2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008, 2009 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"
71 #endif /* def __VMS */
73 extern char *version_string;
77 static char *create_authorization_line (const char *, const char *,
78 const char *, const char *,
79 const char *, bool *);
80 static char *basic_authentication_encode (const char *, const char *);
81 static bool known_authentication_scheme_p (const char *, const char *);
82 static void ensure_extension (struct http_stat *, const char *, int *);
83 static void load_cookies (void);
86 # define MIN(x, y) ((x) > (y) ? (y) : (x))
90 static bool cookies_loaded_p;
91 static struct cookie_jar *wget_cookie_jar;
93 #define TEXTHTML_S "text/html"
94 #define TEXTXHTML_S "application/xhtml+xml"
95 #define TEXTCSS_S "text/css"
97 /* Some status code validation macros: */
98 #define H_20X(x) (((x) >= 200) && ((x) < 300))
99 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
100 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
101 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
102 || (x) == HTTP_STATUS_SEE_OTHER \
103 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
105 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
106 /* Successful 2xx. */
107 #define HTTP_STATUS_OK 200
108 #define HTTP_STATUS_CREATED 201
109 #define HTTP_STATUS_ACCEPTED 202
110 #define HTTP_STATUS_NO_CONTENT 204
111 #define HTTP_STATUS_PARTIAL_CONTENTS 206
113 /* Redirection 3xx. */
114 #define HTTP_STATUS_MULTIPLE_CHOICES 300
115 #define HTTP_STATUS_MOVED_PERMANENTLY 301
116 #define HTTP_STATUS_MOVED_TEMPORARILY 302
117 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
118 #define HTTP_STATUS_NOT_MODIFIED 304
119 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
121 /* Client error 4xx. */
122 #define HTTP_STATUS_BAD_REQUEST 400
123 #define HTTP_STATUS_UNAUTHORIZED 401
124 #define HTTP_STATUS_FORBIDDEN 403
125 #define HTTP_STATUS_NOT_FOUND 404
126 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
128 /* Server errors 5xx. */
129 #define HTTP_STATUS_INTERNAL 500
130 #define HTTP_STATUS_NOT_IMPLEMENTED 501
131 #define HTTP_STATUS_BAD_GATEWAY 502
132 #define HTTP_STATUS_UNAVAILABLE 503
135 rel_none, rel_name, rel_value, rel_both
142 struct request_header {
144 enum rp release_policy;
146 int hcount, hcapacity;
151 /* Create a new, empty request. At least request_set_method must be
152 called before the request can be used. */
154 static struct request *
157 struct request *req = xnew0 (struct request);
159 req->headers = xnew_array (struct request_header, req->hcapacity);
163 /* Set the request's method and its arguments. METH should be a
164 literal string (or it should outlive the request) because it will
165 not be freed. ARG will be freed by request_free. */
168 request_set_method (struct request *req, const char *meth, char *arg)
174 /* Return the method string passed with the last call to
175 request_set_method. */
178 request_method (const struct request *req)
183 /* Free one header according to the release policy specified with
184 request_set_header. */
187 release_header (struct request_header *hdr)
189 switch (hdr->release_policy)
206 /* Set the request named NAME to VALUE. Specifically, this means that
207 a "NAME: VALUE\r\n" header line will be used in the request. If a
208 header with the same name previously existed in the request, its
209 value will be replaced by this one. A NULL value means do nothing.
211 RELEASE_POLICY determines whether NAME and VALUE should be released
212 (freed) with request_free. Allowed values are:
214 - rel_none - don't free NAME or VALUE
215 - rel_name - free NAME when done
216 - rel_value - free VALUE when done
217 - rel_both - free both NAME and VALUE when done
219 Setting release policy is useful when arguments come from different
220 sources. For example:
222 // Don't free literal strings!
223 request_set_header (req, "Pragma", "no-cache", rel_none);
225 // Don't free a global variable, we'll need it later.
226 request_set_header (req, "Referer", opt.referer, rel_none);
228 // Value freshly allocated, free it when done.
229 request_set_header (req, "Range",
230 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
235 request_set_header (struct request *req, char *name, char *value,
236 enum rp release_policy)
238 struct request_header *hdr;
243 /* A NULL value is a no-op; if freeing the name is requested,
244 free it now to avoid leaks. */
245 if (release_policy == rel_name || release_policy == rel_both)
250 for (i = 0; i < req->hcount; i++)
252 hdr = &req->headers[i];
253 if (0 == strcasecmp (name, hdr->name))
255 /* Replace existing header. */
256 release_header (hdr);
259 hdr->release_policy = release_policy;
264 /* Install new header. */
266 if (req->hcount >= req->hcapacity)
268 req->hcapacity <<= 1;
269 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
271 hdr = &req->headers[req->hcount++];
274 hdr->release_policy = release_policy;
277 /* Like request_set_header, but sets the whole header line, as
278 provided by the user using the `--header' option. For example,
279 request_set_user_header (req, "Foo: bar") works just like
280 request_set_header (req, "Foo", "bar"). */
283 request_set_user_header (struct request *req, const char *header)
286 const char *p = strchr (header, ':');
289 BOUNDED_TO_ALLOCA (header, p, name);
291 while (c_isspace (*p))
293 request_set_header (req, xstrdup (name), (char *) p, rel_name);
296 /* Remove the header with specified name from REQ. Returns true if
297 the header was actually removed, false otherwise. */
300 request_remove_header (struct request *req, char *name)
303 for (i = 0; i < req->hcount; i++)
305 struct request_header *hdr = &req->headers[i];
306 if (0 == strcasecmp (name, hdr->name))
308 release_header (hdr);
309 /* Move the remaining headers by one. */
310 if (i < req->hcount - 1)
311 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
319 #define APPEND(p, str) do { \
320 int A_len = strlen (str); \
321 memcpy (p, str, A_len); \
325 /* Construct the request and write it to FD using fd_write. */
328 request_send (const struct request *req, int fd)
330 char *request_string, *p;
331 int i, size, write_error;
333 /* Count the request size. */
336 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
337 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
339 for (i = 0; i < req->hcount; i++)
341 struct request_header *hdr = &req->headers[i];
342 /* NAME ": " VALUE "\r\n" */
343 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
349 p = request_string = alloca_array (char, size);
351 /* Generate the request. */
353 APPEND (p, req->method); *p++ = ' ';
354 APPEND (p, req->arg); *p++ = ' ';
355 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
357 for (i = 0; i < req->hcount; i++)
359 struct request_header *hdr = &req->headers[i];
360 APPEND (p, hdr->name);
361 *p++ = ':', *p++ = ' ';
362 APPEND (p, hdr->value);
363 *p++ = '\r', *p++ = '\n';
366 *p++ = '\r', *p++ = '\n', *p++ = '\0';
367 assert (p - request_string == size);
371 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
373 /* Send the request to the server. */
375 write_error = fd_write (fd, request_string, size - 1, -1);
377 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
382 /* Release the resources used by REQ. */
385 request_free (struct request *req)
388 xfree_null (req->arg);
389 for (i = 0; i < req->hcount; i++)
390 release_header (&req->headers[i]);
391 xfree_null (req->headers);
395 static struct hash_table *basic_authed_hosts;
397 /* Find out if this host has issued a Basic challenge yet; if so, give
398 * it the username, password. A temporary measure until we can get
399 * proper authentication in place. */
402 maybe_send_basic_creds (const char *hostname, const char *user,
403 const char *passwd, struct request *req)
405 bool do_challenge = false;
407 if (opt.auth_without_challenge)
409 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
412 else if (basic_authed_hosts
413 && hash_table_contains(basic_authed_hosts, hostname))
415 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
420 DEBUGP(("Host %s has not issued a general basic challenge.\n",
425 request_set_header (req, "Authorization",
426 basic_authentication_encode (user, passwd),
433 register_basic_auth_host (const char *hostname)
435 if (!basic_authed_hosts)
437 basic_authed_hosts = make_nocase_string_hash_table (1);
439 if (!hash_table_contains(basic_authed_hosts, hostname))
441 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
442 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
447 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
448 PROMISED_SIZE bytes are sent over the wire -- if the file is
449 longer, read only that much; if the file is shorter, report an error. */
452 post_file (int sock, const char *file_name, wgint promised_size)
454 static char chunk[8192];
459 DEBUGP (("[writing POST file %s ... ", file_name));
461 fp = fopen (file_name, "rb");
464 while (!feof (fp) && written < promised_size)
467 int length = fread (chunk, 1, sizeof (chunk), fp);
470 towrite = MIN (promised_size - written, length);
471 write_error = fd_write (sock, chunk, towrite, -1);
481 /* If we've written less than was promised, report a (probably
482 nonsensical) error rather than break the promise. */
483 if (written < promised_size)
489 assert (written == promised_size);
490 DEBUGP (("done]\n"));
494 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
495 If so, return the pointer to the position after the line, otherwise
496 return NULL. This is used as callback to fd_read_hunk. The data
497 between START and PEEKED has been read and cannot be "unread"; the
498 data after PEEKED has only been peeked. */
501 response_head_terminator (const char *start, const char *peeked, int peeklen)
505 /* If at first peek, verify whether HUNK starts with "HTTP". If
506 not, this is a HTTP/0.9 request and we must bail out without
508 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
511 /* Look for "\n[\r]\n", and return the following position if found.
512 Start two chars before the current to cover the possibility that
513 part of the terminator (e.g. "\n\r") arrived in the previous
515 p = peeked - start < 2 ? start : peeked - 2;
516 end = peeked + peeklen;
518 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
519 for (; p < end - 2; p++)
522 if (p[1] == '\r' && p[2] == '\n')
524 else if (p[1] == '\n')
527 /* p==end-2: check for \n\n directly preceding END. */
528 if (p[0] == '\n' && p[1] == '\n')
534 /* The maximum size of a single HTTP response we care to read. Rather
535 than being a limit of the reader implementation, this limit
536 prevents Wget from slurping all available memory upon encountering
537 malicious or buggy server output, thus protecting the user. Define
538 it to 0 to remove the limit. */
540 #define HTTP_RESPONSE_MAX_SIZE 65536
542 /* Read the HTTP request head from FD and return it. The error
543 conditions are the same as with fd_read_hunk.
