2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation,
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
55 # include "http-ntlm.h"
68 #endif /* def __VMS */
70 extern char *version_string;
74 static char *create_authorization_line (const char *, const char *,
75 const char *, const char *,
76 const char *, bool *);
77 static char *basic_authentication_encode (const char *, const char *);
78 static bool known_authentication_scheme_p (const char *, const char *);
79 static void ensure_extension (struct http_stat *, const char *, int *);
80 static void load_cookies (void);
83 # define MIN(x, y) ((x) > (y) ? (y) : (x))
87 static bool cookies_loaded_p;
88 static struct cookie_jar *wget_cookie_jar;
90 #define TEXTHTML_S "text/html"
91 #define TEXTXHTML_S "application/xhtml+xml"
92 #define TEXTCSS_S "text/css"
94 /* Some status code validation macros: */
95 #define H_10X(x) (((x) >= 100) && ((x) < 200))
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);
931 remaining_chunk_size = strtol (line, &endl, 16);
932 if (remaining_chunk_size == 0)
939 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
942 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
944 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
947 /* Don't normally report the error since this is an
948 optimization that should be invisible to the user. */
949 DEBUGP (("] aborting (%s).\n",
950 ret < 0 ? fd_errstr (fd) : "EOF received"));
957 remaining_chunk_size -= ret;
958 if (remaining_chunk_size == 0)
959 if (fd_read_line (fd) == NULL)
963 /* Safe even if %.*s bogusly expects terminating \0 because
964 we've zero-terminated dlbuf above. */
965 DEBUGP (("%.*s", ret, dlbuf));
968 DEBUGP (("] done.\n"));
972 #define NOT_RFC2231 0
973 #define RFC2231_NOENCODING 1
974 #define RFC2231_ENCODING 2
976 /* extract_param extracts the parameter name into NAME.
977 However, if the parameter name is in RFC2231 format then
978 this function adjusts NAME by stripping of the trailing
979 characters that are not part of the name but are present to
980 indicate the presence of encoding information in the value
981 or a fragment of a long parameter value
984 modify_param_name(param_token *name)
986 const char *delim1 = memchr (name->b, '*', name->e - name->b);
987 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
993 result = NOT_RFC2231;
995 else if(delim1 == delim2)
997 if ((name->e - 1) == delim1)
999 result = RFC2231_ENCODING;
1003 result = RFC2231_NOENCODING;
1010 result = RFC2231_ENCODING;
1015 /* extract_param extract the paramater value into VALUE.
1016 Like modify_param_name this function modifies VALUE by
1017 stripping off the encoding information from the actual value
1020 modify_param_value (param_token *value, int encoding_type )
1022 if (RFC2231_ENCODING == encoding_type)
1024 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1025 if ( delim != NULL )
1027 value->b = (delim+1);
1032 /* Extract a parameter from the string (typically an HTTP header) at
1033 **SOURCE and advance SOURCE to the next parameter. Return false
1034 when there are no more parameters to extract. The name of the
1035 parameter is returned in NAME, and the value in VALUE. If the
1036 parameter has no value, the token's value is zeroed out.
1038 For example, if *SOURCE points to the string "attachment;
1039 filename=\"foo bar\"", the first call to this function will return
1040 the token named "attachment" and no value, and the second call will
1041 return the token named "filename" and value "foo bar". The third
1042 call will return false, indicating no more valid tokens. */
1045 extract_param (const char **source, param_token *name, param_token *value,
1048 const char *p = *source;
1050 while (c_isspace (*p)) ++p;
1054 return false; /* no error; nothing more to extract */
1059 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1061 if (name->b == name->e)
1062 return false; /* empty name: error */
1063 while (c_isspace (*p)) ++p;
1064 if (*p == separator || !*p) /* no value */
1067 if (*p == separator) ++p;
1072 return false; /* error */
1074 /* *p is '=', extract value */
1076 while (c_isspace (*p)) ++p;
1077 if (*p == '"') /* quoted */
1080 while (*p && *p != '"') ++p;
1084 /* Currently at closing quote; find the end of param. */
1085 while (c_isspace (*p)) ++p;
1086 while (*p && *p != separator) ++p;
1087 if (*p == separator)
1090 /* garbage after closed quote, e.g. foo="bar"baz */
1096 while (*p && *p != separator) ++p;
1098 while (value->e != value->b && c_isspace (value->e[-1]))
1100 if (*p == separator) ++p;
1104 int param_type = modify_param_name(name);
1105 if (NOT_RFC2231 != param_type)
1107 modify_param_value(value, param_type);
1113 #undef RFC2231_NOENCODING
1114 #undef RFC2231_ENCODING
1116 /* Appends the string represented by VALUE to FILENAME */
1119 append_value_to_filename (char **filename, param_token const * const value)
1121 int original_length = strlen(*filename);
1122 int new_length = strlen(*filename) + (value->e - value->b);
1123 *filename = xrealloc (*filename, new_length+1);
1124 memcpy (*filename + original_length, value->b, (value->e - value->b));
1125 (*filename)[new_length] = '\0';
1129 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1131 /* Parse the contents of the `Content-Disposition' header, extracting
1132 the information useful to Wget. Content-Disposition is a header
1133 borrowed from MIME; when used in HTTP, it typically serves for
1134 specifying the desired file name of the resource. For example:
1136 Content-Disposition: attachment; filename="flora.jpg"
1138 Wget will skip the tokens it doesn't care about, such as
1139 "attachment" in the previous example; it will also skip other
1140 unrecognized params. If the header is syntactically correct and
1141 contains a file name, a copy of the file name is stored in
1142 *filename and true is returned. Otherwise, the function returns
1145 The file name is stripped of directory components and must not be
1148 Historically, this function returned filename prefixed with opt.dir_prefix,
1149 now that logic is handled by the caller, new code should pay attention,
1150 changed by crq, Sep 2010.
1154 parse_content_disposition (const char *hdr, char **filename)
1156 param_token name, value;
1158 while (extract_param (&hdr, &name, &value, ';'))
1160 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1161 if ( isFilename && value.b != NULL)
1163 /* Make the file name begin at the last slash or backslash. */
1164 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1165 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1166 if (last_slash && last_bs)
1167 value.b = 1 + MAX (last_slash, last_bs);
1168 else if (last_slash || last_bs)
1169 value.b = 1 + (last_slash ? last_slash : last_bs);
1170 if (value.b == value.e)
1174 append_value_to_filename (filename, &value);
1176 *filename = strdupdelim (value.b, value.e);
1187 /* Persistent connections. Currently, we cache the most recently used
1188 connection as persistent, provided that the HTTP server agrees to
1189 make it such. The persistence data is stored in the variables
1190 below. Ideally, it should be possible to cache an arbitrary fixed
1191 number of these connections. */
1193 /* Whether a persistent connection is active. */
1194 static bool pconn_active;
1197 /* The socket of the connection. */
1200 /* Host and port of the currently active persistent connection. */
1204 /* Whether a ssl handshake has occoured on this connection. */
1207 /* Whether the connection was authorized. This is only done by
1208 NTLM, which authorizes *connections* rather than individual
1209 requests. (That practice is peculiar for HTTP, but it is a
1210 useful optimization.) */
1214 /* NTLM data of the current connection. */
1215 struct ntlmdata ntlm;
1219 /* Mark the persistent connection as invalid and free the resources it
1220 uses. This is used by the CLOSE_* macros after they forcefully
1221 close a registered persistent connection. */
1224 invalidate_persistent (void)
1226 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1227 pconn_active = false;
1228 fd_close (pconn.socket);
1233 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1234 persistent. This will enable someone to use the same connection
1235 later. In the context of HTTP, this must be called only AFTER the
1236 response has been received and the server has promised that the
1237 connection will remain alive.
