2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010 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. */
56 # include "http-ntlm.h"
69 #endif /* def __VMS */
71 extern char *version_string;
75 static char *create_authorization_line (const char *, const char *,
76 const char *, const char *,
77 const char *, bool *);
78 static char *basic_authentication_encode (const char *, const char *);
79 static bool known_authentication_scheme_p (const char *, const char *);
80 static void ensure_extension (struct http_stat *, const char *, int *);
81 static void load_cookies (void);
84 # define MIN(x, y) ((x) > (y) ? (y) : (x))
88 static bool cookies_loaded_p;
89 static struct cookie_jar *wget_cookie_jar;
91 #define TEXTHTML_S "text/html"
92 #define TEXTXHTML_S "application/xhtml+xml"
93 #define TEXTCSS_S "text/css"
95 /* Some status code validation macros: */
96 #define H_20X(x) (((x) >= 200) && ((x) < 300))
97 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
98 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
99 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
100 || (x) == HTTP_STATUS_SEE_OTHER \
101 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
103 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
104 /* Successful 2xx. */
105 #define HTTP_STATUS_OK 200
106 #define HTTP_STATUS_CREATED 201
107 #define HTTP_STATUS_ACCEPTED 202
108 #define HTTP_STATUS_NO_CONTENT 204
109 #define HTTP_STATUS_PARTIAL_CONTENTS 206
111 /* Redirection 3xx. */
112 #define HTTP_STATUS_MULTIPLE_CHOICES 300
113 #define HTTP_STATUS_MOVED_PERMANENTLY 301
114 #define HTTP_STATUS_MOVED_TEMPORARILY 302
115 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
116 #define HTTP_STATUS_NOT_MODIFIED 304
117 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
119 /* Client error 4xx. */
120 #define HTTP_STATUS_BAD_REQUEST 400
121 #define HTTP_STATUS_UNAUTHORIZED 401
122 #define HTTP_STATUS_FORBIDDEN 403
123 #define HTTP_STATUS_NOT_FOUND 404
124 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
126 /* Server errors 5xx. */
127 #define HTTP_STATUS_INTERNAL 500
128 #define HTTP_STATUS_NOT_IMPLEMENTED 501
129 #define HTTP_STATUS_BAD_GATEWAY 502
130 #define HTTP_STATUS_UNAVAILABLE 503
133 rel_none, rel_name, rel_value, rel_both
140 struct request_header {
142 enum rp release_policy;
144 int hcount, hcapacity;
149 /* Create a new, empty request. At least request_set_method must be
150 called before the request can be used. */
152 static struct request *
155 struct request *req = xnew0 (struct request);
157 req->headers = xnew_array (struct request_header, req->hcapacity);
161 /* Set the request's method and its arguments. METH should be a
162 literal string (or it should outlive the request) because it will
163 not be freed. ARG will be freed by request_free. */
166 request_set_method (struct request *req, const char *meth, char *arg)
172 /* Return the method string passed with the last call to
173 request_set_method. */
176 request_method (const struct request *req)
181 /* Free one header according to the release policy specified with
182 request_set_header. */
185 release_header (struct request_header *hdr)
187 switch (hdr->release_policy)
204 /* Set the request named NAME to VALUE. Specifically, this means that
205 a "NAME: VALUE\r\n" header line will be used in the request. If a
206 header with the same name previously existed in the request, its
207 value will be replaced by this one. A NULL value means do nothing.
209 RELEASE_POLICY determines whether NAME and VALUE should be released
210 (freed) with request_free. Allowed values are:
212 - rel_none - don't free NAME or VALUE
213 - rel_name - free NAME when done
214 - rel_value - free VALUE when done
215 - rel_both - free both NAME and VALUE when done
217 Setting release policy is useful when arguments come from different
218 sources. For example:
220 // Don't free literal strings!
221 request_set_header (req, "Pragma", "no-cache", rel_none);
223 // Don't free a global variable, we'll need it later.
224 request_set_header (req, "Referer", opt.referer, rel_none);
226 // Value freshly allocated, free it when done.
227 request_set_header (req, "Range",
228 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
233 request_set_header (struct request *req, char *name, char *value,
234 enum rp release_policy)
236 struct request_header *hdr;
241 /* A NULL value is a no-op; if freeing the name is requested,
242 free it now to avoid leaks. */
243 if (release_policy == rel_name || release_policy == rel_both)
248 for (i = 0; i < req->hcount; i++)
250 hdr = &req->headers[i];
251 if (0 == strcasecmp (name, hdr->name))
253 /* Replace existing header. */
254 release_header (hdr);
257 hdr->release_policy = release_policy;
262 /* Install new header. */
264 if (req->hcount >= req->hcapacity)
266 req->hcapacity <<= 1;
267 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
269 hdr = &req->headers[req->hcount++];
272 hdr->release_policy = release_policy;
275 /* Like request_set_header, but sets the whole header line, as
276 provided by the user using the `--header' option. For example,
277 request_set_user_header (req, "Foo: bar") works just like
278 request_set_header (req, "Foo", "bar"). */
281 request_set_user_header (struct request *req, const char *header)
284 const char *p = strchr (header, ':');
287 BOUNDED_TO_ALLOCA (header, p, name);
289 while (c_isspace (*p))
291 request_set_header (req, xstrdup (name), (char *) p, rel_name);
294 /* Remove the header with specified name from REQ. Returns true if
295 the header was actually removed, false otherwise. */
298 request_remove_header (struct request *req, char *name)
301 for (i = 0; i < req->hcount; i++)
303 struct request_header *hdr = &req->headers[i];
304 if (0 == strcasecmp (name, hdr->name))
306 release_header (hdr);
307 /* Move the remaining headers by one. */
308 if (i < req->hcount - 1)
309 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
317 #define APPEND(p, str) do { \
318 int A_len = strlen (str); \
319 memcpy (p, str, A_len); \
323 /* Construct the request and write it to FD using fd_write. */
326 request_send (const struct request *req, int fd)
328 char *request_string, *p;
329 int i, size, write_error;
331 /* Count the request size. */
334 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
335 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
337 for (i = 0; i < req->hcount; i++)
339 struct request_header *hdr = &req->headers[i];
340 /* NAME ": " VALUE "\r\n" */
341 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
347 p = request_string = alloca_array (char, size);
349 /* Generate the request. */
351 APPEND (p, req->method); *p++ = ' ';
352 APPEND (p, req->arg); *p++ = ' ';
353 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
355 for (i = 0; i < req->hcount; i++)
357 struct request_header *hdr = &req->headers[i];
358 APPEND (p, hdr->name);
359 *p++ = ':', *p++ = ' ';
360 APPEND (p, hdr->value);
361 *p++ = '\r', *p++ = '\n';
364 *p++ = '\r', *p++ = '\n', *p++ = '\0';
365 assert (p - request_string == size);
369 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
371 /* Send the request to the server. */
373 write_error = fd_write (fd, request_string, size - 1, -1);
375 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
380 /* Release the resources used by REQ. */
383 request_free (struct request *req)
386 xfree_null (req->arg);
387 for (i = 0; i < req->hcount; i++)
388 release_header (&req->headers[i]);
389 xfree_null (req->headers);
393 static struct hash_table *basic_authed_hosts;
395 /* Find out if this host has issued a Basic challenge yet; if so, give
396 * it the username, password. A temporary measure until we can get
397 * proper authentication in place. */
400 maybe_send_basic_creds (const char *hostname, const char *user,
401 const char *passwd, struct request *req)
403 bool do_challenge = false;
405 if (opt.auth_without_challenge)
407 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
410 else if (basic_authed_hosts
411 && hash_table_contains(basic_authed_hosts, hostname))
413 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
418 DEBUGP(("Host %s has not issued a general basic challenge.\n",
423 request_set_header (req, "Authorization",
424 basic_authentication_encode (user, passwd),
431 register_basic_auth_host (const char *hostname)
433 if (!basic_authed_hosts)
435 basic_authed_hosts = make_nocase_string_hash_table (1);
437 if (!hash_table_contains(basic_authed_hosts, hostname))
439 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
440 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
445 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
446 PROMISED_SIZE bytes are sent over the wire -- if the file is
447 longer, read only that much; if the file is shorter, report an error. */
450 post_file (int sock, const char *file_name, wgint promised_size)
452 static char chunk[8192];
457 DEBUGP (("[writing POST file %s ... ", file_name));
459 fp = fopen (file_name, "rb");
462 while (!feof (fp) && written < promised_size)
465 int length = fread (chunk, 1, sizeof (chunk), fp);
468 towrite = MIN (promised_size - written, length);
469 write_error = fd_write (sock, chunk, towrite, -1);
479 /* If we've written less than was promised, report a (probably
480 nonsensical) error rather than break the promise. */
481 if (written < promised_size)
487 assert (written == promised_size);
488 DEBUGP (("done]\n"));
492 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
493 If so, return the pointer to the position after the line, otherwise
494 return NULL. This is used as callback to fd_read_hunk. The data
495 between START and PEEKED has been read and cannot be "unread"; the
496 data after PEEKED has only been peeked. */
499 response_head_terminator (const char *start, const char *peeked, int peeklen)
503 /* If at first peek, verify whether HUNK starts with "HTTP". If
504 not, this is a HTTP/0.9 request and we must bail out without
506 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
509 /* Look for "\n[\r]\n", and return the following position if found.
510 Start two chars before the current to cover the possibility that
511 part of the terminator (e.g. "\n\r") arrived in the previous
513 p = peeked - start < 2 ? start : peeked - 2;
514 end = peeked + peeklen;
516 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
517 for (; p < end - 2; p++)
520 if (p[1] == '\r' && p[2] == '\n')
522 else if (p[1] == '\n')
525 /* p==end-2: check for \n\n directly preceding END. */
526 if (p[0] == '\n' && p[1] == '\n')
532 /* The maximum size of a single HTTP response we care to read. Rather
533 than being a limit of the reader implementation, this limit
534 prevents Wget from slurping all available memory upon encountering
535 malicious or buggy server output, thus protecting the user. Define
536 it to 0 to remove the limit. */
538 #define HTTP_RESPONSE_MAX_SIZE 65536
540 /* Read the HTTP request head from FD and return it. The error
541 conditions are the same as with fd_read_hunk.
