2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
56 # include "http-ntlm.h"
69 extern char *version_string;
73 static char *create_authorization_line (const char *, const char *,
74 const char *, const char *,
75 const char *, bool *);
76 static char *basic_authentication_encode (const char *, const char *);
77 static bool known_authentication_scheme_p (const char *, const char *);
78 static void ensure_extension (struct http_stat *, const char *, int *);
79 static void load_cookies (void);
82 # define MIN(x, y) ((x) > (y) ? (y) : (x))
86 static bool cookies_loaded_p;
87 static struct cookie_jar *wget_cookie_jar;
89 #define TEXTHTML_S "text/html"
90 #define TEXTXHTML_S "application/xhtml+xml"
91 #define TEXTCSS_S "text/css"
93 /* Some status code validation macros: */
94 #define H_20X(x) (((x) >= 200) && ((x) < 300))
95 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
96 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
97 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
98 || (x) == HTTP_STATUS_SEE_OTHER \
99 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
101 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
102 /* Successful 2xx. */
103 #define HTTP_STATUS_OK 200
104 #define HTTP_STATUS_CREATED 201
105 #define HTTP_STATUS_ACCEPTED 202
106 #define HTTP_STATUS_NO_CONTENT 204
107 #define HTTP_STATUS_PARTIAL_CONTENTS 206
109 /* Redirection 3xx. */
110 #define HTTP_STATUS_MULTIPLE_CHOICES 300
111 #define HTTP_STATUS_MOVED_PERMANENTLY 301
112 #define HTTP_STATUS_MOVED_TEMPORARILY 302
113 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
114 #define HTTP_STATUS_NOT_MODIFIED 304
115 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
117 /* Client error 4xx. */
118 #define HTTP_STATUS_BAD_REQUEST 400
119 #define HTTP_STATUS_UNAUTHORIZED 401
120 #define HTTP_STATUS_FORBIDDEN 403
121 #define HTTP_STATUS_NOT_FOUND 404
122 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
124 /* Server errors 5xx. */
125 #define HTTP_STATUS_INTERNAL 500
126 #define HTTP_STATUS_NOT_IMPLEMENTED 501
127 #define HTTP_STATUS_BAD_GATEWAY 502
128 #define HTTP_STATUS_UNAVAILABLE 503
131 rel_none, rel_name, rel_value, rel_both
138 struct request_header {
140 enum rp release_policy;
142 int hcount, hcapacity;
145 /* Create a new, empty request. At least request_set_method must be
146 called before the request can be used. */
148 static struct request *
151 struct request *req = xnew0 (struct request);
153 req->headers = xnew_array (struct request_header, req->hcapacity);
157 /* Set the request's method and its arguments. METH should be a
158 literal string (or it should outlive the request) because it will
159 not be freed. ARG will be freed by request_free. */
162 request_set_method (struct request *req, const char *meth, char *arg)
168 /* Return the method string passed with the last call to
169 request_set_method. */
172 request_method (const struct request *req)
177 /* Free one header according to the release policy specified with
178 request_set_header. */
181 release_header (struct request_header *hdr)
183 switch (hdr->release_policy)
200 /* Set the request named NAME to VALUE. Specifically, this means that
201 a "NAME: VALUE\r\n" header line will be used in the request. If a
202 header with the same name previously existed in the request, its
203 value will be replaced by this one. A NULL value means do nothing.
205 RELEASE_POLICY determines whether NAME and VALUE should be released
206 (freed) with request_free. Allowed values are:
208 - rel_none - don't free NAME or VALUE
209 - rel_name - free NAME when done
210 - rel_value - free VALUE when done
211 - rel_both - free both NAME and VALUE when done
213 Setting release policy is useful when arguments come from different
214 sources. For example:
216 // Don't free literal strings!
217 request_set_header (req, "Pragma", "no-cache", rel_none);
219 // Don't free a global variable, we'll need it later.
220 request_set_header (req, "Referer", opt.referer, rel_none);
222 // Value freshly allocated, free it when done.
223 request_set_header (req, "Range",
224 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
229 request_set_header (struct request *req, char *name, char *value,
230 enum rp release_policy)
232 struct request_header *hdr;
237 /* A NULL value is a no-op; if freeing the name is requested,
238 free it now to avoid leaks. */
239 if (release_policy == rel_name || release_policy == rel_both)
244 for (i = 0; i < req->hcount; i++)
246 hdr = &req->headers[i];
247 if (0 == strcasecmp (name, hdr->name))
249 /* Replace existing header. */
250 release_header (hdr);
253 hdr->release_policy = release_policy;
258 /* Install new header. */
260 if (req->hcount >= req->hcapacity)
262 req->hcapacity <<= 1;
263 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
265 hdr = &req->headers[req->hcount++];
268 hdr->release_policy = release_policy;
271 /* Like request_set_header, but sets the whole header line, as
272 provided by the user using the `--header' option. For example,
273 request_set_user_header (req, "Foo: bar") works just like
274 request_set_header (req, "Foo", "bar"). */
277 request_set_user_header (struct request *req, const char *header)
280 const char *p = strchr (header, ':');
283 BOUNDED_TO_ALLOCA (header, p, name);
285 while (c_isspace (*p))
287 request_set_header (req, xstrdup (name), (char *) p, rel_name);
290 /* Remove the header with specified name from REQ. Returns true if
291 the header was actually removed, false otherwise. */
294 request_remove_header (struct request *req, char *name)
297 for (i = 0; i < req->hcount; i++)
299 struct request_header *hdr = &req->headers[i];
300 if (0 == strcasecmp (name, hdr->name))
302 release_header (hdr);
303 /* Move the remaining headers by one. */
304 if (i < req->hcount - 1)
305 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
313 #define APPEND(p, str) do { \
314 int A_len = strlen (str); \
315 memcpy (p, str, A_len); \
319 /* Construct the request and write it to FD using fd_write. */
322 request_send (const struct request *req, int fd)
324 char *request_string, *p;
325 int i, size, write_error;
327 /* Count the request size. */
330 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
331 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
333 for (i = 0; i < req->hcount; i++)
335 struct request_header *hdr = &req->headers[i];
336 /* NAME ": " VALUE "\r\n" */
337 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
343 p = request_string = alloca_array (char, size);
345 /* Generate the request. */
347 APPEND (p, req->method); *p++ = ' ';
348 APPEND (p, req->arg); *p++ = ' ';
349 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
351 for (i = 0; i < req->hcount; i++)
353 struct request_header *hdr = &req->headers[i];
354 APPEND (p, hdr->name);
355 *p++ = ':', *p++ = ' ';
356 APPEND (p, hdr->value);
357 *p++ = '\r', *p++ = '\n';
360 *p++ = '\r', *p++ = '\n', *p++ = '\0';
361 assert (p - request_string == size);
365 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
367 /* Send the request to the server. */
369 write_error = fd_write (fd, request_string, size - 1, -1);
371 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
376 /* Release the resources used by REQ. */
379 request_free (struct request *req)
382 xfree_null (req->arg);
383 for (i = 0; i < req->hcount; i++)
384 release_header (&req->headers[i]);
385 xfree_null (req->headers);
389 static struct hash_table *basic_authed_hosts;
391 /* Find out if this host has issued a Basic challenge yet; if so, give
392 * it the username, password. A temporary measure until we can get
393 * proper authentication in place. */
396 maybe_send_basic_creds (const char *hostname, const char *user,
397 const char *passwd, struct request *req)
399 bool do_challenge = false;
401 if (opt.auth_without_challenge)
403 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
406 else if (basic_authed_hosts
407 && hash_table_contains(basic_authed_hosts, hostname))
409 DEBUGP(("Found `%s' in basic_authed_hosts.\n", hostname));
414 DEBUGP(("Host `%s' has not issued a general basic challenge.\n",
419 request_set_header (req, "Authorization",
420 basic_authentication_encode (user, passwd),
427 register_basic_auth_host (const char *hostname)
429 if (!basic_authed_hosts)
431 basic_authed_hosts = make_nocase_string_hash_table (1);
433 if (!hash_table_contains(basic_authed_hosts, hostname))
435 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
436 DEBUGP(("Inserted `%s' into basic_authed_hosts\n", hostname));
441 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
442 PROMISED_SIZE bytes are sent over the wire -- if the file is
443 longer, read only that much; if the file is shorter, report an error. */
446 post_file (int sock, const char *file_name, wgint promised_size)
448 static char chunk[8192];
453 DEBUGP (("[writing POST file %s ... ", file_name));
455 fp = fopen (file_name, "rb");
458 while (!feof (fp) && written < promised_size)
461 int length = fread (chunk, 1, sizeof (chunk), fp);
464 towrite = MIN (promised_size - written, length);
465 write_error = fd_write (sock, chunk, towrite, -1);
475 /* If we've written less than was promised, report a (probably
476 nonsensical) error rather than break the promise. */
477 if (written < promised_size)
483 assert (written == promised_size);
484 DEBUGP (("done]\n"));
488 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
489 If so, return the pointer to the position after the line, otherwise
490 return NULL. This is used as callback to fd_read_hunk. The data
491 between START and PEEKED has been read and cannot be "unread"; the
492 data after PEEKED has only been peeked. */
495 response_head_terminator (const char *start, const char *peeked, int peeklen)
499 /* If at first peek, verify whether HUNK starts with "HTTP". If
500 not, this is a HTTP/0.9 request and we must bail out without
502 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
505 /* Look for "\n[\r]\n", and return the following position if found.
