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);
1893 char *tms = datetime_str (time (NULL));
1894 logprintf (LOG_VERBOSE, "%d\n", statcode);
1895 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1896 quotearg_style (escape_quoting_style,
1897 _("Malformed status line")));
1901 hs->message = xstrdup (message);
1904 if (statcode != 200)
1907 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1908 message ? quotearg_style (escape_quoting_style, message) : "?");
1909 xfree_null (message);
1912 xfree_null (message);
1914 /* SOCK is now *really* connected to u->host, so update CONN
1915 to reflect this. That way register_persistent will
1916 register SOCK as being connected to u->host:u->port. */
1920 if (conn->scheme == SCHEME_HTTPS)
1922 if (!ssl_connect_wget (sock))
1927 else if (!ssl_check_certificate (sock, u->host))
1930 return VERIFCERTERR;
1934 #endif /* HAVE_SSL */
1937 /* Send the request to server. */
1938 write_error = request_send (req, sock);
1940 if (write_error >= 0)
1944 DEBUGP (("[POST data: %s]\n", opt.post_data));
1945 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1947 else if (opt.post_file_name && post_data_size != 0)
1948 write_error = post_file (sock, opt.post_file_name, post_data_size);
1951 if (write_error < 0)
1953 CLOSE_INVALIDATE (sock);
1957 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1958 proxy ? "Proxy" : "HTTP");
1963 head = read_http_response_head (sock);
1968 logputs (LOG_NOTQUIET, _("No data received.\n"));
1969 CLOSE_INVALIDATE (sock);
1975 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1977 CLOSE_INVALIDATE (sock);
1982 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1984 resp = resp_new (head);
1986 /* Check for status line. */
1988 statcode = resp_status (resp, &message);
1991 char *tms = datetime_str (time (NULL));
1992 logprintf (LOG_VERBOSE, "%d\n", statcode);
1993 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1994 quotearg_style (escape_quoting_style,
1995 _("Malformed status line")));
1996 CLOSE_INVALIDATE (sock);
2000 hs->message = xstrdup (message);
2001 if (!opt.server_response)
2002 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2003 message ? quotearg_style (escape_quoting_style, message) : "");
2006 logprintf (LOG_VERBOSE, "\n");
2007 print_server_response (resp, " ");
2010 if (!opt.ignore_length
2011 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2015 parsed = str_to_wgint (hdrval, NULL, 10);
2016 if (parsed == WGINT_MAX && errno == ERANGE)
2019 #### If Content-Length is out of range, it most likely
2020 means that the file is larger than 2G and that we're
2021 compiled without LFS. In that case we should probably
2022 refuse to even attempt to download the file. */
2025 else if (parsed < 0)
2027 /* Negative Content-Length; nonsensical, so we can't
2028 assume any information about the content to receive. */
2035 /* Check for keep-alive related responses. */
2036 if (!inhibit_keep_alive && contlen != -1)
2038 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2040 if (0 == strcasecmp (hdrval, "Close"))
2045 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2046 if (0 == strcasecmp (hdrval, "chunked"))
2047 chunked_transfer_encoding = true;
2049 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2053 const char *scbeg, *scend;
2054 /* The jar should have been created by now. */
2055 assert (wget_cookie_jar != NULL);
2057 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2058 &scbeg, &scend)) != -1;
2061 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2062 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2063 u->path, set_cookie);
2068 /* The server has promised that it will not close the connection
2069 when we're done. This means that we can register it. */
2070 register_persistent (conn->host, conn->port, sock, using_ssl);
2072 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2074 /* Authorization is required. */
2075 if (keep_alive && !head_only
2076 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2077 CLOSE_FINISH (sock);
2079 CLOSE_INVALIDATE (sock);
2080 pconn.authorized = false;
2081 if (!auth_finished && (user && passwd))
2083 /* IIS sends multiple copies of WWW-Authenticate, one with
2084 the value "negotiate", and other(s) with data. Loop over
2085 all the occurrences and pick the one we recognize. */
2087 const char *wabeg, *waend;
2088 char *www_authenticate = NULL;
2090 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2091 &wabeg, &waend)) != -1;
2093 if (known_authentication_scheme_p (wabeg, waend))
2095 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2099 if (!www_authenticate)
2101 /* If the authentication header is missing or
2102 unrecognized, there's no sense in retrying. */
2103 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2105 else if (!basic_auth_finished
2106 || !BEGINS_WITH (www_authenticate, "Basic"))
2109 pth = url_full_path (u);
2110 request_set_header (req, "Authorization",
2111 create_authorization_line (www_authenticate,
2113 request_method (req),
2117 if (BEGINS_WITH (www_authenticate, "NTLM"))
2119 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2121 /* Need to register this host as using basic auth,
2122 * so we automatically send creds next time. */
2123 register_basic_auth_host (u->host);
2126 xfree_null (message);
2129 goto retry_with_auth;
2133 /* We already did Basic auth, and it failed. Gotta
2137 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2139 xfree_null (message);
2144 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2146 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2148 pconn.authorized = true;
2151 /* Determine the local filename if needed. Notice that if -O is used
2152 * hstat.local_file is set by http_loop to the argument of -O. */
2153 if (!hs->local_file)
2155 /* Honor Content-Disposition whether possible. */
2156 if (!opt.content_disposition
2157 || !resp_header_copy (resp, "Content-Disposition",
2158 hdrval, sizeof (hdrval))
2159 || !parse_content_disposition (hdrval, &hs->local_file))
2161 /* The Content-Disposition header is missing or broken.