543 To support HTTP/0.9 responses, this function tries to make sure
544 that the data begins with "HTTP". If this is not the case, no data
545 is read and an empty request is returned, so that the remaining
546 data can be treated as body. */
549 read_http_response_head (int fd)
551 return fd_read_hunk (fd, response_head_terminator, 512,
552 HTTP_RESPONSE_MAX_SIZE);
556 /* The response data. */
559 /* The array of pointers that indicate where each header starts.
560 For example, given this HTTP response:
567 The headers are located like this:
569 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
571 headers[0] headers[1] headers[2] headers[3]
573 I.e. headers[0] points to the beginning of the request,
574 headers[1] points to the end of the first header and the
575 beginning of the second one, etc. */
577 const char **headers;
580 /* Create a new response object from the text of the HTTP response,
581 available in HEAD. That text is automatically split into
582 constituent header lines for fast retrieval using
585 static struct response *
586 resp_new (const char *head)
591 struct response *resp = xnew0 (struct response);
596 /* Empty head means that we're dealing with a headerless
597 (HTTP/0.9) response. In that case, don't set HEADERS at
602 /* Split HEAD into header lines, so that resp_header_* functions
603 don't need to do this over and over again. */
609 DO_REALLOC (resp->headers, size, count + 1, const char *);
610 resp->headers[count++] = hdr;
612 /* Break upon encountering an empty line. */
613 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
616 /* Find the end of HDR, including continuations. */
619 const char *end = strchr (hdr, '\n');
625 while (*hdr == ' ' || *hdr == '\t');
627 DO_REALLOC (resp->headers, size, count + 1, const char *);
628 resp->headers[count] = NULL;
633 /* Locate the header named NAME in the request data, starting with
634 position START. This allows the code to loop through the request
635 data, filtering for all requests of a given name. Returns the
636 found position, or -1 for failure. The code that uses this
637 function typically looks like this:
639 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
640 ... do something with header ...
642 If you only care about one header, use resp_header_get instead of
646 resp_header_locate (const struct response *resp, const char *name, int start,
647 const char **begptr, const char **endptr)
650 const char **headers = resp->headers;
653 if (!headers || !headers[1])
656 name_len = strlen (name);
662 for (; headers[i + 1]; i++)
664 const char *b = headers[i];
665 const char *e = headers[i + 1];
667 && b[name_len] == ':'
668 && 0 == strncasecmp (b, name, name_len))
671 while (b < e && c_isspace (*b))
673 while (b < e && c_isspace (e[-1]))
683 /* Find and retrieve the header named NAME in the request data. If
684 found, set *BEGPTR to its starting, and *ENDPTR to its ending
685 position, and return true. Otherwise return false.
687 This function is used as a building block for resp_header_copy
688 and resp_header_strdup. */
691 resp_header_get (const struct response *resp, const char *name,
692 const char **begptr, const char **endptr)
694 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
698 /* Copy the response header named NAME to buffer BUF, no longer than
699 BUFSIZE (BUFSIZE includes the terminating 0). If the header
700 exists, true is returned, false otherwise. If there should be no
701 limit on the size of the header, use resp_header_strdup instead.
703 If BUFSIZE is 0, no data is copied, but the boolean indication of
704 whether the header is present is still returned. */
707 resp_header_copy (const struct response *resp, const char *name,
708 char *buf, int bufsize)
711 if (!resp_header_get (resp, name, &b, &e))
715 int len = MIN (e - b, bufsize - 1);
716 memcpy (buf, b, len);
722 /* Return the value of header named NAME in RESP, allocated with
723 malloc. If such a header does not exist in RESP, return NULL. */
726 resp_header_strdup (const struct response *resp, const char *name)
729 if (!resp_header_get (resp, name, &b, &e))
731 return strdupdelim (b, e);
734 /* Parse the HTTP status line, which is of format:
736 HTTP-Version SP Status-Code SP Reason-Phrase
738 The function returns the status-code, or -1 if the status line
739 appears malformed. The pointer to "reason-phrase" message is
740 returned in *MESSAGE. */
743 resp_status (const struct response *resp, char **message)
750 /* For a HTTP/0.9 response, assume status 200. */
752 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
756 p = resp->headers[0];
757 end = resp->headers[1];
763 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
767 /* Match the HTTP version. This is optional because Gnutella
768 servers have been reported to not specify HTTP version. */
769 if (p < end && *p == '/')
772 while (p < end && c_isdigit (*p))
774 if (p < end && *p == '.')
776 while (p < end && c_isdigit (*p))
780 while (p < end && c_isspace (*p))
782 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
785 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
790 while (p < end && c_isspace (*p))
792 while (p < end && c_isspace (end[-1]))
794 *message = strdupdelim (p, end);
800 /* Release the resources used by RESP. */
803 resp_free (struct response *resp)
805 xfree_null (resp->headers);
809 /* Print a single line of response, the characters [b, e). We tried
811 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
812 but that failed to escape the non-printable characters and, in fact,
813 caused crashes in UTF-8 locales. */
816 print_response_line(const char *prefix, const char *b, const char *e)
819 BOUNDED_TO_ALLOCA(b, e, copy);
820 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
821 quotearg_style (escape_quoting_style, copy));
824 /* Print the server response, line by line, omitting the trailing CRLF
825 from individual header lines, and prefixed with PREFIX. */
828 print_server_response (const struct response *resp, const char *prefix)
833 for (i = 0; resp->headers[i + 1]; i++)
835 const char *b = resp->headers[i];
836 const char *e = resp->headers[i + 1];
838 if (b < e && e[-1] == '\n')
840 if (b < e && e[-1] == '\r')
842 print_response_line(prefix, b, e);
846 /* Parse the `Content-Range' header and extract the information it
847 contains. Returns true if successful, false otherwise. */
849 parse_content_range (const char *hdr, wgint *first_byte_ptr,
850 wgint *last_byte_ptr, wgint *entity_length_ptr)
854 /* Ancient versions of Netscape proxy server, presumably predating
855 rfc2068, sent out `Content-Range' without the "bytes"
857 if (0 == strncasecmp (hdr, "bytes", 5))
860 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
864 while (c_isspace (*hdr))
869 if (!c_isdigit (*hdr))
871 for (num = 0; c_isdigit (*hdr); hdr++)
872 num = 10 * num + (*hdr - '0');
873 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
875 *first_byte_ptr = num;
877 for (num = 0; c_isdigit (*hdr); hdr++)
878 num = 10 * num + (*hdr - '0');
879 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
881 *last_byte_ptr = num;
886 for (num = 0; c_isdigit (*hdr); hdr++)
887 num = 10 * num + (*hdr - '0');
888 *entity_length_ptr = num;
892 /* Read the body of the request, but don't store it anywhere and don't
893 display a progress gauge. This is useful for reading the bodies of
894 administrative responses to which we will soon issue another
895 request. The response is not useful to the user, but reading it
896 allows us to continue using the same connection to the server.
898 If reading fails, false is returned, true otherwise. In debug
899 mode, the body is displayed for debugging purposes. */
902 skip_short_body (int fd, wgint contlen, bool chunked)
905 SKIP_SIZE = 512, /* size of the download buffer */
906 SKIP_THRESHOLD = 4096 /* the largest size we read */
908 wgint remaining_chunk_size = 0;
909 char dlbuf[SKIP_SIZE + 1];
910 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
912 assert (contlen != -1 || contlen);
914 /* If the body is too large, it makes more sense to simply close the
915 connection than to try to read the body. */
916 if (contlen > SKIP_THRESHOLD)
919 while (contlen > 0 || chunked)
924 if (remaining_chunk_size == 0)
926 char *line = fd_read_line (fd);
934 remaining_chunk_size = strtol (line, &endl, 16);
935 if (remaining_chunk_size == 0)
938 if (fd_read_line (fd) == NULL)
944 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
947 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
949 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
952 /* Don't normally report the error since this is an
953 optimization that should be invisible to the user. */
954 DEBUGP (("] aborting (%s).\n",
955 ret < 0 ? fd_errstr (fd) : "EOF received"));
962 remaining_chunk_size -= ret;
963 if (remaining_chunk_size == 0)
964 if (fd_read_line (fd) == NULL)
968 /* Safe even if %.*s bogusly expects terminating \0 because
969 we've zero-terminated dlbuf above. */
970 DEBUGP (("%.*s", ret, dlbuf));
973 DEBUGP (("] done.\n"));
977 #define NOT_RFC2231 0
978 #define RFC2231_NOENCODING 1
979 #define RFC2231_ENCODING 2
981 /* extract_param extracts the parameter name into NAME.
982 However, if the parameter name is in RFC2231 format then
983 this function adjusts NAME by stripping of the trailing
984 characters that are not part of the name but are present to
985 indicate the presence of encoding information in the value
986 or a fragment of a long parameter value
989 modify_param_name(param_token *name)
991 const char *delim1 = memchr (name->b, '*', name->e - name->b);
992 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
998 result = NOT_RFC2231;
1000 else if(delim1 == delim2)
1002 if ((name->e - 1) == delim1)
1004 result = RFC2231_ENCODING;
1008 result = RFC2231_NOENCODING;
1015 result = RFC2231_ENCODING;
1020 /* extract_param extract the paramater value into VALUE.