545 To support HTTP/0.9 responses, this function tries to make sure
546 that the data begins with "HTTP". If this is not the case, no data
547 is read and an empty request is returned, so that the remaining
548 data can be treated as body. */
551 read_http_response_head (int fd)
553 return fd_read_hunk (fd, response_head_terminator, 512,
554 HTTP_RESPONSE_MAX_SIZE);
558 /* The response data. */
561 /* The array of pointers that indicate where each header starts.
562 For example, given this HTTP response:
569 The headers are located like this:
571 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
573 headers[0] headers[1] headers[2] headers[3]
575 I.e. headers[0] points to the beginning of the request,
576 headers[1] points to the end of the first header and the
577 beginning of the second one, etc. */
579 const char **headers;
582 /* Create a new response object from the text of the HTTP response,
583 available in HEAD. That text is automatically split into
584 constituent header lines for fast retrieval using
587 static struct response *
588 resp_new (const char *head)
593 struct response *resp = xnew0 (struct response);
598 /* Empty head means that we're dealing with a headerless
599 (HTTP/0.9) response. In that case, don't set HEADERS at
604 /* Split HEAD into header lines, so that resp_header_* functions
605 don't need to do this over and over again. */
611 DO_REALLOC (resp->headers, size, count + 1, const char *);
612 resp->headers[count++] = hdr;
614 /* Break upon encountering an empty line. */
615 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
618 /* Find the end of HDR, including continuations. */
621 const char *end = strchr (hdr, '\n');
627 while (*hdr == ' ' || *hdr == '\t');
629 DO_REALLOC (resp->headers, size, count + 1, const char *);
630 resp->headers[count] = NULL;
635 /* Locate the header named NAME in the request data, starting with
636 position START. This allows the code to loop through the request
637 data, filtering for all requests of a given name. Returns the
638 found position, or -1 for failure. The code that uses this
639 function typically looks like this:
641 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
642 ... do something with header ...
644 If you only care about one header, use resp_header_get instead of
648 resp_header_locate (const struct response *resp, const char *name, int start,
649 const char **begptr, const char **endptr)
652 const char **headers = resp->headers;
655 if (!headers || !headers[1])
658 name_len = strlen (name);
664 for (; headers[i + 1]; i++)
666 const char *b = headers[i];
667 const char *e = headers[i + 1];
669 && b[name_len] == ':'
670 && 0 == strncasecmp (b, name, name_len))
673 while (b < e && c_isspace (*b))
675 while (b < e && c_isspace (e[-1]))
685 /* Find and retrieve the header named NAME in the request data. If
686 found, set *BEGPTR to its starting, and *ENDPTR to its ending
687 position, and return true. Otherwise return false.
689 This function is used as a building block for resp_header_copy
690 and resp_header_strdup. */
693 resp_header_get (const struct response *resp, const char *name,
694 const char **begptr, const char **endptr)
696 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
700 /* Copy the response header named NAME to buffer BUF, no longer than
701 BUFSIZE (BUFSIZE includes the terminating 0). If the header
702 exists, true is returned, false otherwise. If there should be no
703 limit on the size of the header, use resp_header_strdup instead.
705 If BUFSIZE is 0, no data is copied, but the boolean indication of
706 whether the header is present is still returned. */
709 resp_header_copy (const struct response *resp, const char *name,
710 char *buf, int bufsize)
713 if (!resp_header_get (resp, name, &b, &e))
717 int len = MIN (e - b, bufsize - 1);
718 memcpy (buf, b, len);
724 /* Return the value of header named NAME in RESP, allocated with
725 malloc. If such a header does not exist in RESP, return NULL. */
728 resp_header_strdup (const struct response *resp, const char *name)
731 if (!resp_header_get (resp, name, &b, &e))
733 return strdupdelim (b, e);
736 /* Parse the HTTP status line, which is of format:
738 HTTP-Version SP Status-Code SP Reason-Phrase
740 The function returns the status-code, or -1 if the status line
741 appears malformed. The pointer to "reason-phrase" message is
742 returned in *MESSAGE. */
745 resp_status (const struct response *resp, char **message)
752 /* For a HTTP/0.9 response, assume status 200. */
754 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
758 p = resp->headers[0];
759 end = resp->headers[1];
765 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
769 /* Match the HTTP version. This is optional because Gnutella
770 servers have been reported to not specify HTTP version. */
771 if (p < end && *p == '/')
774 while (p < end && c_isdigit (*p))
776 if (p < end && *p == '.')
778 while (p < end && c_isdigit (*p))
782 while (p < end && c_isspace (*p))
784 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
787 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
792 while (p < end && c_isspace (*p))
794 while (p < end && c_isspace (end[-1]))
796 *message = strdupdelim (p, end);
802 /* Release the resources used by RESP. */
805 resp_free (struct response *resp)
807 xfree_null (resp->headers);
811 /* Print a single line of response, the characters [b, e). We tried
813 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
814 but that failed to escape the non-printable characters and, in fact,
815 caused crashes in UTF-8 locales. */
818 print_response_line(const char *prefix, const char *b, const char *e)
821 BOUNDED_TO_ALLOCA(b, e, copy);
822 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
823 quotearg_style (escape_quoting_style, copy));
826 /* Print the server response, line by line, omitting the trailing CRLF
827 from individual header lines, and prefixed with PREFIX. */
830 print_server_response (const struct response *resp, const char *prefix)
835 for (i = 0; resp->headers[i + 1]; i++)
837 const char *b = resp->headers[i];
838 const char *e = resp->headers[i + 1];
840 if (b < e && e[-1] == '\n')
842 if (b < e && e[-1] == '\r')
844 print_response_line(prefix, b, e);
848 /* Parse the `Content-Range' header and extract the information it
849 contains. Returns true if successful, false otherwise. */
851 parse_content_range (const char *hdr, wgint *first_byte_ptr,
852 wgint *last_byte_ptr, wgint *entity_length_ptr)
856 /* Ancient versions of Netscape proxy server, presumably predating
857 rfc2068, sent out `Content-Range' without the "bytes"
859 if (0 == strncasecmp (hdr, "bytes", 5))
862 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
866 while (c_isspace (*hdr))
871 if (!c_isdigit (*hdr))
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
877 *first_byte_ptr = num;
879 for (num = 0; c_isdigit (*hdr); hdr++)
880 num = 10 * num + (*hdr - '0');
881 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
883 *last_byte_ptr = num;
888 for (num = 0; c_isdigit (*hdr); hdr++)
889 num = 10 * num + (*hdr - '0');
890 *entity_length_ptr = num;
894 /* Read the body of the request, but don't store it anywhere and don't
895 display a progress gauge. This is useful for reading the bodies of
896 administrative responses to which we will soon issue another
897 request. The response is not useful to the user, but reading it
898 allows us to continue using the same connection to the server.
900 If reading fails, false is returned, true otherwise. In debug
901 mode, the body is displayed for debugging purposes. */
904 skip_short_body (int fd, wgint contlen, bool chunked)
907 SKIP_SIZE = 512, /* size of the download buffer */
908 SKIP_THRESHOLD = 4096 /* the largest size we read */
910 wgint remaining_chunk_size = 0;
911 char dlbuf[SKIP_SIZE + 1];
912 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
914 assert (contlen != -1 || contlen);
916 /* If the body is too large, it makes more sense to simply close the
917 connection than to try to read the body. */
918 if (contlen > SKIP_THRESHOLD)
921 while (contlen > 0 || chunked)
926 if (remaining_chunk_size == 0)
928 char *line = fd_read_line (fd);
936 remaining_chunk_size = strtol (line, &endl, 16);
937 if (remaining_chunk_size == 0)
940 if (fd_read_line (fd) == NULL)
946 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
949 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
951 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
954 /* Don't normally report the error since this is an
955 optimization that should be invisible to the user. */
956 DEBUGP (("] aborting (%s).\n",
957 ret < 0 ? fd_errstr (fd) : "EOF received"));
964 remaining_chunk_size -= ret;
965 if (remaining_chunk_size == 0)
966 if (fd_read_line (fd) == NULL)
970 /* Safe even if %.*s bogusly expects terminating \0 because
971 we've zero-terminated dlbuf above. */
972 DEBUGP (("%.*s", ret, dlbuf));
975 DEBUGP (("] done.\n"));
979 #define NOT_RFC2231 0
980 #define RFC2231_NOENCODING 1
981 #define RFC2231_ENCODING 2
983 /* extract_param extracts the parameter name into NAME.
984 However, if the parameter name is in RFC2231 format then
985 this function adjusts NAME by stripping of the trailing
986 characters that are not part of the name but are present to
987 indicate the presence of encoding information in the value
988 or a fragment of a long parameter value
991 modify_param_name(param_token *name)
993 const char *delim1 = memchr (name->b, '*', name->e - name->b);
994 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
1000 result = NOT_RFC2231;
1002 else if(delim1 == delim2)
1004 if ((name->e - 1) == delim1)
1006 result = RFC2231_ENCODING;
1010 result = RFC2231_NOENCODING;
1017 result = RFC2231_ENCODING;
1022 /* extract_param extract the paramater value into VALUE.
1023 Like modify_param_name this function modifies VALUE by
1024 stripping off the encoding information from the actual value
1027 modify_param_value (param_token *value, int encoding_type )
1029 if (RFC2231_ENCODING == encoding_type)
1031 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1032 if ( delim != NULL )
1034 value->b = (delim+1);
1039 /* Extract a parameter from the string (typically an HTTP header) at
1040 **SOURCE and advance SOURCE to the next parameter. Return false
1041 when there are no more parameters to extract. The name of the
1042 parameter is returned in NAME, and the value in VALUE. If the
1043 parameter has no value, the token's value is zeroed out.