1239 If a previous connection was persistent, it is closed. */
1242 register_persistent (const char *host, int port, int fd, bool ssl)
1246 if (pconn.socket == fd)
1248 /* The connection FD is already registered. */
1253 /* The old persistent connection is still active; close it
1254 first. This situation arises whenever a persistent
1255 connection exists, but we then connect to a different
1256 host, and try to register a persistent connection to that
1258 invalidate_persistent ();
1262 pconn_active = true;
1264 pconn.host = xstrdup (host);
1267 pconn.authorized = false;
1269 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1272 /* Return true if a persistent connection is available for connecting
1276 persistent_available_p (const char *host, int port, bool ssl,
1277 bool *host_lookup_failed)
1279 /* First, check whether a persistent connection is active at all. */
1283 /* If we want SSL and the last connection wasn't or vice versa,
1284 don't use it. Checking for host and port is not enough because
1285 HTTP and HTTPS can apparently coexist on the same port. */
1286 if (ssl != pconn.ssl)
1289 /* If we're not connecting to the same port, we're not interested. */
1290 if (port != pconn.port)
1293 /* If the host is the same, we're in business. If not, there is
1294 still hope -- read below. */
1295 if (0 != strcasecmp (host, pconn.host))
1297 /* Check if pconn.socket is talking to HOST under another name.
1298 This happens often when both sites are virtual hosts
1299 distinguished only by name and served by the same network
1300 interface, and hence the same web server (possibly set up by
1301 the ISP and serving many different web sites). This
1302 admittedly unconventional optimization does not contradict
1303 HTTP and works well with popular server software. */
1307 struct address_list *al;
1310 /* Don't try to talk to two different SSL sites over the same
1311 secure connection! (Besides, it's not clear that
1312 name-based virtual hosting is even possible with SSL.) */
1315 /* If pconn.socket's peer is one of the IP addresses HOST
1316 resolves to, pconn.socket is for all intents and purposes
1317 already talking to HOST. */
1319 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1321 /* Can't get the peer's address -- something must be very
1322 wrong with the connection. */
1323 invalidate_persistent ();
1326 al = lookup_host (host, 0);
1329 *host_lookup_failed = true;
1333 found = address_list_contains (al, &ip);
1334 address_list_release (al);
1339 /* The persistent connection's peer address was found among the
1340 addresses HOST resolved to; therefore, pconn.sock is in fact
1341 already talking to HOST -- no need to reconnect. */
1344 /* Finally, check whether the connection is still open. This is
1345 important because most servers implement liberal (short) timeout
1346 on persistent connections. Wget can of course always reconnect
1347 if the connection doesn't work out, but it's nicer to know in
1348 advance. This test is a logical followup of the first test, but
1349 is "expensive" and therefore placed at the end of the list.
1351 (Current implementation of test_socket_open has a nice side
1352 effect that it treats sockets with pending data as "closed".
1353 This is exactly what we want: if a broken server sends message
1354 body in response to HEAD, or if it sends more than conent-length
1355 data, we won't reuse the corrupted connection.) */
1357 if (!test_socket_open (pconn.socket))
1359 /* Oops, the socket is no longer open. Now that we know that,
1360 let's invalidate the persistent connection before returning
1362 invalidate_persistent ();
1369 /* The idea behind these two CLOSE macros is to distinguish between
1370 two cases: one when the job we've been doing is finished, and we
1371 want to close the connection and leave, and two when something is
1372 seriously wrong and we're closing the connection as part of
1375 In case of keep_alive, CLOSE_FINISH should leave the connection
1376 open, while CLOSE_INVALIDATE should still close it.
1378 Note that the semantics of the flag `keep_alive' is "this
1379 connection *will* be reused (the server has promised not to close
1380 the connection once we're done)", while the semantics of
1381 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1382 active, registered connection". */
1384 #define CLOSE_FINISH(fd) do { \
1387 if (pconn_active && (fd) == pconn.socket) \
1388 invalidate_persistent (); \
1397 #define CLOSE_INVALIDATE(fd) do { \
1398 if (pconn_active && (fd) == pconn.socket) \
1399 invalidate_persistent (); \
1407 wgint len; /* received length */
1408 wgint contlen; /* expected length */
1409 wgint restval; /* the restart value */
1410 int res; /* the result of last read */
1411 char *rderrmsg; /* error message from read error */
1412 char *newloc; /* new location (redirection) */
1413 char *remote_time; /* remote time-stamp string */
1414 char *error; /* textual HTTP error */
1415 int statcode; /* status code */
1416 char *message; /* status message */
1417 wgint rd_size; /* amount of data read from socket */
1418 double dltime; /* time it took to download the data */
1419 const char *referer; /* value of the referer header. */
1420 char *local_file; /* local file name. */
1421 bool existence_checked; /* true if we already checked for a file's
1422 existence after having begun to download
1423 (needed in gethttp for when connection is
1424 interrupted/restarted. */
1425 bool timestamp_checked; /* true if pre-download time-stamping checks
1426 * have already been performed */
1427 char *orig_file_name; /* name of file to compare for time-stamping
1428 * (might be != local_file if -K is set) */
1429 wgint orig_file_size; /* size of file to compare for time-stamping */
1430 time_t orig_file_tstamp; /* time-stamp of file to compare for
1435 free_hstat (struct http_stat *hs)
1437 xfree_null (hs->newloc);
1438 xfree_null (hs->remote_time);
1439 xfree_null (hs->error);
1440 xfree_null (hs->rderrmsg);
1441 xfree_null (hs->local_file);
1442 xfree_null (hs->orig_file_name);
1443 xfree_null (hs->message);
1445 /* Guard against being called twice. */
1447 hs->remote_time = NULL;
1451 #define BEGINS_WITH(line, string_constant) \
1452 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1453 && (c_isspace (line[sizeof (string_constant) - 1]) \
1454 || !line[sizeof (string_constant) - 1]))
1457 #define SET_USER_AGENT(req) do { \
1458 if (!opt.useragent) \
1459 request_set_header (req, "User-Agent", \
1460 aprintf ("Wget/%s (VMS %s %s)", \
1461 version_string, vms_arch(), vms_vers()), \
1463 else if (*opt.useragent) \
1464 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1466 #else /* def __VMS */
1467 #define SET_USER_AGENT(req) do { \
1468 if (!opt.useragent) \
1469 request_set_header (req, "User-Agent", \
1470 aprintf ("Wget/%s (%s)", \
1471 version_string, OS_TYPE), \
1473 else if (*opt.useragent) \
1474 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1476 #endif /* def __VMS [else] */
1478 /* The flags that allow clobbering the file (opening with "wb").
1479 Defined here to avoid repetition later. #### This will require
1481 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1482 || opt.dirstruct || opt.output_document)
1484 /* Retrieve a document through HTTP protocol. It recognizes status
1485 code, and correctly handles redirections. It closes the network
1486 socket. If it receives an error from the functions below it, it
1487 will print it if there is enough information to do so (almost
1488 always), returning the error to the caller (i.e. http_loop).