1021 Like modify_param_name this function modifies VALUE by
1022 stripping off the encoding information from the actual value
1025 modify_param_value (param_token *value, int encoding_type )
1027 if (RFC2231_ENCODING == encoding_type)
1029 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1030 if ( delim != NULL )
1032 value->b = (delim+1);
1037 /* Extract a parameter from the string (typically an HTTP header) at
1038 **SOURCE and advance SOURCE to the next parameter. Return false
1039 when there are no more parameters to extract. The name of the
1040 parameter is returned in NAME, and the value in VALUE. If the
1041 parameter has no value, the token's value is zeroed out.
1043 For example, if *SOURCE points to the string "attachment;
1044 filename=\"foo bar\"", the first call to this function will return
1045 the token named "attachment" and no value, and the second call will
1046 return the token named "filename" and value "foo bar". The third
1047 call will return false, indicating no more valid tokens. */
1050 extract_param (const char **source, param_token *name, param_token *value,
1053 const char *p = *source;
1055 while (c_isspace (*p)) ++p;
1059 return false; /* no error; nothing more to extract */
1064 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1066 if (name->b == name->e)
1067 return false; /* empty name: error */
1068 while (c_isspace (*p)) ++p;
1069 if (*p == separator || !*p) /* no value */
1072 if (*p == separator) ++p;
1077 return false; /* error */
1079 /* *p is '=', extract value */
1081 while (c_isspace (*p)) ++p;
1082 if (*p == '"') /* quoted */
1085 while (*p && *p != '"') ++p;
1089 /* Currently at closing quote; find the end of param. */
1090 while (c_isspace (*p)) ++p;
1091 while (*p && *p != separator) ++p;
1092 if (*p == separator)
1095 /* garbage after closed quote, e.g. foo="bar"baz */
1101 while (*p && *p != separator) ++p;
1103 while (value->e != value->b && c_isspace (value->e[-1]))
1105 if (*p == separator) ++p;
1109 int param_type = modify_param_name(name);
1110 if (NOT_RFC2231 != param_type)
1112 modify_param_value(value, param_type);
1118 #undef RFC2231_NOENCODING
1119 #undef RFC2231_ENCODING
1121 /* Appends the string represented by VALUE to FILENAME */
1124 append_value_to_filename (char **filename, param_token const * const value)
1126 int original_length = strlen(*filename);
1127 int new_length = strlen(*filename) + (value->e - value->b);
1128 *filename = xrealloc (*filename, new_length+1);
1129 memcpy (*filename + original_length, value->b, (value->e - value->b));
1130 (*filename)[new_length] = '\0';
1134 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1136 /* Parse the contents of the `Content-Disposition' header, extracting
1137 the information useful to Wget. Content-Disposition is a header
1138 borrowed from MIME; when used in HTTP, it typically serves for
1139 specifying the desired file name of the resource. For example:
1141 Content-Disposition: attachment; filename="flora.jpg"
1143 Wget will skip the tokens it doesn't care about, such as
1144 "attachment" in the previous example; it will also skip other
1145 unrecognized params. If the header is syntactically correct and
1146 contains a file name, a copy of the file name is stored in
1147 *filename and true is returned. Otherwise, the function returns
1150 The file name is stripped of directory components and must not be
1153 parse_content_disposition (const char *hdr, char **filename)
1156 param_token name, value;
1157 while (extract_param (&hdr, &name, &value, ';'))
1159 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1160 if ( isFilename && value.b != NULL)
1162 /* Make the file name begin at the last slash or backslash. */
1163 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1164 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1165 if (last_slash && last_bs)
1166 value.b = 1 + MAX (last_slash, last_bs);
1167 else if (last_slash || last_bs)
1168 value.b = 1 + (last_slash ? last_slash : last_bs);
1169 if (value.b == value.e)
1171 /* Start with the directory prefix, if specified. */
1176 int prefix_length = strlen (opt.dir_prefix);
1177 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1182 total_length = prefix_length + (value.e - value.b);
1183 *filename = xmalloc (total_length + 1);
1184 strcpy (*filename, opt.dir_prefix);
1186 (*filename)[prefix_length - 1] = '/';
1187 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1188 (*filename)[total_length] = '\0';
1192 append_value_to_filename (filename, &value);
1199 append_value_to_filename (filename, &value);
1203 *filename = strdupdelim (value.b, value.e);
1219 /* Persistent connections. Currently, we cache the most recently used
1220 connection as persistent, provided that the HTTP server agrees to
1221 make it such. The persistence data is stored in the variables
1222 below. Ideally, it should be possible to cache an arbitrary fixed
1223 number of these connections. */
1225 /* Whether a persistent connection is active. */
1226 static bool pconn_active;
1229 /* The socket of the connection. */
1232 /* Host and port of the currently active persistent connection. */
1236 /* Whether a ssl handshake has occoured on this connection. */
1239 /* Whether the connection was authorized. This is only done by
1240 NTLM, which authorizes *connections* rather than individual
1241 requests. (That practice is peculiar for HTTP, but it is a
1242 useful optimization.) */
1246 /* NTLM data of the current connection. */
1247 struct ntlmdata ntlm;
1251 /* Mark the persistent connection as invalid and free the resources it
1252 uses. This is used by the CLOSE_* macros after they forcefully
1253 close a registered persistent connection. */
1256 invalidate_persistent (void)
1258 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1259 pconn_active = false;
1260 fd_close (pconn.socket);
1265 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1266 persistent. This will enable someone to use the same connection
1267 later. In the context of HTTP, this must be called only AFTER the
1268 response has been received and the server has promised that the
1269 connection will remain alive.
1271 If a previous connection was persistent, it is closed. */
1274 register_persistent (const char *host, int port, int fd, bool ssl)
1278 if (pconn.socket == fd)
1280 /* The connection FD is already registered. */
1285 /* The old persistent connection is still active; close it
1286 first. This situation arises whenever a persistent
1287 connection exists, but we then connect to a different
1288 host, and try to register a persistent connection to that
1290 invalidate_persistent ();
1294 pconn_active = true;
1296 pconn.host = xstrdup (host);
1299 pconn.authorized = false;
1301 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1304 /* Return true if a persistent connection is available for connecting
1308 persistent_available_p (const char *host, int port, bool ssl,
1309 bool *host_lookup_failed)
1311 /* First, check whether a persistent connection is active at all. */
1315 /* If we want SSL and the last connection wasn't or vice versa,
1316 don't use it. Checking for host and port is not enough because
1317 HTTP and HTTPS can apparently coexist on the same port. */
1318 if (ssl != pconn.ssl)
1321 /* If we're not connecting to the same port, we're not interested. */
1322 if (port != pconn.port)
1325 /* If the host is the same, we're in business. If not, there is
1326 still hope -- read below. */
1327 if (0 != strcasecmp (host, pconn.host))
1329 /* Check if pconn.socket is talking to HOST under another name.
1330 This happens often when both sites are virtual hosts
1331 distinguished only by name and served by the same network
1332 interface, and hence the same web server (possibly set up by
1333 the ISP and serving many different web sites). This
1334 admittedly unconventional optimization does not contradict
1335 HTTP and works well with popular server software. */
1339 struct address_list *al;
1342 /* Don't try to talk to two different SSL sites over the same
1343 secure connection! (Besides, it's not clear that
1344 name-based virtual hosting is even possible with SSL.) */
1347 /* If pconn.socket's peer is one of the IP addresses HOST
1348 resolves to, pconn.socket is for all intents and purposes
1349 already talking to HOST. */
1351 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1353 /* Can't get the peer's address -- something must be very
1354 wrong with the connection. */
1355 invalidate_persistent ();
1358 al = lookup_host (host, 0);
1361 *host_lookup_failed = true;
1365 found = address_list_contains (al, &ip);
1366 address_list_release (al);
1371 /* The persistent connection's peer address was found among the
1372 addresses HOST resolved to; therefore, pconn.sock is in fact
1373 already talking to HOST -- no need to reconnect. */
1376 /* Finally, check whether the connection is still open. This is
1377 important because most servers implement liberal (short) timeout
1378 on persistent connections. Wget can of course always reconnect
1379 if the connection doesn't work out, but it's nicer to know in
1380 advance. This test is a logical followup of the first test, but
1381 is "expensive" and therefore placed at the end of the list.
1383 (Current implementation of test_socket_open has a nice side
1384 effect that it treats sockets with pending data as "closed".
1385 This is exactly what we want: if a broken server sends message
1386 body in response to HEAD, or if it sends more than conent-length
1387 data, we won't reuse the corrupted connection.) */
1389 if (!test_socket_open (pconn.socket))
1391 /* Oops, the socket is no longer open. Now that we know that,
1392 let's invalidate the persistent connection before returning
1394 invalidate_persistent ();
1401 /* The idea behind these two CLOSE macros is to distinguish between
1402 two cases: one when the job we've been doing is finished, and we
1403 want to close the connection and leave, and two when something is
1404 seriously wrong and we're closing the connection as part of
1407 In case of keep_alive, CLOSE_FINISH should leave the connection
1408 open, while CLOSE_INVALIDATE should still close it.