1045 For example, if *SOURCE points to the string "attachment;
1046 filename=\"foo bar\"", the first call to this function will return
1047 the token named "attachment" and no value, and the second call will
1048 return the token named "filename" and value "foo bar". The third
1049 call will return false, indicating no more valid tokens. */
1052 extract_param (const char **source, param_token *name, param_token *value,
1055 const char *p = *source;
1057 while (c_isspace (*p)) ++p;
1061 return false; /* no error; nothing more to extract */
1066 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1068 if (name->b == name->e)
1069 return false; /* empty name: error */
1070 while (c_isspace (*p)) ++p;
1071 if (*p == separator || !*p) /* no value */
1074 if (*p == separator) ++p;
1079 return false; /* error */
1081 /* *p is '=', extract value */
1083 while (c_isspace (*p)) ++p;
1084 if (*p == '"') /* quoted */
1087 while (*p && *p != '"') ++p;
1091 /* Currently at closing quote; find the end of param. */
1092 while (c_isspace (*p)) ++p;
1093 while (*p && *p != separator) ++p;
1094 if (*p == separator)
1097 /* garbage after closed quote, e.g. foo="bar"baz */
1103 while (*p && *p != separator) ++p;
1105 while (value->e != value->b && c_isspace (value->e[-1]))
1107 if (*p == separator) ++p;
1111 int param_type = modify_param_name(name);
1112 if (NOT_RFC2231 != param_type)
1114 modify_param_value(value, param_type);
1120 #undef RFC2231_NOENCODING
1121 #undef RFC2231_ENCODING
1123 /* Appends the string represented by VALUE to FILENAME */
1126 append_value_to_filename (char **filename, param_token const * const value)
1128 int original_length = strlen(*filename);
1129 int new_length = strlen(*filename) + (value->e - value->b);
1130 *filename = xrealloc (*filename, new_length+1);
1131 memcpy (*filename + original_length, value->b, (value->e - value->b));
1132 (*filename)[new_length] = '\0';
1136 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1138 /* Parse the contents of the `Content-Disposition' header, extracting
1139 the information useful to Wget. Content-Disposition is a header
1140 borrowed from MIME; when used in HTTP, it typically serves for
1141 specifying the desired file name of the resource. For example:
1143 Content-Disposition: attachment; filename="flora.jpg"
1145 Wget will skip the tokens it doesn't care about, such as
1146 "attachment" in the previous example; it will also skip other
1147 unrecognized params. If the header is syntactically correct and
1148 contains a file name, a copy of the file name is stored in
1149 *filename and true is returned. Otherwise, the function returns
1152 The file name is stripped of directory components and must not be
1155 parse_content_disposition (const char *hdr, char **filename)
1158 param_token name, value;
1159 while (extract_param (&hdr, &name, &value, ';'))
1161 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1162 if ( isFilename && value.b != NULL)
1164 /* Make the file name begin at the last slash or backslash. */
1165 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1166 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1167 if (last_slash && last_bs)
1168 value.b = 1 + MAX (last_slash, last_bs);
1169 else if (last_slash || last_bs)
1170 value.b = 1 + (last_slash ? last_slash : last_bs);
1171 if (value.b == value.e)
1173 /* Start with the directory prefix, if specified. */
1178 int prefix_length = strlen (opt.dir_prefix);
1179 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1184 total_length = prefix_length + (value.e - value.b);
1185 *filename = xmalloc (total_length + 1);
1186 strcpy (*filename, opt.dir_prefix);
1188 (*filename)[prefix_length - 1] = '/';
1189 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1190 (*filename)[total_length] = '\0';
1194 append_value_to_filename (filename, &value);
1201 append_value_to_filename (filename, &value);
1205 *filename = strdupdelim (value.b, value.e);
1221 /* Persistent connections. Currently, we cache the most recently used
1222 connection as persistent, provided that the HTTP server agrees to
1223 make it such. The persistence data is stored in the variables
1224 below. Ideally, it should be possible to cache an arbitrary fixed
1225 number of these connections. */
1227 /* Whether a persistent connection is active. */
1228 static bool pconn_active;
1231 /* The socket of the connection. */
1234 /* Host and port of the currently active persistent connection. */
1238 /* Whether a ssl handshake has occoured on this connection. */
1241 /* Whether the connection was authorized. This is only done by
1242 NTLM, which authorizes *connections* rather than individual
1243 requests. (That practice is peculiar for HTTP, but it is a
1244 useful optimization.) */
1248 /* NTLM data of the current connection. */
1249 struct ntlmdata ntlm;
1253 /* Mark the persistent connection as invalid and free the resources it
1254 uses. This is used by the CLOSE_* macros after they forcefully
1255 close a registered persistent connection. */
1258 invalidate_persistent (void)
1260 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1261 pconn_active = false;
1262 fd_close (pconn.socket);
1267 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1268 persistent. This will enable someone to use the same connection
1269 later. In the context of HTTP, this must be called only AFTER the
1270 response has been received and the server has promised that the
1271 connection will remain alive.
1273 If a previous connection was persistent, it is closed. */
1276 register_persistent (const char *host, int port, int fd, bool ssl)
1280 if (pconn.socket == fd)
1282 /* The connection FD is already registered. */
1287 /* The old persistent connection is still active; close it
1288 first. This situation arises whenever a persistent
1289 connection exists, but we then connect to a different
1290 host, and try to register a persistent connection to that
1292 invalidate_persistent ();
1296 pconn_active = true;
1298 pconn.host = xstrdup (host);
1301 pconn.authorized = false;
1303 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1306 /* Return true if a persistent connection is available for connecting
1310 persistent_available_p (const char *host, int port, bool ssl,
1311 bool *host_lookup_failed)
1313 /* First, check whether a persistent connection is active at all. */
1317 /* If we want SSL and the last connection wasn't or vice versa,
1318 don't use it. Checking for host and port is not enough because
1319 HTTP and HTTPS can apparently coexist on the same port. */
1320 if (ssl != pconn.ssl)
1323 /* If we're not connecting to the same port, we're not interested. */
1324 if (port != pconn.port)
1327 /* If the host is the same, we're in business. If not, there is
1328 still hope -- read below. */
1329 if (0 != strcasecmp (host, pconn.host))
1331 /* Check if pconn.socket is talking to HOST under another name.
1332 This happens often when both sites are virtual hosts
1333 distinguished only by name and served by the same network
1334 interface, and hence the same web server (possibly set up by
1335 the ISP and serving many different web sites). This
1336 admittedly unconventional optimization does not contradict
1337 HTTP and works well with popular server software. */
1341 struct address_list *al;
1344 /* Don't try to talk to two different SSL sites over the same
1345 secure connection! (Besides, it's not clear that
1346 name-based virtual hosting is even possible with SSL.) */
1349 /* If pconn.socket's peer is one of the IP addresses HOST
1350 resolves to, pconn.socket is for all intents and purposes
1351 already talking to HOST. */
1353 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1355 /* Can't get the peer's address -- something must be very
1356 wrong with the connection. */
1357 invalidate_persistent ();
1360 al = lookup_host (host, 0);
1363 *host_lookup_failed = true;
1367 found = address_list_contains (al, &ip);
1368 address_list_release (al);
1373 /* The persistent connection's peer address was found among the
1374 addresses HOST resolved to; therefore, pconn.sock is in fact
1375 already talking to HOST -- no need to reconnect. */
1378 /* Finally, check whether the connection is still open. This is
1379 important because most servers implement liberal (short) timeout
1380 on persistent connections. Wget can of course always reconnect
1381 if the connection doesn't work out, but it's nicer to know in
1382 advance. This test is a logical followup of the first test, but
1383 is "expensive" and therefore placed at the end of the list.
1385 (Current implementation of test_socket_open has a nice side
1386 effect that it treats sockets with pending data as "closed".
1387 This is exactly what we want: if a broken server sends message
1388 body in response to HEAD, or if it sends more than conent-length
1389 data, we won't reuse the corrupted connection.) */
1391 if (!test_socket_open (pconn.socket))
1393 /* Oops, the socket is no longer open. Now that we know that,
1394 let's invalidate the persistent connection before returning
1396 invalidate_persistent ();
1403 /* The idea behind these two CLOSE macros is to distinguish between
1404 two cases: one when the job we've been doing is finished, and we
1405 want to close the connection and leave, and two when something is
1406 seriously wrong and we're closing the connection as part of
1409 In case of keep_alive, CLOSE_FINISH should leave the connection
1410 open, while CLOSE_INVALIDATE should still close it.