2162 * Choose unique file name according to given URL. */
2163 hs->local_file = url_file_name (u);
2167 /* TODO: perform this check only once. */
2168 if (!hs->existence_checked && file_exists_p (hs->local_file))
2170 if (opt.noclobber && !opt.output_document)
2172 /* If opt.noclobber is turned on and file already exists, do not
2173 retrieve the file. But if the output_document was given, then this
2174 test was already done and the file didn't exist. Hence the !opt.output_document */
2175 logprintf (LOG_VERBOSE, _("\
2176 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2177 /* If the file is there, we suppose it's retrieved OK. */
2180 /* #### Bogusness alert. */
2181 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2182 if (has_html_suffix_p (hs->local_file))
2186 xfree_null (message);
2187 return RETRUNNEEDED;
2189 else if (!ALLOW_CLOBBER)
2191 char *unique = unique_name (hs->local_file, true);
2192 if (unique != hs->local_file)
2193 xfree (hs->local_file);
2194 hs->local_file = unique;
2197 hs->existence_checked = true;
2199 /* Support timestamping */
2200 /* TODO: move this code out of gethttp. */
2201 if (opt.timestamping && !hs->timestamp_checked)
2203 size_t filename_len = strlen (hs->local_file);
2204 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2205 bool local_dot_orig_file_exists = false;
2206 char *local_filename = NULL;
2209 if (opt.backup_converted)
2210 /* If -K is specified, we'll act on the assumption that it was specified
2211 last time these files were downloaded as well, and instead of just
2212 comparing local file X against server file X, we'll compare local
2213 file X.orig (if extant, else X) against server file X. If -K
2214 _wasn't_ specified last time, or the server contains files called
2215 *.orig, -N will be back to not operating correctly with -k. */
2217 /* Would a single s[n]printf() call be faster? --dan
2219 Definitely not. sprintf() is horribly slow. It's a
2220 different question whether the difference between the two
2221 affects a program. Usually I'd say "no", but at one
2222 point I profiled Wget, and found that a measurable and
2223 non-negligible amount of time was lost calling sprintf()
2224 in url.c. Replacing sprintf with inline calls to
2225 strcpy() and number_to_string() made a difference.