1490 Various HTTP parameters are stored to hs.
1492 If PROXY is non-NULL, the connection will be made to the proxy
1493 server, and u->url will be requested. */
1495 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1496 struct iri *iri, int count)
1498 struct request *req;
1501 char *user, *passwd;
1505 wgint contlen, contrange;
1512 /* Set to 1 when the authorization has already been sent and should
1513 not be tried again. */
1514 bool auth_finished = false;
1516 /* Set to 1 when just globally-set Basic authorization has been sent;
1517 * should prevent further Basic negotiations, but not other
1519 bool basic_auth_finished = false;
1521 /* Whether NTLM authentication is used for this request. */
1522 bool ntlm_seen = false;
1524 /* Whether our connection to the remote host is through SSL. */
1525 bool using_ssl = false;
1527 /* Whether a HEAD request will be issued (as opposed to GET or
1529 bool head_only = !!(*dt & HEAD_ONLY);
1532 struct response *resp;
1536 /* Whether this connection will be kept alive after the HTTP request
1540 /* Is the server using the chunked transfer encoding? */
1541 bool chunked_transfer_encoding = false;
1543 /* Whether keep-alive should be inhibited. */
1544 bool inhibit_keep_alive =
1545 !opt.http_keep_alive || opt.ignore_length;
1547 /* Headers sent when using POST. */
1548 wgint post_data_size = 0;
1550 bool host_lookup_failed = false;
1553 if (u->scheme == SCHEME_HTTPS)
1555 /* Initialize the SSL context. After this has once been done,
1556 it becomes a no-op. */
1559 scheme_disable (SCHEME_HTTPS);
1560 logprintf (LOG_NOTQUIET,
1561 _("Disabling SSL due to encountered errors.\n"));
1562 return SSLINITFAILED;
1565 #endif /* HAVE_SSL */
1567 /* Initialize certain elements of struct http_stat. */
1571 hs->rderrmsg = NULL;
1573 hs->remote_time = NULL;
1579 /* Prepare the request to send. */
1581 req = request_new ();
1584 const char *meth = "GET";
1587 else if (opt.post_file_name || opt.post_data)
1589 /* Use the full path, i.e. one that includes the leading slash and
1590 the query string. E.g. if u->path is "foo/bar" and u->query is
1591 "param=value", full_path will be "/foo/bar?param=value". */
1594 /* When using SSL over proxy, CONNECT establishes a direct
1595 connection to the HTTPS server. Therefore use the same
1596 argument as when talking to the server directly. */
1597 && u->scheme != SCHEME_HTTPS
1600 meth_arg = xstrdup (u->url);
1602 meth_arg = url_full_path (u);
1603 request_set_method (req, meth, meth_arg);
1606 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1607 if (*dt & SEND_NOCACHE)
1608 request_set_header (req, "Pragma", "no-cache", rel_none);
1609 if (hs->restval && !opt.timestamping)
1610 request_set_header (req, "Range",
1611 aprintf ("bytes=%s-",
1612 number_to_static_string (hs->restval)),
1614 SET_USER_AGENT (req);
1615 request_set_header (req, "Accept", "*/*", rel_none);
1617 /* Find the username and password for authentication. */
1620 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1621 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1622 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1624 /* We only do "site-wide" authentication with "global" user/password
1625 * values unless --auth-no-challange has been requested; URL user/password
1626 * info overrides. */
1627 if (user && passwd && (!u->user || opt.auth_without_challenge))
1629 /* If this is a host for which we've already received a Basic
1630 * challenge, we'll go ahead and send Basic authentication creds. */
1631 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1634 /* Generate the Host header, HOST:PORT. Take into account that:
1636 - Broken server-side software often doesn't recognize the PORT
1637 argument, so we must generate "Host: www.server.com" instead of
1638 "Host: www.server.com:80" (and likewise for https port).
1640 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1641 becomes ambiguous and needs to be rewritten as "Host:
1642 [3ffe:8100:200:2::2]:1234". */
1644 /* Formats arranged for hfmt[add_port][add_squares]. */
1645 static const char *hfmt[][2] = {
1646 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1648 int add_port = u->port != scheme_default_port (u->scheme);
1649 int add_squares = strchr (u->host, ':') != NULL;
1650 request_set_header (req, "Host",
1651 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1655 if (inhibit_keep_alive)
1656 request_set_header (req, "Connection", "Close", rel_none);
1660 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1663 request_set_header (req, "Connection", "Close", rel_none);
1664 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1668 if (opt.post_data || opt.post_file_name)
1670 request_set_header (req, "Content-Type",
1671 "application/x-www-form-urlencoded", rel_none);
1673 post_data_size = strlen (opt.post_data);
1676 post_data_size = file_size (opt.post_file_name);
1677 if (post_data_size == -1)
1679 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1680 quote (opt.post_file_name), strerror (errno));
1684 request_set_header (req, "Content-Length",
1685 xstrdup (number_to_static_string (post_data_size)),
1690 /* We need to come back here when the initial attempt to retrieve
1691 without authorization header fails. (Expected to happen at least
1692 for the Digest authorization scheme.) */
1695 request_set_header (req, "Cookie",
1696 cookie_header (wget_cookie_jar,
1697 u->host, u->port, u->path,
1699 u->scheme == SCHEME_HTTPS
1706 /* Add the user headers. */
1707 if (opt.user_headers)
1710 for (i = 0; opt.user_headers[i]; i++)
1711 request_set_user_header (req, opt.user_headers[i]);
1717 char *proxy_user, *proxy_passwd;
1718 /* For normal username and password, URL components override
1719 command-line/wgetrc parameters. With proxy
1720 authentication, it's the reverse, because proxy URLs are
1721 normally the "permanent" ones, so command-line args
1722 should take precedence. */
1723 if (opt.proxy_user && opt.proxy_passwd)
1725 proxy_user = opt.proxy_user;
1726 proxy_passwd = opt.proxy_passwd;
1730 proxy_user = proxy->user;
1731 proxy_passwd = proxy->passwd;
1733 /* #### This does not appear right. Can't the proxy request,
1734 say, `Digest' authentication? */
1735 if (proxy_user && proxy_passwd)
1736 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1738 /* If we're using a proxy, we will be connecting to the proxy
1742 /* Proxy authorization over SSL is handled below. */
1744 if (u->scheme != SCHEME_HTTPS)
1746 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1751 /* Establish the connection. */
1753 if (inhibit_keep_alive)
1757 /* Look for a persistent connection to target host, unless a
1758 proxy is used. The exception is when SSL is in use, in which
1759 case the proxy is nothing but a passthrough to the target
1760 host, registered as a connection to the latter. */
1761 struct url *relevant = conn;
1763 if (u->scheme == SCHEME_HTTPS)
1767 if (persistent_available_p (relevant->host, relevant->port,
1769 relevant->scheme == SCHEME_HTTPS,
1773 &host_lookup_failed))
1775 sock = pconn.socket;
1776 using_ssl = pconn.ssl;
1777 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1778 quotearg_style (escape_quoting_style, pconn.host),
1780 DEBUGP (("Reusing fd %d.\n", sock));
1781 if (pconn.authorized)
1782 /* If the connection is already authorized, the "Basic"
1783 authorization added by code above is unnecessary and
1785 request_remove_header (req, "Authorization");
1787 else if (host_lookup_failed)
1790 logprintf(LOG_NOTQUIET,