1412 Note that the semantics of the flag `keep_alive' is "this
1413 connection *will* be reused (the server has promised not to close
1414 the connection once we're done)", while the semantics of
1415 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1416 active, registered connection". */
1418 #define CLOSE_FINISH(fd) do { \
1421 if (pconn_active && (fd) == pconn.socket) \
1422 invalidate_persistent (); \
1431 #define CLOSE_INVALIDATE(fd) do { \
1432 if (pconn_active && (fd) == pconn.socket) \
1433 invalidate_persistent (); \
1441 wgint len; /* received length */
1442 wgint contlen; /* expected length */
1443 wgint restval; /* the restart value */
1444 int res; /* the result of last read */
1445 char *rderrmsg; /* error message from read error */
1446 char *newloc; /* new location (redirection) */
1447 char *remote_time; /* remote time-stamp string */
1448 char *error; /* textual HTTP error */
1449 int statcode; /* status code */
1450 char *message; /* status message */
1451 wgint rd_size; /* amount of data read from socket */
1452 double dltime; /* time it took to download the data */
1453 const char *referer; /* value of the referer header. */
1454 char *local_file; /* local file name. */
1455 bool existence_checked; /* true if we already checked for a file's
1456 existence after having begun to download
1457 (needed in gethttp for when connection is
1458 interrupted/restarted. */
1459 bool timestamp_checked; /* true if pre-download time-stamping checks
1460 * have already been performed */
1461 char *orig_file_name; /* name of file to compare for time-stamping
1462 * (might be != local_file if -K is set) */
1463 wgint orig_file_size; /* size of file to compare for time-stamping */
1464 time_t orig_file_tstamp; /* time-stamp of file to compare for
1469 free_hstat (struct http_stat *hs)
1471 xfree_null (hs->newloc);
1472 xfree_null (hs->remote_time);
1473 xfree_null (hs->error);
1474 xfree_null (hs->rderrmsg);
1475 xfree_null (hs->local_file);
1476 xfree_null (hs->orig_file_name);
1477 xfree_null (hs->message);
1479 /* Guard against being called twice. */
1481 hs->remote_time = NULL;
1485 #define BEGINS_WITH(line, string_constant) \
1486 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1487 && (c_isspace (line[sizeof (string_constant) - 1]) \
1488 || !line[sizeof (string_constant) - 1]))
1491 #define SET_USER_AGENT(req) do { \
1492 if (!opt.useragent) \
1493 request_set_header (req, "User-Agent", \
1494 aprintf ("Wget/%s (VMS %s %s)", \
1495 version_string, vms_arch(), vms_vers()), \
1497 else if (*opt.useragent) \
1498 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1500 #else /* def __VMS */
1501 #define SET_USER_AGENT(req) do { \
1502 if (!opt.useragent) \
1503 request_set_header (req, "User-Agent", \
1504 aprintf ("Wget/%s (%s)", \
1505 version_string, OS_TYPE), \
1507 else if (*opt.useragent) \
1508 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1510 #endif /* def __VMS [else] */
1512 /* The flags that allow clobbering the file (opening with "wb").
1513 Defined here to avoid repetition later. #### This will require
1515 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1516 || opt.dirstruct || opt.output_document)
1518 /* Retrieve a document through HTTP protocol. It recognizes status
1519 code, and correctly handles redirections. It closes the network
1520 socket. If it receives an error from the functions below it, it
1521 will print it if there is enough information to do so (almost
1522 always), returning the error to the caller (i.e. http_loop).
1524 Various HTTP parameters are stored to hs.
1526 If PROXY is non-NULL, the connection will be made to the proxy
1527 server, and u->url will be requested. */
1529 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1532 struct request *req;
1535 char *user, *passwd;
1539 wgint contlen, contrange;
1546 /* Set to 1 when the authorization has already been sent and should
1547 not be tried again. */
1548 bool auth_finished = false;
1550 /* Set to 1 when just globally-set Basic authorization has been sent;
1551 * should prevent further Basic negotiations, but not other
1553 bool basic_auth_finished = false;
1555 /* Whether NTLM authentication is used for this request. */
1556 bool ntlm_seen = false;
1558 /* Whether our connection to the remote host is through SSL. */
1559 bool using_ssl = false;
1561 /* Whether a HEAD request will be issued (as opposed to GET or
1563 bool head_only = !!(*dt & HEAD_ONLY);
1566 struct response *resp;
1570 /* Whether this connection will be kept alive after the HTTP request
1574 /* Is the server using the chunked transfer encoding? */
1575 bool chunked_transfer_encoding = false;
1577 /* Whether keep-alive should be inhibited.
1579 RFC 2068 requests that 1.0 clients not send keep-alive requests
1580 to proxies. This is because many 1.0 proxies do not interpret
1581 the Connection header and transfer it to the remote server,
1582 causing it to not close the connection and leave both the proxy
1583 and the client hanging. */
1584 bool inhibit_keep_alive =
1585 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1587 /* Headers sent when using POST. */
1588 wgint post_data_size = 0;
1590 bool host_lookup_failed = false;
1593 if (u->scheme == SCHEME_HTTPS)
1595 /* Initialize the SSL context. After this has once been done,
1596 it becomes a no-op. */
1599 scheme_disable (SCHEME_HTTPS);
1600 logprintf (LOG_NOTQUIET,
1601 _("Disabling SSL due to encountered errors.\n"));
1602 return SSLINITFAILED;
1605 #endif /* HAVE_SSL */
1607 /* Initialize certain elements of struct http_stat. */
1611 hs->rderrmsg = NULL;
1613 hs->remote_time = NULL;
1619 /* Prepare the request to send. */
1621 req = request_new ();
1624 const char *meth = "GET";
1627 else if (opt.post_file_name || opt.post_data)
1629 /* Use the full path, i.e. one that includes the leading slash and
1630 the query string. E.g. if u->path is "foo/bar" and u->query is
1631 "param=value", full_path will be "/foo/bar?param=value". */
1634 /* When using SSL over proxy, CONNECT establishes a direct
1635 connection to the HTTPS server. Therefore use the same
1636 argument as when talking to the server directly. */
1637 && u->scheme != SCHEME_HTTPS
1640 meth_arg = xstrdup (u->url);
1642 meth_arg = url_full_path (u);
1643 request_set_method (req, meth, meth_arg);
1646 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1647 if (*dt & SEND_NOCACHE)
1648 request_set_header (req, "Pragma", "no-cache", rel_none);
1650 request_set_header (req, "Range",
1651 aprintf ("bytes=%s-",
1652 number_to_static_string (hs->restval)),
1654 SET_USER_AGENT (req);
1655 request_set_header (req, "Accept", "*/*", rel_none);
1657 /* Find the username and password for authentication. */
1660 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1661 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1662 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1664 /* We only do "site-wide" authentication with "global" user/password
1665 * values unless --auth-no-challange has been requested; URL user/password
1666 * info overrides. */
1667 if (user && passwd && (!u->user || opt.auth_without_challenge))
1669 /* If this is a host for which we've already received a Basic
1670 * challenge, we'll go ahead and send Basic authentication creds. */
1671 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1674 /* Generate the Host header, HOST:PORT. Take into account that:
1676 - Broken server-side software often doesn't recognize the PORT
1677 argument, so we must generate "Host: www.server.com" instead of
1678 "Host: www.server.com:80" (and likewise for https port).
1680 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1681 becomes ambiguous and needs to be rewritten as "Host:
1682 [3ffe:8100:200:2::2]:1234". */
1684 /* Formats arranged for hfmt[add_port][add_squares]. */
1685 static const char *hfmt[][2] = {
1686 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1688 int add_port = u->port != scheme_default_port (u->scheme);
1689 int add_squares = strchr (u->host, ':') != NULL;
1690 request_set_header (req, "Host",
1691 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1695 if (!inhibit_keep_alive)
1696 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1699 request_set_header (req, "Cookie",
1700 cookie_header (wget_cookie_jar,
1701 u->host, u->port, u->path,
1703 u->scheme == SCHEME_HTTPS
1710 if (opt.post_data || opt.post_file_name)
1712 request_set_header (req, "Content-Type",
1713 "application/x-www-form-urlencoded", rel_none);
1715 post_data_size = strlen (opt.post_data);
1718 post_data_size = file_size (opt.post_file_name);
1719 if (post_data_size == -1)
1721 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1722 quote (opt.post_file_name), strerror (errno));
1726 request_set_header (req, "Content-Length",
1727 xstrdup (number_to_static_string (post_data_size)),
1731 /* Add the user headers. */
1732 if (opt.user_headers)
1735 for (i = 0; opt.user_headers[i]; i++)
1736 request_set_user_header (req, opt.user_headers[i]);
1740 /* We need to come back here when the initial attempt to retrieve
1741 without authorization header fails. (Expected to happen at least
1742 for the Digest authorization scheme.) */
1747 char *proxy_user, *proxy_passwd;
1748 /* For normal username and password, URL components override
1749 command-line/wgetrc parameters. With proxy
1750 authentication, it's the reverse, because proxy URLs are
1751 normally the "permanent" ones, so command-line args
1752 should take precedence. */
1753 if (opt.proxy_user && opt.proxy_passwd)
1755 proxy_user = opt.proxy_user;
1756 proxy_passwd = opt.proxy_passwd;
1760 proxy_user = proxy->user;
1761 proxy_passwd = proxy->passwd;
1763 /* #### This does not appear right. Can't the proxy request,
1764 say, `Digest' authentication? */
1765 if (proxy_user && proxy_passwd)
1766 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1768 /* If we're using a proxy, we will be connecting to the proxy
1772 /* Proxy authorization over SSL is handled below. */
1774 if (u->scheme != SCHEME_HTTPS)
1776 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1781 /* Establish the connection. */