2227 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2228 memcpy (filename_plus_orig_suffix + filename_len,
2229 ORIG_SFX, sizeof (ORIG_SFX));
2231 /* Try to stat() the .orig file. */
2232 if (stat (filename_plus_orig_suffix, &st) == 0)
2234 local_dot_orig_file_exists = true;
2235 local_filename = filename_plus_orig_suffix;
2239 if (!local_dot_orig_file_exists)
2240 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2241 if (stat (hs->local_file, &st) == 0)
2242 local_filename = hs->local_file;
2244 if (local_filename != NULL)
2245 /* There was a local file, so we'll check later to see if the version
2246 the server has is the same version we already have, allowing us to
2249 hs->orig_file_name = xstrdup (local_filename);
2250 hs->orig_file_size = st.st_size;
2251 hs->orig_file_tstamp = st.st_mtime;
2253 /* Modification time granularity is 2 seconds for Windows, so
2254 increase local time by 1 second for later comparison. */
2255 ++hs->orig_file_tstamp;
2262 hs->statcode = statcode;
2264 hs->error = xstrdup (_("Malformed status line"));
2266 hs->error = xstrdup (_("(no description)"));
2268 hs->error = xstrdup (message);
2269 xfree_null (message);
2271 type = resp_header_strdup (resp, "Content-Type");
2274 char *tmp = strchr (type, ';');
2277 /* sXXXav: only needed if IRI support is enabled */
2278 char *tmp2 = tmp + 1;
2280 while (tmp > type && c_isspace (tmp[-1]))
2284 /* Try to get remote encoding if needed */
2285 if (opt.enable_iri && !opt.encoding_remote)
2287 tmp = parse_charset (tmp2);
2289 set_content_encoding (iri, tmp);
2293 hs->newloc = resp_header_strdup (resp, "Location");
2294 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2296 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2298 wgint first_byte_pos, last_byte_pos, entity_length;
2299 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2302 contrange = first_byte_pos;
2303 contlen = last_byte_pos - first_byte_pos + 1;
2308 /* 20x responses are counted among successful by default. */
2309 if (H_20X (statcode))
2312 /* Return if redirected. */
2313 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2315 /* RFC2068 says that in case of the 300 (multiple choices)
2316 response, the server can output a preferred URL through
2317 `Location' header; otherwise, the request should be treated
2318 like GET. So, if the location is set, it will be a
2319 redirection; otherwise, just proceed normally. */
2320 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2324 logprintf (LOG_VERBOSE,
2325 _("Location: %s%s\n"),
2326 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2327 hs->newloc ? _(" [following]") : "");
2328 if (keep_alive && !head_only
2329 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2330 CLOSE_FINISH (sock);
2332 CLOSE_INVALIDATE (sock);
2339 /* If content-type is not given, assume text/html. This is because
2340 of the multitude of broken CGI's that "forget" to generate the
2343 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2344 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2350 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2355 if (opt.adjust_extension)
2358 /* -E / --adjust-extension / adjust_extension = on was specified,
2359 and this is a text/html file. If some case-insensitive
2360 variation on ".htm[l]" isn't already the file's suffix,
2363 ensure_extension (hs, ".html", dt);
2365 else if (*dt & TEXTCSS)
2367 ensure_extension (hs, ".css", dt);
2371 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2372 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2373 && contrange == 0 && hs->restval >= contlen)
2376 /* If `-c' is in use and the file has been fully downloaded (or
2377 the remote file has shrunk), Wget effectively requests bytes
2378 after the end of file and the server response with 416
2379 (or 200 with a <= Content-Length. */
2380 logputs (LOG_VERBOSE, _("\
2381 \n The file is already fully retrieved; nothing to do.\n\n"));
2382 /* In case the caller inspects. */
2385 /* Mark as successfully retrieved. */
2388 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2389 might be more bytes in the body. */
2391 return RETRUNNEEDED;
2393 if ((contrange != 0 && contrange != hs->restval)
2394 || (H_PARTIAL (statcode) && !contrange))
2396 /* The Range request was somehow misunderstood by the server.
2399 CLOSE_INVALIDATE (sock);
2406 hs->contlen = contlen + contrange;
2412 /* No need to print this output if the body won't be
2413 downloaded at all, or if the original server response is
2415 logputs (LOG_VERBOSE, _("Length: "));
2418 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2419 if (contlen + contrange >= 1024)
2420 logprintf (LOG_VERBOSE, " (%s)",
2421 human_readable (contlen + contrange));
2424 if (contlen >= 1024)
2425 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2426 number_to_static_string (contlen),
2427 human_readable (contlen));
2429 logprintf (LOG_VERBOSE, _(", %s remaining"),
2430 number_to_static_string (contlen));
2434 logputs (LOG_VERBOSE,
2435 opt.ignore_length ? _("ignored") : _("unspecified"));
2437 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2439 logputs (LOG_VERBOSE, "\n");
2443 type = NULL; /* We don't need it any more. */
2445 /* Return if we have no intention of further downloading. */
2446 if (!(*dt & RETROKF) || head_only)
2448 /* In case the caller cares to look... */
2453 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2454 servers not to send body in response to a HEAD request, and
2455 those that do will likely be caught by test_socket_open.
2456 If not, they can be worked around using
2457 `--no-http-keep-alive'. */
2458 CLOSE_FINISH (sock);
2460 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2461 /* Successfully skipped the body; also keep using the socket. */
2462 CLOSE_FINISH (sock);
2464 CLOSE_INVALIDATE (sock);
2466 return RETRFINISHED;
2470 For VMS, define common fopen() optional arguments.
2473 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2474 # define FOPEN_BIN_FLAG 3
2475 #else /* def __VMS */
2476 # define FOPEN_BIN_FLAG true
2477 #endif /* def __VMS [else] */
2479 /* Open the local file. */
2482 mkalldirs (hs->local_file);
2484 rotate_backups (hs->local_file);
2491 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2492 #else /* def __VMS */
2493 fp = fopen (hs->local_file, "ab");
2494 #endif /* def __VMS [else] */
2496 else if (ALLOW_CLOBBER)
2502 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2503 #else /* def __VMS */
2504 fp = fopen (hs->local_file, "wb");
2505 #endif /* def __VMS [else] */
2509 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2510 if (!fp && errno == EEXIST)
2512 /* We cannot just invent a new name and use it (which is
2513 what functions like unique_create typically do)
2514 because we told the user we'd use this name.