1791 _("%s: unable to resolve host address %s\n"),
1792 exec_name, quote (relevant->host));
1799 sock = connect_to_host (conn->host, conn->port);
1808 return (retryable_socket_connect_error (errno)
1809 ? CONERROR : CONIMPOSSIBLE);
1813 if (proxy && u->scheme == SCHEME_HTTPS)
1815 /* When requesting SSL URLs through proxies, use the
1816 CONNECT method to request passthrough. */
1817 struct request *connreq = request_new ();
1818 request_set_method (connreq, "CONNECT",
1819 aprintf ("%s:%d", u->host, u->port));
1820 SET_USER_AGENT (connreq);
1823 request_set_header (connreq, "Proxy-Authorization",
1824 proxyauth, rel_value);
1825 /* Now that PROXYAUTH is part of the CONNECT request,
1826 zero it out so we don't send proxy authorization with
1827 the regular request below. */
1830 /* Examples in rfc2817 use the Host header in CONNECT
1831 requests. I don't see how that gains anything, given
1832 that the contents of Host would be exactly the same as
1833 the contents of CONNECT. */
1835 write_error = request_send (connreq, sock);
1836 request_free (connreq);
1837 if (write_error < 0)
1839 CLOSE_INVALIDATE (sock);
1843 head = read_http_response_head (sock);
1846 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1848 CLOSE_INVALIDATE (sock);
1857 DEBUGP (("proxy responded with: [%s]\n", head));
1859 resp = resp_new (head);
1860 statcode = resp_status (resp, &message);
1863 char *tms = datetime_str (time (NULL));
1864 logprintf (LOG_VERBOSE, "%d\n", statcode);
1865 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1866 quotearg_style (escape_quoting_style,
1867 _("Malformed status line")));
1871 hs->message = xstrdup (message);
1874 if (statcode != 200)
1877 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1878 message ? quotearg_style (escape_quoting_style, message) : "?");
1879 xfree_null (message);
1882 xfree_null (message);
1884 /* SOCK is now *really* connected to u->host, so update CONN
1885 to reflect this. That way register_persistent will
1886 register SOCK as being connected to u->host:u->port. */
1890 if (conn->scheme == SCHEME_HTTPS)
1892 if (!ssl_connect_wget (sock))
1897 else if (!ssl_check_certificate (sock, u->host))
1900 return VERIFCERTERR;
1904 #endif /* HAVE_SSL */
1907 /* Send the request to server. */
1908 write_error = request_send (req, sock);
1910 if (write_error >= 0)
1914 DEBUGP (("[POST data: %s]\n", opt.post_data));
1915 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1917 else if (opt.post_file_name && post_data_size != 0)
1918 write_error = post_file (sock, opt.post_file_name, post_data_size);
1921 if (write_error < 0)
1923 CLOSE_INVALIDATE (sock);
1927 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1928 proxy ? "Proxy" : "HTTP");
1934 head = read_http_response_head (sock);
1939 logputs (LOG_NOTQUIET, _("No data received.\n"));
1940 CLOSE_INVALIDATE (sock);
1946 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1948 CLOSE_INVALIDATE (sock);
1953 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1955 resp = resp_new (head);
1957 /* Check for status line. */
1959 statcode = resp_status (resp, &message);
1962 char *tms = datetime_str (time (NULL));
1963 logprintf (LOG_VERBOSE, "%d\n", statcode);
1964 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1965 quotearg_style (escape_quoting_style,
1966 _("Malformed status line")));
1967 CLOSE_INVALIDATE (sock);
1972 if (H_10X (statcode))
1974 DEBUGP (("Ignoring response\n"));
1978 hs->message = xstrdup (message);
1979 if (!opt.server_response)
1980 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1981 message ? quotearg_style (escape_quoting_style, message) : "");
1984 logprintf (LOG_VERBOSE, "\n");
1985 print_server_response (resp, " ");
1988 if (!opt.ignore_length
1989 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1993 parsed = str_to_wgint (hdrval, NULL, 10);
1994 if (parsed == WGINT_MAX && errno == ERANGE)
1997 #### If Content-Length is out of range, it most likely
1998 means that the file is larger than 2G and that we're
1999 compiled without LFS. In that case we should probably
2000 refuse to even attempt to download the file. */
2003 else if (parsed < 0)
2005 /* Negative Content-Length; nonsensical, so we can't
2006 assume any information about the content to receive. */
2013 /* Check for keep-alive related responses. */
2014 if (!inhibit_keep_alive && contlen != -1)
2016 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2018 if (0 == strcasecmp (hdrval, "Close"))
2023 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2024 if (0 == strcasecmp (hdrval, "chunked"))
2025 chunked_transfer_encoding = true;
2027 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2031 const char *scbeg, *scend;
2032 /* The jar should have been created by now. */
2033 assert (wget_cookie_jar != NULL);
2035 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2036 &scbeg, &scend)) != -1;
2039 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2040 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2041 u->path, set_cookie);
2046 /* The server has promised that it will not close the connection
2047 when we're done. This means that we can register it. */
2048 register_persistent (conn->host, conn->port, sock, using_ssl);
2050 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2052 /* Authorization is required. */
2053 if (keep_alive && !head_only
2054 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2055 CLOSE_FINISH (sock);
2057 CLOSE_INVALIDATE (sock);
2058 pconn.authorized = false;
2059 if (!auth_finished && (user && passwd))
2061 /* IIS sends multiple copies of WWW-Authenticate, one with
2062 the value "negotiate", and other(s) with data. Loop over
2063 all the occurrences and pick the one we recognize. */
2065 const char *wabeg, *waend;
2066 char *www_authenticate = NULL;
2068 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2069 &wabeg, &waend)) != -1;
2071 if (known_authentication_scheme_p (wabeg, waend))
2073 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2077 if (!www_authenticate)
2079 /* If the authentication header is missing or
2080 unrecognized, there's no sense in retrying. */
2081 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2083 else if (!basic_auth_finished
2084 || !BEGINS_WITH (www_authenticate, "Basic"))
2087 pth = url_full_path (u);
2088 request_set_header (req, "Authorization",
2089 create_authorization_line (www_authenticate,
2091 request_method (req),
2095 if (BEGINS_WITH (www_authenticate, "NTLM"))
2097 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2099 /* Need to register this host as using basic auth,
2100 * so we automatically send creds next time. */
2101 register_basic_auth_host (u->host);
2104 xfree_null (message);
2107 goto retry_with_auth;
2111 /* We already did Basic auth, and it failed. Gotta
2115 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2117 xfree_null (message);
2122 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2124 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2126 pconn.authorized = true;
2129 /* Determine the local filename if needed. Notice that if -O is used
2130 * hstat.local_file is set by http_loop to the argument of -O. */
2131 if (!hs->local_file)
2133 char *local_file = NULL;
2135 /* Honor Content-Disposition whether possible. */
2136 if (!opt.content_disposition
2137 || !resp_header_copy (resp, "Content-Disposition",
2138 hdrval, sizeof (hdrval))
2139 || !parse_content_disposition (hdrval, &local_file))
2141 /* The Content-Disposition header is missing or broken.