1783 if (inhibit_keep_alive)
1787 /* Look for a persistent connection to target host, unless a
1788 proxy is used. The exception is when SSL is in use, in which
1789 case the proxy is nothing but a passthrough to the target
1790 host, registered as a connection to the latter. */
1791 struct url *relevant = conn;
1793 if (u->scheme == SCHEME_HTTPS)
1797 if (persistent_available_p (relevant->host, relevant->port,
1799 relevant->scheme == SCHEME_HTTPS,
1803 &host_lookup_failed))
1805 sock = pconn.socket;
1806 using_ssl = pconn.ssl;
1807 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1808 quotearg_style (escape_quoting_style, pconn.host),
1810 DEBUGP (("Reusing fd %d.\n", sock));
1811 if (pconn.authorized)
1812 /* If the connection is already authorized, the "Basic"
1813 authorization added by code above is unnecessary and
1815 request_remove_header (req, "Authorization");
1817 else if (host_lookup_failed)
1820 logprintf(LOG_NOTQUIET,
1821 _("%s: unable to resolve host address %s\n"),
1822 exec_name, quote (relevant->host));
1829 sock = connect_to_host (conn->host, conn->port);
1838 return (retryable_socket_connect_error (errno)
1839 ? CONERROR : CONIMPOSSIBLE);
1843 if (proxy && u->scheme == SCHEME_HTTPS)
1845 /* When requesting SSL URLs through proxies, use the
1846 CONNECT method to request passthrough. */
1847 struct request *connreq = request_new ();
1848 request_set_method (connreq, "CONNECT",
1849 aprintf ("%s:%d", u->host, u->port));
1850 SET_USER_AGENT (connreq);
1853 request_set_header (connreq, "Proxy-Authorization",
1854 proxyauth, rel_value);
1855 /* Now that PROXYAUTH is part of the CONNECT request,
1856 zero it out so we don't send proxy authorization with
1857 the regular request below. */
1860 /* Examples in rfc2817 use the Host header in CONNECT
1861 requests. I don't see how that gains anything, given
1862 that the contents of Host would be exactly the same as
1863 the contents of CONNECT. */
1865 write_error = request_send (connreq, sock);
1866 request_free (connreq);
1867 if (write_error < 0)
1869 CLOSE_INVALIDATE (sock);
1873 head = read_http_response_head (sock);
1876 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1878 CLOSE_INVALIDATE (sock);
1887 DEBUGP (("proxy responded with: [%s]\n", head));
1889 resp = resp_new (head);
1890 statcode = resp_status (resp, &message);
1891 hs->message = xstrdup (message);
1894 if (statcode != 200)
1897 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1898 message ? quotearg_style (escape_quoting_style, message) : "?");
1899 xfree_null (message);
1902 xfree_null (message);
1904 /* SOCK is now *really* connected to u->host, so update CONN
1905 to reflect this. That way register_persistent will
1906 register SOCK as being connected to u->host:u->port. */
1910 if (conn->scheme == SCHEME_HTTPS)
1912 if (!ssl_connect_wget (sock))
1917 else if (!ssl_check_certificate (sock, u->host))
1920 return VERIFCERTERR;
1924 #endif /* HAVE_SSL */
1927 /* Send the request to server. */
1928 write_error = request_send (req, sock);
1930 if (write_error >= 0)
1934 DEBUGP (("[POST data: %s]\n", opt.post_data));
1935 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1937 else if (opt.post_file_name && post_data_size != 0)
1938 write_error = post_file (sock, opt.post_file_name, post_data_size);
1941 if (write_error < 0)
1943 CLOSE_INVALIDATE (sock);
1947 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1948 proxy ? "Proxy" : "HTTP");
1953 head = read_http_response_head (sock);
1958 logputs (LOG_NOTQUIET, _("No data received.\n"));
1959 CLOSE_INVALIDATE (sock);
1965 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1967 CLOSE_INVALIDATE (sock);
1972 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1974 resp = resp_new (head);
1976 /* Check for status line. */
1978 statcode = resp_status (resp, &message);
1979 hs->message = xstrdup (message);
1980 if (!opt.server_response)
1981 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1982 message ? quotearg_style (escape_quoting_style, message) : "");
1985 logprintf (LOG_VERBOSE, "\n");
1986 print_server_response (resp, " ");
1989 if (!opt.ignore_length
1990 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1994 parsed = str_to_wgint (hdrval, NULL, 10);
1995 if (parsed == WGINT_MAX && errno == ERANGE)
1998 #### If Content-Length is out of range, it most likely
1999 means that the file is larger than 2G and that we're
2000 compiled without LFS. In that case we should probably
2001 refuse to even attempt to download the file. */
2004 else if (parsed < 0)
2006 /* Negative Content-Length; nonsensical, so we can't
2007 assume any information about the content to receive. */
2014 /* Check for keep-alive related responses. */
2015 if (!inhibit_keep_alive && contlen != -1)
2017 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2019 if (0 == strcasecmp (hdrval, "Close"))
2024 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2025 if (0 == strcasecmp (hdrval, "chunked"))
2026 chunked_transfer_encoding = true;
2028 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2032 const char *scbeg, *scend;
2033 /* The jar should have been created by now. */
2034 assert (wget_cookie_jar != NULL);
2036 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2037 &scbeg, &scend)) != -1;
2040 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2041 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2042 u->path, set_cookie);
2047 /* The server has promised that it will not close the connection
2048 when we're done. This means that we can register it. */
2049 register_persistent (conn->host, conn->port, sock, using_ssl);
2051 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2053 /* Authorization is required. */
2054 if (keep_alive && !head_only
2055 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2056 CLOSE_FINISH (sock);
2058 CLOSE_INVALIDATE (sock);
2059 pconn.authorized = false;
2060 if (!auth_finished && (user && passwd))
2062 /* IIS sends multiple copies of WWW-Authenticate, one with
2063 the value "negotiate", and other(s) with data. Loop over
2064 all the occurrences and pick the one we recognize. */
2066 const char *wabeg, *waend;
2067 char *www_authenticate = NULL;
2069 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2070 &wabeg, &waend)) != -1;
2072 if (known_authentication_scheme_p (wabeg, waend))
2074 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2078 if (!www_authenticate)
2080 /* If the authentication header is missing or
2081 unrecognized, there's no sense in retrying. */
2082 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2084 else if (!basic_auth_finished
2085 || !BEGINS_WITH (www_authenticate, "Basic"))
2088 pth = url_full_path (u);
2089 request_set_header (req, "Authorization",
2090 create_authorization_line (www_authenticate,
2092 request_method (req),
2096 if (BEGINS_WITH (www_authenticate, "NTLM"))
2098 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2100 /* Need to register this host as using basic auth,
2101 * so we automatically send creds next time. */
2102 register_basic_auth_host (u->host);
2105 xfree_null (message);
2108 goto retry_with_auth;
2112 /* We already did Basic auth, and it failed. Gotta
2116 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2118 xfree_null (message);
2123 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2125 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2127 pconn.authorized = true;
2130 /* Determine the local filename if needed. Notice that if -O is used
2131 * hstat.local_file is set by http_loop to the argument of -O. */
2132 if (!hs->local_file)
2134 /* Honor Content-Disposition whether possible. */
2135 if (!opt.content_disposition
2136 || !resp_header_copy (resp, "Content-Disposition",
2137 hdrval, sizeof (hdrval))
2138 || !parse_content_disposition (hdrval, &hs->local_file))
2140 /* The Content-Disposition header is missing or broken.
2141 * Choose unique file name according to given URL. */
2142 hs->local_file = url_file_name (u);
2146 /* TODO: perform this check only once. */
2147 if (!hs->existence_checked && file_exists_p (hs->local_file))
2149 if (opt.noclobber && !opt.output_document)
2151 /* If opt.noclobber is turned on and file already exists, do not
2152 retrieve the file. But if the output_document was given, then this
2153 test was already done and the file didn't exist. Hence the !opt.output_document */
2154 logprintf (LOG_VERBOSE, _("\
2155 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2156 /* If the file is there, we suppose it's retrieved OK. */
2159 /* #### Bogusness alert. */
2160 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2161 if (has_html_suffix_p (hs->local_file))
2165 xfree_null (message);
2166 return RETRUNNEEDED;
2168 else if (!ALLOW_CLOBBER)
2170 char *unique = unique_name (hs->local_file, true);
2171 if (unique != hs->local_file)
2172 xfree (hs->local_file);
2173 hs->local_file = unique;
2176 hs->existence_checked = true;
2178 /* Support timestamping */
2179 /* TODO: move this code out of gethttp. */
2180 if (opt.timestamping && !hs->timestamp_checked)
2182 size_t filename_len = strlen (hs->local_file);
2183 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2184 bool local_dot_orig_file_exists = false;
2185 char *local_filename = NULL;
2188 if (opt.backup_converted)
2189 /* If -K is specified, we'll act on the assumption that it was specified
2190 last time these files were downloaded as well, and instead of just
2191 comparing local file X against server file X, we'll compare local
2192 file X.orig (if extant, else X) against server file X. If -K
2193 _wasn't_ specified last time, or the server contains files called
2194 *.orig, -N will be back to not operating correctly with -k. */
2196 /* Would a single s[n]printf() call be faster? --dan
2198 Definitely not. sprintf() is horribly slow. It's a
2199 different question whether the difference between the two
2200 affects a program. Usually I'd say "no", but at one
2201 point I profiled Wget, and found that a measurable and
2202 non-negligible amount of time was lost calling sprintf()
2203 in url.c. Replacing sprintf with inline calls to
2204 strcpy() and number_to_string() made a difference.