2515 Instead, return and retry the download. */
2516 logprintf (LOG_NOTQUIET,
2517 _("%s has sprung into existence.\n"),
2519 CLOSE_INVALIDATE (sock);
2521 return FOPEN_EXCL_ERR;
2526 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2527 CLOSE_INVALIDATE (sock);
2535 /* Print fetch message, if opt.verbose. */
2538 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2539 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2542 /* This confuses the timestamping code that checks for file size.
2543 #### The timestamping code should be smarter about file size. */
2544 if (opt.save_headers && hs->restval == 0)
2545 fwrite (head, 1, strlen (head), fp);
2547 /* Now we no longer need to store the response header. */
2550 /* Download the request body. */
2553 /* If content-length is present, read that much; otherwise, read
2554 until EOF. The HTTP spec doesn't require the server to
2555 actually close the connection when it's done sending data. */
2556 flags |= rb_read_exactly;
2557 if (hs->restval > 0 && contrange == 0)
2558 /* If the server ignored our range request, instruct fd_read_body
2559 to skip the first RESTVAL bytes of body. */
2560 flags |= rb_skip_startpos;
2562 if (chunked_transfer_encoding)
2563 flags |= rb_chunked_transfer_encoding;
2565 hs->len = hs->restval;
2567 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2568 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2572 CLOSE_FINISH (sock);
2576 hs->rderrmsg = xstrdup (fd_errstr (sock));
2577 CLOSE_INVALIDATE (sock);
2584 return RETRFINISHED;
2587 /* The genuine HTTP loop! This is the part where the retrieval is
2588 retried, and retried, and retried, and... */
2590 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2591 int *dt, struct url *proxy, struct iri *iri)
2594 bool got_head = false; /* used for time-stamping and filename detection */
2595 bool time_came_from_head = false;
2596 bool got_name = false;
2599 uerr_t err, ret = TRYLIMEXC;
2600 time_t tmr = -1; /* remote time-stamp */
2601 struct http_stat hstat; /* HTTP status */
2603 bool send_head_first = true;
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)
2637 hstat.local_file = url_file_name (u);
2641 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2642 * screaming for some refactoring. */
2643 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2645 /* If opt.noclobber is turned on and file already exists, do not
2646 retrieve the file. But if the output_document was given, then this
2647 test was already done and the file didn't exist. Hence the !opt.output_document */
2648 logprintf (LOG_VERBOSE, _("\
2649 File %s already there; not retrieving.\n\n"),
2650 quote (hstat.local_file));
2651 /* If the file is there, we suppose it's retrieved OK. */
2654 /* #### Bogusness alert. */
2655 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2656 if (has_html_suffix_p (hstat.local_file))
2663 /* Reset the counter. */
2666 /* Reset the document type. */
2669 /* Skip preliminary HEAD request if we're not in spider mode AND
2670 * if -O was given or HTTP Content-Disposition support is disabled. */
2672 && (got_name || !opt.content_disposition))
2673 send_head_first = false;
2675 /* Send preliminary HEAD request if -N is given and we have an existing
2676 * destination file. */
2677 file_name = url_file_name (u);
2678 if (opt.timestamping
2679 && !opt.content_disposition
2680 && file_exists_p (file_name))
2681 send_head_first = true;
2687 /* Increment the pass counter. */
2689 sleep_between_retrievals (count);
2691 /* Get the current time string. */
2692 tms = datetime_str (time (NULL));
2694 if (opt.spider && !got_head)
2695 logprintf (LOG_VERBOSE, _("\
2696 Spider mode enabled. Check if remote file exists.\n"));
2698 /* Print fetch message, if opt.verbose. */
2701 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2706 sprintf (tmp, _("(try:%2d)"), count);
2707 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2712 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2717 ws_changetitle (hurl);
2722 /* Default document type is empty. However, if spider mode is
2723 on or time-stamping is employed, HEAD_ONLY commands is
2724 encoded within *dt. */
2725 if (send_head_first && !got_head)
2730 /* Decide whether or not to restart. */
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);
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 /* Return the new location to the caller. */