2142 * Choose unique file name according to given URL. */
2143 hs->local_file = url_file_name (u, NULL);
2147 DEBUGP (("Parsed filename from Content-Disposition: %s\n",
2149 hs->local_file = url_file_name (u, local_file);
2153 /* TODO: perform this check only once. */
2154 if (!hs->existence_checked && file_exists_p (hs->local_file))
2156 if (opt.noclobber && !opt.output_document)
2158 /* If opt.noclobber is turned on and file already exists, do not
2159 retrieve the file. But if the output_document was given, then this
2160 test was already done and the file didn't exist. Hence the !opt.output_document */
2161 logprintf (LOG_VERBOSE, _("\
2162 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2163 /* If the file is there, we suppose it's retrieved OK. */
2166 /* #### Bogusness alert. */
2167 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2168 if (has_html_suffix_p (hs->local_file))
2172 xfree_null (message);
2173 return RETRUNNEEDED;
2175 else if (!ALLOW_CLOBBER)
2177 char *unique = unique_name (hs->local_file, true);
2178 if (unique != hs->local_file)
2179 xfree (hs->local_file);
2180 hs->local_file = unique;
2183 hs->existence_checked = true;
2185 /* Support timestamping */
2186 /* TODO: move this code out of gethttp. */
2187 if (opt.timestamping && !hs->timestamp_checked)
2189 size_t filename_len = strlen (hs->local_file);
2190 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2191 bool local_dot_orig_file_exists = false;
2192 char *local_filename = NULL;
2195 if (opt.backup_converted)
2196 /* If -K is specified, we'll act on the assumption that it was specified
2197 last time these files were downloaded as well, and instead of just
2198 comparing local file X against server file X, we'll compare local
2199 file X.orig (if extant, else X) against server file X. If -K
2200 _wasn't_ specified last time, or the server contains files called
2201 *.orig, -N will be back to not operating correctly with -k. */
2203 /* Would a single s[n]printf() call be faster? --dan
2205 Definitely not. sprintf() is horribly slow. It's a
2206 different question whether the difference between the two
2207 affects a program. Usually I'd say "no", but at one
2208 point I profiled Wget, and found that a measurable and
2209 non-negligible amount of time was lost calling sprintf()
2210 in url.c. Replacing sprintf with inline calls to
2211 strcpy() and number_to_string() made a difference.
2213 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2214 memcpy (filename_plus_orig_suffix + filename_len,
2215 ORIG_SFX, sizeof (ORIG_SFX));
2217 /* Try to stat() the .orig file. */
2218 if (stat (filename_plus_orig_suffix, &st) == 0)
2220 local_dot_orig_file_exists = true;
2221 local_filename = filename_plus_orig_suffix;
2225 if (!local_dot_orig_file_exists)
2226 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2227 if (stat (hs->local_file, &st) == 0)
2228 local_filename = hs->local_file;
2230 if (local_filename != NULL)
2231 /* There was a local file, so we'll check later to see if the version
2232 the server has is the same version we already have, allowing us to
2235 hs->orig_file_name = xstrdup (local_filename);
2236 hs->orig_file_size = st.st_size;
2237 hs->orig_file_tstamp = st.st_mtime;
2239 /* Modification time granularity is 2 seconds for Windows, so
2240 increase local time by 1 second for later comparison. */
2241 ++hs->orig_file_tstamp;
2248 hs->statcode = statcode;
2250 hs->error = xstrdup (_("Malformed status line"));
2252 hs->error = xstrdup (_("(no description)"));
2254 hs->error = xstrdup (message);
2255 xfree_null (message);
2257 type = resp_header_strdup (resp, "Content-Type");
2260 char *tmp = strchr (type, ';');
2263 /* sXXXav: only needed if IRI support is enabled */
2264 char *tmp2 = tmp + 1;
2266 while (tmp > type && c_isspace (tmp[-1]))
2270 /* Try to get remote encoding if needed */
2271 if (opt.enable_iri && !opt.encoding_remote)
2273 tmp = parse_charset (tmp2);
2275 set_content_encoding (iri, tmp);
2279 hs->newloc = resp_header_strdup (resp, "Location");
2280 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2282 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2284 wgint first_byte_pos, last_byte_pos, entity_length;
2285 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2288 contrange = first_byte_pos;
2289 contlen = last_byte_pos - first_byte_pos + 1;
2294 /* 20x responses are counted among successful by default. */
2295 if (H_20X (statcode))
2298 /* Return if redirected. */
2299 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2301 /* RFC2068 says that in case of the 300 (multiple choices)
2302 response, the server can output a preferred URL through
2303 `Location' header; otherwise, the request should be treated
2304 like GET. So, if the location is set, it will be a
2305 redirection; otherwise, just proceed normally. */
2306 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2310 logprintf (LOG_VERBOSE,
2311 _("Location: %s%s\n"),
2312 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2313 hs->newloc ? _(" [following]") : "");
2314 if (keep_alive && !head_only
2315 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2316 CLOSE_FINISH (sock);
2318 CLOSE_INVALIDATE (sock);
2325 /* If content-type is not given, assume text/html. This is because
2326 of the multitude of broken CGI's that "forget" to generate the
2329 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2330 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2336 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2341 if (opt.adjust_extension)
2344 /* -E / --adjust-extension / adjust_extension = on was specified,
2345 and this is a text/html file. If some case-insensitive
2346 variation on ".htm[l]" isn't already the file's suffix,
2349 ensure_extension (hs, ".html", dt);
2351 else if (*dt & TEXTCSS)
2353 ensure_extension (hs, ".css", dt);
2357 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2358 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2359 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2361 /* If `-c' is in use and the file has been fully downloaded (or
2362 the remote file has shrunk), Wget effectively requests bytes
2363 after the end of file and the server response with 416
2364 (or 200 with a <= Content-Length. */
2365 logputs (LOG_VERBOSE, _("\
2366 \n The file is already fully retrieved; nothing to do.\n\n"));
2367 /* In case the caller inspects. */
2370 /* Mark as successfully retrieved. */
2373 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2374 might be more bytes in the body. */
2376 return RETRUNNEEDED;
2378 if ((contrange != 0 && contrange != hs->restval)
2379 || (H_PARTIAL (statcode) && !contrange))
2381 /* The Range request was somehow misunderstood by the server.
2384 CLOSE_INVALIDATE (sock);
2391 hs->contlen = contlen + contrange;
2397 /* No need to print this output if the body won't be
2398 downloaded at all, or if the original server response is
2400 logputs (LOG_VERBOSE, _("Length: "));
2403 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2404 if (contlen + contrange >= 1024)
2405 logprintf (LOG_VERBOSE, " (%s)",
2406 human_readable (contlen + contrange));
2409 if (contlen >= 1024)
2410 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2411 number_to_static_string (contlen),
2412 human_readable (contlen));
2414 logprintf (LOG_VERBOSE, _(", %s remaining"),
2415 number_to_static_string (contlen));
2419 logputs (LOG_VERBOSE,
2420 opt.ignore_length ? _("ignored") : _("unspecified"));
2422 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2424 logputs (LOG_VERBOSE, "\n");
2428 type = NULL; /* We don't need it any more. */
2430 /* Return if we have no intention of further downloading. */
2431 if (!(*dt & RETROKF) || head_only)
2433 /* In case the caller cares to look... */
2438 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2439 servers not to send body in response to a HEAD request, and
2440 those that do will likely be caught by test_socket_open.