506 Start two chars before the current to cover the possibility that
507 part of the terminator (e.g. "\n\r") arrived in the previous
509 p = peeked - start < 2 ? start : peeked - 2;
510 end = peeked + peeklen;
512 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
513 for (; p < end - 2; p++)
516 if (p[1] == '\r' && p[2] == '\n')
518 else if (p[1] == '\n')
521 /* p==end-2: check for \n\n directly preceding END. */
522 if (p[0] == '\n' && p[1] == '\n')
528 /* The maximum size of a single HTTP response we care to read. Rather
529 than being a limit of the reader implementation, this limit
530 prevents Wget from slurping all available memory upon encountering
531 malicious or buggy server output, thus protecting the user. Define
532 it to 0 to remove the limit. */
534 #define HTTP_RESPONSE_MAX_SIZE 65536
536 /* Read the HTTP request head from FD and return it. The error
537 conditions are the same as with fd_read_hunk.
539 To support HTTP/0.9 responses, this function tries to make sure
540 that the data begins with "HTTP". If this is not the case, no data
541 is read and an empty request is returned, so that the remaining
542 data can be treated as body. */
545 read_http_response_head (int fd)
547 return fd_read_hunk (fd, response_head_terminator, 512,
548 HTTP_RESPONSE_MAX_SIZE);
552 /* The response data. */
555 /* The array of pointers that indicate where each header starts.
556 For example, given this HTTP response:
563 The headers are located like this:
565 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
567 headers[0] headers[1] headers[2] headers[3]
569 I.e. headers[0] points to the beginning of the request,
570 headers[1] points to the end of the first header and the
571 beginning of the second one, etc. */
573 const char **headers;
576 /* Create a new response object from the text of the HTTP response,
577 available in HEAD. That text is automatically split into
578 constituent header lines for fast retrieval using
581 static struct response *
582 resp_new (const char *head)
587 struct response *resp = xnew0 (struct response);
592 /* Empty head means that we're dealing with a headerless
593 (HTTP/0.9) response. In that case, don't set HEADERS at
598 /* Split HEAD into header lines, so that resp_header_* functions
599 don't need to do this over and over again. */
605 DO_REALLOC (resp->headers, size, count + 1, const char *);
606 resp->headers[count++] = hdr;
608 /* Break upon encountering an empty line. */
609 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
612 /* Find the end of HDR, including continuations. */
615 const char *end = strchr (hdr, '\n');
621 while (*hdr == ' ' || *hdr == '\t');
623 DO_REALLOC (resp->headers, size, count + 1, const char *);
624 resp->headers[count] = NULL;
629 /* Locate the header named NAME in the request data, starting with
630 position START. This allows the code to loop through the request
631 data, filtering for all requests of a given name. Returns the
632 found position, or -1 for failure. The code that uses this
633 function typically looks like this:
635 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
636 ... do something with header ...
638 If you only care about one header, use resp_header_get instead of
642 resp_header_locate (const struct response *resp, const char *name, int start,
643 const char **begptr, const char **endptr)
646 const char **headers = resp->headers;
649 if (!headers || !headers[1])
652 name_len = strlen (name);
658 for (; headers[i + 1]; i++)
660 const char *b = headers[i];
661 const char *e = headers[i + 1];
663 && b[name_len] == ':'
664 && 0 == strncasecmp (b, name, name_len))
667 while (b < e && c_isspace (*b))
669 while (b < e && c_isspace (e[-1]))
679 /* Find and retrieve the header named NAME in the request data. If
680 found, set *BEGPTR to its starting, and *ENDPTR to its ending
681 position, and return true. Otherwise return false.
683 This function is used as a building block for resp_header_copy
684 and resp_header_strdup. */
687 resp_header_get (const struct response *resp, const char *name,
688 const char **begptr, const char **endptr)
690 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
694 /* Copy the response header named NAME to buffer BUF, no longer than
695 BUFSIZE (BUFSIZE includes the terminating 0). If the header
696 exists, true is returned, false otherwise. If there should be no
697 limit on the size of the header, use resp_header_strdup instead.
699 If BUFSIZE is 0, no data is copied, but the boolean indication of
700 whether the header is present is still returned. */
703 resp_header_copy (const struct response *resp, const char *name,
704 char *buf, int bufsize)
707 if (!resp_header_get (resp, name, &b, &e))
711 int len = MIN (e - b, bufsize - 1);
712 memcpy (buf, b, len);
718 /* Return the value of header named NAME in RESP, allocated with
719 malloc. If such a header does not exist in RESP, return NULL. */
722 resp_header_strdup (const struct response *resp, const char *name)
725 if (!resp_header_get (resp, name, &b, &e))
727 return strdupdelim (b, e);
730 /* Parse the HTTP status line, which is of format:
732 HTTP-Version SP Status-Code SP Reason-Phrase
734 The function returns the status-code, or -1 if the status line
735 appears malformed. The pointer to "reason-phrase" message is
736 returned in *MESSAGE. */
739 resp_status (const struct response *resp, char **message)
746 /* For a HTTP/0.9 response, assume status 200. */
748 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
752 p = resp->headers[0];
753 end = resp->headers[1];
759 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
763 /* Match the HTTP version. This is optional because Gnutella
764 servers have been reported to not specify HTTP version. */
765 if (p < end && *p == '/')
768 while (p < end && c_isdigit (*p))
770 if (p < end && *p == '.')
772 while (p < end && c_isdigit (*p))
776 while (p < end && c_isspace (*p))
778 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
781 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
786 while (p < end && c_isspace (*p))
788 while (p < end && c_isspace (end[-1]))
790 *message = strdupdelim (p, end);
796 /* Release the resources used by RESP. */
799 resp_free (struct response *resp)
801 xfree_null (resp->headers);
805 /* Print a single line of response, the characters [b, e). We tried
807 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
808 but that failed to escape the non-printable characters and, in fact,
809 caused crashes in UTF-8 locales. */
812 print_response_line(const char *prefix, const char *b, const char *e)
815 BOUNDED_TO_ALLOCA(b, e, copy);
816 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
819 /* Print the server response, line by line, omitting the trailing CRLF
820 from individual header lines, and prefixed with PREFIX. */
823 print_server_response (const struct response *resp, const char *prefix)
828 for (i = 0; resp->headers[i + 1]; i++)
830 const char *b = resp->headers[i];
831 const char *e = resp->headers[i + 1];
833 if (b < e && e[-1] == '\n')
835 if (b < e && e[-1] == '\r')
837 print_response_line(prefix, b, e);
841 /* Parse the `Content-Range' header and extract the information it
842 contains. Returns true if successful, false otherwise. */
844 parse_content_range (const char *hdr, wgint *first_byte_ptr,
845 wgint *last_byte_ptr, wgint *entity_length_ptr)
849 /* Ancient versions of Netscape proxy server, presumably predating
850 rfc2068, sent out `Content-Range' without the "bytes"
852 if (0 == strncasecmp (hdr, "bytes", 5))
855 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
859 while (c_isspace (*hdr))
864 if (!c_isdigit (*hdr))
866 for (num = 0; c_isdigit (*hdr); hdr++)
867 num = 10 * num + (*hdr - '0');
868 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
870 *first_byte_ptr = num;
872 for (num = 0; c_isdigit (*hdr); hdr++)
873 num = 10 * num + (*hdr - '0');
874 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
876 *last_byte_ptr = num;
881 for (num = 0; c_isdigit (*hdr); hdr++)
882 num = 10 * num + (*hdr - '0');
883 *entity_length_ptr = num;
887 /* Read the body of the request, but don't store it anywhere and don't
888 display a progress gauge. This is useful for reading the bodies of
889 administrative responses to which we will soon issue another
890 request. The response is not useful to the user, but reading it
891 allows us to continue using the same connection to the server.
893 If reading fails, false is returned, true otherwise. In debug
894 mode, the body is displayed for debugging purposes. */
897 skip_short_body (int fd, wgint contlen)
900 SKIP_SIZE = 512, /* size of the download buffer */
901 SKIP_THRESHOLD = 4096 /* the largest size we read */
903 char dlbuf[SKIP_SIZE + 1];
904 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
906 /* We shouldn't get here with unknown contlen. (This will change
907 with HTTP/1.1, which supports "chunked" transfer.) */
908 assert (contlen != -1);
910 /* If the body is too large, it makes more sense to simply close the
911 connection than to try to read the body. */
912 if (contlen > SKIP_THRESHOLD)
915 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
919 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
922 /* Don't normally report the error since this is an
923 optimization that should be invisible to the user. */
924 DEBUGP (("] aborting (%s).\n",
925 ret < 0 ? fd_errstr (fd) : "EOF received"));
929 /* Safe even if %.*s bogusly expects terminating \0 because
930 we've zero-terminated dlbuf above. */
931 DEBUGP (("%.*s", ret, dlbuf));
934 DEBUGP (("] done.\n"));
938 /* Extract a parameter from the string (typically an HTTP header) at
939 **SOURCE and advance SOURCE to the next parameter. Return false
940 when there are no more parameters to extract. The name of the
941 parameter is returned in NAME, and the value in VALUE. If the
942 parameter has no value, the token's value is zeroed out.