2796 logprintf (LOG_NOTQUIET,
2797 _("ERROR: Redirection (%d) without location.\n"),
2807 /* The file was already fully retrieved. */
2811 /* Deal with you later. */
2814 /* All possibilities should have been exhausted. */
2818 if (!(*dt & RETROKF))
2823 /* #### Ugly ugly ugly! */
2824 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2825 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2828 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2830 && (hstat.statcode == 500 || hstat.statcode == 501))
2835 /* Maybe we should always keep track of broken links, not just in
2837 * Don't log error if it was UTF-8 encoded because we will try
2838 * once unencoded. */
2839 else if (opt.spider && !iri->utf8_encode)
2841 /* #### Again: ugly ugly ugly! */
2843 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2844 nonexisting_url (hurl);
2845 logprintf (LOG_NOTQUIET, _("\
2846 Remote file does not exist -- broken link!!!\n"));
2850 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2851 tms, hstat.statcode,
2852 quotearg_style (escape_quoting_style, hstat.error));
2854 logputs (LOG_VERBOSE, "\n");
2860 /* Did we get the time-stamp? */
2863 got_head = true; /* no more time-stamping */
2865 if (opt.timestamping && !hstat.remote_time)
2867 logputs (LOG_NOTQUIET, _("\
2868 Last-modified header missing -- time-stamps turned off.\n"));
2870 else if (hstat.remote_time)
2872 /* Convert the date-string into struct tm. */
2873 tmr = http_atotm (hstat.remote_time);
2874 if (tmr == (time_t) (-1))
2875 logputs (LOG_VERBOSE, _("\
2876 Last-modified header invalid -- time-stamp ignored.\n"));
2877 if (*dt & HEAD_ONLY)
2878 time_came_from_head = true;
2881 if (send_head_first)
2883 /* The time-stamping section. */
2884 if (opt.timestamping)
2886 if (hstat.orig_file_name) /* Perform the following
2887 checks only if the file
2889 download already exists. */
2891 if (hstat.remote_time &&
2892 tmr != (time_t) (-1))
2894 /* Now time-stamping can be used validly.
2895 Time-stamping means that if the sizes of
2896 the local and remote file match, and local
2897 file is newer than the remote file, it will
2898 not be retrieved. Otherwise, the normal
2899 download procedure is resumed. */
2900 if (hstat.orig_file_tstamp >= tmr)
2902 if (hstat.contlen == -1
2903 || hstat.orig_file_size == hstat.contlen)
2905 logprintf (LOG_VERBOSE, _("\
2906 Server file no newer than local file %s -- not retrieving.\n\n"),
2907 quote (hstat.orig_file_name));
2913 logprintf (LOG_VERBOSE, _("\
2914 The sizes do not match (local %s) -- retrieving.\n"),
2915 number_to_static_string (hstat.orig_file_size));
2919 logputs (LOG_VERBOSE,
2920 _("Remote file is newer, retrieving.\n"));
2922 logputs (LOG_VERBOSE, "\n");
2926 /* free_hstat (&hstat); */
2927 hstat.timestamp_checked = true;
2932 bool finished = true;
2937 logputs (LOG_VERBOSE, _("\
2938 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2943 logprintf (LOG_VERBOSE, _("\
2944 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2945 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2952 logprintf (LOG_VERBOSE, _("\
2953 Remote file exists and could contain further links,\n\
2954 but recursion is disabled -- not retrieving.\n\n"));
2958 logprintf (LOG_VERBOSE, _("\
2959 Remote file exists.\n\n"));
2961 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2966 logprintf (LOG_NONVERBOSE,
2967 _("%s URL: %s %2d %s\n"),
2968 tms, u->url, hstat.statcode,
2969 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2976 count = 0; /* the retrieve count for HEAD is reset */
2978 } /* send_head_first */
2981 if (opt.useservertimestamps
2982 && (tmr != (time_t) (-1))
2983 && ((hstat.len == hstat.contlen) ||
2984 ((hstat.res == 0) && (hstat.contlen == -1))))
2986 const char *fl = NULL;
2987 set_local_file (&fl, hstat.local_file);
2991 /* Reparse time header, in case it's changed. */
2992 if (time_came_from_head
2993 && hstat.remote_time && hstat.remote_time[0])
2995 newtmr = http_atotm (hstat.remote_time);
2996 if (newtmr != (time_t)-1)
3002 /* End of time-stamping section. */
3004 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3005 total_download_time += hstat.dltime;
3007 if (hstat.len == hstat.contlen)
3011 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3013 logprintf (LOG_VERBOSE,
3015 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3016 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3018 write_to_stdout ? "" : quote (hstat.local_file),