2441 If not, they can be worked around using
2442 `--no-http-keep-alive'. */
2443 CLOSE_FINISH (sock);
2445 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2446 /* Successfully skipped the body; also keep using the socket. */
2447 CLOSE_FINISH (sock);
2449 CLOSE_INVALIDATE (sock);
2451 return RETRFINISHED;
2455 For VMS, define common fopen() optional arguments.
2458 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2459 # define FOPEN_BIN_FLAG 3
2460 #else /* def __VMS */
2461 # define FOPEN_BIN_FLAG true
2462 #endif /* def __VMS [else] */
2464 /* Open the local file. */
2467 mkalldirs (hs->local_file);
2469 rotate_backups (hs->local_file);
2476 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2477 #else /* def __VMS */
2478 fp = fopen (hs->local_file, "ab");
2479 #endif /* def __VMS [else] */
2481 else if (ALLOW_CLOBBER || count > 0)
2483 if (opt.unlink && file_exists_p (hs->local_file))
2485 int res = unlink (hs->local_file);
2488 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file,
2490 CLOSE_INVALIDATE (sock);
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, struct url *original_url, char **newloc,
2589 char **local_file, const char *referer, int *dt, struct url *proxy,
2593 bool got_head = false; /* used for time-stamping and filename detection */
2594 bool time_came_from_head = false;
2595 bool got_name = false;
2598 uerr_t err, ret = TRYLIMEXC;
2599 time_t tmr = -1; /* remote time-stamp */
2600 struct http_stat hstat; /* HTTP status */
2602 bool send_head_first = true;
2604 bool force_full_retrieve = false;
2606 /* Assert that no value for *LOCAL_FILE was passed. */
2607 assert (local_file == NULL || *local_file == NULL);
2609 /* Set LOCAL_FILE parameter. */
2610 if (local_file && opt.output_document)
2611 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2613 /* Reset NEWLOC parameter. */
2616 /* This used to be done in main(), but it's a better idea to do it
2617 here so that we don't go through the hoops if we're just using
2622 /* Warn on (likely bogus) wildcard usage in HTTP. */
2623 if (opt.ftp_glob && has_wildcards_p (u->path))
2624 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2626 /* Setup hstat struct. */
2628 hstat.referer = referer;
2630 if (opt.output_document)
2632 hstat.local_file = xstrdup (opt.output_document);
2635 else if (!opt.content_disposition)
2638 url_file_name (opt.trustservernames ? u : original_url, NULL);
2642 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2643 * screaming for some refactoring. */
2644 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2646 /* If opt.noclobber is turned on and file already exists, do not
2647 retrieve the file. But if the output_document was given, then this
2648 test was already done and the file didn't exist. Hence the !opt.output_document */
2649 logprintf (LOG_VERBOSE, _("\
2650 File %s already there; not retrieving.\n\n"),
2651 quote (hstat.local_file));
2652 /* If the file is there, we suppose it's retrieved OK. */
2655 /* #### Bogusness alert. */
2656 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2657 if (has_html_suffix_p (hstat.local_file))
2664 /* Reset the counter. */
2667 /* Reset the document type. */
2670 /* Skip preliminary HEAD request if we're not in spider mode. */
2672 send_head_first = false;
2674 /* Send preliminary HEAD request if -N is given and we have an existing
2675 * destination file. */
2676 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL);
2677 if (opt.timestamping && (file_exists_p (file_name)
2678 || opt.content_disposition))
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. */
2729 if (force_full_retrieve)
2730 hstat.restval = hstat.len;
2731 else if (opt.always_rest
2733 && stat (hstat.local_file, &st) == 0
2734 && S_ISREG (st.st_mode))
2735 /* When -c is used, continue from on-disk size. (Can't use
2736 hstat.len even if count>1 because we don't want a failed
2737 first attempt to clobber existing data.) */
2738 hstat.restval = st.st_size;
2740 /* otherwise, continue where the previous try left off */
2741 hstat.restval = hstat.len;
2745 /* Decide whether to send the no-cache directive. We send it in
2747 a) we're using a proxy, and we're past our first retrieval.
2748 Some proxies are notorious for caching incomplete data, so
2749 we require a fresh get.
2750 b) caching is explicitly inhibited. */
2751 if ((proxy && count > 1) /* a */
2752 || !opt.allow_cache) /* b */
2753 *dt |= SEND_NOCACHE;
2755 *dt &= ~SEND_NOCACHE;
2757 /* Try fetching the document, or at least its head. */
2758 err = gethttp (u, &hstat, dt, proxy, iri, count);
2761 tms = datetime_str (time (NULL));
2763 /* Get the new location (with or without the redirection). */
2765 *newloc = xstrdup (hstat.newloc);
2769 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2770 case CONERROR: case READERR: case WRITEFAILED:
2771 case RANGEERR: case FOPEN_EXCL_ERR:
2772 /* Non-fatal errors continue executing the loop, which will
2773 bring them to "while" statement at the end, to judge
2774 whether the number of tries was exceeded. */
2775 printwhat (count, opt.ntry);
2777 case FWRITEERR: case FOPENERR:
2778 /* Another fatal error. */
2779 logputs (LOG_VERBOSE, "\n");
2780 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2781 quote (hstat.local_file), strerror (errno));
2782 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2783 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2784 /* Fatal errors just return from the function. */
2788 /* Another fatal error. */
2789 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2793 /* Another fatal error. */
2794 logputs (LOG_VERBOSE, "\n");
2795 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"),
2796 quote (hstat.local_file), strerror (errno));
2800 /* Return the new location to the caller. */
2803 logprintf (LOG_NOTQUIET,
2804 _("ERROR: Redirection (%d) without location.\n"),
2814 /* The file was already fully retrieved. */
2818 /* Deal with you later. */
2821 /* All possibilities should have been exhausted. */
2825 if (!(*dt & RETROKF))
2830 /* #### Ugly ugly ugly! */
2831 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2832 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2835 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2837 && (hstat.statcode == 500 || hstat.statcode == 501))
2842 /* Maybe we should always keep track of broken links, not just in
2844 * Don't log error if it was UTF-8 encoded because we will try
2845 * once unencoded. */
2846 else if (opt.spider && !iri->utf8_encode)
2848 /* #### Again: ugly ugly ugly! */
2850 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2851 nonexisting_url (hurl);
2852 logprintf (LOG_NOTQUIET, _("\
2853 Remote file does not exist -- broken link!!!\n"));
2857 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2858 tms, hstat.statcode,
2859 quotearg_style (escape_quoting_style, hstat.error));
2861 logputs (LOG_VERBOSE, "\n");
2867 /* Did we get the time-stamp? */
2870 got_head = true; /* no more time-stamping */
2872 if (opt.timestamping && !hstat.remote_time)
2874 logputs (LOG_NOTQUIET, _("\
2875 Last-modified header missing -- time-stamps turned off.\n"));
2877 else if (hstat.remote_time)
2879 /* Convert the date-string into struct tm. */
2880 tmr = http_atotm (hstat.remote_time);
2881 if (tmr == (time_t) (-1))
2882 logputs (LOG_VERBOSE, _("\
2883 Last-modified header invalid -- time-stamp ignored.\n"));
2884 if (*dt & HEAD_ONLY)
2885 time_came_from_head = true;
2888 if (send_head_first)
2890 /* The time-stamping section. */
2891 if (opt.timestamping)
2893 if (hstat.orig_file_name) /* Perform the following
2894 checks only if the file
2896 download already exists. */
2898 if (hstat.remote_time &&
2899 tmr != (time_t) (-1))
2901 /* Now time-stamping can be used validly.
