the response header named NAME to buffer BUF, no longer than
685 BUFSIZE (BUFSIZE includes the terminating 0). If the header
686 exists, true is returned, false otherwise. If there should be no
687 limit on the size of the header, use resp_header_strdup instead.
689 If BUFSIZE is 0, no data is copied, but the boolean indication of
690 whether the header is present is still returned. */
693 resp_header_copy (const struct response *resp, const char *name,
694 char *buf, int bufsize)
697 if (!resp_header_get (resp, name, &b, &e))
701 int len = MIN (e - b, bufsize - 1);
702 memcpy (buf, b, len);
708 /* Return the value of header named NAME in RESP, allocated with
709 malloc. If such a header does not exist in RESP, return NULL. */
712 resp_header_strdup (const struct response *resp, const char *name)
715 if (!resp_header_get (resp, name, &b, &e))
717 return strdupdelim (b, e);
720 /* Parse the HTTP status line, which is of format:
722 HTTP-Version SP Status-Code SP Reason-Phrase
724 The function returns the status-code, or -1 if the status line
725 appears malformed. The pointer to "reason-phrase" message is
726 returned in *MESSAGE. */
729 resp_status (const struct response *resp, char **message)
736 /* For a HTTP/0.9 response, assume status 200. */
738 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
742 p = resp->headers[0];
743 end = resp->headers[1];
749 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
753 /* Match the HTTP version. This is optional because Gnutella
754 servers have been reported to not specify HTTP version. */
755 if (p < end && *p == '/')
758 while (p < end && ISDIGIT (*p))
760 if (p < end && *p == '.')
762 while (p < end && ISDIGIT (*p))
766 while (p < end && ISSPACE (*p))
768 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
771 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
776 while (p < end && ISSPACE (*p))
778 while (p < end && ISSPACE (end[-1]))
780 *message = strdupdelim (p, end);
786 /* Release the resources used by RESP. */
789 resp_free (struct response *resp)
791 xfree_null (resp->headers);
795 /* Print a single line of response, the characters [b, e). We tried
797 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
798 but that failed to escape the non-printable characters and, in fact,
799 caused crashes in UTF-8 locales. */
802 print_response_line(const char *prefix, const char *b, const char *e)
805 BOUNDED_TO_ALLOCA(b, e, copy);
806 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
809 /* Print the server response, line by line, omitting the trailing CRLF
810 from individual header lines, and prefixed with PREFIX. */
813 print_server_response (const struct response *resp, const char *prefix)
818 for (i = 0; resp->headers[i + 1]; i++)
820 const char *b = resp->headers[i];
821 const char *e = resp->headers[i + 1];
823 if (b < e && e[-1] == '\n')
825 if (b < e && e[-1] == '\r')
827 print_response_line(prefix, b, e);
831 /* Parse the `Content-Range' header and extract the information it
832 contains. Returns true if successful, false otherwise. */
834 parse_content_range (const char *hdr, wgint *first_byte_ptr,
835 wgint *last_byte_ptr, wgint *entity_length_ptr)
839 /* Ancient versions of Netscape proxy server, presumably predating
840 rfc2068, sent out `Content-Range' without the "bytes"
842 if (0 == strncasecmp (hdr, "bytes", 5))
845 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
849 while (ISSPACE (*hdr))
856 for (num = 0; ISDIGIT (*hdr); hdr++)
857 num = 10 * num + (*hdr - '0');
858 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
860 *first_byte_ptr = num;
862 for (num = 0; ISDIGIT (*hdr); hdr++)
863 num = 10 * num + (*hdr - '0');
864 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
866 *last_byte_ptr = num;
868 for (num = 0; ISDIGIT (*hdr); hdr++)
869 num = 10 * num + (*hdr - '0');
870 *entity_length_ptr = num;
874 /* Read the body of the request, but don't store it anywhere and don't
875 display a progress gauge. This is useful for reading the bodies of
876 administrative responses to which we will soon issue another
877 request. The response is not useful to the user, but reading it
878 allows us to continue using the same connection to the server.
880 If reading fails, false is returned, true otherwise. In debug
881 mode, the body is displayed for debugging purposes. */
884 skip_short_body (int fd, wgint contlen)
887 SKIP_SIZE = 512, /* size of the download buffer */
888 SKIP_THRESHOLD = 4096 /* the largest size we read */
890 char dlbuf[SKIP_SIZE + 1];
891 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
893 /* We shouldn't get here with unknown contlen. (This will change
894 with HTTP/1.1, which supports "chunked" transfer.) */
895 assert (contlen != -1);
897 /* If the body is too large, it makes more sense to simply close the
898 connection than to try to read the body. */
899 if (contlen > SKIP_THRESHOLD)
902 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
906 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
909 /* Don't normally report the error since this is an
910 optimization that should be invisible to the user. */
911 DEBUGP (("] aborting (%s).\n",
912 ret < 0 ? fd_errstr (fd) : "EOF received"));
916 /* Safe even if %.*s bogusly expects terminating \0 because
917 we've zero-terminated dlbuf above. */
918 DEBUGP (("%.*s", ret, dlbuf));
921 DEBUGP (("] done.\n"));
925 /* Extract a parameter from the string (typically an HTTP header) at
926 **SOURCE and advance SOURCE to the next parameter. Return false
927 when there are no more parameters to extract. The name of the
928 parameter is returned in NAME, and the value in VALUE. If the
929 parameter has no value, the token's value is zeroed out.