3019 number_to_static_string (hstat.len),
3020 number_to_static_string (hstat.contlen));
3021 logprintf (LOG_NONVERBOSE,
3022 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3024 number_to_static_string (hstat.len),
3025 number_to_static_string (hstat.contlen),
3026 hstat.local_file, count);
3029 total_downloaded_bytes += hstat.len;
3031 /* Remember that we downloaded the file for later ".orig" code. */
3032 if (*dt & ADDED_HTML_EXTENSION)
3033 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3035 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3040 else if (hstat.res == 0) /* No read error */
3042 if (hstat.contlen == -1) /* We don't know how much we were supposed
3043 to get, so assume we succeeded. */
3047 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3049 logprintf (LOG_VERBOSE,
3051 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3052 : _("%s (%s) - %s saved [%s]\n\n"),
3054 write_to_stdout ? "" : quote (hstat.local_file),
3055 number_to_static_string (hstat.len));
3056 logprintf (LOG_NONVERBOSE,
3057 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3058 tms, u->url, number_to_static_string (hstat.len),
3059 hstat.local_file, count);
3062 total_downloaded_bytes += hstat.len;
3064 /* Remember that we downloaded the file for later ".orig" code. */
3065 if (*dt & ADDED_HTML_EXTENSION)
3066 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3068 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3073 else if (hstat.len < hstat.contlen) /* meaning we lost the
3074 connection too soon */
3076 logprintf (LOG_VERBOSE,
3077 _("%s (%s) - Connection closed at byte %s. "),
3078 tms, tmrate, number_to_static_string (hstat.len));
3079 printwhat (count, opt.ntry);
3082 else if (hstat.len != hstat.restval)
3083 /* Getting here would mean reading more data than
3084 requested with content-length, which we never do. */
3088 /* Getting here probably means that the content-length was
3089 * _less_ than the original, local size. We should probably
3090 * truncate or re-read, or something. FIXME */
3095 else /* from now on hstat.res can only be -1 */
3097 if (hstat.contlen == -1)
3099 logprintf (LOG_VERBOSE,
3100 _("%s (%s) - Read error at byte %s (%s)."),
3101 tms, tmrate, number_to_static_string (hstat.len),
3103 printwhat (count, opt.ntry);
3106 else /* hstat.res == -1 and contlen is given */
3108 logprintf (LOG_VERBOSE,
3109 _("%s (%s) - Read error at byte %s/%s (%s). "),
3111 number_to_static_string (hstat.len),
3112 number_to_static_string (hstat.contlen),
3114 printwhat (count, opt.ntry);
3120 while (!opt.ntry || (count < opt.ntry));
3124 *local_file = xstrdup (hstat.local_file);
3125 free_hstat (&hstat);
3130 /* Check whether the result of strptime() indicates success.
3131 strptime() returns the pointer to how far it got to in the string.
3132 The processing has been successful if the string is at `GMT' or
3133 `+X', or at the end of the string.
3135 In extended regexp parlance, the function returns 1 if P matches
3136 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3137 can return) is considered a failure and 0 is returned. */
3139 check_end (const char *p)
3143 while (c_isspace (*p))
3146 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3147 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3153 /* Convert the textual specification of time in TIME_STRING to the
3154 number of seconds since the Epoch.
3156 TIME_STRING can be in any of the three formats RFC2616 allows the
3157 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3158 as well as the time format used in the Set-Cookie header.
3159 Timezones are ignored, and should be GMT.
3161 Return the computed time_t representation, or -1 if the conversion
3164 This function uses strptime with various string formats for parsing
3165 TIME_STRING. This results in a parser that is not as lenient in
3166 interpreting TIME_STRING as I would like it to be. Being based on
3167 strptime, it always allows shortened months, one-digit days, etc.,
3168 but due to the multitude of formats in which time can be
3169 represented, an ideal HTTP time parser would be even more
3170 forgiving. It should completely ignore things like week days and
3171 concentrate only on the various forms of representing years,
3172 months, days, hours, minutes, and seconds. For example, it would
3173 be nice if it accepted ISO 8601 out of the box.