1410 Note that the semantics of the flag `keep_alive' is "this
1411 connection *will* be reused (the server has promised not to close
1412 the connection once we're done)", while the semantics of
1413 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1414 active, registered connection". */
1416 #define CLOSE_FINISH(fd) do { \
1419 if (pconn_active && (fd) == pconn.socket) \
1420 invalidate_persistent (); \
1429 #define CLOSE_INVALIDATE(fd) do { \
1430 if (pconn_active && (fd) == pconn.socket) \
1431 invalidate_persistent (); \
1439 wgint len; /* received length */
1440 wgint contlen; /* expected length */
1441 wgint restval; /* the restart value */
1442 int res; /* the result of last read */
1443 char *rderrmsg; /* error message from read error */
1444 char *newloc; /* new location (redirection) */
1445 char *remote_time; /* remote time-stamp string */
1446 char *error; /* textual HTTP error */
1447 int statcode; /* status code */
1448 char *message; /* status message */
1449 wgint rd_size; /* amount of data read from socket */
1450 double dltime; /* time it took to download the data */
1451 const char *referer; /* value of the referer header. */
1452 char *local_file; /* local file name. */
1453 bool existence_checked; /* true if we already checked for a file's
1454 existence after having begun to download
1455 (needed in gethttp for when connection is
1456 interrupted/restarted. */
1457 bool timestamp_checked; /* true if pre-download time-stamping checks
1458 * have already been performed */
1459 char *orig_file_name; /* name of file to compare for time-stamping
1460 * (might be != local_file if -K is set) */
1461 wgint orig_file_size; /* size of file to compare for time-stamping */
1462 time_t orig_file_tstamp; /* time-stamp of file to compare for
1467 free_hstat (struct http_stat *hs)
1469 xfree_null (hs->newloc);
1470 xfree_null (hs->remote_time);
1471 xfree_null (hs->error);
1472 xfree_null (hs->rderrmsg);
1473 xfree_null (hs->local_file);
1474 xfree_null (hs->orig_file_name);
1475 xfree_null (hs->message);
1477 /* Guard against being called twice. */
1479 hs->remote_time = NULL;
1483 #define BEGINS_WITH(line, string_constant) \
1484 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1485 && (c_isspace (line[sizeof (string_constant) - 1]) \
1486 || !line[sizeof (string_constant) - 1]))
1489 #define SET_USER_AGENT(req) do { \
1490 if (!opt.useragent) \
1491 request_set_header (req, "User-Agent", \
1492 aprintf ("Wget/%s (VMS %s %s)", \
1493 version_string, vms_arch(), vms_vers()), \
1495 else if (*opt.useragent) \
1496 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1498 #else /* def __VMS */
1499 #define SET_USER_AGENT(req) do { \
1500 if (!opt.useragent) \
1501 request_set_header (req, "User-Agent", \
1502 aprintf ("Wget/%s (%s)", \
1503 version_string, OS_TYPE), \
1505 else if (*opt.useragent) \
1506 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1508 #endif /* def __VMS [else] */
1510 /* The flags that allow clobbering the file (opening with "wb").
1511 Defined here to avoid repetition later. #### This will require
1513 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1514 || opt.dirstruct || opt.output_document)
1516 /* Retrieve a document through HTTP protocol. It recognizes status
1517 code, and correctly handles redirections. It closes the network
1518 socket. If it receives an error from the functions below it, it
1519 will print it if there is enough information to do so (almost
1520 always), returning the error to the caller (i.e. http_loop).
1522 Various HTTP parameters are stored to hs.
1524 If PROXY is non-NULL, the connection will be made to the proxy
1525 server, and u->url will be requested. */
1527 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1530 struct request *req;
1533 char *user, *passwd;
1537 wgint contlen, contrange;
1544 /* Set to 1 when the authorization has already been sent and should
1545 not be tried again. */
1546 bool auth_finished = false;
1548 /* Set to 1 when just globally-set Basic authorization has been sent;
1549 * should prevent further Basic negotiations, but not other
1551 bool basic_auth_finished = false;
1553 /* Whether NTLM authentication is used for this request. */
1554 bool ntlm_seen = false;
1556 /* Whether our connection to the remote host is through SSL. */
1557 bool using_ssl = false;
1559 /* Whether a HEAD request will be issued (as opposed to GET or
1561 bool head_only = !!(*dt & HEAD_ONLY);
1564 struct response *resp;
1568 /* Whether this connection will be kept alive after the HTTP request
1572 /* Is the server using the chunked transfer encoding? */
1573 bool chunked_transfer_encoding = false;
1575 /* Whether keep-alive should be inhibited.
1577 RFC 2068 requests that 1.0 clients not send keep-alive requests
1578 to proxies. This is because many 1.0 proxies do not interpret
1579 the Connection header and transfer it to the remote server,
1580 causing it to not close the connection and leave both the proxy
1581 and the client hanging. */
1582 bool inhibit_keep_alive =
1583 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1585 /* Headers sent when using POST. */
1586 wgint post_data_size = 0;
1588 bool host_lookup_failed = false;
1591 if (u->scheme == SCHEME_HTTPS)
1593 /* Initialize the SSL context. After this has once been done,
1594 it becomes a no-op. */
1597 scheme_disable (SCHEME_HTTPS);
1598 logprintf (LOG_NOTQUIET,
1599 _("Disabling SSL due to encountered errors.\n"));
1600 return SSLINITFAILED;
1603 #endif /* HAVE_SSL */
1605 /* Initialize certain elements of struct http_stat. */
1609 hs->rderrmsg = NULL;
1611 hs->remote_time = NULL;
1617 /* Prepare the request to send. */
1619 req = request_new ();
1622 const char *meth = "GET";
1625 else if (opt.post_file_name || opt.post_data)
1627 /* Use the full path, i.e. one that includes the leading slash and
1628 the query string. E.g. if u->path is "foo/bar" and u->query is
1629 "param=value", full_path will be "/foo/bar?param=value". */
1632 /* When using SSL over proxy, CONNECT establishes a direct
1633 connection to the HTTPS server. Therefore use the same
1634 argument as when talking to the server directly. */
1635 && u->scheme != SCHEME_HTTPS
1638 meth_arg = xstrdup (u->url);
1640 meth_arg = url_full_path (u);
1641 request_set_method (req, meth, meth_arg);
1644 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1645 if (*dt & SEND_NOCACHE)
1646 request_set_header (req, "Pragma", "no-cache", rel_none);
1648 request_set_header (req, "Range",
1649 aprintf ("bytes=%s-",
1650 number_to_static_string (hs->restval)),
1652 SET_USER_AGENT (req);
1653 request_set_header (req, "Accept", "*/*", rel_none);
1655 /* Find the username and password for authentication. */
1658 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1659 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1660 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1662 /* We only do "site-wide" authentication with "global" user/password
1663 * values unless --auth-no-challange has been requested; URL user/password
1664 * info overrides. */
1665 if (user && passwd && (!u->user || opt.auth_without_challenge))
1667 /* If this is a host for which we've already received a Basic
1668 * challenge, we'll go ahead and send Basic authentication creds. */
1669 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1672 /* Generate the Host header, HOST:PORT. Take into account that:
1674 - Broken server-side software often doesn't recognize the PORT
1675 argument, so we must generate "Host: www.server.com" instead of
1676 "Host: www.server.com:80" (and likewise for https port).
1678 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1679 becomes ambiguous and needs to be rewritten as "Host:
1680 [3ffe:8100:200:2::2]:1234". */
1682 /* Formats arranged for hfmt[add_port][add_squares]. */
1683 static const char *hfmt[][2] = {
1684 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1686 int add_port = u->port != scheme_default_port (u->scheme);
1687 int add_squares = strchr (u->host, ':') != NULL;
1688 request_set_header (req, "Host",
1689 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1693 if (!inhibit_keep_alive)
1694 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1697 request_set_header (req, "Cookie",
1698 cookie_header (wget_cookie_jar,
1699 u->host, u->port, u->path,
1701 u->scheme == SCHEME_HTTPS
1708 if (opt.post_data || opt.post_file_name)
1710 request_set_header (req, "Content-Type",
1711 "application/x-www-form-urlencoded", rel_none);
1713 post_data_size = strlen (opt.post_data);
1716 post_data_size = file_size (opt.post_file_name);
1717 if (post_data_size == -1)
1719 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1720 quote (opt.post_file_name), strerror (errno));
1724 request_set_header (req, "Content-Length",
1725 xstrdup (number_to_static_string (post_data_size)),
1729 /* Add the user headers. */
1730 if (opt.user_headers)
1733 for (i = 0; opt.user_headers[i]; i++)
1734 request_set_user_header (req, opt.user_headers[i]);
1738 /* We need to come back here when the initial attempt to retrieve
1739 without authorization header fails. (Expected to happen at least
1740 for the Digest authorization scheme.) */
1745 char *proxy_user, *proxy_passwd;
1746 /* For normal username and password, URL components override
1747 command-line/wgetrc parameters. With proxy
1748 authentication, it's the reverse, because proxy URLs are
1749 normally the "permanent" ones, so command-line args
1750 should take precedence. */
1751 if (opt.proxy_user && opt.proxy_passwd)
1753 proxy_user = opt.proxy_user;
1754 proxy_passwd = opt.proxy_passwd;
1758 proxy_user = proxy->user;
1759 proxy_passwd = proxy->passwd;
1761 /* #### This does not appear right. Can't the proxy request,
1762 say, `Digest' authentication? */
1763 if (proxy_user && proxy_passwd)
1764 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1766 /* If we're using a proxy, we will be connecting to the proxy
1770 /* Proxy authorization over SSL is handled below. */
1772 if (u->scheme != SCHEME_HTTPS)
1774 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1779 /* Establish the connection. */
1781 if (inhibit_keep_alive)
1785 /* Look for a persistent connection to target host, unless a
1786 proxy is used. The exception is when SSL is in use, in which
1787 case the proxy is nothing but a passthrough to the target
1788 host, registered as a connection to the latter. */