931 For example, if *SOURCE points to the string "attachment;
932 filename=\"foo bar\"", the first call to this function will return
933 the token named "attachment" and no value, and the second call will
934 return the token named "filename" and value "foo bar". The third
935 call will return false, indicating no more valid tokens. */
938 extract_param (const char **source, param_token *name, param_token *value,
941 const char *p = *source;
943 while (ISSPACE (*p)) ++p;
947 return false; /* no error; nothing more to extract */
952 while (*p && !ISSPACE (*p) && *p != '=' && *p != separator) ++p;
954 if (name->b == name->e)
955 return false; /* empty name: error */
956 while (ISSPACE (*p)) ++p;
957 if (*p == separator || !*p) /* no value */
960 if (*p == separator) ++p;
965 return false; /* error */
967 /* *p is '=', extract value */
969 while (ISSPACE (*p)) ++p;
970 if (*p == '"') /* quoted */
973 while (*p && *p != '"') ++p;
977 /* Currently at closing quote; find the end of param. */
978 while (ISSPACE (*p)) ++p;
979 while (*p && *p != separator) ++p;
983 /* garbage after closed quote, e.g. foo="bar"baz */
989 while (*p && *p != separator) ++p;
991 while (value->e != value->b && ISSPACE (value->e[-1]))
993 if (*p == separator) ++p;
1000 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1002 /* Parse the contents of the `Content-Disposition' header, extracting
1003 the information useful to Wget. Content-Disposition is a header
1004 borrowed from MIME; when used in HTTP, it typically serves for
1005 specifying the desired file name of the resource. For example:
1007 Content-Disposition: attachment; filename="flora.jpg"
1009 Wget will skip the tokens it doesn't care about, such as
1010 "attachment" in the previous example; it will also skip other
1011 unrecognized params. If the header is syntactically correct and
1012 contains a file name, a copy of the file name is stored in
1013 *filename and true is returned. Otherwise, the function returns
1016 The file name is stripped of directory components and must not be
1020 parse_content_disposition (const char *hdr, char **filename)
1022 param_token name, value;
1023 while (extract_param (&hdr, &name, &value, ';'))
1024 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1026 /* Make the file name begin at the last slash or backslash. */
1027 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1028 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1029 if (last_slash && last_bs)
1030 value.b = 1 + MAX (last_slash, last_bs);
1031 else if (last_slash || last_bs)
1032 value.b = 1 + (last_slash ? last_slash : last_bs);
1033 if (value.b == value.e)
1035 /* Start with the directory prefix, if specified. */
1038 int prefix_length = strlen (opt.dir_prefix);
1039 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1044 total_length = prefix_length + (value.e - value.b);
1045 *filename = xmalloc (total_length + 1);
1046 strcpy (*filename, opt.dir_prefix);
1048 (*filename)[prefix_length - 1] = '/';
1049 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1050 (*filename)[total_length] = '\0';
1053 *filename = strdupdelim (value.b, value.e);
1059 /* Persistent connections. Currently, we cache the most recently used
1060 connection as persistent, provided that the HTTP server agrees to
1061 make it such. The persistence data is stored in the variables
1062 below. Ideally, it should be possible to cache an arbitrary fixed
1063 number of these connections. */
1065 /* Whether a persistent connection is active. */
1066 static bool pconn_active;
1069 /* The socket of the connection. */
1072 /* Host and port of the currently active persistent connection. */
1076 /* Whether a ssl handshake has occoured on this connection. */
1079 /* Whether the connection was authorized. This is only done by
1080 NTLM, which authorizes *connections* rather than individual
1081 requests. (That practice is peculiar for HTTP, but it is a
1082 useful optimization.) */
1086 /* NTLM data of the current connection. */
1087 struct ntlmdata ntlm;
1091 /* Mark the persistent connection as invalid and free the resources it
1092 uses. This is used by the CLOSE_* macros after they forcefully
1093 close a registered persistent connection. */
1096 invalidate_persistent (void)
1098 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1099 pconn_active = false;
1100 fd_close (pconn.socket);
1105 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1106 persistent. This will enable someone to use the same connection
1107 later. In the context of HTTP, this must be called only AFTER the
1108 response has been received and the server has promised that the
1109 connection will remain alive.
1111 If a previous connection was persistent, it is closed. */
1114 register_persistent (const char *host, int port, int fd, bool ssl)
1118 if (pconn.socket == fd)
1120 /* The connection FD is already registered. */
1125 /* The old persistent connection is still active; close it
1126 first. This situation arises whenever a persistent
1127 connection exists, but we then connect to a different
1128 host, and try to register a persistent connection to that
1130 invalidate_persistent ();
1134 pconn_active = true;
1136 pconn.host = xstrdup (host);
1139 pconn.authorized = false;
1141 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1144 /* Return true if a persistent connection is available for connecting
1148 persistent_available_p (const char *host, int port, bool ssl,
1149 bool *host_lookup_failed)
1151 /* First, check whether a persistent connection is active at all. */
1155 /* If we want SSL and the last connection wasn't or vice versa,
1156 don't use it. Checking for host and port is not enough because
1157 HTTP and HTTPS can apparently coexist on the same port. */
1158 if (ssl != pconn.ssl)
1161 /* If we're not connecting to the same port, we're not interested. */
1162 if (port != pconn.port)
1165 /* If the host is the same, we're in business. If not, there is
1166 still hope -- read below. */
1167 if (0 != strcasecmp (host, pconn.host))
1169 /* Check if pconn.socket is talking to HOST under another name.
1170 This happens often when both sites are virtual hosts
1171 distinguished only by name and served by the same network
1172 interface, and hence the same web server (possibly set up by
1173 the ISP and serving many different web sites). This
1174 admittedly unconventional optimization does not contradict
1175 HTTP and works well with popular server software. */
1179 struct address_list *al;
1182 /* Don't try to talk to two different SSL sites over the same
1183 secure connection! (Besides, it's not clear that
1184 name-based virtual hosting is even possible with SSL.) */
1187 /* If pconn.socket's peer is one of the IP addresses HOST
1188 resolves to, pconn.socket is for all intents and purposes
1189 already talking to HOST. */
1191 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1193 /* Can't get the peer's address -- something must be very
1194 wrong with the connection. */
1195 invalidate_persistent ();
1198 al = lookup_host (host, 0);
1201 *host_lookup_failed = true;
1205 found = address_list_contains (al, &ip);
1206 address_list_release (al);
1211 /* The persistent connection's peer address was found among the
1212 addresses HOST resolved to; therefore, pconn.sock is in fact
1213 already talking to HOST -- no need to reconnect. */
1216 /* Finally, check whether the connection is still open. This is
1217 important because most servers implement liberal (short) timeout
1218 on persistent connections. Wget can of course always reconnect
1219 if the connection doesn't work out, but it's nicer to know in
1220 advance. This test is a logical followup of the first test, but
1221 is "expensive" and therefore placed at the end of the list.
1223 (Current implementation of test_socket_open has a nice side
1224 effect that it treats sockets with pending data as "closed".
1225 This is exactly what we want: if a broken server sends message
1226 body in response to HEAD, or if it sends more than conent-length
1227 data, we won't reuse the corrupted connection.) */
1229 if (!test_socket_open (pconn.socket))
1231 /* Oops, the socket is no longer open. Now that we know that,
1232 let's invalidate the persistent connection before returning
1234 invalidate_persistent ();
1241 /* The idea behind these two CLOSE macros is to distinguish between
1242 two cases: one when the job we've been doing is finished, and we
1243 want to close the connection and leave, and two when something is
1244 seriously wrong and we're closing the connection as part of
1247 In case of keep_alive, CLOSE_FINISH should leave the connection
1248 open, while CLOSE_INVALIDATE should still close it.
1250 Note that the semantics of the flag `keep_alive' is "this
1251 connection *will* be reused (the server has promised not to close
1252 the connection once we're done)", while the semantics of
1253 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1254 active, registered connection". */
1256 #define CLOSE_FINISH(fd) do { \
1259 if (pconn_active && (fd) == pconn.socket) \
1260 invalidate_persistent (); \
1269 #define CLOSE_INVALIDATE(fd) do { \
1270 if (pconn_active && (fd) == pconn.socket) \
1271 invalidate_persistent (); \
1279 wgint len; /* received length */
1280 wgint contlen; /* expected length */
1281 wgint restval; /* the restart value */
1282 int res; /* the result of last read */
1283 char *rderrmsg; /* error message from read error */
1284 char *newloc; /* new location (redirection) */
1285 char *remote_time; /* remote time-stamp string */
1286 char *error; /* textual HTTP error */
1287 int statcode; /* status code */
1288 wgint rd_size; /* amount of data read from socket */
1289 double dltime; /* time it took to download the data */
1290 const char *referer; /* value of the referer header. */
1291 char *local_file; /* local file name. */
1292 bool timestamp_checked; /* true if pre-download time-stamping checks
1293 * have already been performed */
1294 char *orig_file_name; /* name of file to compare for time-stamping
1295 * (might be != local_file if -K is set) */
1296 wgint orig_file_size; /* size of file to compare for time-stamping */
1297 time_t orig_file_tstamp; /* time-stamp of file to compare for
1302 free_hstat (struct http_stat *hs)
1304 xfree_null (hs->newloc);
1305 xfree_null (hs->remote_time);
1306 xfree_null (hs->error);
1307 xfree_null (hs->rderrmsg);
1308 xfree_null (hs->local_file);
1309 xfree_null (hs->orig_file_name);
1311 /* Guard against being called twice. */
1313 hs->remote_time = NULL;
1317 #define BEGINS_WITH(line, string_constant) \
1318 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1319 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1320 || !line[sizeof (string_constant) - 1]))
1322 #define SET_USER_AGENT(req) do { \
1323 if (!opt.useragent) \
1324 request_set_header (req, "User-Agent", \
1325 aprintf ("Wget/%s", version_string), rel_value); \
1326 else if (*opt.useragent) \
1327 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1330 /* The flags that allow clobbering the file (opening with "wb").