944 For example, if *SOURCE points to the string "attachment;
945 filename=\"foo bar\"", the first call to this function will return
946 the token named "attachment" and no value, and the second call will
947 return the token named "filename" and value "foo bar". The third
948 call will return false, indicating no more valid tokens. */
951 extract_param (const char **source, param_token *name, param_token *value,
954 const char *p = *source;
956 while (c_isspace (*p)) ++p;
960 return false; /* no error; nothing more to extract */
965 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
967 if (name->b == name->e)
968 return false; /* empty name: error */
969 while (c_isspace (*p)) ++p;
970 if (*p == separator || !*p) /* no value */
973 if (*p == separator) ++p;
978 return false; /* error */
980 /* *p is '=', extract value */
982 while (c_isspace (*p)) ++p;
983 if (*p == '"') /* quoted */
986 while (*p && *p != '"') ++p;
990 /* Currently at closing quote; find the end of param. */
991 while (c_isspace (*p)) ++p;
992 while (*p && *p != separator) ++p;
996 /* garbage after closed quote, e.g. foo="bar"baz */
1002 while (*p && *p != separator) ++p;
1004 while (value->e != value->b && c_isspace (value->e[-1]))
1006 if (*p == separator) ++p;
1013 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1015 /* Parse the contents of the `Content-Disposition' header, extracting
1016 the information useful to Wget. Content-Disposition is a header
1017 borrowed from MIME; when used in HTTP, it typically serves for
1018 specifying the desired file name of the resource. For example:
1020 Content-Disposition: attachment; filename="flora.jpg"
1022 Wget will skip the tokens it doesn't care about, such as
1023 "attachment" in the previous example; it will also skip other
1024 unrecognized params. If the header is syntactically correct and
1025 contains a file name, a copy of the file name is stored in
1026 *filename and true is returned. Otherwise, the function returns
1029 The file name is stripped of directory components and must not be
1033 parse_content_disposition (const char *hdr, char **filename)
1035 param_token name, value;
1036 while (extract_param (&hdr, &name, &value, ';'))
1037 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1039 /* Make the file name begin at the last slash or backslash. */
1040 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1041 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1042 if (last_slash && last_bs)
1043 value.b = 1 + MAX (last_slash, last_bs);
1044 else if (last_slash || last_bs)
1045 value.b = 1 + (last_slash ? last_slash : last_bs);
1046 if (value.b == value.e)
1048 /* Start with the directory prefix, if specified. */
1051 int prefix_length = strlen (opt.dir_prefix);
1052 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1057 total_length = prefix_length + (value.e - value.b);
1058 *filename = xmalloc (total_length + 1);
1059 strcpy (*filename, opt.dir_prefix);
1061 (*filename)[prefix_length - 1] = '/';
1062 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1063 (*filename)[total_length] = '\0';
1066 *filename = strdupdelim (value.b, value.e);
1072 /* Persistent connections. Currently, we cache the most recently used
1073 connection as persistent, provided that the HTTP server agrees to
1074 make it such. The persistence data is stored in the variables
1075 below. Ideally, it should be possible to cache an arbitrary fixed
1076 number of these connections. */
1078 /* Whether a persistent connection is active. */
1079 static bool pconn_active;
1082 /* The socket of the connection. */
1085 /* Host and port of the currently active persistent connection. */
1089 /* Whether a ssl handshake has occoured on this connection. */
1092 /* Whether the connection was authorized. This is only done by
1093 NTLM, which authorizes *connections* rather than individual
1094 requests. (That practice is peculiar for HTTP, but it is a
1095 useful optimization.) */
1099 /* NTLM data of the current connection. */
1100 struct ntlmdata ntlm;
1104 /* Mark the persistent connection as invalid and free the resources it
1105 uses. This is used by the CLOSE_* macros after they forcefully
1106 close a registered persistent connection. */
1109 invalidate_persistent (void)
1111 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1112 pconn_active = false;
1113 fd_close (pconn.socket);
1118 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1119 persistent. This will enable someone to use the same connection
1120 later. In the context of HTTP, this must be called only AFTER the
1121 response has been received and the server has promised that the
1122 connection will remain alive.
1124 If a previous connection was persistent, it is closed. */
1127 register_persistent (const char *host, int port, int fd, bool ssl)
1131 if (pconn.socket == fd)
1133 /* The connection FD is already registered. */
1138 /* The old persistent connection is still active; close it
1139 first. This situation arises whenever a persistent
1140 connection exists, but we then connect to a different
1141 host, and try to register a persistent connection to that
1143 invalidate_persistent ();
1147 pconn_active = true;
1149 pconn.host = xstrdup (host);
1152 pconn.authorized = false;
1154 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1157 /* Return true if a persistent connection is available for connecting
1161 persistent_available_p (const char *host, int port, bool ssl,
1162 bool *host_lookup_failed)
1164 /* First, check whether a persistent connection is active at all. */
1168 /* If we want SSL and the last connection wasn't or vice versa,
1169 don't use it. Checking for host and port is not enough because
1170 HTTP and HTTPS can apparently coexist on the same port. */
1171 if (ssl != pconn.ssl)
1174 /* If we're not connecting to the same port, we're not interested. */
1175 if (port != pconn.port)
1178 /* If the host is the same, we're in business. If not, there is
1179 still hope -- read below. */
1180 if (0 != strcasecmp (host, pconn.host))
1182 /* Check if pconn.socket is talking to HOST under another name.
1183 This happens often when both sites are virtual hosts
1184 distinguished only by name and served by the same network
1185 interface, and hence the same web server (possibly set up by
1186 the ISP and serving many different web sites). This
1187 admittedly unconventional optimization does not contradict
1188 HTTP and works well with popular server software. */
1192 struct address_list *al;
1195 /* Don't try to talk to two different SSL sites over the same
1196 secure connection! (Besides, it's not clear that
1197 name-based virtual hosting is even possible with SSL.) */
1200 /* If pconn.socket's peer is one of the IP addresses HOST
1201 resolves to, pconn.socket is for all intents and purposes
1202 already talking to HOST. */
1204 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1206 /* Can't get the peer's address -- something must be very
1207 wrong with the connection. */
1208 invalidate_persistent ();
1211 al = lookup_host (host, 0);
1214 *host_lookup_failed = true;
1218 found = address_list_contains (al, &ip);
1219 address_list_release (al);
1224 /* The persistent connection's peer address was found among the
1225 addresses HOST resolved to; therefore, pconn.sock is in fact
1226 already talking to HOST -- no need to reconnect. */
1229 /* Finally, check whether the connection is still open. This is
1230 important because most servers implement liberal (short) timeout
1231 on persistent connections. Wget can of course always reconnect
1232 if the connection doesn't work out, but it's nicer to know in
1233 advance. This test is a logical followup of the first test, but
1234 is "expensive" and therefore placed at the end of the list.
1236 (Current implementation of test_socket_open has a nice side
1237 effect that it treats sockets with pending data as "closed".
1238 This is exactly what we want: if a broken server sends message
1239 body in response to HEAD, or if it sends more than conent-length
1240 data, we won't reuse the corrupted connection.) */
1242 if (!test_socket_open (pconn.socket))
1244 /* Oops, the socket is no longer open. Now that we know that,
1245 let's invalidate the persistent connection before returning
1247 invalidate_persistent ();
1254 /* The idea behind these two CLOSE macros is to distinguish between
1255 two cases: one when the job we've been doing is finished, and we
1256 want to close the connection and leave, and two when something is
1257 seriously wrong and we're closing the connection as part of
1260 In case of keep_alive, CLOSE_FINISH should leave the connection
1261 open, while CLOSE_INVALIDATE should still close it.