1789 struct url *relevant = conn;
1791 if (u->scheme == SCHEME_HTTPS)
1795 if (persistent_available_p (relevant->host, relevant->port,
1797 relevant->scheme == SCHEME_HTTPS,
1801 &host_lookup_failed))
1803 sock = pconn.socket;
1804 using_ssl = pconn.ssl;
1805 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1806 quotearg_style (escape_quoting_style, pconn.host),
1808 DEBUGP (("Reusing fd %d.\n", sock));
1809 if (pconn.authorized)
1810 /* If the connection is already authorized, the "Basic"
1811 authorization added by code above is unnecessary and
1813 request_remove_header (req, "Authorization");
1815 else if (host_lookup_failed)
1818 logprintf(LOG_NOTQUIET,
1819 _("%s: unable to resolve host address %s\n"),
1820 exec_name, quote (relevant->host));
1827 sock = connect_to_host (conn->host, conn->port);
1836 return (retryable_socket_connect_error (errno)
1837 ? CONERROR : CONIMPOSSIBLE);
1841 if (proxy && u->scheme == SCHEME_HTTPS)
1843 /* When requesting SSL URLs through proxies, use the
1844 CONNECT method to request passthrough. */
1845 struct request *connreq = request_new ();
1846 request_set_method (connreq, "CONNECT",
1847 aprintf ("%s:%d", u->host, u->port));
1848 SET_USER_AGENT (connreq);
1851 request_set_header (connreq, "Proxy-Authorization",
1852 proxyauth, rel_value);
1853 /* Now that PROXYAUTH is part of the CONNECT request,
1854 zero it out so we don't send proxy authorization with
1855 the regular request below. */
1858 /* Examples in rfc2817 use the Host header in CONNECT
1859 requests. I don't see how that gains anything, given
1860 that the contents of Host would be exactly the same as
1861 the contents of CONNECT. */
1863 write_error = request_send (connreq, sock);
1864 request_free (connreq);
1865 if (write_error < 0)
1867 CLOSE_INVALIDATE (sock);
1871 head = read_http_response_head (sock);
1874 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1876 CLOSE_INVALIDATE (sock);
1885 DEBUGP (("proxy responded with: [%s]\n", head));
1887 resp = resp_new (head);
1888 statcode = resp_status (resp, &message);
1891 char *tms = datetime_str (time (NULL));
1892 logprintf (LOG_VERBOSE, "%d\n", statcode);
1893 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1894 quotearg_style (escape_quoting_style,
1895 _("Malformed status line")));
1899 hs->message = xstrdup (message);
1902 if (statcode != 200)
1905 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1906 message ? quotearg_style (escape_quoting_style, message) : "?");
1907 xfree_null (message);
1910 xfree_null (message);
1912 /* SOCK is now *really* connected to u->host, so update CONN
1913 to reflect this. That way register_persistent will
1914 register SOCK as being connected to u->host:u->port. */
1918 if (conn->scheme == SCHEME_HTTPS)
1920 if (!ssl_connect_wget (sock))
1925 else if (!ssl_check_certificate (sock, u->host))
1928 return VERIFCERTERR;
1932 #endif /* HAVE_SSL */
1935 /* Send the request to server. */
1936 write_error = request_send (req, sock);
1938 if (write_error >= 0)
1942 DEBUGP (("[POST data: %s]\n", opt.post_data));
1943 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1945 else if (opt.post_file_name && post_data_size != 0)
1946 write_error = post_file (sock, opt.post_file_name, post_data_size);
1949 if (write_error < 0)
1951 CLOSE_INVALIDATE (sock);
1955 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1956 proxy ? "Proxy" : "HTTP");
1961 head = read_http_response_head (sock);
1966 logputs (LOG_NOTQUIET, _("No data received.\n"));
1967 CLOSE_INVALIDATE (sock);
1973 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1975 CLOSE_INVALIDATE (sock);
1980 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1982 resp = resp_new (head);
1984 /* Check for status line. */
1986 statcode = resp_status (resp, &message);
1989 char *tms = datetime_str (time (NULL));
1990 logprintf (LOG_VERBOSE, "%d\n", statcode);
1991 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1992 quotearg_style (escape_quoting_style,
1993 _("Malformed status line")));
1994 CLOSE_INVALIDATE (sock);
1998 hs->message = xstrdup (message);
1999 if (!opt.server_response)
2000 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2001 message ? quotearg_style (escape_quoting_style, message) : "");
2004 logprintf (LOG_VERBOSE, "\n");
2005 print_server_response (resp, " ");
2008 if (!opt.ignore_length
2009 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2013 parsed = str_to_wgint (hdrval, NULL, 10);
2014 if (parsed == WGINT_MAX && errno == ERANGE)
2017 #### If Content-Length is out of range, it most likely
2018 means that the file is larger than 2G and that we're
2019 compiled without LFS. In that case we should probably
2020 refuse to even attempt to download the file. */
2023 else if (parsed < 0)
2025 /* Negative Content-Length; nonsensical, so we can't
2026 assume any information about the content to receive. */
2033 /* Check for keep-alive related responses. */
2034 if (!inhibit_keep_alive && contlen != -1)
2036 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2038 if (0 == strcasecmp (hdrval, "Close"))
2043 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2044 if (0 == strcasecmp (hdrval, "chunked"))
2045 chunked_transfer_encoding = true;
2047 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2051 const char *scbeg, *scend;
2052 /* The jar should have been created by now. */
2053 assert (wget_cookie_jar != NULL);
2055 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2056 &scbeg, &scend)) != -1;
2059 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2060 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2061 u->path, set_cookie);
2066 /* The server has promised that it will not close the connection
2067 when we're done. This means that we can register it. */
2068 register_persistent (conn->host, conn->port, sock, using_ssl);
2070 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2072 /* Authorization is required. */
2073 if (keep_alive && !head_only
2074 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2075 CLOSE_FINISH (sock);
2077 CLOSE_INVALIDATE (sock);
2078 pconn.authorized = false;
2079 if (!auth_finished && (user && passwd))
2081 /* IIS sends multiple copies of WWW-Authenticate, one with
2082 the value "negotiate", and other(s) with data. Loop over
2083 all the occurrences and pick the one we recognize. */
2085 const char *wabeg, *waend;
2086 char *www_authenticate = NULL;
2088 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2089 &wabeg, &waend)) != -1;
2091 if (known_authentication_scheme_p (wabeg, waend))
2093 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2097 if (!www_authenticate)
2099 /* If the authentication header is missing or
2100 unrecognized, there's no sense in retrying. */
2101 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2103 else if (!basic_auth_finished
2104 || !BEGINS_WITH (www_authenticate, "Basic"))
2107 pth = url_full_path (u);
2108 request_set_header (req, "Authorization",
2109 create_authorization_line (www_authenticate,
2111 request_method (req),
2115 if (BEGINS_WITH (www_authenticate, "NTLM"))
2117 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2119 /* Need to register this host as using basic auth,
2120 * so we automatically send creds next time. */
2121 register_basic_auth_host (u->host);
2124 xfree_null (message);
2127 goto retry_with_auth;
2131 /* We already did Basic auth, and it failed. Gotta
2135 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2137 xfree_null (message);
2142 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2144 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2146 pconn.authorized = true;
2149 /* Determine the local filename if needed. Notice that if -O is used
2150 * hstat.local_file is set by http_loop to the argument of -O. */
2151 if (!hs->local_file)
2153 /* Honor Content-Disposition whether possible. */
2154 if (!opt.content_disposition
2155 || !resp_header_copy (resp, "Content-Disposition",
2156 hdrval, sizeof (hdrval))
2157 || !parse_content_disposition (hdrval, &hs->local_file))
2159 /* The Content-Disposition header is missing or broken.
2160 * Choose unique file name according to given URL. */
2161 hs->local_file = url_file_name (u);
2165 /* TODO: perform this check only once. */
2166 if (!hs->existence_checked && file_exists_p (hs->local_file))
2168 if (opt.noclobber && !opt.output_document)
2170 /* If opt.noclobber is turned on and file already exists, do not
2171 retrieve the file. But if the output_document was given, then this
2172 test was already done and the file didn't exist. Hence the !opt.output_document */
2173 logprintf (LOG_VERBOSE, _("\
2174 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2175 /* If the file is there, we suppose it's retrieved OK. */
2178 /* #### Bogusness alert. */
2179 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2180 if (has_html_suffix_p (hs->local_file))
2184 xfree_null (message);
2185 return RETRUNNEEDED;
2187 else if (!ALLOW_CLOBBER)
2189 char *unique = unique_name (hs->local_file, true);
2190 if (unique != hs->local_file)
2191 xfree (hs->local_file);
2192 hs->local_file = unique;
2195 hs->existence_checked = true;
2197 /* Support timestamping */
2198 /* TODO: move this code out of gethttp. */
2199 if (opt.timestamping && !hs->timestamp_checked)
2201 size_t filename_len = strlen (hs->local_file);
2202 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2203 bool local_dot_orig_file_exists = false;
2204 char *local_filename = NULL;
2207 if (opt.backup_converted)
2208 /* If -K is specified, we'll act on the assumption that it was specified
2209 last time these files were downloaded as well, and instead of just
2210 comparing local file X against server file X, we'll compare local
2211 file X.orig (if extant, else X) against server file X. If -K
2212 _wasn't_ specified last time, or the server contains files called
2213 *.orig, -N will be back to not operating correctly with -k. */
2215 /* Would a single s[n]printf() call be faster? --dan
2217 Definitely not. sprintf() is horribly slow. It's a
2218 different question whether the difference between the two
2219 affects a program. Usually I'd say "no", but at one
2220 point I profiled Wget, and found that a measurable and
2221 non-negligible amount of time was lost calling sprintf()
2222 in url.c. Replacing sprintf with inline calls to
2223 strcpy() and number_to_string() made a difference.