2206 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2207 memcpy (filename_plus_orig_suffix + filename_len,
2208 ORIG_SFX, sizeof (ORIG_SFX));
2210 /* Try to stat() the .orig file. */
2211 if (stat (filename_plus_orig_suffix, &st) == 0)
2213 local_dot_orig_file_exists = true;
2214 local_filename = filename_plus_orig_suffix;
2218 if (!local_dot_orig_file_exists)
2219 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2220 if (stat (hs->local_file, &st) == 0)
2221 local_filename = hs->local_file;
2223 if (local_filename != NULL)
2224 /* There was a local file, so we'll check later to see if the version
2225 the server has is the same version we already have, allowing us to
2228 hs->orig_file_name = xstrdup (local_filename);
2229 hs->orig_file_size = st.st_size;
2230 hs->orig_file_tstamp = st.st_mtime;
2232 /* Modification time granularity is 2 seconds for Windows, so
2233 increase local time by 1 second for later comparison. */
2234 ++hs->orig_file_tstamp;
2241 hs->statcode = statcode;
2243 hs->error = xstrdup (_("Malformed status line"));
2245 hs->error = xstrdup (_("(no description)"));
2247 hs->error = xstrdup (message);
2248 xfree_null (message);
2250 type = resp_header_strdup (resp, "Content-Type");
2253 char *tmp = strchr (type, ';');
2256 /* sXXXav: only needed if IRI support is enabled */
2257 char *tmp2 = tmp + 1;
2259 while (tmp > type && c_isspace (tmp[-1]))
2263 /* Try to get remote encoding if needed */
2264 if (opt.enable_iri && !opt.encoding_remote)
2266 tmp = parse_charset (tmp2);
2268 set_content_encoding (iri, tmp);
2272 hs->newloc = resp_header_strdup (resp, "Location");
2273 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2275 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2277 wgint first_byte_pos, last_byte_pos, entity_length;
2278 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2281 contrange = first_byte_pos;
2282 contlen = last_byte_pos - first_byte_pos + 1;
2287 /* 20x responses are counted among successful by default. */
2288 if (H_20X (statcode))
2291 /* Return if redirected. */
2292 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2294 /* RFC2068 says that in case of the 300 (multiple choices)
2295 response, the server can output a preferred URL through
2296 `Location' header; otherwise, the request should be treated
2297 like GET. So, if the location is set, it will be a
2298 redirection; otherwise, just proceed normally. */
2299 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2303 logprintf (LOG_VERBOSE,
2304 _("Location: %s%s\n"),
2305 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2306 hs->newloc ? _(" [following]") : "");
2307 if (keep_alive && !head_only
2308 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2309 CLOSE_FINISH (sock);
2311 CLOSE_INVALIDATE (sock);
2318 /* If content-type is not given, assume text/html. This is because
2319 of the multitude of broken CGI's that "forget" to generate the
2322 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2323 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2329 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2334 if (opt.adjust_extension)
2337 /* -E / --adjust-extension / adjust_extension = on was specified,
2338 and this is a text/html file. If some case-insensitive
2339 variation on ".htm[l]" isn't already the file's suffix,
2342 ensure_extension (hs, ".html", dt);
2344 else if (*dt & TEXTCSS)
2346 ensure_extension (hs, ".css", dt);
2350 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2351 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2352 && contrange == 0 && hs->restval >= contlen)
2355 /* If `-c' is in use and the file has been fully downloaded (or
2356 the remote file has shrunk), Wget effectively requests bytes
2357 after the end of file and the server response with 416
2358 (or 200 with a <= Content-Length. */
2359 logputs (LOG_VERBOSE, _("\
2360 \n The file is already fully retrieved; nothing to do.\n\n"));
2361 /* In case the caller inspects. */
2364 /* Mark as successfully retrieved. */
2367 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2368 might be more bytes in the body. */
2370 return RETRUNNEEDED;
2372 if ((contrange != 0 && contrange != hs->restval)
2373 || (H_PARTIAL (statcode) && !contrange))
2375 /* The Range request was somehow misunderstood by the server.
2378 CLOSE_INVALIDATE (sock);
2385 hs->contlen = contlen + contrange;
2391 /* No need to print this output if the body won't be
2392 downloaded at all, or if the original server response is
2394 logputs (LOG_VERBOSE, _("Length: "));
2397 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2398 if (contlen + contrange >= 1024)
2399 logprintf (LOG_VERBOSE, " (%s)",
2400 human_readable (contlen + contrange));
2403 if (contlen >= 1024)
2404 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2405 number_to_static_string (contlen),
2406 human_readable (contlen));
2408 logprintf (LOG_VERBOSE, _(", %s remaining"),
2409 number_to_static_string (contlen));
2413 logputs (LOG_VERBOSE,
2414 opt.ignore_length ? _("ignored") : _("unspecified"));
2416 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2418 logputs (LOG_VERBOSE, "\n");
2422 type = NULL; /* We don't need it any more. */
2424 /* Return if we have no intention of further downloading. */
2425 if (!(*dt & RETROKF) || head_only)
2427 /* In case the caller cares to look... */
2432 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2433 servers not to send body in response to a HEAD request, and
2434 those that do will likely be caught by test_socket_open.
2435 If not, they can be worked around using
2436 `--no-http-keep-alive'. */
2437 CLOSE_FINISH (sock);
2439 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2440 /* Successfully skipped the body; also keep using the socket. */
2441 CLOSE_FINISH (sock);
2443 CLOSE_INVALIDATE (sock);
2445 return RETRFINISHED;
2449 For VMS, define common fopen() optional arguments.
2452 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2453 # define FOPEN_BIN_FLAG 3
2454 #else /* def __VMS */
2455 # define FOPEN_BIN_FLAG true
2456 #endif /* def __VMS [else] */
2458 /* Open the local file. */
2461 mkalldirs (hs->local_file);
2463 rotate_backups (hs->local_file);
2470 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2471 #else /* def __VMS */
2472 fp = fopen (hs->local_file, "ab");
2473 #endif /* def __VMS [else] */
2475 else if (ALLOW_CLOBBER)
2481 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2482 #else /* def __VMS */
2483 fp = fopen (hs->local_file, "wb");
2484 #endif /* def __VMS [else] */
2488 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2489 if (!fp && errno == EEXIST)
2491 /* We cannot just invent a new name and use it (which is
2492 what functions like unique_create typically do)
2493 because we told the user we'd use this name.
2494 Instead, return and retry the download. */
2495 logprintf (LOG_NOTQUIET,
2496 _("%s has sprung into existence.\n"),
2498 CLOSE_INVALIDATE (sock);
2500 return FOPEN_EXCL_ERR;
2505 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2506 CLOSE_INVALIDATE (sock);
2514 /* Print fetch message, if opt.verbose. */
2517 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2518 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2521 /* This confuses the timestamping code that checks for file size.
2522 #### The timestamping code should be smarter about file size. */
2523 if (opt.save_headers && hs->restval == 0)
2524 fwrite (head, 1, strlen (head), fp);
2526 /* Now we no longer need to store the response header. */
2529 /* Download the request body. */
2532 /* If content-length is present, read that much; otherwise, read
2533 until EOF. The HTTP spec doesn't require the server to
2534 actually close the connection when it's done sending data. */
2535 flags |= rb_read_exactly;
2536 if (hs->restval > 0 && contrange == 0)
2537 /* If the server ignored our range request, instruct fd_read_body
2538 to skip the first RESTVAL bytes of body. */
2539 flags |= rb_skip_startpos;
2541 if (chunked_transfer_encoding)
2542 flags |= rb_chunked_transfer_encoding;
2544 hs->len = hs->restval;
2546 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2547 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2551 CLOSE_FINISH (sock);
2555 hs->rderrmsg = xstrdup (fd_errstr (sock));
2556 CLOSE_INVALIDATE (sock);
2563 return RETRFINISHED;
2566 /* The genuine HTTP loop! This is the part where the retrieval is
2567 retried, and retried, and retried, and... */
2569 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2570 int *dt, struct url *proxy, struct iri *iri)
2573 bool got_head = false; /* used for time-stamping and filename detection */
2574 bool time_came_from_head = false;
2575 bool got_name = false;
2578 uerr_t err, ret = TRYLIMEXC;
2579 time_t tmr = -1; /* remote time-stamp */
2580 struct http_stat hstat; /* HTTP status */
2582 bool send_head_first = true;
2585 /* Assert that no value for *LOCAL_FILE was passed. */
2586 assert (local_file == NULL || *local_file == NULL);
2588 /* Set LOCAL_FILE parameter. */
2589 if (local_file && opt.output_document)
2590 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2592 /* Reset NEWLOC parameter. */
2595 /* This used to be done in main(), but it's a better idea to do it
2596 here so that we don't go through the hoops if we're just using
2601 /* Warn on (likely bogus) wildcard usage in HTTP. */
2602 if (opt.ftp_glob && has_wildcards_p (u->path))
2603 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2605 /* Setup hstat struct. */
2607 hstat.referer = referer;
2609 if (opt.output_document)
2611 hstat.local_file = xstrdup (opt.output_document);
2614 else if (!opt.content_disposition)
2616 hstat.local_file = url_file_name (u);
2620 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2621 * screaming for some refactoring. */
2622 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2624 /* If opt.noclobber is turned on and file already exists, do not
2625 retrieve the file. But if the output_document was given, then this
2626 test was already done and the file didn't exist. Hence the !opt.output_document */
2627 logprintf (LOG_VERBOSE, _("\
2628 File %s already there; not retrieving.\n\n"),
2629 quote (hstat.local_file));
2630 /* If the file is there, we suppose it's retrieved OK. */
2633 /* #### Bogusness alert. */
2634 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2635 if (has_html_suffix_p (hstat.local_file))
2642 /* Reset the counter. */
2645 /* Reset the document type. */
2648 /* Skip preliminary HEAD request if we're not in spider mode AND
2649 * if -O was given or HTTP Content-Disposition support is disabled. */
2651 && (got_name || !opt.content_disposition))
2652 send_head_first = false;
2654 /* Send preliminary HEAD request if -N is given and we have an existing
2655 * destination file. */
2656 file_name = url_file_name (u);
2657 if (opt.timestamping
2658 && !opt.content_disposition
2659 && file_exists_p (file_name))
2660 send_head_first = true;
2666 /* Increment the pass counter. */
2668 sleep_between_retrievals (count);
2670 /* Get the current time string. */
2671 tms = datetime_str (time (NULL));
2673 if (opt.spider && !got_head)
2674 logprintf (LOG_VERBOSE, _("\
2675 Spider mode enabled. Check if remote file exists.\n"));
2677 /* Print fetch message, if opt.verbose. */
2680 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2685 sprintf (tmp, _("(try:%2d)"), count);
2686 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2691 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2696 ws_changetitle (hurl);
2701 /* Default document type is empty. However, if spider mode is
2702 on or time-stamping is employed, HEAD_ONLY commands is
2703 encoded within *dt. */
2704 if (send_head_first && !got_head)
2709 /* Decide whether or not to restart. */
2712 && stat (hstat.local_file, &st) == 0
2713 && S_ISREG (st.st_mode))
2714 /* When -c is used, continue from on-disk size. (Can't use
2715 hstat.len even if count>1 because we don't want a failed
2716 first attempt to clobber existing data.) */
2717 hstat.restval = st.st_size;
2719 /* otherwise, continue where the previous try left off */
2720 hstat.restval = hstat.len;
2724 /* Decide whether to send the no-cache directive. We send it in
2726 a) we're using a proxy, and we're past our first retrieval.