1263 Note that the semantics of the flag `keep_alive' is "this
1264 connection *will* be reused (the server has promised not to close
1265 the connection once we're done)", while the semantics of
1266 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1267 active, registered connection". */
1269 #define CLOSE_FINISH(fd) do { \
1272 if (pconn_active && (fd) == pconn.socket) \
1273 invalidate_persistent (); \
1282 #define CLOSE_INVALIDATE(fd) do { \
1283 if (pconn_active && (fd) == pconn.socket) \
1284 invalidate_persistent (); \
1292 wgint len; /* received length */
1293 wgint contlen; /* expected length */
1294 wgint restval; /* the restart value */
1295 int res; /* the result of last read */
1296 char *rderrmsg; /* error message from read error */
1297 char *newloc; /* new location (redirection) */
1298 char *remote_time; /* remote time-stamp string */
1299 char *error; /* textual HTTP error */
1300 int statcode; /* status code */
1301 wgint rd_size; /* amount of data read from socket */
1302 double dltime; /* time it took to download the data */
1303 const char *referer; /* value of the referer header. */
1304 char *local_file; /* local file name. */
1305 bool existence_checked; /* true if we already checked for a file's
1306 existence after having begun to download
1307 (needed in gethttp for when connection is
1308 interrupted/restarted. */
1309 bool timestamp_checked; /* true if pre-download time-stamping checks
1310 * have already been performed */
1311 char *orig_file_name; /* name of file to compare for time-stamping
1312 * (might be != local_file if -K is set) */
1313 wgint orig_file_size; /* size of file to compare for time-stamping */
1314 time_t orig_file_tstamp; /* time-stamp of file to compare for
1319 free_hstat (struct http_stat *hs)
1321 xfree_null (hs->newloc);
1322 xfree_null (hs->remote_time);
1323 xfree_null (hs->error);
1324 xfree_null (hs->rderrmsg);
1325 xfree_null (hs->local_file);
1326 xfree_null (hs->orig_file_name);
1328 /* Guard against being called twice. */
1330 hs->remote_time = NULL;
1334 #define BEGINS_WITH(line, string_constant) \
1335 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1336 && (c_isspace (line[sizeof (string_constant) - 1]) \
1337 || !line[sizeof (string_constant) - 1]))
1339 #define SET_USER_AGENT(req) do { \
1340 if (!opt.useragent) \
1341 request_set_header (req, "User-Agent", \
1342 aprintf ("Wget/%s", version_string), rel_value); \
1343 else if (*opt.useragent) \
1344 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1347 /* The flags that allow clobbering the file (opening with "wb").
1348 Defined here to avoid repetition later. #### This will require
1350 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1351 || opt.dirstruct || opt.output_document)
1353 /* Retrieve a document through HTTP protocol. It recognizes status
1354 code, and correctly handles redirections. It closes the network
1355 socket. If it receives an error from the functions below it, it
1356 will print it if there is enough information to do so (almost
1357 always), returning the error to the caller (i.e. http_loop).
1359 Various HTTP parameters are stored to hs.
1361 If PROXY is non-NULL, the connection will be made to the proxy
1362 server, and u->url will be requested. */
1364 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1366 struct request *req;
1369 char *user, *passwd;
1373 wgint contlen, contrange;
1380 /* Set to 1 when the authorization has already been sent and should
1381 not be tried again. */
1382 bool auth_finished = false;
1384 /* Set to 1 when just globally-set Basic authorization has been sent;
1385 * should prevent further Basic negotiations, but not other
1387 bool basic_auth_finished = false;
1389 /* Whether NTLM authentication is used for this request. */
1390 bool ntlm_seen = false;
1392 /* Whether our connection to the remote host is through SSL. */
1393 bool using_ssl = false;
1395 /* Whether a HEAD request will be issued (as opposed to GET or
1397 bool head_only = !!(*dt & HEAD_ONLY);
1400 struct response *resp;
1404 /* Whether this connection will be kept alive after the HTTP request
1408 /* Whether keep-alive should be inhibited.
1410 RFC 2068 requests that 1.0 clients not send keep-alive requests
1411 to proxies. This is because many 1.0 proxies do not interpret
1412 the Connection header and transfer it to the remote server,
1413 causing it to not close the connection and leave both the proxy
1414 and the client hanging. */
1415 bool inhibit_keep_alive =
1416 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1418 /* Headers sent when using POST. */
1419 wgint post_data_size = 0;
1421 bool host_lookup_failed = false;
1424 if (u->scheme == SCHEME_HTTPS)
1426 /* Initialize the SSL context. After this has once been done,
1427 it becomes a no-op. */
1430 scheme_disable (SCHEME_HTTPS);
1431 logprintf (LOG_NOTQUIET,
1432 _("Disabling SSL due to encountered errors.\n"));
1433 return SSLINITFAILED;
1436 #endif /* HAVE_SSL */
1438 /* Initialize certain elements of struct http_stat. */
1442 hs->rderrmsg = NULL;
1444 hs->remote_time = NULL;
1449 /* Prepare the request to send. */
1451 req = request_new ();
1454 const char *meth = "GET";
1457 else if (opt.post_file_name || opt.post_data)
1459 /* Use the full path, i.e. one that includes the leading slash and
1460 the query string. E.g. if u->path is "foo/bar" and u->query is
1461 "param=value", full_path will be "/foo/bar?param=value". */
1464 /* When using SSL over proxy, CONNECT establishes a direct
1465 connection to the HTTPS server. Therefore use the same
1466 argument as when talking to the server directly. */
1467 && u->scheme != SCHEME_HTTPS
1470 meth_arg = xstrdup (u->url);
1472 meth_arg = url_full_path (u);
1473 request_set_method (req, meth, meth_arg);
1476 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1477 if (*dt & SEND_NOCACHE)
1478 request_set_header (req, "Pragma", "no-cache", rel_none);
1480 request_set_header (req, "Range",
1481 aprintf ("bytes=%s-",
1482 number_to_static_string (hs->restval)),
1484 SET_USER_AGENT (req);
1485 request_set_header (req, "Accept", "*/*", rel_none);
1487 /* Find the username and password for authentication. */
1490 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1491 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1492 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1495 && !u->user) /* We only do "site-wide" authentication with "global"
1496 user/password values; URL user/password info overrides. */
1498 /* If this is a host for which we've already received a Basic
1499 * challenge, we'll go ahead and send Basic authentication creds. */
1500 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1506 char *proxy_user, *proxy_passwd;
1507 /* For normal username and password, URL components override
1508 command-line/wgetrc parameters. With proxy
1509 authentication, it's the reverse, because proxy URLs are
1510 normally the "permanent" ones, so command-line args
1511 should take precedence. */
1512 if (opt.proxy_user && opt.proxy_passwd)
1514 proxy_user = opt.proxy_user;
1515 proxy_passwd = opt.proxy_passwd;
1519 proxy_user = proxy->user;
1520 proxy_passwd = proxy->passwd;
1522 /* #### This does not appear right. Can't the proxy request,
1523 say, `Digest' authentication? */
1524 if (proxy_user && proxy_passwd)
1525 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1527 /* If we're using a proxy, we will be connecting to the proxy
1531 /* Proxy authorization over SSL is handled below. */
1533 if (u->scheme != SCHEME_HTTPS)
1535 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1538 /* Generate the Host header, HOST:PORT. Take into account that:
1540 - Broken server-side software often doesn't recognize the PORT
1541 argument, so we must generate "Host: www.server.com" instead of
1542 "Host: www.server.com:80" (and likewise for https port).
1544 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1545 becomes ambiguous and needs to be rewritten as "Host:
1546 [3ffe:8100:200:2::2]:1234". */
1548 /* Formats arranged for hfmt[add_port][add_squares]. */
1549 static const char *hfmt[][2] = {
1550 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1552 int add_port = u->port != scheme_default_port (u->scheme);
1553 int add_squares = strchr (u->host, ':') != NULL;
1554 request_set_header (req, "Host",
1555 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1559 if (!inhibit_keep_alive)
1560 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1563 request_set_header (req, "Cookie",
1564 cookie_header (wget_cookie_jar,
1565 u->host, u->port, u->path,
1567 u->scheme == SCHEME_HTTPS
1574 if (opt.post_data || opt.post_file_name)
1576 request_set_header (req, "Content-Type",
1577 "application/x-www-form-urlencoded", rel_none);
1579 post_data_size = strlen (opt.post_data);
1582 post_data_size = file_size (opt.post_file_name);
1583 if (post_data_size == -1)
1585 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1586 opt.post_file_name, strerror (errno));
1590 request_set_header (req, "Content-Length",
1591 xstrdup (number_to_static_string (post_data_size)),
1595 /* Add the user headers. */
1596 if (opt.user_headers)
1599 for (i = 0; opt.user_headers[i]; i++)
1600 request_set_user_header (req, opt.user_headers[i]);
1604 /* We need to come back here when the initial attempt to retrieve
1605 without authorization header fails. (Expected to happen at least
1606 for the Digest authorization scheme.) */
1610 /* Establish the connection. */
1612 if (!inhibit_keep_alive)
1614 /* Look for a persistent connection to target host, unless a
1615 proxy is used. The exception is when SSL is in use, in which
1616 case the proxy is nothing but a passthrough to the target
1617 host, registered as a connection to the latter. */
1618 struct url *relevant = conn;
1620 if (u->scheme == SCHEME_HTTPS)
1624 if (persistent_available_p (relevant->host, relevant->port,
1626 relevant->scheme == SCHEME_HTTPS,
1630 &host_lookup_failed))
1632 sock = pconn.socket;
1633 using_ssl = pconn.ssl;
1634 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1635 escnonprint (pconn.host), pconn.port);
1636 DEBUGP (("Reusing fd %d.\n", sock));
1637 if (pconn.authorized)
1638 /* If the connection is already authorized, the "Basic"
1639 authorization added by code above is unnecessary and
1641 request_remove_header (req, "Authorization");
1643 else if (host_lookup_failed)
1646 logprintf(LOG_NOTQUIET,
1647 _("%s: unable to resolve host address `%s'\n"),
1648 exec_name, relevant->host);
1655 sock = connect_to_host (conn->host, conn->port);
1664 return (retryable_socket_connect_error (errno)
1665 ? CONERROR : CONIMPOSSIBLE);
1669 if (proxy && u->scheme == SCHEME_HTTPS)
1671 /* When requesting SSL URLs through proxies, use the
1672 CONNECT method to request passthrough. */
1673 struct request *connreq = request_new ();
1674 request_set_method (connreq, "CONNECT",
1675 aprintf ("%s:%d", u->host, u->port));
1676 SET_USER_AGENT (connreq);
1679 request_set_header (connreq, "Proxy-Authorization",
1680 proxyauth, rel_value);
1681 /* Now that PROXYAUTH is part of the CONNECT request,
1682 zero it out so we don't send proxy authorization with
1683 the regular request below. */
1686 /* Examples in rfc2817 use the Host header in CONNECT
1687 requests. I don't see how that gains anything, given
1688 that the contents of Host would be exactly the same as
1689 the contents of CONNECT. */
1691 write_error = request_send (connreq, sock);
1692 request_free (connreq);