2225 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2226 memcpy (filename_plus_orig_suffix + filename_len,
2227 ORIG_SFX, sizeof (ORIG_SFX));
2229 /* Try to stat() the .orig file. */
2230 if (stat (filename_plus_orig_suffix, &st) == 0)
2232 local_dot_orig_file_exists = true;
2233 local_filename = filename_plus_orig_suffix;
2237 if (!local_dot_orig_file_exists)
2238 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2239 if (stat (hs->local_file, &st) == 0)
2240 local_filename = hs->local_file;
2242 if (local_filename != NULL)
2243 /* There was a local file, so we'll check later to see if the version
2244 the server has is the same version we already have, allowing us to
2247 hs->orig_file_name = xstrdup (local_filename);
2248 hs->orig_file_size = st.st_size;
2249 hs->orig_file_tstamp = st.st_mtime;
2251 /* Modification time granularity is 2 seconds for Windows, so
2252 increase local time by 1 second for later comparison. */
2253 ++hs->orig_file_tstamp;
2260 hs->statcode = statcode;
2262 hs->error = xstrdup (_("Malformed status line"));
2264 hs->error = xstrdup (_("(no description)"));
2266 hs->error = xstrdup (message);
2267 xfree_null (message);
2269 type = resp_header_strdup (resp, "Content-Type");
2272 char *tmp = strchr (type, ';');
2275 /* sXXXav: only needed if IRI support is enabled */
2276 char *tmp2 = tmp + 1;
2278 while (tmp > type && c_isspace (tmp[-1]))
2282 /* Try to get remote encoding if needed */
2283 if (opt.enable_iri && !opt.encoding_remote)
2285 tmp = parse_charset (tmp2);
2287 set_content_encoding (iri, tmp);
2291 hs->newloc = resp_header_strdup (resp, "Location");
2292 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2294 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2296 wgint first_byte_pos, last_byte_pos, entity_length;
2297 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2300 contrange = first_byte_pos;
2301 contlen = last_byte_pos - first_byte_pos + 1;
2306 /* 20x responses are counted among successful by default. */
2307 if (H_20X (statcode))
2310 /* Return if redirected. */
2311 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2313 /* RFC2068 says that in case of the 300 (multiple choices)
2314 response, the server can output a preferred URL through
2315 `Location' header; otherwise, the request should be treated
2316 like GET. So, if the location is set, it will be a
2317 redirection; otherwise, just proceed normally. */
2318 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2322 logprintf (LOG_VERBOSE,
2323 _("Location: %s%s\n"),
2324 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2325 hs->newloc ? _(" [following]") : "");
2326 if (keep_alive && !head_only
2327 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2328 CLOSE_FINISH (sock);
2330 CLOSE_INVALIDATE (sock);
2337 /* If content-type is not given, assume text/html. This is because
2338 of the multitude of broken CGI's that "forget" to generate the
2341 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2342 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2348 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2353 if (opt.adjust_extension)
2356 /* -E / --adjust-extension / adjust_extension = on was specified,
2357 and this is a text/html file. If some case-insensitive
2358 variation on ".htm[l]" isn't already the file's suffix,
2361 ensure_extension (hs, ".html", dt);
2363 else if (*dt & TEXTCSS)
2365 ensure_extension (hs, ".css", dt);
2369 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2370 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2371 && contrange == 0 && hs->restval >= contlen)
2374 /* If `-c' is in use and the file has been fully downloaded (or
2375 the remote file has shrunk), Wget effectively requests bytes
2376 after the end of file and the server response with 416
2377 (or 200 with a <= Content-Length. */
2378 logputs (LOG_VERBOSE, _("\
2379 \n The file is already fully retrieved; nothing to do.\n\n"));
2380 /* In case the caller inspects. */
2383 /* Mark as successfully retrieved. */
2386 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2387 might be more bytes in the body. */
2389 return RETRUNNEEDED;
2391 if ((contrange != 0 && contrange != hs->restval)
2392 || (H_PARTIAL (statcode) && !contrange))
2394 /* The Range request was somehow misunderstood by the server.
2397 CLOSE_INVALIDATE (sock);
2404 hs->contlen = contlen + contrange;
2410 /* No need to print this output if the body won't be
2411 downloaded at all, or if the original server response is
2413 logputs (LOG_VERBOSE, _("Length: "));
2416 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2417 if (contlen + contrange >= 1024)
2418 logprintf (LOG_VERBOSE, " (%s)",
2419 human_readable (contlen + contrange));
2422 if (contlen >= 1024)
2423 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2424 number_to_static_string (contlen),
2425 human_readable (contlen));
2427 logprintf (LOG_VERBOSE, _(", %s remaining"),
2428 number_to_static_string (contlen));
2432 logputs (LOG_VERBOSE,
2433 opt.ignore_length ? _("ignored") : _("unspecified"));
2435 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2437 logputs (LOG_VERBOSE, "\n");
2441 type = NULL; /* We don't need it any more. */
2443 /* Return if we have no intention of further downloading. */
2444 if (!(*dt & RETROKF) || head_only)
2446 /* In case the caller cares to look... */
2451 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2452 servers not to send body in response to a HEAD request, and
2453 those that do will likely be caught by test_socket_open.
2454 If not, they can be worked around using
2455 `--no-http-keep-alive'. */
2456 CLOSE_FINISH (sock);
2458 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2459 /* Successfully skipped the body; also keep using the socket. */
2460 CLOSE_FINISH (sock);
2462 CLOSE_INVALIDATE (sock);
2464 return RETRFINISHED;
2468 For VMS, define common fopen() optional arguments.
2471 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2472 # define FOPEN_BIN_FLAG 3
2473 #else /* def __VMS */
2474 # define FOPEN_BIN_FLAG true
2475 #endif /* def __VMS [else] */
2477 /* Open the local file. */
2480 mkalldirs (hs->local_file);
2482 rotate_backups (hs->local_file);
2489 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2490 #else /* def __VMS */
2491 fp = fopen (hs->local_file, "ab");
2492 #endif /* def __VMS [else] */
2494 else if (ALLOW_CLOBBER)
2500 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2501 #else /* def __VMS */
2502 fp = fopen (hs->local_file, "wb");
2503 #endif /* def __VMS [else] */
2507 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2508 if (!fp && errno == EEXIST)
2510 /* We cannot just invent a new name and use it (which is
2511 what functions like unique_create typically do)
2512 because we told the user we'd use this name.
2513 Instead, return and retry the download. */
2514 logprintf (LOG_NOTQUIET,
2515 _("%s has sprung into existence.\n"),
2517 CLOSE_INVALIDATE (sock);
2519 return FOPEN_EXCL_ERR;
2524 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2525 CLOSE_INVALIDATE (sock);
2533 /* Print fetch message, if opt.verbose. */
2536 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2537 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2540 /* This confuses the timestamping code that checks for file size.
2541 #### The timestamping code should be smarter about file size. */
2542 if (opt.save_headers && hs->restval == 0)
2543 fwrite (head, 1, strlen (head), fp);
2545 /* Now we no longer need to store the response header. */
2548 /* Download the request body. */
2551 /* If content-length is present, read that much; otherwise, read
2552 until EOF. The HTTP spec doesn't require the server to
2553 actually close the connection when it's done sending data. */
2554 flags |= rb_read_exactly;
2555 if (hs->restval > 0 && contrange == 0)
2556 /* If the server ignored our range request, instruct fd_read_body
2557 to skip the first RESTVAL bytes of body. */
2558 flags |= rb_skip_startpos;
2560 if (chunked_transfer_encoding)
2561 flags |= rb_chunked_transfer_encoding;
2563 hs->len = hs->restval;
2565 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2566 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2570 CLOSE_FINISH (sock);
2574 hs->rderrmsg = xstrdup (fd_errstr (sock));
2575 CLOSE_INVALIDATE (sock);
2582 return RETRFINISHED;
2585 /* The genuine HTTP loop! This is the part where the retrieval is
2586 retried, and retried, and retried, and... */
2588 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2589 int *dt, struct url *proxy, struct iri *iri)
2592 bool got_head = false; /* used for time-stamping and filename detection */
2593 bool time_came_from_head = false;
2594 bool got_name = false;
2597 uerr_t err, ret = TRYLIMEXC;
2598 time_t tmr = -1; /* remote time-stamp */
2599 struct http_stat hstat; /* HTTP status */
2601 bool send_head_first = true;
2604 /* Assert that no value for *LOCAL_FILE was passed. */
2605 assert (local_file == NULL || *local_file == NULL);
2607 /* Set LOCAL_FILE parameter. */
2608 if (local_file && opt.output_document)
2609 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2611 /* Reset NEWLOC parameter. */
2614 /* This used to be done in main(), but it's a better idea to do it
2615 here so that we don't go through the hoops if we're just using
2620 /* Warn on (likely bogus) wildcard usage in HTTP. */
2621 if (opt.ftp_glob && has_wildcards_p (u->path))
2622 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2624 /* Setup hstat struct. */
2626 hstat.referer = referer;
2628 if (opt.output_document)
2630 hstat.local_file = xstrdup (opt.output_document);
2633 else if (!opt.content_disposition)
2635 hstat.local_file = url_file_name (u);
2639 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2640 * screaming for some refactoring. */
2641 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2643 /* If opt.noclobber is turned on and file already exists, do not
2644 retrieve the file. But if the output_document was given, then this
2645 test was already done and the file didn't exist. Hence the !opt.output_document */
2646 logprintf (LOG_VERBOSE, _("\
2647 File %s already there; not retrieving.\n\n"),
2648 quote (hstat.local_file));
2649 /* If the file is there, we suppose it's retrieved OK. */
2652 /* #### Bogusness alert. */
2653 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2654 if (has_html_suffix_p (hstat.local_file))
2661 /* Reset the counter. */
2664 /* Reset the document type. */
2667 /* Skip preliminary HEAD request if we're not in spider mode AND
2668 * if -O was given or HTTP Content-Disposition support is disabled. */
2670 && (got_name || !opt.content_disposition))
2671 send_head_first = false;
2673 /* Send preliminary HEAD request if -N is given and we have an existing
2674 * destination file. */
2675 file_name = url_file_name (u);
2676 if (opt.timestamping
2677 && !opt.content_disposition
2678 && file_exists_p (file_name))
2679 send_head_first = true;
2685 /* Increment the pass counter. */
2687 sleep_between_retrievals (count);
2689 /* Get the current time string. */
2690 tms = datetime_str (time (NULL));
2692 if (opt.spider && !got_head)
2693 logprintf (LOG_VERBOSE, _("\
2694 Spider mode enabled. Check if remote file exists.\n"));
2696 /* Print fetch message, if opt.verbose. */
2699 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2704 sprintf (tmp, _("(try:%2d)"), count);
2705 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2710 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2715 ws_changetitle (hurl);
2720 /* Default document type is empty. However, if spider mode is
2721 on or time-stamping is employed, HEAD_ONLY commands is
2722 encoded within *dt. */
2723 if (send_head_first && !got_head)
2728 /* Decide whether or not to restart. */
2731 && stat (hstat.local_file, &st) == 0
2732 && S_ISREG (st.st_mode))
2733 /* When -c is used, continue from on-disk size. (Can't use
2734 hstat.len even if count>1 because we don't want a failed
2735 first attempt to clobber existing data.) */
2736 hstat.restval = st.st_size;
2738 /* otherwise, continue where the previous try left off */
2739 hstat.restval = hstat.len;
2743 /* Decide whether to send the no-cache directive. We send it in
2745 a) we're using a proxy, and we're past our first retrieval.