3175 I've investigated free and PD code for this purpose, but none was
3176 usable. getdate was big and unwieldy, and had potential copyright
3177 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3178 distributed with phttpd, is excellent, but we cannot use it because
3179 it is not assigned to the FSF. So I stuck it with strptime. */
3182 http_atotm (const char *time_string)
3184 /* NOTE: Solaris strptime man page claims that %n and %t match white
3185 space, but that's not universally available. Instead, we simply
3186 use ` ' to mean "skip all WS", which works under all strptime
3187 implementations I've tested. */
3189 static const char *time_formats[] = {
3190 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3191 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3192 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3193 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3194 (used in Set-Cookie, defined in the
3195 Netscape cookie specification.) */
3197 const char *oldlocale;
3198 char savedlocale[256];
3200 time_t ret = (time_t) -1;
3202 /* Solaris strptime fails to recognize English month names in
3203 non-English locales, which we work around by temporarily setting
3204 locale to C before invoking strptime. */
3205 oldlocale = setlocale (LC_TIME, NULL);
3208 size_t l = strlen (oldlocale);
3209 if (l >= sizeof savedlocale)
3210 savedlocale[0] = '\0';
3212 memcpy (savedlocale, oldlocale, l);
3214 else savedlocale[0] = '\0';
3216 setlocale (LC_TIME, "C");
3218 for (i = 0; i < countof (time_formats); i++)
3222 /* Some versions of strptime use the existing contents of struct
3223 tm to recalculate the date according to format. Zero it out
3224 to prevent stack garbage from influencing strptime. */
3227 if (check_end (strptime (time_string, time_formats[i], &t)))
3234 /* Restore the previous locale. */
3236 setlocale (LC_TIME, savedlocale);
3241 /* Authorization support: We support three authorization schemes:
3243 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3245 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3246 consisting of answering to the server's challenge with the proper
3249 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3250 Stenberg for libcurl. Like digest, NTLM is based on a
3251 challenge-response mechanism, but unlike digest, it is non-standard
3252 (authenticates TCP connections rather than requests), undocumented
3253 and Microsoft-specific. */
3255 /* Create the authentication header contents for the `Basic' scheme.
3256 This is done by encoding the string "USER:PASS" to base64 and
3257 prepending the string "Basic " in front of it. */
3260 basic_authentication_encode (const char *user, const char *passwd)
3263 int len1 = strlen (user) + 1 + strlen (passwd);
3265 t1 = (char *)alloca (len1 + 1);
3266 sprintf (t1, "%s:%s", user, passwd);
3268 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3269 base64_encode (t1, len1, t2);
3271 return concat_strings ("Basic ", t2, (char *) 0);
3274 #define SKIP_WS(x) do { \
3275 while (c_isspace (*(x))) \
3279 #ifdef ENABLE_DIGEST
3280 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3281 an array of 16 bytes containing the hash keys, and BUF should be a
3282 buffer of 33 writable characters (32 for hex digits plus one for
3283 zero termination). */
3285 dump_hash (char *buf, const unsigned char *hash)
3289 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3291 *buf++ = XNUM_TO_digit (*hash >> 4);
3292 *buf++ = XNUM_TO_digit (*hash & 0xf);
3297 /* Take the line apart to find the challenge, and compose a digest
3298 authorization header. See RFC2069 section 2.1.2. */
3300 digest_authentication_encode (const char *au, const char *user,
3301 const char *passwd, const char *method,
3304 static char *realm, *opaque, *nonce;
3309 { "realm", &realm },
3310 { "opaque", &opaque },
3314 param_token name, value;
3316 realm = opaque = nonce = NULL;
3318 au += 6; /* skip over `Digest' */
3319 while (extract_param (&au, &name, &value, ','))
3322 size_t namelen = name.e - name.b;
3323 for (i = 0; i < countof (options); i++)
3324 if (namelen == strlen (options[i].name)
3325 && 0 == strncmp (name.b, options[i].name,
3328 *options[i].variable = strdupdelim (value.b, value.e);
3332 if (!realm || !nonce || !user || !passwd || !path || !method)
3335 xfree_null (opaque);
3340 /* Calculate the digest value. */
3342 ALLOCA_MD5_CONTEXT (ctx);
3343 unsigned char hash[MD5_HASHLEN];
3344 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3345 char response_digest[MD5_HASHLEN * 2 + 1];
3347 /* A1BUF = H(user ":" realm ":" password) */
3349 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3350 gen_md5_update ((unsigned char *)":", 1, ctx);
3351 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3352 gen_md5_update ((unsigned char *)":", 1, ctx);
3353 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3354 gen_md5_finish (ctx, hash);
3355 dump_hash (a1buf, hash);
3357 /* A2BUF = H(method ":" path) */
3359 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3360 gen_md5_update ((unsigned char *)":", 1, ctx);
3361 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3362 gen_md5_finish (ctx, hash);
3363 dump_hash (a2buf, hash);
3365 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3367 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3368 gen_md5_update ((unsigned char *)":", 1, ctx);
3369 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3370 gen_md5_update ((unsigned char *)":", 1, ctx);
3371 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3372 gen_md5_finish (ctx, hash);
3373 dump_hash (response_digest, hash);
3375 res = xmalloc (strlen (user)
3380 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3381 + (opaque ? strlen (opaque) : 0)
3383 sprintf (res, "Digest \
3384 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3385 user, realm, nonce, path, response_digest);
3388 char *p = res + strlen (res);
3389 strcat (p, ", opaque=\"");
3396 #endif /* ENABLE_DIGEST */
3398 /* Computing the size of a string literal must take into account that
3399 value returned by sizeof includes the terminating \0. */
3400 #define STRSIZE(literal) (sizeof (literal) - 1)
3402 /* Whether chars in [b, e) begin with the literal string provided as
3403 first argument and are followed by whitespace or terminating \0.