1693 if (write_error < 0)
1695 CLOSE_INVALIDATE (sock);
1699 head = read_http_response_head (sock);
1702 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1704 CLOSE_INVALIDATE (sock);
1713 DEBUGP (("proxy responded with: [%s]\n", head));
1715 resp = resp_new (head);
1716 statcode = resp_status (resp, &message);
1719 if (statcode != 200)
1722 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1723 message ? escnonprint (message) : "?");
1724 xfree_null (message);
1727 xfree_null (message);
1729 /* SOCK is now *really* connected to u->host, so update CONN
1730 to reflect this. That way register_persistent will
1731 register SOCK as being connected to u->host:u->port. */
1735 if (conn->scheme == SCHEME_HTTPS)
1737 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1744 #endif /* HAVE_SSL */
1747 /* Send the request to server. */
1748 write_error = request_send (req, sock);
1750 if (write_error >= 0)
1754 DEBUGP (("[POST data: %s]\n", opt.post_data));
1755 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1757 else if (opt.post_file_name && post_data_size != 0)
1758 write_error = post_file (sock, opt.post_file_name, post_data_size);
1761 if (write_error < 0)
1763 CLOSE_INVALIDATE (sock);
1767 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1768 proxy ? "Proxy" : "HTTP");
1773 head = read_http_response_head (sock);
1778 logputs (LOG_NOTQUIET, _("No data received.\n"));
1779 CLOSE_INVALIDATE (sock);
1785 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1787 CLOSE_INVALIDATE (sock);
1792 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1794 resp = resp_new (head);
1796 /* Check for status line. */
1798 statcode = resp_status (resp, &message);
1799 if (!opt.server_response)
1800 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1801 message ? escnonprint (message) : "");
1804 logprintf (LOG_VERBOSE, "\n");
1805 print_server_response (resp, " ");
1808 /* Determine the local filename if needed. Notice that if -O is used
1809 * hstat.local_file is set by http_loop to the argument of -O. */
1810 if (!hs->local_file)
1812 /* Honor Content-Disposition whether possible. */
1813 if (!opt.content_disposition
1814 || !resp_header_copy (resp, "Content-Disposition",
1815 hdrval, sizeof (hdrval))
1816 || !parse_content_disposition (hdrval, &hs->local_file))
1818 /* The Content-Disposition header is missing or broken.
1819 * Choose unique file name according to given URL. */
1820 hs->local_file = url_file_name (u);
1824 /* TODO: perform this check only once. */
1825 if (!hs->existence_checked && file_exists_p (hs->local_file))
1829 /* If opt.noclobber is turned on and file already exists, do not
1830 retrieve the file */
1831 logprintf (LOG_VERBOSE, _("\
1832 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1833 /* If the file is there, we suppose it's retrieved OK. */
1836 /* #### Bogusness alert. */
1837 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1838 if (has_html_suffix_p (hs->local_file))
1841 return RETRUNNEEDED;
1843 else if (!ALLOW_CLOBBER)
1845 char *unique = unique_name (hs->local_file, true);
1846 if (unique != hs->local_file)
1847 xfree (hs->local_file);
1848 hs->local_file = unique;
1851 hs->existence_checked = true;
1853 /* Support timestamping */
1854 /* TODO: move this code out of gethttp. */
1855 if (opt.timestamping && !hs->timestamp_checked)
1857 size_t filename_len = strlen (hs->local_file);
1858 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1859 bool local_dot_orig_file_exists = false;
1860 char *local_filename = NULL;
1863 if (opt.backup_converted)
1864 /* If -K is specified, we'll act on the assumption that it was specified
1865 last time these files were downloaded as well, and instead of just
1866 comparing local file X against server file X, we'll compare local
1867 file X.orig (if extant, else X) against server file X. If -K
1868 _wasn't_ specified last time, or the server contains files called
1869 *.orig, -N will be back to not operating correctly with -k. */
1871 /* Would a single s[n]printf() call be faster? --dan
1873 Definitely not. sprintf() is horribly slow. It's a
1874 different question whether the difference between the two
1875 affects a program. Usually I'd say "no", but at one
1876 point I profiled Wget, and found that a measurable and
1877 non-negligible amount of time was lost calling sprintf()
1878 in url.c. Replacing sprintf with inline calls to
1879 strcpy() and number_to_string() made a difference.
1881 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1882 memcpy (filename_plus_orig_suffix + filename_len,
1883 ".orig", sizeof (".orig"));
1885 /* Try to stat() the .orig file. */
1886 if (stat (filename_plus_orig_suffix, &st) == 0)
1888 local_dot_orig_file_exists = true;
1889 local_filename = filename_plus_orig_suffix;
1893 if (!local_dot_orig_file_exists)
1894 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1895 if (stat (hs->local_file, &st) == 0)
1896 local_filename = hs->local_file;
1898 if (local_filename != NULL)
1899 /* There was a local file, so we'll check later to see if the version
1900 the server has is the same version we already have, allowing us to
1903 hs->orig_file_name = xstrdup (local_filename);
1904 hs->orig_file_size = st.st_size;
1905 hs->orig_file_tstamp = st.st_mtime;
1907 /* Modification time granularity is 2 seconds for Windows, so
1908 increase local time by 1 second for later comparison. */
1909 ++hs->orig_file_tstamp;
1914 if (!opt.ignore_length
1915 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1919 parsed = str_to_wgint (hdrval, NULL, 10);
1920 if (parsed == WGINT_MAX && errno == ERANGE)
1923 #### If Content-Length is out of range, it most likely
1924 means that the file is larger than 2G and that we're
1925 compiled without LFS. In that case we should probably
1926 refuse to even attempt to download the file. */
1929 else if (parsed < 0)
1931 /* Negative Content-Length; nonsensical, so we can't
1932 assume any information about the content to receive. */
1939 /* Check for keep-alive related responses. */
1940 if (!inhibit_keep_alive && contlen != -1)
1942 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1944 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1946 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1951 /* The server has promised that it will not close the connection
1952 when we're done. This means that we can register it. */
1953 register_persistent (conn->host, conn->port, sock, using_ssl);
1955 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1957 /* Authorization is required. */
1958 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1959 CLOSE_FINISH (sock);
1961 CLOSE_INVALIDATE (sock);
1962 pconn.authorized = false;
1963 if (!auth_finished && (user && passwd))
1965 /* IIS sends multiple copies of WWW-Authenticate, one with
1966 the value "negotiate", and other(s) with data. Loop over
1967 all the occurrences and pick the one we recognize. */
1969 const char *wabeg, *waend;
1970 char *www_authenticate = NULL;
1972 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1973 &wabeg, &waend)) != -1;
1975 if (known_authentication_scheme_p (wabeg, waend))
1977 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1981 if (!www_authenticate)
1983 /* If the authentication header is missing or
1984 unrecognized, there's no sense in retrying. */
1985 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1987 else if (!basic_auth_finished
1988 || !BEGINS_WITH (www_authenticate, "Basic"))
1991 pth = url_full_path (u);
1992 request_set_header (req, "Authorization",
1993 create_authorization_line (www_authenticate,
1995 request_method (req),
1999 if (BEGINS_WITH (www_authenticate, "NTLM"))
2001 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2003 /* Need to register this host as using basic auth,
2004 * so we automatically send creds next time. */
2005 register_basic_auth_host (u->host);
2008 goto retry_with_auth;
2012 /* We already did Basic auth, and it failed. Gotta
2016 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2020 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2022 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2024 pconn.authorized = true;
2028 hs->statcode = statcode;
2030 hs->error = xstrdup (_("Malformed status line"));
2032 hs->error = xstrdup (_("(no description)"));
2034 hs->error = xstrdup (message);
2035 xfree_null (message);
2037 type = resp_header_strdup (resp, "Content-Type");
2040 char *tmp = strchr (type, ';');
2043 while (tmp > type && c_isspace (tmp[-1]))
2048 hs->newloc = resp_header_strdup (resp, "Location");
2049 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2051 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2055 const char *scbeg, *scend;
2056 /* The jar should have been created by now. */
2057 assert (wget_cookie_jar != NULL);
2059 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2060 &scbeg, &scend)) != -1;
2063 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2064 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2065 u->path, set_cookie);
2069 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2071 wgint first_byte_pos, last_byte_pos, entity_length;
2072 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2075 contrange = first_byte_pos;
2076 contlen = last_byte_pos - first_byte_pos + 1;
2081 /* 20x responses are counted among successful by default. */
2082 if (H_20X (statcode))
2085 /* Return if redirected. */
2086 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2088 /* RFC2068 says that in case of the 300 (multiple choices)
2089 response, the server can output a preferred URL through
2090 `Location' header; otherwise, the request should be treated
2091 like GET. So, if the location is set, it will be a
2092 redirection; otherwise, just proceed normally. */
2093 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2097 logprintf (LOG_VERBOSE,
2098 _("Location: %s%s\n"),
2099 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2100 hs->newloc ? _(" [following]") : "");
2101 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2102 CLOSE_FINISH (sock);
2104 CLOSE_INVALIDATE (sock);
2110 /* If content-type is not given, assume text/html. This is because
2111 of the multitude of broken CGI's that "forget" to generate the
2114 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2115 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2121 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2126 if (opt.html_extension)
2129 /* -E / --html-extension / html_extension = on was specified,
2130 and this is a text/html file. If some case-insensitive
2131 variation on ".htm[l]" isn't already the file's suffix,
2134 ensure_extension (hs, ".html", dt);
2136 else if (*dt & TEXTCSS)
2138 ensure_extension (hs, ".css", dt);
2142 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2144 /* If `-c' is in use and the file has been fully downloaded (or
2145 the remote file has shrunk), Wget effectively requests bytes
2146 after the end of file and the server response with 416. */
2147 logputs (LOG_VERBOSE, _("\
2148 \n The file is already fully retrieved; nothing to do.\n\n"));
2149 /* In case the caller inspects. */
2152 /* Mark as successfully retrieved. */
2155 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2156 might be more bytes in the body. */
2157 return RETRUNNEEDED;
2159 if ((contrange != 0 && contrange != hs->restval)
2160 || (H_PARTIAL (statcode) && !contrange))
2162 /* The Range request was somehow misunderstood by the server.