1331 Defined here to avoid repetition later. #### This will require
1333 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1334 || opt.dirstruct || opt.output_document)
1336 /* Retrieve a document through HTTP protocol. It recognizes status
1337 code, and correctly handles redirections. It closes the network
1338 socket. If it receives an error from the functions below it, it
1339 will print it if there is enough information to do so (almost
1340 always), returning the error to the caller (i.e. http_loop).
1342 Various HTTP parameters are stored to hs.
1344 If PROXY is non-NULL, the connection will be made to the proxy
1345 server, and u->url will be requested. */
1347 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1349 struct request *req;
1352 char *user, *passwd;
1356 wgint contlen, contrange;
1363 /* Set to 1 when the authorization has already been sent and should
1364 not be tried again. */
1365 bool auth_finished = false;
1367 /* Set to 1 when just globally-set Basic authorization has been sent;
1368 * should prevent further Basic negotiations, but not other
1370 bool basic_auth_finished = false;
1372 /* Whether NTLM authentication is used for this request. */
1373 bool ntlm_seen = false;
1375 /* Whether our connection to the remote host is through SSL. */
1376 bool using_ssl = false;
1378 /* Whether a HEAD request will be issued (as opposed to GET or
1380 bool head_only = !!(*dt & HEAD_ONLY);
1383 struct response *resp;
1387 /* Whether this connection will be kept alive after the HTTP request
1391 /* Whether keep-alive should be inhibited.
1393 RFC 2068 requests that 1.0 clients not send keep-alive requests
1394 to proxies. This is because many 1.0 proxies do not interpret
1395 the Connection header and transfer it to the remote server,
1396 causing it to not close the connection and leave both the proxy
1397 and the client hanging. */
1398 bool inhibit_keep_alive =
1399 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1401 /* Headers sent when using POST. */
1402 wgint post_data_size = 0;
1404 bool host_lookup_failed = false;
1407 if (u->scheme == SCHEME_HTTPS)
1409 /* Initialize the SSL context. After this has once been done,
1410 it becomes a no-op. */
1413 scheme_disable (SCHEME_HTTPS);
1414 logprintf (LOG_NOTQUIET,
1415 _("Disabling SSL due to encountered errors.\n"));
1416 return SSLINITFAILED;
1419 #endif /* HAVE_SSL */
1421 /* Initialize certain elements of struct http_stat. */
1425 hs->rderrmsg = NULL;
1427 hs->remote_time = NULL;
1432 /* Prepare the request to send. */
1434 req = request_new ();
1437 const char *meth = "GET";
1440 else if (opt.post_file_name || opt.post_data)
1442 /* Use the full path, i.e. one that includes the leading slash and
1443 the query string. E.g. if u->path is "foo/bar" and u->query is
1444 "param=value", full_path will be "/foo/bar?param=value". */
1447 /* When using SSL over proxy, CONNECT establishes a direct
1448 connection to the HTTPS server. Therefore use the same
1449 argument as when talking to the server directly. */
1450 && u->scheme != SCHEME_HTTPS
1453 meth_arg = xstrdup (u->url);
1455 meth_arg = url_full_path (u);
1456 request_set_method (req, meth, meth_arg);
1459 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1460 if (*dt & SEND_NOCACHE)
1461 request_set_header (req, "Pragma", "no-cache", rel_none);
1463 request_set_header (req, "Range",
1464 aprintf ("bytes=%s-",
1465 number_to_static_string (hs->restval)),
1467 SET_USER_AGENT (req);
1468 request_set_header (req, "Accept", "*/*", rel_none);
1470 /* Find the username and password for authentication. */
1473 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1474 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1475 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1478 && !u->user) /* We only do "site-wide" authentication with "global"
1479 user/password values; URL user/password info overrides. */
1481 /* If this is a host for which we've already received a Basic
1482 * challenge, we'll go ahead and send Basic authentication creds. */
1483 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1489 char *proxy_user, *proxy_passwd;
1490 /* For normal username and password, URL components override
1491 command-line/wgetrc parameters. With proxy
1492 authentication, it's the reverse, because proxy URLs are
1493 normally the "permanent" ones, so command-line args
1494 should take precedence. */
1495 if (opt.proxy_user && opt.proxy_passwd)
1497 proxy_user = opt.proxy_user;
1498 proxy_passwd = opt.proxy_passwd;
1502 proxy_user = proxy->user;
1503 proxy_passwd = proxy->passwd;
1505 /* #### This does not appear right. Can't the proxy request,
1506 say, `Digest' authentication? */
1507 if (proxy_user && proxy_passwd)
1508 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1510 /* If we're using a proxy, we will be connecting to the proxy
1514 /* Proxy authorization over SSL is handled below. */
1516 if (u->scheme != SCHEME_HTTPS)
1518 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1521 /* Generate the Host header, HOST:PORT. Take into account that:
1523 - Broken server-side software often doesn't recognize the PORT
1524 argument, so we must generate "Host: www.server.com" instead of
1525 "Host: www.server.com:80" (and likewise for https port).