2746 Some proxies are notorious for caching incomplete data, so
2747 we require a fresh get.
2748 b) caching is explicitly inhibited. */
2749 if ((proxy && count > 1) /* a */
2750 || !opt.allow_cache) /* b */
2751 *dt |= SEND_NOCACHE;
2753 *dt &= ~SEND_NOCACHE;
2755 /* Try fetching the document, or at least its head. */
2756 err = gethttp (u, &hstat, dt, proxy, iri);
2759 tms = datetime_str (time (NULL));
2761 /* Get the new location (with or without the redirection). */
2763 *newloc = xstrdup (hstat.newloc);
2767 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2768 case CONERROR: case READERR: case WRITEFAILED:
2769 case RANGEERR: case FOPEN_EXCL_ERR:
2770 /* Non-fatal errors continue executing the loop, which will
2771 bring them to "while" statement at the end, to judge
2772 whether the number of tries was exceeded. */
2773 printwhat (count, opt.ntry);
2775 case FWRITEERR: case FOPENERR:
2776 /* Another fatal error. */
2777 logputs (LOG_VERBOSE, "\n");
2778 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2779 quote (hstat.local_file), strerror (errno));
2780 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2781 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2782 /* Fatal errors just return from the function. */
2786 /* Another fatal error. */
2787 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2791 /* Return the new location to the caller. */
2794 logprintf (LOG_NOTQUIET,
2795 _("ERROR: Redirection (%d) without location.\n"),
2805 /* The file was already fully retrieved. */
2809 /* Deal with you later. */
2812 /* All possibilities should have been exhausted. */
2816 if (!(*dt & RETROKF))
2821 /* #### Ugly ugly ugly! */
2822 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2823 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2826 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2828 && (hstat.statcode == 500 || hstat.statcode == 501))
2833 /* Maybe we should always keep track of broken links, not just in
2835 * Don't log error if it was UTF-8 encoded because we will try
2836 * once unencoded. */
2837 else if (opt.spider && !iri->utf8_encode)
2839 /* #### Again: ugly ugly ugly! */
2841 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2842 nonexisting_url (hurl);
2843 logprintf (LOG_NOTQUIET, _("\
2844 Remote file does not exist -- broken link!!!\n"));
2848 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2849 tms, hstat.statcode,
2850 quotearg_style (escape_quoting_style, hstat.error));
2852 logputs (LOG_VERBOSE, "\n");
2858 /* Did we get the time-stamp? */
2861 got_head = true; /* no more time-stamping */
2863 if (opt.timestamping && !hstat.remote_time)
2865 logputs (LOG_NOTQUIET, _("\
2866 Last-modified header missing -- time-stamps turned off.\n"));
2868 else if (hstat.remote_time)
2870 /* Convert the date-string into struct tm. */
2871 tmr = http_atotm (hstat.remote_time);
2872 if (tmr == (time_t) (-1))
2873 logputs (LOG_VERBOSE, _("\
2874 Last-modified header invalid -- time-stamp ignored.\n"));
2875 if (*dt & HEAD_ONLY)
2876 time_came_from_head = true;
2879 if (send_head_first)
2881 /* The time-stamping section. */
2882 if (opt.timestamping)
2884 if (hstat.orig_file_name) /* Perform the following
2885 checks only if the file
2887 download already exists. */
2889 if (hstat.remote_time &&
2890 tmr != (time_t) (-1))
2892 /* Now time-stamping can be used validly.
2893 Time-stamping means that if the sizes of
2894 the local and remote file match, and local
2895 file is newer than the remote file, it will
2896 not be retrieved. Otherwise, the normal
2897 download procedure is resumed. */
2898 if (hstat.orig_file_tstamp >= tmr)
2900 if (hstat.contlen == -1
2901 || hstat.orig_file_size == hstat.contlen)
2903 logprintf (LOG_VERBOSE, _("\
2904 Server file no newer than local file %s -- not retrieving.\n\n"),
2905 quote (hstat.orig_file_name));
2911 logprintf (LOG_VERBOSE, _("\
2912 The sizes do not match (local %s) -- retrieving.\n"),
2913 number_to_static_string (hstat.orig_file_size));
2917 logputs (LOG_VERBOSE,
2918 _("Remote file is newer, retrieving.\n"));
2920 logputs (LOG_VERBOSE, "\n");
2924 /* free_hstat (&hstat); */
2925 hstat.timestamp_checked = true;
2930 bool finished = true;
2935 logputs (LOG_VERBOSE, _("\
2936 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2941 logprintf (LOG_VERBOSE, _("\
2942 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2943 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2950 logprintf (LOG_VERBOSE, _("\
2951 Remote file exists and could contain further links,\n\
2952 but recursion is disabled -- not retrieving.\n\n"));
2956 logprintf (LOG_VERBOSE, _("\
2957 Remote file exists.\n\n"));
2959 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2964 logprintf (LOG_NONVERBOSE,
2965 _("%s URL: %s %2d %s\n"),
2966 tms, u->url, hstat.statcode,
2967 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2974 count = 0; /* the retrieve count for HEAD is reset */
2976 } /* send_head_first */
2979 if (opt.useservertimestamps
2980 && (tmr != (time_t) (-1))
2981 && ((hstat.len == hstat.contlen) ||
2982 ((hstat.res == 0) && (hstat.contlen == -1))))
2984 const char *fl = NULL;
2985 set_local_file (&fl, hstat.local_file);
2989 /* Reparse time header, in case it's changed. */
2990 if (time_came_from_head
2991 && hstat.remote_time && hstat.remote_time[0])
2993 newtmr = http_atotm (hstat.remote_time);
2994 if (newtmr != (time_t)-1)
3000 /* End of time-stamping section. */
3002 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3003 total_download_time += hstat.dltime;
3005 if (hstat.len == hstat.contlen)
3009 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3011 logprintf (LOG_VERBOSE,
3013 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3014 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3016 write_to_stdout ? "" : quote (hstat.local_file),
3017 number_to_static_string (hstat.len),
3018 number_to_static_string (hstat.contlen));
3019 logprintf (LOG_NONVERBOSE,
3020 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3022 number_to_static_string (hstat.len),
3023 number_to_static_string (hstat.contlen),
3024 hstat.local_file, count);
3027 total_downloaded_bytes += hstat.len;
3029 /* Remember that we downloaded the file for later ".orig" code. */
3030 if (*dt & ADDED_HTML_EXTENSION)
3031 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3033 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3038 else if (hstat.res == 0) /* No read error */
3040 if (hstat.contlen == -1) /* We don't know how much we were supposed
3041 to get, so assume we succeeded. */
3045 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3047 logprintf (LOG_VERBOSE,
3049 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3050 : _("%s (%s) - %s saved [%s]\n\n"),
3052 write_to_stdout ? "" : quote (hstat.local_file),
3053 number_to_static_string (hstat.len));
3054 logprintf (LOG_NONVERBOSE,
3055 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3056 tms, u->url, number_to_static_string (hstat.len),
3057 hstat.local_file, count);
3060 total_downloaded_bytes += hstat.len;
3062 /* Remember that we downloaded the file for later ".orig" code. */
3063 if (*dt & ADDED_HTML_EXTENSION)
3064 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3066 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3071 else if (hstat.len < hstat.contlen) /* meaning we lost the
3072 connection too soon */
3074 logprintf (LOG_VERBOSE,
3075 _("%s (%s) - Connection closed at byte %s. "),
3076 tms, tmrate, number_to_static_string (hstat.len));
3077 printwhat (count, opt.ntry);
3080 else if (hstat.len != hstat.restval)
3081 /* Getting here would mean reading more data than
3082 requested with content-length, which we never do. */
3086 /* Getting here probably means that the content-length was
3087 * _less_ than the original, local size. We should probably
3088 * truncate or re-read, or something. FIXME */
3093 else /* from now on hstat.res can only be -1 */
3095 if (hstat.contlen == -1)
3097 logprintf (LOG_VERBOSE,
3098 _("%s (%s) - Read error at byte %s (%s)."),
3099 tms, tmrate, number_to_static_string (hstat.len),
3101 printwhat (count, opt.ntry);
3104 else /* hstat.res == -1 and contlen is given */
3106 logprintf (LOG_VERBOSE,
3107 _("%s (%s) - Read error at byte %s/%s (%s). "),
3109 number_to_static_string (hstat.len),
3110 number_to_static_string (hstat.contlen),
3112 printwhat (count, opt.ntry);
3118 while (!opt.ntry || (count < opt.ntry));
3122 *local_file = xstrdup (hstat.local_file);
3123 free_hstat (&hstat);
3128 /* Check whether the result of strptime() indicates success.
3129 strptime() returns the pointer to how far it got to in the string.
3130 The processing has been successful if the string is at `GMT' or
3131 `+X', or at the end of the string.
3133 In extended regexp parlance, the function returns 1 if P matches
3134 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3135 can return) is considered a failure and 0 is returned. */
3137 check_end (const char *p)
3141 while (c_isspace (*p))
3144 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3145 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3151 /* Convert the textual specification of time in TIME_STRING to the
3152 number of seconds since the Epoch.