2165 CLOSE_INVALIDATE (sock);
2171 hs->contlen = contlen + contrange;
2177 /* No need to print this output if the body won't be
2178 downloaded at all, or if the original server response is
2180 logputs (LOG_VERBOSE, _("Length: "));
2183 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2184 if (contlen + contrange >= 1024)
2185 logprintf (LOG_VERBOSE, " (%s)",
2186 human_readable (contlen + contrange));
2189 if (contlen >= 1024)
2190 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2191 number_to_static_string (contlen),
2192 human_readable (contlen));
2194 logprintf (LOG_VERBOSE, _(", %s remaining"),
2195 number_to_static_string (contlen));
2199 logputs (LOG_VERBOSE,
2200 opt.ignore_length ? _("ignored") : _("unspecified"));
2202 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2204 logputs (LOG_VERBOSE, "\n");
2208 type = NULL; /* We don't need it any more. */
2210 /* Return if we have no intention of further downloading. */
2211 if (!(*dt & RETROKF) || head_only)
2213 /* In case the caller cares to look... */
2218 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2219 servers not to send body in response to a HEAD request, and
2220 those that do will likely be caught by test_socket_open.
2221 If not, they can be worked around using
2222 `--no-http-keep-alive'. */
2223 CLOSE_FINISH (sock);
2224 else if (keep_alive && skip_short_body (sock, contlen))
2225 /* Successfully skipped the body; also keep using the socket. */
2226 CLOSE_FINISH (sock);
2228 CLOSE_INVALIDATE (sock);
2229 return RETRFINISHED;
2232 /* Open the local file. */
2235 mkalldirs (hs->local_file);
2237 rotate_backups (hs->local_file);
2239 fp = fopen (hs->local_file, "ab");
2240 else if (ALLOW_CLOBBER)
2241 fp = fopen (hs->local_file, "wb");
2244 fp = fopen_excl (hs->local_file, true);
2245 if (!fp && errno == EEXIST)
2247 /* We cannot just invent a new name and use it (which is
2248 what functions like unique_create typically do)
2249 because we told the user we'd use this name.
2250 Instead, return and retry the download. */
2251 logprintf (LOG_NOTQUIET,
2252 _("%s has sprung into existence.\n"),
2254 CLOSE_INVALIDATE (sock);
2255 return FOPEN_EXCL_ERR;
2260 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2261 CLOSE_INVALIDATE (sock);
2268 /* Print fetch message, if opt.verbose. */
2271 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2272 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2275 /* This confuses the timestamping code that checks for file size.
2276 #### The timestamping code should be smarter about file size. */
2277 if (opt.save_headers && hs->restval == 0)
2278 fwrite (head, 1, strlen (head), fp);
2280 /* Now we no longer need to store the response header. */
2283 /* Download the request body. */
2286 /* If content-length is present, read that much; otherwise, read
2287 until EOF. The HTTP spec doesn't require the server to
2288 actually close the connection when it's done sending data. */
2289 flags |= rb_read_exactly;
2290 if (hs->restval > 0 && contrange == 0)
2291 /* If the server ignored our range request, instruct fd_read_body
2292 to skip the first RESTVAL bytes of body. */
2293 flags |= rb_skip_startpos;
2294 hs->len = hs->restval;
2296 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2297 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2301 CLOSE_FINISH (sock);
2305 hs->rderrmsg = xstrdup (fd_errstr (sock));
2306 CLOSE_INVALIDATE (sock);
2313 return RETRFINISHED;
2316 /* The genuine HTTP loop! This is the part where the retrieval is
2317 retried, and retried, and retried, and... */
2319 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2320 int *dt, struct url *proxy)
2323 bool got_head = false; /* used for time-stamping and filename detection */
2324 bool time_came_from_head = false;
2325 bool got_name = false;
2328 uerr_t err, ret = TRYLIMEXC;
2329 time_t tmr = -1; /* remote time-stamp */
2330 struct http_stat hstat; /* HTTP status */
2332 bool send_head_first = true;
2334 /* Assert that no value for *LOCAL_FILE was passed. */
2335 assert (local_file == NULL || *local_file == NULL);
2337 /* Set LOCAL_FILE parameter. */
2338 if (local_file && opt.output_document)
2339 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2341 /* Reset NEWLOC parameter. */
2344 /* This used to be done in main(), but it's a better idea to do it
2345 here so that we don't go through the hoops if we're just using
2350 /* Warn on (likely bogus) wildcard usage in HTTP. */
2351 if (opt.ftp_glob && has_wildcards_p (u->path))
2352 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2354 /* Setup hstat struct. */
2356 hstat.referer = referer;
2358 if (opt.output_document)
2360 hstat.local_file = xstrdup (opt.output_document);
2363 else if (!opt.content_disposition)
2365 hstat.local_file = url_file_name (u);
2369 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2370 * screaming for some refactoring. */
2371 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber)
2373 /* If opt.noclobber is turned on and file already exists, do not
2374 retrieve the file */
2375 logprintf (LOG_VERBOSE, _("\
2376 File `%s' already there; not retrieving.\n\n"),
2378 /* If the file is there, we suppose it's retrieved OK. */
2381 /* #### Bogusness alert. */
2382 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2383 if (has_html_suffix_p (hstat.local_file))
2386 return RETRUNNEEDED;
2389 /* Reset the counter. */
2392 /* Reset the document type. */
2395 /* Skip preliminary HEAD request if we're not in spider mode AND
2396 * if -O was given or HTTP Content-Disposition support is disabled. */
2398 && (got_name || !opt.content_disposition))
2399 send_head_first = false;
2401 /* Send preliminary HEAD request if -N is given and we have an existing
2402 * destination file. */
2403 if (opt.timestamping
2404 && !opt.content_disposition
2405 && file_exists_p (url_file_name (u)))
2406 send_head_first = true;
2411 /* Increment the pass counter. */
2413 sleep_between_retrievals (count);
2415 /* Get the current time string. */
2416 tms = datetime_str (time (NULL));
2418 if (opt.spider && !got_head)
2419 logprintf (LOG_VERBOSE, _("\
2420 Spider mode enabled. Check if remote file exists.\n"));
2422 /* Print fetch message, if opt.verbose. */
2425 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2430 sprintf (tmp, _("(try:%2d)"), count);
2431 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2436 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2441 ws_changetitle (hurl);
2446 /* Default document type is empty. However, if spider mode is
2447 on or time-stamping is employed, HEAD_ONLY commands is
2448 encoded within *dt. */
2449 if (send_head_first && !got_head)
2454 /* Decide whether or not to restart. */
2457 && stat (hstat.local_file, &st) == 0
2458 && S_ISREG (st.st_mode))
2459 /* When -c is used, continue from on-disk size. (Can't use
2460 hstat.len even if count>1 because we don't want a failed
2461 first attempt to clobber existing data.) */
2462 hstat.restval = st.st_size;
2464 /* otherwise, continue where the previous try left off */
2465 hstat.restval = hstat.len;
2469 /* Decide whether to send the no-cache directive. We send it in
2471 a) we're using a proxy, and we're past our first retrieval.