1527 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1528 becomes ambiguous and needs to be rewritten as "Host:
1529 [3ffe:8100:200:2::2]:1234". */
1531 /* Formats arranged for hfmt[add_port][add_squares]. */
1532 static const char *hfmt[][2] = {
1533 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1535 int add_port = u->port != scheme_default_port (u->scheme);
1536 int add_squares = strchr (u->host, ':') != NULL;
1537 request_set_header (req, "Host",
1538 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1542 if (!inhibit_keep_alive)
1543 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1546 request_set_header (req, "Cookie",
1547 cookie_header (wget_cookie_jar,
1548 u->host, u->port, u->path,
1550 u->scheme == SCHEME_HTTPS
1557 if (opt.post_data || opt.post_file_name)
1559 request_set_header (req, "Content-Type",
1560 "application/x-www-form-urlencoded", rel_none);
1562 post_data_size = strlen (opt.post_data);
1565 post_data_size = file_size (opt.post_file_name);
1566 if (post_data_size == -1)
1568 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1569 opt.post_file_name, strerror (errno));
1573 request_set_header (req, "Content-Length",
1574 xstrdup (number_to_static_string (post_data_size)),
1578 /* Add the user headers. */
1579 if (opt.user_headers)
1582 for (i = 0; opt.user_headers[i]; i++)
1583 request_set_user_header (req, opt.user_headers[i]);
1587 /* We need to come back here when the initial attempt to retrieve
1588 without authorization header fails. (Expected to happen at least
1589 for the Digest authorization scheme.) */
1593 /* Establish the connection. */
1595 if (!inhibit_keep_alive)
1597 /* Look for a persistent connection to target host, unless a
1598 proxy is used. The exception is when SSL is in use, in which
1599 case the proxy is nothing but a passthrough to the target
1600 host, registered as a connection to the latter. */
1601 struct url *relevant = conn;
1603 if (u->scheme == SCHEME_HTTPS)
1607 if (persistent_available_p (relevant->host, relevant->port,
1609 relevant->scheme == SCHEME_HTTPS,
1613 &host_lookup_failed))
1615 sock = pconn.socket;
1616 using_ssl = pconn.ssl;
1617 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1618 escnonprint (pconn.host), pconn.port);
1619 DEBUGP (("Reusing fd %d.\n", sock));
1620 if (pconn.authorized)
1621 /* If the connection is already authorized, the "Basic"
1622 authorization added by code above is unnecessary and
1624 request_remove_header (req, "Authorization");
1626 else if (host_lookup_failed)
1629 logprintf(LOG_NOTQUIET,
1630 _("%s: unable to resolve host address `%s'\n"),
1631 exec_name, relevant->host);
1638 sock = connect_to_host (conn->host, conn->port);
1647 return (retryable_socket_connect_error (errno)
1648 ? CONERROR : CONIMPOSSIBLE);
1652 if (proxy && u->scheme == SCHEME_HTTPS)
1654 /* When requesting SSL URLs through proxies, use the
1655 CONNECT method to request passthrough. */
1656 struct request *connreq = request_new ();
1657 request_set_method (connreq, "CONNECT",
1658 aprintf ("%s:%d", u->host, u->port));
1659 SET_USER_AGENT (connreq);
1662 request_set_header (connreq, "Proxy-Authorization",
1663 proxyauth, rel_value);
1664 /* Now that PROXYAUTH is part of the CONNECT request,
1665 zero it out so we don't send proxy authorization with
1666 the regular request below. */
1669 /* Examples in rfc2817 use the Host header in CONNECT
1670 requests. I don't see how that gains anything, given
1671 that the contents of Host would be exactly the same as
1672 the contents of CONNECT. */
1674 write_error = request_send (connreq, sock);
1675 request_free (connreq);
1676 if (write_error < 0)
1678 CLOSE_INVALIDATE (sock);
1682 head = read_http_response_head (sock);
1685 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1687 CLOSE_INVALIDATE (sock);
1696 DEBUGP (("proxy responded with: [%s]\n", head));
1698 resp = resp_new (head);
1699 statcode = resp_status (resp, &message);
1702 if (statcode != 200)
1705 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1706 message ? escnonprint (message) : "?");
1707 xfree_null (message);
1710 xfree_null (message);
1712 /* SOCK is now *really* connected to u->host, so update CONN
1713 to reflect this. That way register_persistent will
1714 register SOCK as being connected to u->host:u->port. */
1718 if (conn->scheme == SCHEME_HTTPS)
1720 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1727 #endif /* HAVE_SSL */
1730 /* Send the request to server. */
1731 write_error = request_send (req, sock);
1733 if (write_error >= 0)
1737 DEBUGP (("[POST data: %s]\n", opt.post_data));
1738 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1740 else if (opt.post_file_name && post_data_size != 0)
1741 write_error = post_file (sock, opt.post_file_name, post_data_size);
1744 if (write_error < 0)
1746 CLOSE_INVALIDATE (sock);
1750 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1751 proxy ? "Proxy" : "HTTP");
1756 head = read_http_response_head (sock);
1761 logputs (LOG_NOTQUIET, _("No data received.\n"));
1762 CLOSE_INVALIDATE (sock);
1768 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1770 CLOSE_INVALIDATE (sock);
1775 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1777 resp = resp_new (head);
1779 /* Check for status line. */
1781 statcode = resp_status (resp, &message);
1782 if (!opt.server_response)
1783 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1784 message ? escnonprint (message) : "");
1787 logprintf (LOG_VERBOSE, "\n");
1788 print_server_response (resp, " ");
1791 /* Determine the local filename if needed. Notice that if -O is used
1792 * hstat.local_file is set by http_loop to the argument of -O. */
1793 if (!hs->local_file)
1795 /* Honor Content-Disposition whether possible. */
1796 if (!opt.content_disposition
1797 || !resp_header_copy (resp, "Content-Disposition",
1798 hdrval, sizeof (hdrval))
1799 || !parse_content_disposition (hdrval, &hs->local_file))
1801 /* The Content-Disposition header is missing or broken.
1802 * Choose unique file name according to given URL. */
1803 hs->local_file = url_file_name (u);
1807 /* TODO: perform this check only once. */
1808 if (file_exists_p (hs->local_file))
1812 /* If opt.noclobber is turned on and file already exists, do not
1813 retrieve the file */
1814 logprintf (LOG_VERBOSE, _("\
1815 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1816 /* If the file is there, we suppose it's retrieved OK. */
1819 /* #### Bogusness alert. */
1820 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1821 if (has_html_suffix_p (hs->local_file))
1824 return RETRUNNEEDED;
1826 else if (!ALLOW_CLOBBER)
1828 char *unique = unique_name (hs->local_file, true);
1829 if (unique != hs->local_file)
1830 xfree (hs->local_file);
1831 hs->local_file = unique;
1835 /* Support timestamping */
1836 /* TODO: move this code out of gethttp. */
1837 if (opt.timestamping && !hs->timestamp_checked)
1839 size_t filename_len = strlen (hs->local_file);
1840 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1841 bool local_dot_orig_file_exists = false;
1842 char *local_filename = NULL;
1845 if (opt.backup_converted)
1846 /* If -K is specified, we'll act on the assumption that it was specified
1847 last time these files were downloaded as well, and instead of just
1848 comparing local file X against server file X, we'll compare local
1849 file X.orig (if extant, else X) against server file X. If -K
1850 _wasn't_ specified last time, or the server contains files called
1851 *.orig, -N will be back to not operating correctly with -k. */
1853 /* Would a single s[n]printf() call be faster? --dan
1855 Definitely not. sprintf() is horribly slow. It's a
1856 different question whether the difference between the two
1857 affects a program. Usually I'd say "no", but at one
1858 point I profiled Wget, and found that a measurable and
1859 non-negligible amount of time was lost calling sprintf()
1860 in url.c. Replacing sprintf with inline calls to
1861 strcpy() and number_to_string() made a difference.