2902 Time-stamping means that if the sizes of
2903 the local and remote file match, and local
2904 file is newer than the remote file, it will
2905 not be retrieved. Otherwise, the normal
2906 download procedure is resumed. */
2907 if (hstat.orig_file_tstamp >= tmr)
2909 if (hstat.contlen == -1
2910 || hstat.orig_file_size == hstat.contlen)
2912 logprintf (LOG_VERBOSE, _("\
2913 Server file no newer than local file %s -- not retrieving.\n\n"),
2914 quote (hstat.orig_file_name));
2920 logprintf (LOG_VERBOSE, _("\
2921 The sizes do not match (local %s) -- retrieving.\n"),
2922 number_to_static_string (hstat.orig_file_size));
2927 force_full_retrieve = true;
2928 logputs (LOG_VERBOSE,
2929 _("Remote file is newer, retrieving.\n"));
2932 logputs (LOG_VERBOSE, "\n");
2936 /* free_hstat (&hstat); */
2937 hstat.timestamp_checked = true;
2942 bool finished = true;
2947 logputs (LOG_VERBOSE, _("\
2948 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2953 logprintf (LOG_VERBOSE, _("\
2954 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2955 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2962 logprintf (LOG_VERBOSE, _("\
2963 Remote file exists and could contain further links,\n\
2964 but recursion is disabled -- not retrieving.\n\n"));
2968 logprintf (LOG_VERBOSE, _("\
2969 Remote file exists.\n\n"));
2971 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2976 logprintf (LOG_NONVERBOSE,
2977 _("%s URL: %s %2d %s\n"),
2978 tms, u->url, hstat.statcode,
2979 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2986 count = 0; /* the retrieve count for HEAD is reset */
2988 } /* send_head_first */
2991 if (opt.useservertimestamps
2992 && (tmr != (time_t) (-1))
2993 && ((hstat.len == hstat.contlen) ||
2994 ((hstat.res == 0) && (hstat.contlen == -1))))
2996 const char *fl = NULL;
2997 set_local_file (&fl, hstat.local_file);
3001 /* Reparse time header, in case it's changed. */
3002 if (time_came_from_head
3003 && hstat.remote_time && hstat.remote_time[0])
3005 newtmr = http_atotm (hstat.remote_time);
3006 if (newtmr != (time_t)-1)
3012 /* End of time-stamping section. */
3014 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3015 total_download_time += hstat.dltime;
3017 if (hstat.len == hstat.contlen)
3021 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3023 logprintf (LOG_VERBOSE,
3025 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3026 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3028 write_to_stdout ? "" : quote (hstat.local_file),
3029 number_to_static_string (hstat.len),
3030 number_to_static_string (hstat.contlen));
3031 logprintf (LOG_NONVERBOSE,
3032 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3034 number_to_static_string (hstat.len),
3035 number_to_static_string (hstat.contlen),
3036 hstat.local_file, count);
3039 total_downloaded_bytes += hstat.rd_size;
3041 /* Remember that we downloaded the file for later ".orig" code. */
3042 if (*dt & ADDED_HTML_EXTENSION)
3043 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3045 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3050 else if (hstat.res == 0) /* No read error */
3052 if (hstat.contlen == -1) /* We don't know how much we were supposed
3053 to get, so assume we succeeded. */
3057 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3059 logprintf (LOG_VERBOSE,
3061 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3062 : _("%s (%s) - %s saved [%s]\n\n"),
3064 write_to_stdout ? "" : quote (hstat.local_file),
3065 number_to_static_string (hstat.len));
3066 logprintf (LOG_NONVERBOSE,
3067 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3068 tms, u->url, number_to_static_string (hstat.len),
3069 hstat.local_file, count);
3072 total_downloaded_bytes += hstat.rd_size;
3074 /* Remember that we downloaded the file for later ".orig" code. */
3075 if (*dt & ADDED_HTML_EXTENSION)
3076 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3078 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3083 else if (hstat.len < hstat.contlen) /* meaning we lost the
3084 connection too soon */
3086 logprintf (LOG_VERBOSE,
3087 _("%s (%s) - Connection closed at byte %s. "),
3088 tms, tmrate, number_to_static_string (hstat.len));
3089 printwhat (count, opt.ntry);
3092 else if (hstat.len != hstat.restval)
3093 /* Getting here would mean reading more data than
3094 requested with content-length, which we never do. */
3098 /* Getting here probably means that the content-length was
3099 * _less_ than the original, local size. We should probably
3100 * truncate or re-read, or something. FIXME */
3105 else /* from now on hstat.res can only be -1 */
3107 if (hstat.contlen == -1)
3109 logprintf (LOG_VERBOSE,
3110 _("%s (%s) - Read error at byte %s (%s)."),
3111 tms, tmrate, number_to_static_string (hstat.len),
3113 printwhat (count, opt.ntry);
3116 else /* hstat.res == -1 and contlen is given */
3118 logprintf (LOG_VERBOSE,
3119 _("%s (%s) - Read error at byte %s/%s (%s). "),
3121 number_to_static_string (hstat.len),
3122 number_to_static_string (hstat.contlen),
3124 printwhat (count, opt.ntry);
3130 while (!opt.ntry || (count < opt.ntry));
3133 if (ret == RETROK && local_file)
3134 *local_file = xstrdup (hstat.local_file);
3135 free_hstat (&hstat);
3140 /* Check whether the result of strptime() indicates success.
3141 strptime() returns the pointer to how far it got to in the string.
3142 The processing has been successful if the string is at `GMT' or
3143 `+X', or at the end of the string.
3145 In extended regexp parlance, the function returns 1 if P matches
3146 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3147 can return) is considered a failure and 0 is returned. */
3149 check_end (const char *p)
3153 while (c_isspace (*p))
3156 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3157 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3163 /* Convert the textual specification of time in TIME_STRING to the
3164 number of seconds since the Epoch.
3166 TIME_STRING can be in any of the three formats RFC2616 allows the
3167 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3168 as well as the time format used in the Set-Cookie header.
3169 Timezones are ignored, and should be GMT.
3171 Return the computed time_t representation, or -1 if the conversion
3174 This function uses strptime with various string formats for parsing
3175 TIME_STRING. This results in a parser that is not as lenient in
3176 interpreting TIME_STRING as I would like it to be. Being based on
3177 strptime, it always allows shortened months, one-digit days, etc.,
3178 but due to the multitude of formats in which time can be
3179 represented, an ideal HTTP time parser would be even more
3180 forgiving. It should completely ignore things like week days and
3181 concentrate only on the various forms of representing years,
3182 months, days, hours, minutes, and seconds. For example, it would
3183 be nice if it accepted ISO 8601 out of the box.
3185 I've investigated free and PD code for this purpose, but none was
3186 usable. getdate was big and unwieldy, and had potential copyright
3187 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3188 distributed with phttpd, is excellent, but we cannot use it because
3189 it is not assigned to the FSF. So I stuck it with strptime. */
3192 http_atotm (const char *time_string)
3194 /* NOTE: Solaris strptime man page claims that %n and %t match white
3195 space, but that's not universally available. Instead, we simply
3196 use ` ' to mean "skip all WS", which works under all strptime
3197 implementations I've tested. */
3199 static const char *time_formats[] = {
3200 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3201 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3202 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3203 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3204 (used in Set-Cookie, defined in the
3205 Netscape cookie specification.) */
3207 const char *oldlocale;
3208 char savedlocale[256];
3210 time_t ret = (time_t) -1;
3212 /* Solaris strptime fails to recognize English month names in
3213 non-English locales, which we work around by temporarily setting
3214 locale to C before invoking strptime. */
3215 oldlocale = setlocale (LC_TIME, NULL);
3218 size_t l = strlen (oldlocale) + 1;
3219 if (l >= sizeof savedlocale)
3220 savedlocale[0] = '\0';
3222 memcpy (savedlocale, oldlocale, l);
3224 else savedlocale[0] = '\0';
3226 setlocale (LC_TIME, "C");
3228 for (i = 0; i < countof (time_formats); i++)
3232 /* Some versions of strptime use the existing contents of struct
3233 tm to recalculate the date according to format. Zero it out
3234 to prevent stack garbage from influencing strptime. */
3237 if (check_end (strptime (time_string, time_formats[i], &t)))
3244 /* Restore the previous locale. */
3246 setlocale (LC_TIME, savedlocale);
3251 /* Authorization support: We support three authorization schemes:
3253 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3255 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3256 consisting of answering to the server's challenge with the proper
3259 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3260 Stenberg for libcurl. Like digest, NTLM is based on a
3261 challenge-response mechanism, but unlike digest, it is non-standard
3262 (authenticates TCP connections rather than requests), undocumented
3263 and Microsoft-specific. */
3265 /* Create the authentication header contents for the `Basic' scheme.