2472 Some proxies are notorious for caching incomplete data, so
2473 we require a fresh get.
2474 b) caching is explicitly inhibited. */
2475 if ((proxy && count > 1) /* a */
2476 || !opt.allow_cache) /* b */
2477 *dt |= SEND_NOCACHE;
2479 *dt &= ~SEND_NOCACHE;
2481 /* Try fetching the document, or at least its head. */
2482 err = gethttp (u, &hstat, dt, proxy);
2485 tms = datetime_str (time (NULL));
2487 /* Get the new location (with or without the redirection). */
2489 *newloc = xstrdup (hstat.newloc);
2493 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2494 case CONERROR: case READERR: case WRITEFAILED:
2495 case RANGEERR: case FOPEN_EXCL_ERR:
2496 /* Non-fatal errors continue executing the loop, which will
2497 bring them to "while" statement at the end, to judge
2498 whether the number of tries was exceeded. */
2499 printwhat (count, opt.ntry);
2501 case FWRITEERR: case FOPENERR:
2502 /* Another fatal error. */
2503 logputs (LOG_VERBOSE, "\n");
2504 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2505 hstat.local_file, strerror (errno));
2506 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2507 case SSLINITFAILED: case CONTNOTSUPPORTED:
2508 /* Fatal errors just return from the function. */
2512 /* Another fatal error. */
2513 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2517 /* Return the new location to the caller. */
2520 logprintf (LOG_NOTQUIET,
2521 _("ERROR: Redirection (%d) without location.\n"),
2531 /* The file was already fully retrieved. */
2535 /* Deal with you later. */
2538 /* All possibilities should have been exhausted. */
2542 if (!(*dt & RETROKF))
2547 /* #### Ugly ugly ugly! */
2548 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2549 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2552 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2554 && (hstat.statcode == 500 || hstat.statcode == 501))
2559 /* Maybe we should always keep track of broken links, not just in
2561 else if (opt.spider)
2563 /* #### Again: ugly ugly ugly! */
2565 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2566 nonexisting_url (hurl);
2567 logprintf (LOG_NOTQUIET, _("\
2568 Remote file does not exist -- broken link!!!\n"));
2572 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2573 tms, hstat.statcode, escnonprint (hstat.error));
2575 logputs (LOG_VERBOSE, "\n");
2581 /* Did we get the time-stamp? */
2584 got_head = true; /* no more time-stamping */
2586 if (opt.timestamping && !hstat.remote_time)
2588 logputs (LOG_NOTQUIET, _("\
2589 Last-modified header missing -- time-stamps turned off.\n"));
2591 else if (hstat.remote_time)
2593 /* Convert the date-string into struct tm. */
2594 tmr = http_atotm (hstat.remote_time);
2595 if (tmr == (time_t) (-1))
2596 logputs (LOG_VERBOSE, _("\
2597 Last-modified header invalid -- time-stamp ignored.\n"));
2598 if (*dt & HEAD_ONLY)
2599 time_came_from_head = true;
2602 if (send_head_first)
2604 /* The time-stamping section. */
2605 if (opt.timestamping)
2607 if (hstat.orig_file_name) /* Perform the following
2608 checks only if the file
2610 download already exists. */
2612 if (hstat.remote_time &&
2613 tmr != (time_t) (-1))
2615 /* Now time-stamping can be used validly.
2616 Time-stamping means that if the sizes of
2617 the local and remote file match, and local
2618 file is newer than the remote file, it will
2619 not be retrieved. Otherwise, the normal
2620 download procedure is resumed. */
2621 if (hstat.orig_file_tstamp >= tmr)
2623 if (hstat.contlen == -1
2624 || hstat.orig_file_size == hstat.contlen)
2626 logprintf (LOG_VERBOSE, _("\
2627 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2628 hstat.orig_file_name);
2634 logprintf (LOG_VERBOSE, _("\
2635 The sizes do not match (local %s) -- retrieving.\n"),
2636 number_to_static_string (hstat.orig_file_size));
2640 logputs (LOG_VERBOSE,
2641 _("Remote file is newer, retrieving.\n"));
2643 logputs (LOG_VERBOSE, "\n");
2647 /* free_hstat (&hstat); */
2648 hstat.timestamp_checked = true;
2657 logputs (LOG_VERBOSE, _("\
2658 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2662 logprintf (LOG_VERBOSE, _("\
2663 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2664 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2672 logprintf (LOG_VERBOSE, _("\
2673 Remote file exists and could contain further links,\n\
2674 but recursion is disabled -- not retrieving.\n\n"));
2678 logprintf (LOG_VERBOSE, _("\
2679 Remote file exists.\n\n"));
2681 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2688 count = 0; /* the retrieve count for HEAD is reset */
2690 } /* send_head_first */
2693 if ((tmr != (time_t) (-1))
2694 && ((hstat.len == hstat.contlen) ||
2695 ((hstat.res == 0) && (hstat.contlen == -1))))
2697 /* #### This code repeats in http.c and ftp.c. Move it to a
2699 const char *fl = NULL;
2700 if (opt.output_document)
2702 if (output_stream_regular)
2703 fl = opt.output_document;
2706 fl = hstat.local_file;
2710 /* Reparse time header, in case it's changed. */
2711 if (time_came_from_head
2712 && hstat.remote_time && hstat.remote_time[0])
2714 newtmr = http_atotm (hstat.remote_time);
2721 /* End of time-stamping section. */
2723 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2724 total_download_time += hstat.dltime;
2726 if (hstat.len == hstat.contlen)
2730 logprintf (LOG_VERBOSE,
2731 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2732 tms, tmrate, hstat.local_file,
2733 number_to_static_string (hstat.len),
2734 number_to_static_string (hstat.contlen));
2735 logprintf (LOG_NONVERBOSE,
2736 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2738 number_to_static_string (hstat.len),
2739 number_to_static_string (hstat.contlen),
2740 hstat.local_file, count);
2743 total_downloaded_bytes += hstat.len;
2745 /* Remember that we downloaded the file for later ".orig" code. */
2746 if (*dt & ADDED_HTML_EXTENSION)
2747 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2749 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2754 else if (hstat.res == 0) /* No read error */
2756 if (hstat.contlen == -1) /* We don't know how much we were supposed
2757 to get, so assume we succeeded. */
2761 logprintf (LOG_VERBOSE,
2762 _("%s (%s) - `%s' saved [%s]\n\n"),
2763 tms, tmrate, hstat.local_file,
2764 number_to_static_string (hstat.len));
2765 logprintf (LOG_NONVERBOSE,
2766 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2767 tms, u->url, number_to_static_string (hstat.len),
2768 hstat.local_file, count);
2771 total_downloaded_bytes += hstat.len;
2773 /* Remember that we downloaded the file for later ".orig" code. */
2774 if (*dt & ADDED_HTML_EXTENSION)
2775 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2777 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2782 else if (hstat.len < hstat.contlen) /* meaning we lost the
2783 connection too soon */
2785 logprintf (LOG_VERBOSE,
2786 _("%s (%s) - Connection closed at byte %s. "),
2787 tms, tmrate, number_to_static_string (hstat.len));
2788 printwhat (count, opt.ntry);
2792 /* Getting here would mean reading more data than
2793 requested with content-length, which we never do. */
2796 else /* from now on hstat.res can only be -1 */
2798 if (hstat.contlen == -1)
2800 logprintf (LOG_VERBOSE,
2801 _("%s (%s) - Read error at byte %s (%s)."),
2802 tms, tmrate, number_to_static_string (hstat.len),
2804 printwhat (count, opt.ntry);
2807 else /* hstat.res == -1 and contlen is given */
2809 logprintf (LOG_VERBOSE,
2810 _("%s (%s) - Read error at byte %s/%s (%s). "),
2812 number_to_static_string (hstat.len),
2813 number_to_static_string (hstat.contlen),
2815 printwhat (count, opt.ntry);
2821 while (!opt.ntry || (count < opt.ntry));
2825 *local_file = xstrdup (hstat.local_file);
2826 free_hstat (&hstat);
2831 /* Check whether the result of strptime() indicates success.
2832 strptime() returns the pointer to how far it got to in the string.
2833 The processing has been successful if the string is at `GMT' or
2834 `+X', or at the end of the string.
2836 In extended regexp parlance, the function returns 1 if P matches
2837 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2838 can return) is considered a failure and 0 is returned. */
2840 check_end (const char *p)
2844 while (c_isspace (*p))
2847 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2848 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2854 /* Convert the textual specification of time in TIME_STRING to the
2855 number of seconds since the Epoch.
2857 TIME_STRING can be in any of the three formats RFC2616 allows the
2858 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2859 as well as the time format used in the Set-Cookie header.
2860 Timezones are ignored, and should be GMT.
2862 Return the computed time_t representation, or -1 if the conversion
2865 This function uses strptime with various string formats for parsing
2866 TIME_STRING. This results in a parser that is not as lenient in
2867 interpreting TIME_STRING as I would like it to be. Being based on
2868 strptime, it always allows shortened months, one-digit days, etc.,
2869 but due to the multitude of formats in which time can be
2870 represented, an ideal HTTP time parser would be even more
2871 forgiving. It should completely ignore things like week days and
2872 concentrate only on the various forms of representing years,
2873 months, days, hours, minutes, and seconds. For example, it would
2874 be nice if it accepted ISO 8601 out of the box.