1863 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1864 memcpy (filename_plus_orig_suffix + filename_len,
1865 ".orig", sizeof (".orig"));
1867 /* Try to stat() the .orig file. */
1868 if (stat (filename_plus_orig_suffix, &st) == 0)
1870 local_dot_orig_file_exists = true;
1871 local_filename = filename_plus_orig_suffix;
1875 if (!local_dot_orig_file_exists)
1876 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1877 if (stat (hs->local_file, &st) == 0)
1878 local_filename = hs->local_file;
1880 if (local_filename != NULL)
1881 /* There was a local file, so we'll check later to see if the version
1882 the server has is the same version we already have, allowing us to
1885 hs->orig_file_name = xstrdup (local_filename);
1886 hs->orig_file_size = st.st_size;
1887 hs->orig_file_tstamp = st.st_mtime;
1889 /* Modification time granularity is 2 seconds for Windows, so
1890 increase local time by 1 second for later comparison. */
1891 ++hs->orig_file_tstamp;
1896 if (!opt.ignore_length
1897 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1901 parsed = str_to_wgint (hdrval, NULL, 10);
1902 if (parsed == WGINT_MAX && errno == ERANGE)
1905 #### If Content-Length is out of range, it most likely
1906 means that the file is larger than 2G and that we're
1907 compiled without LFS. In that case we should probably
1908 refuse to even attempt to download the file. */
1911 else if (parsed < 0)
1913 /* Negative Content-Length; nonsensical, so we can't
1914 assume any information about the content to receive. */
1921 /* Check for keep-alive related responses. */
1922 if (!inhibit_keep_alive && contlen != -1)
1924 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1926 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1928 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1933 /* The server has promised that it will not close the connection
1934 when we're done. This means that we can register it. */
1935 register_persistent (conn->host, conn->port, sock, using_ssl);
1937 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1939 /* Authorization is required. */
1940 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1941 CLOSE_FINISH (sock);
1943 CLOSE_INVALIDATE (sock);
1944 pconn.authorized = false;
1945 if (!auth_finished && (user && passwd))
1947 /* IIS sends multiple copies of WWW-Authenticate, one with
1948 the value "negotiate", and other(s) with data. Loop over
1949 all the occurrences and pick the one we recognize. */
1951 const char *wabeg, *waend;
1952 char *www_authenticate = NULL;
1954 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1955 &wabeg, &waend)) != -1;
1957 if (known_authentication_scheme_p (wabeg, waend))
1959 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1963 if (!www_authenticate)
1965 /* If the authentication header is missing or
1966 unrecognized, there's no sense in retrying. */
1967 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1969 else if (!basic_auth_finished
1970 || !BEGINS_WITH (www_authenticate, "Basic"))
1973 pth = url_full_path (u);
1974 request_set_header (req, "Authorization",
1975 create_authorization_line (www_authenticate,
1977 request_method (req),
1981 if (BEGINS_WITH (www_authenticate, "NTLM"))
1983 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1985 /* Need to register this host as using basic auth,
1986 * so we automatically send creds next time. */
1987 register_basic_auth_host (u->host);
1990 goto retry_with_auth;
1994 /* We already did Basic auth, and it failed. Gotta
1998 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2002 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2004 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2006 pconn.authorized = true;
2010 hs->statcode = statcode;
2012 hs->error = xstrdup (_("Malformed status line"));
2014 hs->error = xstrdup (_("(no description)"));
2016 hs->error = xstrdup (message);
2017 xfree_null (message);
2019 type = resp_header_strdup (resp, "Content-Type");
2022 char *tmp = strchr (type, ';');
2025 while (tmp > type && ISSPACE (tmp[-1]))
2030 hs->newloc = resp_header_strdup (resp, "Location");
2031 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2033 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2037 const char *scbeg, *scend;
2038 /* The jar should have been created by now. */
2039 assert (wget_cookie_jar != NULL);
2041 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2042 &scbeg, &scend)) != -1;
2045 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2046 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2047 u->path, set_cookie);
2051 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2053 wgint first_byte_pos, last_byte_pos, entity_length;
2054 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2056 contrange = first_byte_pos;
2060 /* 20x responses are counted among successful by default. */
2061 if (H_20X (statcode))
2064 /* Return if redirected. */
2065 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2067 /* RFC2068 says that in case of the 300 (multiple choices)
2068 response, the server can output a preferred URL through
2069 `Location' header; otherwise, the request should be treated
2070 like GET. So, if the location is set, it will be a
2071 redirection; otherwise, just proceed normally. */
2072 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2076 logprintf (LOG_VERBOSE,
2077 _("Location: %s%s\n"),
2078 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2079 hs->newloc ? _(" [following]") : "");
2080 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2081 CLOSE_FINISH (sock);
2083 CLOSE_INVALIDATE (sock);
2089 /* If content-type is not given, assume text/html. This is because
2090 of the multitude of broken CGI's that "forget" to generate the
2093 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2094 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2099 if (opt.html_extension && (*dt & TEXTHTML))
2100 /* -E / --html-extension / html_extension = on was specified, and this is a
2101 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2102 already the file's suffix, tack on ".html". */
2104 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2106 if (last_period_in_local_filename == NULL
2107 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2108 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2110 int local_filename_len = strlen (hs->local_file);
2111 /* Resize the local file, allowing for ".html" preceded by
2112 optional ".NUMBER". */
2113 hs->local_file = xrealloc (hs->local_file,
2114 local_filename_len + 24 + sizeof (".html"));
2115 strcpy(hs->local_file + local_filename_len, ".html");
2116 /* If clobbering is not allowed and the file, as named,
2117 exists, tack on ".NUMBER.html" instead. */
2118 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2122 sprintf (hs->local_file + local_filename_len,
2123 ".%d.html", ext_num++);
2124 while (file_exists_p (hs->local_file));
2126 *dt |= ADDED_HTML_EXTENSION;
2130 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2132 /* If `-c' is in use and the file has been fully downloaded (or
2133 the remote file has shrunk), Wget effectively requests bytes
2134 after the end of file and the server response with 416. */
2135 logputs (LOG_VERBOSE, _("\
2136 \n The file is already fully retrieved; nothing to do.\n\n"));
2137 /* In case the caller inspects. */
2140 /* Mark as successfully retrieved. */
2143 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2144 might be more bytes in the body. */
2145 return RETRUNNEEDED;
2147 if ((contrange != 0 && contrange != hs->restval)
2148 || (H_PARTIAL (statcode) && !contrange))
2150 /* The Range request was somehow misunderstood by the server.
2153 CLOSE_INVALIDATE (sock);
2156 hs->contlen = contlen + contrange;
2162 /* No need to print this output if the body won't be
2163 downloaded at all, or if the original server response is
2165 logputs (LOG_VERBOSE, _("Length: "));
2168 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2169 if (contlen + contrange >= 1024)
2170 logprintf (LOG_VERBOSE, " (%s)",
2171 human_readable (contlen + contrange));
2174 if (contlen >= 1024)
2175 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2176 number_to_static_string (contlen),
2177 human_readable (contlen));
2179 logprintf (LOG_VERBOSE, _(", %s remaining"),
2180 number_to_static_string (contlen));
2184 logputs (LOG_VERBOSE,
2185 opt.ignore_length ? _("ignored") : _("unspecified"));
2187 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2189 logputs (LOG_VERBOSE, "\n");
2193 type = NULL; /* We don't need it any more. */
2195 /* Return if we have no intention of further downloading. */
2196 if (!(*dt & RETROKF) || head_only)
2198 /* In case the caller cares to look... */
2203 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2204 servers not to send body in response to a HEAD request, and
2205 those that do will likely be caught by test_socket_open.
2206 If not, they can be worked around using
2207 `--no-http-keep-alive'. */
2208 CLOSE_FINISH (sock);
2209 else if (keep_alive && skip_short_body (sock, contlen))
2210 /* Successfully skipped the body; also keep using the socket. */
2211 CLOSE_FINISH (sock);
2213 CLOSE_INVALIDATE (sock);
2214 return RETRFINISHED;
2217 /* Open the local file. */
2220 mkalldirs (hs->local_file);
2222 rotate_backups (hs->local_file);
2224 fp = fopen (hs->local_file, "ab");
2225 else if (ALLOW_CLOBBER)
2226 fp = fopen (hs->local_file, "wb");
2229 fp = fopen_excl (hs->local_file, true);
2230 if (!fp && errno == EEXIST)
2232 /* We cannot just invent a new name and use it (which is
2233 what functions like unique_create typically do)
2234 because we told the user we'd use this name.