2727 Some proxies are notorious for caching incomplete data, so
2728 we require a fresh get.
2729 b) caching is explicitly inhibited. */
2730 if ((proxy && count > 1) /* a */
2731 || !opt.allow_cache) /* b */
2732 *dt |= SEND_NOCACHE;
2734 *dt &= ~SEND_NOCACHE;
2736 /* Try fetching the document, or at least its head. */
2737 err = gethttp (u, &hstat, dt, proxy, iri);
2740 tms = datetime_str (time (NULL));
2742 /* Get the new location (with or without the redirection). */
2744 *newloc = xstrdup (hstat.newloc);
2748 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2749 case CONERROR: case READERR: case WRITEFAILED:
2750 case RANGEERR: case FOPEN_EXCL_ERR:
2751 /* Non-fatal errors continue executing the loop, which will
2752 bring them to "while" statement at the end, to judge
2753 whether the number of tries was exceeded. */
2754 printwhat (count, opt.ntry);
2756 case FWRITEERR: case FOPENERR:
2757 /* Another fatal error. */
2758 logputs (LOG_VERBOSE, "\n");
2759 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2760 quote (hstat.local_file), strerror (errno));
2761 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2762 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2763 /* Fatal errors just return from the function. */
2767 /* Another fatal error. */
2768 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2772 /* Return the new location to the caller. */
2775 logprintf (LOG_NOTQUIET,
2776 _("ERROR: Redirection (%d) without location.\n"),
2786 /* The file was already fully retrieved. */
2790 /* Deal with you later. */
2793 /* All possibilities should have been exhausted. */
2797 if (!(*dt & RETROKF))
2802 /* #### Ugly ugly ugly! */
2803 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2804 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2807 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2809 && (hstat.statcode == 500 || hstat.statcode == 501))
2814 /* Maybe we should always keep track of broken links, not just in
2816 * Don't log error if it was UTF-8 encoded because we will try
2817 * once unencoded. */
2818 else if (opt.spider && !iri->utf8_encode)
2820 /* #### Again: ugly ugly ugly! */
2822 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2823 nonexisting_url (hurl);
2824 logprintf (LOG_NOTQUIET, _("\
2825 Remote file does not exist -- broken link!!!\n"));
2829 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2830 tms, hstat.statcode,
2831 quotearg_style (escape_quoting_style, hstat.error));
2833 logputs (LOG_VERBOSE, "\n");
2839 /* Did we get the time-stamp? */
2842 got_head = true; /* no more time-stamping */
2844 if (opt.timestamping && !hstat.remote_time)
2846 logputs (LOG_NOTQUIET, _("\
2847 Last-modified header missing -- time-stamps turned off.\n"));
2849 else if (hstat.remote_time)
2851 /* Convert the date-string into struct tm. */
2852 tmr = http_atotm (hstat.remote_time);
2853 if (tmr == (time_t) (-1))
2854 logputs (LOG_VERBOSE, _("\
2855 Last-modified header invalid -- time-stamp ignored.\n"));
2856 if (*dt & HEAD_ONLY)
2857 time_came_from_head = true;
2860 if (send_head_first)
2862 /* The time-stamping section. */
2863 if (opt.timestamping)
2865 if (hstat.orig_file_name) /* Perform the following
2866 checks only if the file
2868 download already exists. */
2870 if (hstat.remote_time &&
2871 tmr != (time_t) (-1))
2873 /* Now time-stamping can be used validly.
2874 Time-stamping means that if the sizes of
2875 the local and remote file match, and local
2876 file is newer than the remote file, it will
2877 not be retrieved. Otherwise, the normal
2878 download procedure is resumed. */
2879 if (hstat.orig_file_tstamp >= tmr)
2881 if (hstat.contlen == -1
2882 || hstat.orig_file_size == hstat.contlen)
2884 logprintf (LOG_VERBOSE, _("\
2885 Server file no newer than local file %s -- not retrieving.\n\n"),
2886 quote (hstat.orig_file_name));
2892 logprintf (LOG_VERBOSE, _("\
2893 The sizes do not match (local %s) -- retrieving.\n"),
2894 number_to_static_string (hstat.orig_file_size));
2898 logputs (LOG_VERBOSE,
2899 _("Remote file is newer, retrieving.\n"));
2901 logputs (LOG_VERBOSE, "\n");
2905 /* free_hstat (&hstat); */
2906 hstat.timestamp_checked = true;
2911 bool finished = true;
2916 logputs (LOG_VERBOSE, _("\
2917 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2922 logprintf (LOG_VERBOSE, _("\
2923 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2924 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2931 logprintf (LOG_VERBOSE, _("\
2932 Remote file exists and could contain further links,\n\
2933 but recursion is disabled -- not retrieving.\n\n"));
2937 logprintf (LOG_VERBOSE, _("\
2938 Remote file exists.\n\n"));
2940 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2945 logprintf (LOG_NONVERBOSE,
2946 _("%s URL: %s %2d %s\n"),
2947 tms, u->url, hstat.statcode,
2948 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2955 count = 0; /* the retrieve count for HEAD is reset */
2957 } /* send_head_first */
2960 if (opt.useservertimestamps
2961 && (tmr != (time_t) (-1))
2962 && ((hstat.len == hstat.contlen) ||
2963 ((hstat.res == 0) && (hstat.contlen == -1))))
2965 const char *fl = NULL;
2966 set_local_file (&fl, hstat.local_file);
2970 /* Reparse time header, in case it's changed. */
2971 if (time_came_from_head
2972 && hstat.remote_time && hstat.remote_time[0])
2974 newtmr = http_atotm (hstat.remote_time);
2975 if (newtmr != (time_t)-1)
2981 /* End of time-stamping section. */
2983 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2984 total_download_time += hstat.dltime;
2986 if (hstat.len == hstat.contlen)
2990 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
2992 logprintf (LOG_VERBOSE,
2994 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
2995 : _("%s (%s) - %s saved [%s/%s]\n\n"),
2997 write_to_stdout ? "" : quote (hstat.local_file),
2998 number_to_static_string (hstat.len),
2999 number_to_static_string (hstat.contlen));
3000 logprintf (LOG_NONVERBOSE,
3001 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3003 number_to_static_string (hstat.len),
3004 number_to_static_string (hstat.contlen),
3005 hstat.local_file, count);
3008 total_downloaded_bytes += hstat.len;
3010 /* Remember that we downloaded the file for later ".orig" code. */
3011 if (*dt & ADDED_HTML_EXTENSION)
3012 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3014 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3019 else if (hstat.res == 0) /* No read error */
3021 if (hstat.contlen == -1) /* We don't know how much we were supposed
3022 to get, so assume we succeeded. */
3026 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3028 logprintf (LOG_VERBOSE,
3030 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3031 : _("%s (%s) - %s saved [%s]\n\n"),
3033 write_to_stdout ? "" : quote (hstat.local_file),
3034 number_to_static_string (hstat.len));
3035 logprintf (LOG_NONVERBOSE,
3036 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3037 tms, u->url, number_to_static_string (hstat.len),
3038 hstat.local_file, count);
3041 total_downloaded_bytes += hstat.len;
3043 /* Remember that we downloaded the file for later ".orig" code. */
3044 if (*dt & ADDED_HTML_EXTENSION)
3045 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3047 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3052 else if (hstat.len < hstat.contlen) /* meaning we lost the
3053 connection too soon */
3055 logprintf (LOG_VERBOSE,
3056 _("%s (%s) - Connection closed at byte %s. "),
3057 tms, tmrate, number_to_static_string (hstat.len));
3058 printwhat (count, opt.ntry);
3061 else if (hstat.len != hstat.restval)
3062 /* Getting here would mean reading more data than
3063 requested with content-length, which we never do. */
3067 /* Getting here probably means that the content-length was
3068 * _less_ than the original, local size. We should probably
3069 * truncate or re-read, or something. FIXME */
3074 else /* from now on hstat.res can only be -1 */
3076 if (hstat.contlen == -1)
3078 logprintf (LOG_VERBOSE,
3079 _("%s (%s) - Read error at byte %s (%s)."),
3080 tms, tmrate, number_to_static_string (hstat.len),
3082 printwhat (count, opt.ntry);
3085 else /* hstat.res == -1 and contlen is given */
3087 logprintf (LOG_VERBOSE,
3088 _("%s (%s) - Read error at byte %s/%s (%s). "),
3090 number_to_static_string (hstat.len),
3091 number_to_static_string (hstat.contlen),
3093 printwhat (count, opt.ntry);
3099 while (!opt.ntry || (count < opt.ntry));
3103 *local_file = xstrdup (hstat.local_file);
3104 free_hstat (&hstat);
3109 /* Check whether the result of strptime() indicates success.
3110 strptime() returns the pointer to how far it got to in the string.
3111 The processing has been successful if the string is at `GMT' or
3112 `+X', or at the end of the string.
3114 In extended regexp parlance, the function returns 1 if P matches
3115 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3116 can return) is considered a failure and 0 is returned. */
3118 check_end (const char *p)
3122 while (c_isspace (*p))
3125 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3126 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3132 /* Convert the textual specification of time in TIME_STRING to the
3133 number of seconds since the Epoch.
3135 TIME_STRING can be in any of the three formats RFC2616 allows the
3136 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3137 as well as the time format used in the Set-Cookie header.