3404 The comparison is case-insensitive. */
3405 #define STARTS(literal, b, e) \
3407 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3408 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3409 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3410 || c_isspace (b[STRSIZE (literal)])))
3413 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3415 return STARTS ("Basic", hdrbeg, hdrend)
3416 #ifdef ENABLE_DIGEST
3417 || STARTS ("Digest", hdrbeg, hdrend)
3420 || STARTS ("NTLM", hdrbeg, hdrend)
3427 /* Create the HTTP authorization request header. When the
3428 `WWW-Authenticate' response header is seen, according to the
3429 authorization scheme specified in that header (`Basic' and `Digest'
3430 are supported by the current implementation), produce an
3431 appropriate HTTP authorization request header. */
3433 create_authorization_line (const char *au, const char *user,
3434 const char *passwd, const char *method,
3435 const char *path, bool *finished)
3437 /* We are called only with known schemes, so we can dispatch on the
3439 switch (c_toupper (*au))
3441 case 'B': /* Basic */
3443 return basic_authentication_encode (user, passwd);
3444 #ifdef ENABLE_DIGEST
3445 case 'D': /* Digest */
3447 return digest_authentication_encode (au, user, passwd, method, path);
3450 case 'N': /* NTLM */
3451 if (!ntlm_input (&pconn.ntlm, au))
3456 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3459 /* We shouldn't get here -- this function should be only called
3460 with values approved by known_authentication_scheme_p. */
3468 if (!wget_cookie_jar)
3469 wget_cookie_jar = cookie_jar_new ();
3470 if (opt.cookies_input && !cookies_loaded_p)
3472 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3473 cookies_loaded_p = true;
3480 if (wget_cookie_jar)
3481 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3487 xfree_null (pconn.host);
3488 if (wget_cookie_jar)
3489 cookie_jar_delete (wget_cookie_jar);
3493 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3495 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3497 int len = strlen (ext);
3500 strncpy (shortext, ext, len - 1);
3501 shortext[len - 2] = '\0';
3504 if (last_period_in_local_filename == NULL
3505 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3506 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3508 int local_filename_len = strlen (hs->local_file);
3509 /* Resize the local file, allowing for ".html" preceded by
3510 optional ".NUMBER". */
3511 hs->local_file = xrealloc (hs->local_file,
3512 local_filename_len + 24 + len);
3513 strcpy (hs->local_file + local_filename_len, ext);
3514 /* If clobbering is not allowed and the file, as named,
3515 exists, tack on ".NUMBER.html" instead. */
3516 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3520 sprintf (hs->local_file + local_filename_len,
3521 ".%d%s", ext_num++, ext);
3522 while (file_exists_p (hs->local_file));
3524 *dt |= ADDED_HTML_EXTENSION;
3532 test_parse_content_disposition()
3537 char *opt_dir_prefix;
3541 { "filename=\"file.ext\"", NULL, "file.ext", true },
3542 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3543 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3544 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3545 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3546 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3547 { "attachment", NULL, NULL, false },
3548 { "attachment", "somedir", NULL, false },
3549 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3550 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3553 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3558 opt.dir_prefix = test_array[i].opt_dir_prefix;
3559 res = parse_content_disposition (test_array[i].hdrval, &filename);
3561 mu_assert ("test_parse_content_disposition: wrong result",
3562 res == test_array[i].result
3564 || 0 == strcmp (test_array[i].filename, filename)));
3570 #endif /* TESTING */
3573 * vim: et sts=2 sw=2 cino+={s