2235 Instead, return and retry the download. */
2236 logprintf (LOG_NOTQUIET,
2237 _("%s has sprung into existence.\n"),
2239 CLOSE_INVALIDATE (sock);
2240 return FOPEN_EXCL_ERR;
2245 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2246 CLOSE_INVALIDATE (sock);
2253 /* Print fetch message, if opt.verbose. */
2256 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2257 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2260 /* This confuses the timestamping code that checks for file size.
2261 #### The timestamping code should be smarter about file size. */
2262 if (opt.save_headers && hs->restval == 0)
2263 fwrite (head, 1, strlen (head), fp);
2265 /* Now we no longer need to store the response header. */
2268 /* Download the request body. */
2271 /* If content-length is present, read that much; otherwise, read
2272 until EOF. The HTTP spec doesn't require the server to
2273 actually close the connection when it's done sending data. */
2274 flags |= rb_read_exactly;
2275 if (hs->restval > 0 && contrange == 0)
2276 /* If the server ignored our range request, instruct fd_read_body
2277 to skip the first RESTVAL bytes of body. */
2278 flags |= rb_skip_startpos;
2279 hs->len = hs->restval;
2281 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2282 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2286 CLOSE_FINISH (sock);
2290 hs->rderrmsg = xstrdup (fd_errstr (sock));
2291 CLOSE_INVALIDATE (sock);
2298 return RETRFINISHED;
2301 /* The genuine HTTP loop! This is the part where the retrieval is
2302 retried, and retried, and retried, and... */
2304 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2305 int *dt, struct url *proxy)
2308 bool got_head = false; /* used for time-stamping and filename detection */
2309 bool time_came_from_head = false;
2310 bool got_name = false;
2313 uerr_t err, ret = TRYLIMEXC;
2314 time_t tmr = -1; /* remote time-stamp */
2315 struct http_stat hstat; /* HTTP status */
2317 bool send_head_first = true;
2319 /* Assert that no value for *LOCAL_FILE was passed. */
2320 assert (local_file == NULL || *local_file == NULL);
2322 /* Set LOCAL_FILE parameter. */
2323 if (local_file && opt.output_document)
2324 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2326 /* Reset NEWLOC parameter. */
2329 /* This used to be done in main(), but it's a better idea to do it
2330 here so that we don't go through the hoops if we're just using
2335 /* Warn on (likely bogus) wildcard usage in HTTP. */
2336 if (opt.ftp_glob && has_wildcards_p (u->path))
2337 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2339 /* Setup hstat struct. */
2341 hstat.referer = referer;
2343 if (opt.output_document)
2345 hstat.local_file = xstrdup (opt.output_document);
2348 else if (!opt.content_disposition)
2350 hstat.local_file = url_file_name (u);
2354 /* Reset the counter. */
2357 /* Reset the document type. */
2360 /* Skip preliminary HEAD request if we're not in spider mode AND
2361 * if -O was given or HTTP Content-Disposition support is disabled. */
2363 && (got_name || !opt.content_disposition))
2364 send_head_first = false;
2366 /* Send preliminary HEAD request if -N is given and we have an existing
2367 * destination file. */
2368 if (opt.timestamping
2369 && !opt.content_disposition
2370 && file_exists_p (url_file_name (u)))
2371 send_head_first = true;
2376 /* Increment the pass counter. */
2378 sleep_between_retrievals (count);
2380 /* Get the current time string. */
2381 tms = datetime_str (time (NULL));
2383 if (opt.spider && !got_head)
2384 logprintf (LOG_VERBOSE, _("\
2385 Spider mode enabled. Check if remote file exists.\n"));
2387 /* Print fetch message, if opt.verbose. */
2390 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2395 sprintf (tmp, _("(try:%2d)"), count);
2396 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2401 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2406 ws_changetitle (hurl);
2411 /* Default document type is empty. However, if spider mode is
2412 on or time-stamping is employed, HEAD_ONLY commands is
2413 encoded within *dt. */
2414 if (send_head_first && !got_head)
2419 /* Decide whether or not to restart. */
2422 && stat (hstat.local_file, &st) == 0
2423 && S_ISREG (st.st_mode))
2424 /* When -c is used, continue from on-disk size. (Can't use
2425 hstat.len even if count>1 because we don't want a failed
2426 first attempt to clobber existing data.) */
2427 hstat.restval = st.st_size;
2429 /* otherwise, continue where the previous try left off */
2430 hstat.restval = hstat.len;
2434 /* Decide whether to send the no-cache directive. We send it in
2436 a) we're using a proxy, and we're past our first retrieval.
2437 Some proxies are notorious for caching incomplete data, so
2438 we require a fresh get.
2439 b) caching is explicitly inhibited. */
2440 if ((proxy && count > 1) /* a */
2441 || !opt.allow_cache) /* b */
2442 *dt |= SEND_NOCACHE;
2444 *dt &= ~SEND_NOCACHE;
2446 /* Try fetching the document, or at least its head. */
2447 err = gethttp (u, &hstat, dt, proxy);
2450 tms = datetime_str (time (NULL));
2452 /* Get the new location (with or without the redirection). */
2454 *newloc = xstrdup (hstat.newloc);
2458 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2459 case CONERROR: case READERR: case WRITEFAILED:
2460 case RANGEERR: case FOPEN_EXCL_ERR:
2461 /* Non-fatal errors continue executing the loop, which will
2462 bring them to "while" statement at the end, to judge
2463 whether the number of tries was exceeded. */
2464 printwhat (count, opt.ntry);
2466 case FWRITEERR: case FOPENERR:
2467 /* Another fatal error. */
2468 logputs (LOG_VERBOSE, "\n");
2469 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2470 hstat.local_file, strerror (errno));
2471 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2472 case SSLINITFAILED: case CONTNOTSUPPORTED:
2473 /* Fatal errors just return from the function. */
2477 /* Another fatal error. */
2478 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2482 /* Return the new location to the caller. */
2485 logprintf (LOG_NOTQUIET,
2486 _("ERROR: Redirection (%d) without location.\n"),
2496 /* The file was already fully retrieved. */
2500 /* Deal with you later. */
2503 /* All possibilities should have been exhausted. */
2507 if (!(*dt & RETROKF))
2512 /* #### Ugly ugly ugly! */
2513 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2514 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2517 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2519 && (hstat.statcode == 500 || hstat.statcode == 501))
2524 /* Maybe we should always keep track of broken links, not just in
2526 else if (opt.spider)
2528 /* #### Again: ugly ugly ugly! */
2530 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2531 nonexisting_url (hurl);
2532 logprintf (LOG_NOTQUIET, _("\
2533 Remote file does not exist -- broken link!!!\n"));
2537 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2538 tms, hstat.statcode, escnonprint (hstat.error));
2540 logputs (LOG_VERBOSE, "\n");
2546 /* Did we get the time-stamp? */