3266 This is done by encoding the string "USER:PASS" to base64 and
3267 prepending the string "Basic " in front of it. */
3270 basic_authentication_encode (const char *user, const char *passwd)
3273 int len1 = strlen (user) + 1 + strlen (passwd);
3275 t1 = (char *)alloca (len1 + 1);
3276 sprintf (t1, "%s:%s", user, passwd);
3278 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3279 base64_encode (t1, len1, t2);
3281 return concat_strings ("Basic ", t2, (char *) 0);
3284 #define SKIP_WS(x) do { \
3285 while (c_isspace (*(x))) \
3289 #ifdef ENABLE_DIGEST
3290 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3291 an array of 16 bytes containing the hash keys, and BUF should be a
3292 buffer of 33 writable characters (32 for hex digits plus one for
3293 zero termination). */
3295 dump_hash (char *buf, const unsigned char *hash)
3299 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3301 *buf++ = XNUM_TO_digit (*hash >> 4);
3302 *buf++ = XNUM_TO_digit (*hash & 0xf);
3307 /* Take the line apart to find the challenge, and compose a digest
3308 authorization header. See RFC2069 section 2.1.2. */
3310 digest_authentication_encode (const char *au, const char *user,
3311 const char *passwd, const char *method,
3314 static char *realm, *opaque, *nonce;
3319 { "realm", &realm },
3320 { "opaque", &opaque },
3324 param_token name, value;
3326 realm = opaque = nonce = NULL;
3328 au += 6; /* skip over `Digest' */
3329 while (extract_param (&au, &name, &value, ','))
3332 size_t namelen = name.e - name.b;
3333 for (i = 0; i < countof (options); i++)
3334 if (namelen == strlen (options[i].name)
3335 && 0 == strncmp (name.b, options[i].name,
3338 *options[i].variable = strdupdelim (value.b, value.e);
3342 if (!realm || !nonce || !user || !passwd || !path || !method)
3345 xfree_null (opaque);
3350 /* Calculate the digest value. */
3353 unsigned char hash[MD5_DIGEST_SIZE];
3354 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3355 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3357 /* A1BUF = H(user ":" realm ":" password) */
3358 md5_init_ctx (&ctx);
3359 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3360 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3361 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3362 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3363 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3364 md5_finish_ctx (&ctx, hash);
3365 dump_hash (a1buf, hash);
3367 /* A2BUF = H(method ":" path) */
3368 md5_init_ctx (&ctx);
3369 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3370 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3371 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3372 md5_finish_ctx (&ctx, hash);
3373 dump_hash (a2buf, hash);
3375 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3376 md5_init_ctx (&ctx);
3377 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3378 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3379 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3380 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3381 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3382 md5_finish_ctx (&ctx, hash);
3383 dump_hash (response_digest, hash);
3385 res = xmalloc (strlen (user)
3390 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3391 + (opaque ? strlen (opaque) : 0)
3393 sprintf (res, "Digest \
3394 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3395 user, realm, nonce, path, response_digest);
3398 char *p = res + strlen (res);
3399 strcat (p, ", opaque=\"");
3406 #endif /* ENABLE_DIGEST */
3408 /* Computing the size of a string literal must take into account that
3409 value returned by sizeof includes the terminating \0. */
3410 #define STRSIZE(literal) (sizeof (literal) - 1)
3412 /* Whether chars in [b, e) begin with the literal string provided as
3413 first argument and are followed by whitespace or terminating \0.
3414 The comparison is case-insensitive. */
3415 #define STARTS(literal, b, e) \
3417 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3418 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3419 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3420 || c_isspace (b[STRSIZE (literal)])))
3423 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3425 return STARTS ("Basic", hdrbeg, hdrend)
3426 #ifdef ENABLE_DIGEST
3427 || STARTS ("Digest", hdrbeg, hdrend)
3430 || STARTS ("NTLM", hdrbeg, hdrend)
3437 /* Create the HTTP authorization request header. When the
3438 `WWW-Authenticate' response header is seen, according to the
3439 authorization scheme specified in that header (`Basic' and `Digest'
3440 are supported by the current implementation), produce an
3441 appropriate HTTP authorization request header. */
3443 create_authorization_line (const char *au, const char *user,
3444 const char *passwd, const char *method,
3445 const char *path, bool *finished)
3447 /* We are called only with known schemes, so we can dispatch on the
3449 switch (c_toupper (*au))
3451 case 'B': /* Basic */
3453 return basic_authentication_encode (user, passwd);
3454 #ifdef ENABLE_DIGEST
3455 case 'D': /* Digest */
3457 return digest_authentication_encode (au, user, passwd, method, path);
3460 case 'N': /* NTLM */
3461 if (!ntlm_input (&pconn.ntlm, au))
3466 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3469 /* We shouldn't get here -- this function should be only called
3470 with values approved by known_authentication_scheme_p. */
3478 if (!wget_cookie_jar)
3479 wget_cookie_jar = cookie_jar_new ();
3480 if (opt.cookies_input && !cookies_loaded_p)
3482 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3483 cookies_loaded_p = true;
3490 if (wget_cookie_jar)
3491 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3497 xfree_null (pconn.host);
3498 if (wget_cookie_jar)
3499 cookie_jar_delete (wget_cookie_jar);
3503 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3505 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3507 int len = strlen (ext);
3510 strncpy (shortext, ext, len - 1);
3511 shortext[len - 2] = '\0';
3514 if (last_period_in_local_filename == NULL
3515 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3516 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3518 int local_filename_len = strlen (hs->local_file);
3519 /* Resize the local file, allowing for ".html" preceded by
3520 optional ".NUMBER". */
3521 hs->local_file = xrealloc (hs->local_file,
3522 local_filename_len + 24 + len);
3523 strcpy (hs->local_file + local_filename_len, ext);
3524 /* If clobbering is not allowed and the file, as named,
3525 exists, tack on ".NUMBER.html" instead. */
3526 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3530 sprintf (hs->local_file + local_filename_len,
3531 ".%d%s", ext_num++, ext);
3532 while (file_exists_p (hs->local_file));
3534 *dt |= ADDED_HTML_EXTENSION;
3542 test_parse_content_disposition()
3550 { "filename=\"file.ext\"", "file.ext", true },
3551 { "attachment; filename=\"file.ext\"", "file.ext", true },
3552 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
3553 { "attachment", NULL, false },
3554 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true },
3555 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true },
3558 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3563 res = parse_content_disposition (test_array[i].hdrval, &filename);
3565 mu_assert ("test_parse_content_disposition: wrong result",
3566 res == test_array[i].result
3568 || 0 == strcmp (test_array[i].filename, filename)));
3574 #endif /* TESTING */
3577 * vim: et sts=2 sw=2 cino+={s