2876 I've investigated free and PD code for this purpose, but none was
2877 usable. getdate was big and unwieldy, and had potential copyright
2878 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2879 distributed with phttpd, is excellent, but we cannot use it because
2880 it is not assigned to the FSF. So I stuck it with strptime. */
2883 http_atotm (const char *time_string)
2885 /* NOTE: Solaris strptime man page claims that %n and %t match white
2886 space, but that's not universally available. Instead, we simply
2887 use ` ' to mean "skip all WS", which works under all strptime
2888 implementations I've tested. */
2890 static const char *time_formats[] = {
2891 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2892 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2893 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2894 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2895 (used in Set-Cookie, defined in the
2896 Netscape cookie specification.) */
2898 const char *oldlocale;
2900 time_t ret = (time_t) -1;
2902 /* Solaris strptime fails to recognize English month names in
2903 non-English locales, which we work around by temporarily setting
2904 locale to C before invoking strptime. */
2905 oldlocale = setlocale (LC_TIME, NULL);
2906 setlocale (LC_TIME, "C");
2908 for (i = 0; i < countof (time_formats); i++)
2912 /* Some versions of strptime use the existing contents of struct
2913 tm to recalculate the date according to format. Zero it out
2914 to prevent stack garbage from influencing strptime. */
2917 if (check_end (strptime (time_string, time_formats[i], &t)))
2924 /* Restore the previous locale. */
2925 setlocale (LC_TIME, oldlocale);
2930 /* Authorization support: We support three authorization schemes:
2932 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2934 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2935 consisting of answering to the server's challenge with the proper
2938 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2939 Stenberg for libcurl. Like digest, NTLM is based on a
2940 challenge-response mechanism, but unlike digest, it is non-standard
2941 (authenticates TCP connections rather than requests), undocumented
2942 and Microsoft-specific. */
2944 /* Create the authentication header contents for the `Basic' scheme.
2945 This is done by encoding the string "USER:PASS" to base64 and
2946 prepending the string "Basic " in front of it. */
2949 basic_authentication_encode (const char *user, const char *passwd)
2952 int len1 = strlen (user) + 1 + strlen (passwd);
2954 t1 = (char *)alloca (len1 + 1);
2955 sprintf (t1, "%s:%s", user, passwd);
2957 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2958 base64_encode (t1, len1, t2);
2960 return concat_strings ("Basic ", t2, (char *) 0);
2963 #define SKIP_WS(x) do { \
2964 while (c_isspace (*(x))) \
2968 #ifdef ENABLE_DIGEST
2969 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2970 an array of 16 bytes containing the hash keys, and BUF should be a
2971 buffer of 33 writable characters (32 for hex digits plus one for
2972 zero termination). */
2974 dump_hash (char *buf, const unsigned char *hash)
2978 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2980 *buf++ = XNUM_TO_digit (*hash >> 4);
2981 *buf++ = XNUM_TO_digit (*hash & 0xf);
2986 /* Take the line apart to find the challenge, and compose a digest
2987 authorization header. See RFC2069 section 2.1.2. */
2989 digest_authentication_encode (const char *au, const char *user,
2990 const char *passwd, const char *method,
2993 static char *realm, *opaque, *nonce;
2998 { "realm", &realm },
2999 { "opaque", &opaque },
3003 param_token name, value;
3005 realm = opaque = nonce = NULL;
3007 au += 6; /* skip over `Digest' */
3008 while (extract_param (&au, &name, &value, ','))
3011 for (i = 0; i < countof (options); i++)
3012 if (name.e - name.b == strlen (options[i].name)
3013 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
3015 *options[i].variable = strdupdelim (value.b, value.e);
3019 if (!realm || !nonce || !user || !passwd || !path || !method)
3022 xfree_null (opaque);
3027 /* Calculate the digest value. */
3029 ALLOCA_MD5_CONTEXT (ctx);
3030 unsigned char hash[MD5_HASHLEN];
3031 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3032 char response_digest[MD5_HASHLEN * 2 + 1];
3034 /* A1BUF = H(user ":" realm ":" password) */
3036 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3037 gen_md5_update ((unsigned char *)":", 1, ctx);
3038 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3039 gen_md5_update ((unsigned char *)":", 1, ctx);
3040 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3041 gen_md5_finish (ctx, hash);
3042 dump_hash (a1buf, hash);
3044 /* A2BUF = H(method ":" path) */
3046 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3047 gen_md5_update ((unsigned char *)":", 1, ctx);
3048 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3049 gen_md5_finish (ctx, hash);
3050 dump_hash (a2buf, hash);
3052 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3054 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3055 gen_md5_update ((unsigned char *)":", 1, ctx);
3056 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3057 gen_md5_update ((unsigned char *)":", 1, ctx);
3058 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3059 gen_md5_finish (ctx, hash);
3060 dump_hash (response_digest, hash);
3062 res = xmalloc (strlen (user)
3067 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3068 + (opaque ? strlen (opaque) : 0)
3070 sprintf (res, "Digest \
3071 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3072 user, realm, nonce, path, response_digest);
3075 char *p = res + strlen (res);
3076 strcat (p, ", opaque=\"");
3083 #endif /* ENABLE_DIGEST */
3085 /* Computing the size of a string literal must take into account that
3086 value returned by sizeof includes the terminating \0. */
3087 #define STRSIZE(literal) (sizeof (literal) - 1)
3089 /* Whether chars in [b, e) begin with the literal string provided as
3090 first argument and are followed by whitespace or terminating \0.
3091 The comparison is case-insensitive. */
3092 #define STARTS(literal, b, e) \
3093 ((e) - (b) >= STRSIZE (literal) \
3094 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3095 && ((e) - (b) == STRSIZE (literal) \
3096 || c_isspace (b[STRSIZE (literal)])))
3099 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3101 return STARTS ("Basic", hdrbeg, hdrend)
3102 #ifdef ENABLE_DIGEST
3103 || STARTS ("Digest", hdrbeg, hdrend)
3106 || STARTS ("NTLM", hdrbeg, hdrend)
3113 /* Create the HTTP authorization request header. When the
3114 `WWW-Authenticate' response header is seen, according to the
3115 authorization scheme specified in that header (`Basic' and `Digest'
3116 are supported by the current implementation), produce an
3117 appropriate HTTP authorization request header. */
3119 create_authorization_line (const char *au, const char *user,
3120 const char *passwd, const char *method,
3121 const char *path, bool *finished)
3123 /* We are called only with known schemes, so we can dispatch on the
3125 switch (c_toupper (*au))
3127 case 'B': /* Basic */
3129 return basic_authentication_encode (user, passwd);
3130 #ifdef ENABLE_DIGEST
3131 case 'D': /* Digest */
3133 return digest_authentication_encode (au, user, passwd, method, path);
3136 case 'N': /* NTLM */
3137 if (!ntlm_input (&pconn.ntlm, au))
3142 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3145 /* We shouldn't get here -- this function should be only called
3146 with values approved by known_authentication_scheme_p. */
3154 if (!wget_cookie_jar)
3155 wget_cookie_jar = cookie_jar_new ();
3156 if (opt.cookies_input && !cookies_loaded_p)
3158 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3159 cookies_loaded_p = true;
3166 if (wget_cookie_jar)
3167 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3173 xfree_null (pconn.host);
3174 if (wget_cookie_jar)
3175 cookie_jar_delete (wget_cookie_jar);
3179 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3181 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3183 int len = strlen (ext);
3186 strncpy (shortext, ext, len - 1);
3187 shortext[len - 2] = '\0';
3190 if (last_period_in_local_filename == NULL
3191 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3192 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3194 int local_filename_len = strlen (hs->local_file);
3195 /* Resize the local file, allowing for ".html" preceded by
3196 optional ".NUMBER". */
3197 hs->local_file = xrealloc (hs->local_file,
3198 local_filename_len + 24 + len);
3199 strcpy (hs->local_file + local_filename_len, ext);
3200 /* If clobbering is not allowed and the file, as named,
3201 exists, tack on ".NUMBER.html" instead. */
3202 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3206 sprintf (hs->local_file + local_filename_len,
3207 ".%d%s", ext_num++, ext);
3208 while (file_exists_p (hs->local_file));
3210 *dt |= ADDED_HTML_EXTENSION;
3218 test_parse_content_disposition()
3223 char *opt_dir_prefix;
3227 { "filename=\"file.ext\"", NULL, "file.ext", true },
3228 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3229 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3230 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3231 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3232 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3233 { "attachment", NULL, NULL, false },
3234 { "attachment", "somedir", NULL, false },
3237 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3242 opt.dir_prefix = test_array[i].opt_dir_prefix;
3243 res = parse_content_disposition (test_array[i].hdrval, &filename);
3245 mu_assert ("test_parse_content_disposition: wrong result",
3246 res == test_array[i].result
3248 || 0 == strcmp (test_array[i].filename, filename)));
3254 #endif /* TESTING */
3257 * vim: et sts=2 sw=2 cino+={s