2549 bool restart_loop = false;
2551 if (opt.timestamping && !hstat.remote_time)
2553 logputs (LOG_NOTQUIET, _("\
2554 Last-modified header missing -- time-stamps turned off.\n"));
2556 else if (hstat.remote_time)
2558 /* Convert the date-string into struct tm. */
2559 tmr = http_atotm (hstat.remote_time);
2560 if (tmr == (time_t) (-1))
2561 logputs (LOG_VERBOSE, _("\
2562 Last-modified header invalid -- time-stamp ignored.\n"));
2563 if (*dt & HEAD_ONLY)
2564 time_came_from_head = true;
2567 /* The time-stamping section. */
2568 if (opt.timestamping)
2570 if (hstat.orig_file_name) /* Perform the following checks only
2571 if the file we're supposed to
2572 download already exists. */
2574 if (hstat.remote_time &&
2575 tmr != (time_t) (-1))
2577 /* Now time-stamping can be used validly. Time-stamping
2578 means that if the sizes of the local and remote file
2579 match, and local file is newer than the remote file,
2580 it will not be retrieved. Otherwise, the normal
2581 download procedure is resumed. */
2582 if (hstat.orig_file_tstamp >= tmr)
2584 if (hstat.contlen == -1
2585 || hstat.orig_file_size == hstat.contlen)
2587 logprintf (LOG_VERBOSE, _("\
2588 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2589 hstat.orig_file_name);
2595 logprintf (LOG_VERBOSE, _("\
2596 The sizes do not match (local %s) -- retrieving.\n"),
2597 number_to_static_string (hstat.orig_file_size));
2601 logputs (LOG_VERBOSE,
2602 _("Remote file is newer, retrieving.\n"));
2604 logputs (LOG_VERBOSE, "\n");
2608 /* free_hstat (&hstat); */
2609 hstat.timestamp_checked = true;
2610 restart_loop = true;
2619 logputs (LOG_VERBOSE, _("\
2620 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2621 restart_loop = true;
2625 logprintf (LOG_VERBOSE, _("\
2626 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2627 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2633 logprintf (LOG_VERBOSE, _("\
2634 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2635 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2640 if (send_head_first)
2643 restart_loop = true;
2646 got_head = true; /* no more time-stamping */
2648 count = 0; /* the retrieve count for HEAD is reset */
2654 if ((tmr != (time_t) (-1))
2655 && ((hstat.len == hstat.contlen) ||
2656 ((hstat.res == 0) && (hstat.contlen == -1))))
2658 /* #### This code repeats in http.c and ftp.c. Move it to a
2660 const char *fl = NULL;
2661 if (opt.output_document)
2663 if (output_stream_regular)
2664 fl = opt.output_document;
2667 fl = hstat.local_file;
2671 /* Reparse time header, in case it's changed. */
2672 if (time_came_from_head
2673 && hstat.remote_time && hstat.remote_time[0])
2675 newtmr = http_atotm (hstat.remote_time);
2682 /* End of time-stamping section. */
2684 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2685 total_download_time += hstat.dltime;
2687 if (hstat.len == hstat.contlen)
2691 logprintf (LOG_VERBOSE,
2692 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2693 tms, tmrate, hstat.local_file,
2694 number_to_static_string (hstat.len),
2695 number_to_static_string (hstat.contlen));
2696 logprintf (LOG_NONVERBOSE,
2697 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2699 number_to_static_string (hstat.len),
2700 number_to_static_string (hstat.contlen),
2701 hstat.local_file, count);
2704 total_downloaded_bytes += hstat.len;
2706 /* Remember that we downloaded the file for later ".orig" code. */
2707 if (*dt & ADDED_HTML_EXTENSION)
2708 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2710 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2715 else if (hstat.res == 0) /* No read error */
2717 if (hstat.contlen == -1) /* We don't know how much we were supposed
2718 to get, so assume we succeeded. */
2722 logprintf (LOG_VERBOSE,
2723 _("%s (%s) - `%s' saved [%s]\n\n"),
2724 tms, tmrate, hstat.local_file,
2725 number_to_static_string (hstat.len));
2726 logprintf (LOG_NONVERBOSE,
2727 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2728 tms, u->url, number_to_static_string (hstat.len),
2729 hstat.local_file, count);
2732 total_downloaded_bytes += hstat.len;
2734 /* Remember that we downloaded the file for later ".orig" code. */
2735 if (*dt & ADDED_HTML_EXTENSION)
2736 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2738 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2743 else if (hstat.len < hstat.contlen) /* meaning we lost the
2744 connection too soon */
2746 logprintf (LOG_VERBOSE,
2747 _("%s (%s) - Connection closed at byte %s. "),
2748 tms, tmrate, number_to_static_string (hstat.len));
2749 printwhat (count, opt.ntry);
2753 /* Getting here would mean reading more data than
2754 requested with content-length, which we never do. */
2757 else /* from now on hstat.res can only be -1 */
2759 if (hstat.contlen == -1)
2761 logprintf (LOG_VERBOSE,
2762 _("%s (%s) - Read error at byte %s (%s)."),
2763 tms, tmrate, number_to_static_string (hstat.len),
2765 printwhat (count, opt.ntry);
2768 else /* hstat.res == -1 and contlen is given */
2770 logprintf (LOG_VERBOSE,
2771 _("%s (%s) - Read error at byte %s/%s (%s). "),
2773 number_to_static_string (hstat.len),
2774 number_to_static_string (hstat.contlen),
2776 printwhat (count, opt.ntry);
2782 while (!opt.ntry || (count < opt.ntry));
2786 *local_file = xstrdup (hstat.local_file);
2787 free_hstat (&hstat);
2792 /* Check whether the result of strptime() indicates success.
2793 strptime() returns the pointer to how far it got to in the string.
2794 The processing has been successful if the string is at `GMT' or
2795 `+X', or at the end of the string.
2797 In extended regexp parlance, the function returns 1 if P matches
2798 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2799 can return) is considered a failure and 0 is returned. */
2801 check_end (const char *p)
2805 while (ISSPACE (*p))
2808 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2809 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2815 /* Convert the textual specification of time in TIME_STRING to the
2816 number of seconds since the Epoch.
2818 TIME_STRING can be in any of the three formats RFC2616 allows the
2819 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2820 as well as the time format used in the Set-Cookie header.
2821 Timezones are ignored, and should be GMT.
2823 Return the computed time_t representation, or -1 if the conversion
2826 This function uses strptime with various string formats for parsing
2827 TIME_STRING. This results in a parser that is not as lenient in
2828 interpreting TIME_STRING as I would like it to be. Being based on
2829 strptime, it always allows shortened months, one-digit days, etc.,
2830 but due to the multitude of formats in which time can be
2831 represented, an ideal HTTP time parser would be even more
2832 forgiving. It should completely ignore things like week days and
2833 concentrate only on the various forms of representing years,
2834 months, days, hours, minutes, and seconds. For example, it would
2835 be nice if it accepted ISO 8601 out of the box.