3138 Timezones are ignored, and should be GMT.
3140 Return the computed time_t representation, or -1 if the conversion
3143 This function uses strptime with various string formats for parsing
3144 TIME_STRING. This results in a parser that is not as lenient in
3145 interpreting TIME_STRING as I would like it to be. Being based on
3146 strptime, it always allows shortened months, one-digit days, etc.,
3147 but due to the multitude of formats in which time can be
3148 represented, an ideal HTTP time parser would be even more
3149 forgiving. It should completely ignore things like week days and
3150 concentrate only on the various forms of representing years,
3151 months, days, hours, minutes, and seconds. For example, it would
3152 be nice if it accepted ISO 8601 out of the box.
3154 I've investigated free and PD code for this purpose, but none was
3155 usable. getdate was big and unwieldy, and had potential copyright
3156 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3157 distributed with phttpd, is excellent, but we cannot use it because
3158 it is not assigned to the FSF. So I stuck it with strptime. */
3161 http_atotm (const char *time_string)
3163 /* NOTE: Solaris strptime man page claims that %n and %t match white
3164 space, but that's not universally available. Instead, we simply
3165 use ` ' to mean "skip all WS", which works under all strptime
3166 implementations I've tested. */
3168 static const char *time_formats[] = {
3169 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3170 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3171 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3172 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3173 (used in Set-Cookie, defined in the
3174 Netscape cookie specification.) */
3176 const char *oldlocale;
3177 char savedlocale[256];
3179 time_t ret = (time_t) -1;
3181 /* Solaris strptime fails to recognize English month names in
3182 non-English locales, which we work around by temporarily setting
3183 locale to C before invoking strptime. */
3184 oldlocale = setlocale (LC_TIME, NULL);
3187 size_t l = strlen (oldlocale);
3188 if (l >= sizeof savedlocale)
3189 savedlocale[0] = '\0';
3191 memcpy (savedlocale, oldlocale, l);
3193 else savedlocale[0] = '\0';
3195 setlocale (LC_TIME, "C");
3197 for (i = 0; i < countof (time_formats); i++)
3201 /* Some versions of strptime use the existing contents of struct
3202 tm to recalculate the date according to format. Zero it out
3203 to prevent stack garbage from influencing strptime. */
3206 if (check_end (strptime (time_string, time_formats[i], &t)))
3213 /* Restore the previous locale. */
3215 setlocale (LC_TIME, savedlocale);
3220 /* Authorization support: We support three authorization schemes:
3222 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3224 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3225 consisting of answering to the server's challenge with the proper
3228 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3229 Stenberg for libcurl. Like digest, NTLM is based on a
3230 challenge-response mechanism, but unlike digest, it is non-standard
3231 (authenticates TCP connections rather than requests), undocumented
3232 and Microsoft-specific. */
3234 /* Create the authentication header contents for the `Basic' scheme.
3235 This is done by encoding the string "USER:PASS" to base64 and
3236 prepending the string "Basic " in front of it. */
3239 basic_authentication_encode (const char *user, const char *passwd)
3242 int len1 = strlen (user) + 1 + strlen (passwd);
3244 t1 = (char *)alloca (len1 + 1);
3245 sprintf (t1, "%s:%s", user, passwd);
3247 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3248 base64_encode (t1, len1, t2);
3250 return concat_strings ("Basic ", t2, (char *) 0);
3253 #define SKIP_WS(x) do { \
3254 while (c_isspace (*(x))) \
3258 #ifdef ENABLE_DIGEST
3259 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3260 an array of 16 bytes containing the hash keys, and BUF should be a
3261 buffer of 33 writable characters (32 for hex digits plus one for
3262 zero termination). */
3264 dump_hash (char *buf, const unsigned char *hash)
3268 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3270 *buf++ = XNUM_TO_digit (*hash >> 4);
3271 *buf++ = XNUM_TO_digit (*hash & 0xf);
3276 /* Take the line apart to find the challenge, and compose a digest
3277 authorization header. See RFC2069 section 2.1.2. */
3279 digest_authentication_encode (const char *au, const char *user,
3280 const char *passwd, const char *method,
3283 static char *realm, *opaque, *nonce;
3288 { "realm", &realm },
3289 { "opaque", &opaque },
3293 param_token name, value;
3295 realm = opaque = nonce = NULL;
3297 au += 6; /* skip over `Digest' */
3298 while (extract_param (&au, &name, &value, ','))
3301 size_t namelen = name.e - name.b;
3302 for (i = 0; i < countof (options); i++)
3303 if (namelen == strlen (options[i].name)
3304 && 0 == strncmp (name.b, options[i].name,
3307 *options[i].variable = strdupdelim (value.b, value.e);
3311 if (!realm || !nonce || !user || !passwd || !path || !method)
3314 xfree_null (opaque);
3319 /* Calculate the digest value. */
3321 ALLOCA_MD5_CONTEXT (ctx);
3322 unsigned char hash[MD5_HASHLEN];
3323 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3324 char response_digest[MD5_HASHLEN * 2 + 1];
3326 /* A1BUF = H(user ":" realm ":" password) */
3328 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3329 gen_md5_update ((unsigned char *)":", 1, ctx);
3330 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3331 gen_md5_update ((unsigned char *)":", 1, ctx);
3332 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3333 gen_md5_finish (ctx, hash);
3334 dump_hash (a1buf, hash);
3336 /* A2BUF = H(method ":" path) */
3338 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3339 gen_md5_update ((unsigned char *)":", 1, ctx);
3340 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3341 gen_md5_finish (ctx, hash);
3342 dump_hash (a2buf, hash);
3344 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3346 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3347 gen_md5_update ((unsigned char *)":", 1, ctx);
3348 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3349 gen_md5_update ((unsigned char *)":", 1, ctx);
3350 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3351 gen_md5_finish (ctx, hash);
3352 dump_hash (response_digest, hash);
3354 res = xmalloc (strlen (user)
3359 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3360 + (opaque ? strlen (opaque) : 0)
3362 sprintf (res, "Digest \
3363 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3364 user, realm, nonce, path, response_digest);
3367 char *p = res + strlen (res);
3368 strcat (p, ", opaque=\"");
3375 #endif /* ENABLE_DIGEST */
3377 /* Computing the size of a string literal must take into account that
3378 value returned by sizeof includes the terminating \0. */
3379 #define STRSIZE(literal) (sizeof (literal) - 1)
3381 /* Whether chars in [b, e) begin with the literal string provided as
3382 first argument and are followed by whitespace or terminating \0.
3383 The comparison is case-insensitive. */
3384 #define STARTS(literal, b, e) \
3386 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3387 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3388 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3389 || c_isspace (b[STRSIZE (literal)])))
3392 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3394 return STARTS ("Basic", hdrbeg, hdrend)
3395 #ifdef ENABLE_DIGEST
3396 || STARTS ("Digest", hdrbeg, hdrend)
3399 || STARTS ("NTLM", hdrbeg, hdrend)
3406 /* Create the HTTP authorization request header. When the
3407 `WWW-Authenticate' response header is seen, according to the
3408 authorization scheme specified in that header (`Basic' and `Digest'
3409 are supported by the current implementation), produce an
3410 appropriate HTTP authorization request header. */
3412 create_authorization_line (const char *au, const char *user,
3413 const char *passwd, const char *method,
3414 const char *path, bool *finished)
3416 /* We are called only with known schemes, so we can dispatch on the
3418 switch (c_toupper (*au))
3420 case 'B': /* Basic */
3422 return basic_authentication_encode (user, passwd);
3423 #ifdef ENABLE_DIGEST
3424 case 'D': /* Digest */
3426 return digest_authentication_encode (au, user, passwd, method, path);
3429 case 'N': /* NTLM */
3430 if (!ntlm_input (&pconn.ntlm, au))
3435 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3438 /* We shouldn't get here -- this function should be only called
3439 with values approved by known_authentication_scheme_p. */
3447 if (!wget_cookie_jar)
3448 wget_cookie_jar = cookie_jar_new ();
3449 if (opt.cookies_input && !cookies_loaded_p)
3451 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3452 cookies_loaded_p = true;
3459 if (wget_cookie_jar)
3460 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3466 xfree_null (pconn.host);
3467 if (wget_cookie_jar)
3468 cookie_jar_delete (wget_cookie_jar);
3472 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3474 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3476 int len = strlen (ext);
3479 strncpy (shortext, ext, len - 1);
3480 shortext[len - 2] = '\0';
3483 if (last_period_in_local_filename == NULL
3484 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3485 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3487 int local_filename_len = strlen (hs->local_file);
3488 /* Resize the local file, allowing for ".html" preceded by
3489 optional ".NUMBER". */
3490 hs->local_file = xrealloc (hs->local_file,
3491 local_filename_len + 24 + len);
3492 strcpy (hs->local_file + local_filename_len, ext);
3493 /* If clobbering is not allowed and the file, as named,
3494 exists, tack on ".NUMBER.html" instead. */
3495 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3499 sprintf (hs->local_file + local_filename_len,
3500 ".%d%s", ext_num++, ext);
3501 while (file_exists_p (hs->local_file));
3503 *dt |= ADDED_HTML_EXTENSION;
3511 test_parse_content_disposition()
3516 char *opt_dir_prefix;
3520 { "filename=\"file.ext\"", NULL, "file.ext", true },
3521 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3522 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3523 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3524 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3525 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3526 { "attachment", NULL, NULL, false },
3527 { "attachment", "somedir", NULL, false },
3528 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3529 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3532 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3537 opt.dir_prefix = test_array[i].opt_dir_prefix;
3538 res = parse_content_disposition (test_array[i].hdrval, &filename);
3540 mu_assert ("test_parse_content_disposition: wrong result",
3541 res == test_array[i].result
3543 || 0 == strcmp (test_array[i].filename, filename)));
3549 #endif /* TESTING */
3552 * vim: et sts=2 sw=2 cino+={s