3154 TIME_STRING can be in any of the three formats RFC2616 allows the
3155 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3156 as well as the time format used in the Set-Cookie header.
3157 Timezones are ignored, and should be GMT.
3159 Return the computed time_t representation, or -1 if the conversion
3162 This function uses strptime with various string formats for parsing
3163 TIME_STRING. This results in a parser that is not as lenient in
3164 interpreting TIME_STRING as I would like it to be. Being based on
3165 strptime, it always allows shortened months, one-digit days, etc.,
3166 but due to the multitude of formats in which time can be
3167 represented, an ideal HTTP time parser would be even more
3168 forgiving. It should completely ignore things like week days and
3169 concentrate only on the various forms of representing years,
3170 months, days, hours, minutes, and seconds. For example, it would
3171 be nice if it accepted ISO 8601 out of the box.
3173 I've investigated free and PD code for this purpose, but none was
3174 usable. getdate was big and unwieldy, and had potential copyright
3175 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3176 distributed with phttpd, is excellent, but we cannot use it because
3177 it is not assigned to the FSF. So I stuck it with strptime. */
3180 http_atotm (const char *time_string)
3182 /* NOTE: Solaris strptime man page claims that %n and %t match white
3183 space, but that's not universally available. Instead, we simply
3184 use ` ' to mean "skip all WS", which works under all strptime
3185 implementations I've tested. */
3187 static const char *time_formats[] = {
3188 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3189 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3190 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3191 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3192 (used in Set-Cookie, defined in the
3193 Netscape cookie specification.) */
3195 const char *oldlocale;
3196 char savedlocale[256];
3198 time_t ret = (time_t) -1;
3200 /* Solaris strptime fails to recognize English month names in
3201 non-English locales, which we work around by temporarily setting
3202 locale to C before invoking strptime. */
3203 oldlocale = setlocale (LC_TIME, NULL);
3206 size_t l = strlen (oldlocale);
3207 if (l >= sizeof savedlocale)
3208 savedlocale[0] = '\0';
3210 memcpy (savedlocale, oldlocale, l);
3212 else savedlocale[0] = '\0';
3214 setlocale (LC_TIME, "C");
3216 for (i = 0; i < countof (time_formats); i++)
3220 /* Some versions of strptime use the existing contents of struct
3221 tm to recalculate the date according to format. Zero it out
3222 to prevent stack garbage from influencing strptime. */
3225 if (check_end (strptime (time_string, time_formats[i], &t)))
3232 /* Restore the previous locale. */
3234 setlocale (LC_TIME, savedlocale);
3239 /* Authorization support: We support three authorization schemes:
3241 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3243 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3244 consisting of answering to the server's challenge with the proper
3247 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3248 Stenberg for libcurl. Like digest, NTLM is based on a
3249 challenge-response mechanism, but unlike digest, it is non-standard
3250 (authenticates TCP connections rather than requests), undocumented
3251 and Microsoft-specific. */
3253 /* Create the authentication header contents for the `Basic' scheme.
3254 This is done by encoding the string "USER:PASS" to base64 and
3255 prepending the string "Basic " in front of it. */
3258 basic_authentication_encode (const char *user, const char *passwd)
3261 int len1 = strlen (user) + 1 + strlen (passwd);
3263 t1 = (char *)alloca (len1 + 1);
3264 sprintf (t1, "%s:%s", user, passwd);
3266 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3267 base64_encode (t1, len1, t2);
3269 return concat_strings ("Basic ", t2, (char *) 0);
3272 #define SKIP_WS(x) do { \
3273 while (c_isspace (*(x))) \
3277 #ifdef ENABLE_DIGEST
3278 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3279 an array of 16 bytes containing the hash keys, and BUF should be a
3280 buffer of 33 writable characters (32 for hex digits plus one for
3281 zero termination). */
3283 dump_hash (char *buf, const unsigned char *hash)
3287 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3289 *buf++ = XNUM_TO_digit (*hash >> 4);
3290 *buf++ = XNUM_TO_digit (*hash & 0xf);
3295 /* Take the line apart to find the challenge, and compose a digest
3296 authorization header. See RFC2069 section 2.1.2. */
3298 digest_authentication_encode (const char *au, const char *user,
3299 const char *passwd, const char *method,
3302 static char *realm, *opaque, *nonce;
3307 { "realm", &realm },
3308 { "opaque", &opaque },
3312 param_token name, value;
3314 realm = opaque = nonce = NULL;
3316 au += 6; /* skip over `Digest' */
3317 while (extract_param (&au, &name, &value, ','))
3320 size_t namelen = name.e - name.b;
3321 for (i = 0; i < countof (options); i++)
3322 if (namelen == strlen (options[i].name)
3323 && 0 == strncmp (name.b, options[i].name,
3326 *options[i].variable = strdupdelim (value.b, value.e);
3330 if (!realm || !nonce || !user || !passwd || !path || !method)
3333 xfree_null (opaque);
3338 /* Calculate the digest value. */
3341 unsigned char hash[MD5_DIGEST_SIZE];
3342 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3343 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3345 /* A1BUF = H(user ":" realm ":" password) */
3346 md5_init_ctx (&ctx);
3347 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3348 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3349 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3350 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3351 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3352 md5_finish_ctx (&ctx, hash);
3353 dump_hash (a1buf, hash);
3355 /* A2BUF = H(method ":" path) */
3356 md5_init_ctx (&ctx);
3357 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3358 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3359 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3360 md5_finish_ctx (&ctx, hash);
3361 dump_hash (a2buf, hash);
3363 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3364 md5_init_ctx (&ctx);
3365 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3366 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3367 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3368 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3369 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3370 md5_finish_ctx (&ctx, hash);
3371 dump_hash (response_digest, hash);
3373 res = xmalloc (strlen (user)
3378 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3379 + (opaque ? strlen (opaque) : 0)
3381 sprintf (res, "Digest \
3382 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3383 user, realm, nonce, path, response_digest);
3386 char *p = res + strlen (res);
3387 strcat (p, ", opaque=\"");
3394 #endif /* ENABLE_DIGEST */
3396 /* Computing the size of a string literal must take into account that
3397 value returned by sizeof includes the terminating \0. */
3398 #define STRSIZE(literal) (sizeof (literal) - 1)
3400 /* Whether chars in [b, e) begin with the literal string provided as
3401 first argument and are followed by whitespace or terminating \0.
3402 The comparison is case-insensitive. */
3403 #define STARTS(literal, b, e) \
3405 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3406 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3407 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3408 || c_isspace (b[STRSIZE (literal)])))
3411 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3413 return STARTS ("Basic", hdrbeg, hdrend)
3414 #ifdef ENABLE_DIGEST
3415 || STARTS ("Digest", hdrbeg, hdrend)
3418 || STARTS ("NTLM", hdrbeg, hdrend)
3425 /* Create the HTTP authorization request header. When the
3426 `WWW-Authenticate' response header is seen, according to the
3427 authorization scheme specified in that header (`Basic' and `Digest'
3428 are supported by the current implementation), produce an
3429 appropriate HTTP authorization request header. */
3431 create_authorization_line (const char *au, const char *user,
3432 const char *passwd, const char *method,
3433 const char *path, bool *finished)
3435 /* We are called only with known schemes, so we can dispatch on the
3437 switch (c_toupper (*au))
3439 case 'B': /* Basic */
3441 return basic_authentication_encode (user, passwd);
3442 #ifdef ENABLE_DIGEST
3443 case 'D': /* Digest */
3445 return digest_authentication_encode (au, user, passwd, method, path);
3448 case 'N': /* NTLM */
3449 if (!ntlm_input (&pconn.ntlm, au))
3454 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3457 /* We shouldn't get here -- this function should be only called
3458 with values approved by known_authentication_scheme_p. */
3466 if (!wget_cookie_jar)
3467 wget_cookie_jar = cookie_jar_new ();
3468 if (opt.cookies_input && !cookies_loaded_p)
3470 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3471 cookies_loaded_p = true;
3478 if (wget_cookie_jar)
3479 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3485 xfree_null (pconn.host);
3486 if (wget_cookie_jar)
3487 cookie_jar_delete (wget_cookie_jar);
3491 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3493 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3495 int len = strlen (ext);
3498 strncpy (shortext, ext, len - 1);
3499 shortext[len - 2] = '\0';
3502 if (last_period_in_local_filename == NULL
3503 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3504 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3506 int local_filename_len = strlen (hs->local_file);
3507 /* Resize the local file, allowing for ".html" preceded by
3508 optional ".NUMBER". */
3509 hs->local_file = xrealloc (hs->local_file,
3510 local_filename_len + 24 + len);
3511 strcpy (hs->local_file + local_filename_len, ext);
3512 /* If clobbering is not allowed and the file, as named,
3513 exists, tack on ".NUMBER.html" instead. */
3514 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3518 sprintf (hs->local_file + local_filename_len,
3519 ".%d%s", ext_num++, ext);
3520 while (file_exists_p (hs->local_file));
3522 *dt |= ADDED_HTML_EXTENSION;
3530 test_parse_content_disposition()
3535 char *opt_dir_prefix;
3539 { "filename=\"file.ext\"", NULL, "file.ext", true },
3540 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3541 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3542 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3543 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3544 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3545 { "attachment", NULL, NULL, false },
3546 { "attachment", "somedir", NULL, false },
3547 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3548 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3551 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3556 opt.dir_prefix = test_array[i].opt_dir_prefix;
3557 res = parse_content_disposition (test_array[i].hdrval, &filename);
3559 mu_assert ("test_parse_content_disposition: wrong result",
3560 res == test_array[i].result
3562 || 0 == strcmp (test_array[i].filename, filename)));
3568 #endif /* TESTING */
3571 * vim: et sts=2 sw=2 cino+={s