2837 I've investigated free and PD code for this purpose, but none was
2838 usable. getdate was big and unwieldy, and had potential copyright
2839 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2840 distributed with phttpd, is excellent, but we cannot use it because
2841 it is not assigned to the FSF. So I stuck it with strptime. */
2844 http_atotm (const char *time_string)
2846 /* NOTE: Solaris strptime man page claims that %n and %t match white
2847 space, but that's not universally available. Instead, we simply
2848 use ` ' to mean "skip all WS", which works under all strptime
2849 implementations I've tested. */
2851 static const char *time_formats[] = {
2852 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2853 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2854 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2855 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2856 (used in Set-Cookie, defined in the
2857 Netscape cookie specification.) */
2859 const char *oldlocale;
2861 time_t ret = (time_t) -1;
2863 /* Solaris strptime fails to recognize English month names in
2864 non-English locales, which we work around by temporarily setting
2865 locale to C before invoking strptime. */
2866 oldlocale = setlocale (LC_TIME, NULL);
2867 setlocale (LC_TIME, "C");
2869 for (i = 0; i < countof (time_formats); i++)
2873 /* Some versions of strptime use the existing contents of struct
2874 tm to recalculate the date according to format. Zero it out
2875 to prevent stack garbage from influencing strptime. */
2878 if (check_end (strptime (time_string, time_formats[i], &t)))
2885 /* Restore the previous locale. */
2886 setlocale (LC_TIME, oldlocale);
2891 /* Authorization support: We support three authorization schemes:
2893 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2895 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2896 consisting of answering to the server's challenge with the proper
2899 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2900 Stenberg for libcurl. Like digest, NTLM is based on a
2901 challenge-response mechanism, but unlike digest, it is non-standard
2902 (authenticates TCP connections rather than requests), undocumented
2903 and Microsoft-specific. */
2905 /* Create the authentication header contents for the `Basic' scheme.
2906 This is done by encoding the string "USER:PASS" to base64 and
2907 prepending the string "Basic " in front of it. */
2910 basic_authentication_encode (const char *user, const char *passwd)
2913 int len1 = strlen (user) + 1 + strlen (passwd);
2915 t1 = (char *)alloca (len1 + 1);
2916 sprintf (t1, "%s:%s", user, passwd);
2918 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2919 base64_encode (t1, len1, t2);
2921 return concat_strings ("Basic ", t2, (char *) 0);
2924 #define SKIP_WS(x) do { \
2925 while (ISSPACE (*(x))) \
2929 #ifdef ENABLE_DIGEST
2930 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2931 an array of 16 bytes containing the hash keys, and BUF should be a
2932 buffer of 33 writable characters (32 for hex digits plus one for
2933 zero termination). */
2935 dump_hash (char *buf, const unsigned char *hash)
2939 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2941 *buf++ = XNUM_TO_digit (*hash >> 4);
2942 *buf++ = XNUM_TO_digit (*hash & 0xf);
2947 /* Take the line apart to find the challenge, and compose a digest
2948 authorization header. See RFC2069 section 2.1.2. */
2950 digest_authentication_encode (const char *au, const char *user,
2951 const char *passwd, const char *method,
2954 static char *realm, *opaque, *nonce;
2959 { "realm", &realm },
2960 { "opaque", &opaque },
2964 param_token name, value;
2966 realm = opaque = nonce = NULL;
2968 au += 6; /* skip over `Digest' */
2969 while (extract_param (&au, &name, &value, ','))
2972 for (i = 0; i < countof (options); i++)
2973 if (name.e - name.b == strlen (options[i].name)
2974 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2976 *options[i].variable = strdupdelim (value.b, value.e);
2980 if (!realm || !nonce || !user || !passwd || !path || !method)
2983 xfree_null (opaque);
2988 /* Calculate the digest value. */
2990 ALLOCA_MD5_CONTEXT (ctx);
2991 unsigned char hash[MD5_HASHLEN];
2992 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2993 char response_digest[MD5_HASHLEN * 2 + 1];
2995 /* A1BUF = H(user ":" realm ":" password) */
2997 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2998 gen_md5_update ((unsigned char *)":", 1, ctx);
2999 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3000 gen_md5_update ((unsigned char *)":", 1, ctx);
3001 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3002 gen_md5_finish (ctx, hash);
3003 dump_hash (a1buf, hash);
3005 /* A2BUF = H(method ":" path) */
3007 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3008 gen_md5_update ((unsigned char *)":", 1, ctx);
3009 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3010 gen_md5_finish (ctx, hash);
3011 dump_hash (a2buf, hash);
3013 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3015 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3016 gen_md5_update ((unsigned char *)":", 1, ctx);
3017 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3018 gen_md5_update ((unsigned char *)":", 1, ctx);
3019 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3020 gen_md5_finish (ctx, hash);
3021 dump_hash (response_digest, hash);
3023 res = xmalloc (strlen (user)
3028 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3029 + (opaque ? strlen (opaque) : 0)
3031 sprintf (res, "Digest \
3032 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3033 user, realm, nonce, path, response_digest);
3036 char *p = res + strlen (res);
3037 strcat (p, ", opaque=\"");
3044 #endif /* ENABLE_DIGEST */
3046 /* Computing the size of a string literal must take into account that
3047 value returned by sizeof includes the terminating \0. */
3048 #define STRSIZE(literal) (sizeof (literal) - 1)
3050 /* Whether chars in [b, e) begin with the literal string provided as
3051 first argument and are followed by whitespace or terminating \0.
3052 The comparison is case-insensitive. */
3053 #define STARTS(literal, b, e) \
3054 ((e) - (b) >= STRSIZE (literal) \
3055 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3056 && ((e) - (b) == STRSIZE (literal) \
3057 || ISSPACE (b[STRSIZE (literal)])))
3060 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3062 return STARTS ("Basic", hdrbeg, hdrend)
3063 #ifdef ENABLE_DIGEST
3064 || STARTS ("Digest", hdrbeg, hdrend)
3067 || STARTS ("NTLM", hdrbeg, hdrend)
3074 /* Create the HTTP authorization request header. When the
3075 `WWW-Authenticate' response header is seen, according to the
3076 authorization scheme specified in that header (`Basic' and `Digest'
3077 are supported by the current implementation), produce an
3078 appropriate HTTP authorization request header. */
3080 create_authorization_line (const char *au, const char *user,
3081 const char *passwd, const char *method,
3082 const char *path, bool *finished)
3084 /* We are called only with known schemes, so we can dispatch on the
3086 switch (TOUPPER (*au))
3088 case 'B': /* Basic */
3090 return basic_authentication_encode (user, passwd);
3091 #ifdef ENABLE_DIGEST
3092 case 'D': /* Digest */
3094 return digest_authentication_encode (au, user, passwd, method, path);
3097 case 'N': /* NTLM */
3098 if (!ntlm_input (&pconn.ntlm, au))
3103 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3106 /* We shouldn't get here -- this function should be only called
3107 with values approved by known_authentication_scheme_p. */
3115 if (!wget_cookie_jar)
3116 wget_cookie_jar = cookie_jar_new ();
3117 if (opt.cookies_input && !cookies_loaded_p)
3119 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3120 cookies_loaded_p = true;
3127 if (wget_cookie_jar)
3128 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3134 xfree_null (pconn.host);
3135 if (wget_cookie_jar)
3136 cookie_jar_delete (wget_cookie_jar);
3143 test_parse_content_disposition()
3148 char *opt_dir_prefix;
3152 { "filename=\"file.ext\"", NULL, "file.ext", true },
3153 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3154 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3155 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3156 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3157 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3158 { "attachment", NULL, NULL, false },
3159 { "attachment", "somedir", NULL, false },
3162 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3167 opt.dir_prefix = test_array[i].opt_dir_prefix;
3168 res = parse_content_disposition (test_array[i].hdrval, &filename);
3170 mu_assert ("test_parse_content_disposition: wrong result",
3171 res == test_array[i].result
3173 || 0 == strcmp (test_array[i].filename, filename)));
3179 #endif /* TESTING */
3182 * vim: et sts=2 sw=2 cino+={s