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", quote (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", quote (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_ALWAYS, "%s%s\n", prefix,
817 quotearg_style (escape_quoting_style, copy));
820 /* Print the server response, line by line, omitting the trailing CRLF
821 from individual header lines, and prefixed with PREFIX. */
824 print_server_response (const struct response *resp, const char *prefix)
829 for (i = 0; resp->headers[i + 1]; i++)
831 const char *b = resp->headers[i];
832 const char *e = resp->headers[i + 1];
834 if (b < e && e[-1] == '\n')
836 if (b < e && e[-1] == '\r')
838 print_response_line(prefix, b, e);
842 /* Parse the `Content-Range' header and extract the information it
843 contains. Returns true if successful, false otherwise. */
845 parse_content_range (const char *hdr, wgint *first_byte_ptr,
846 wgint *last_byte_ptr, wgint *entity_length_ptr)
850 /* Ancient versions of Netscape proxy server, presumably predating
851 rfc2068, sent out `Content-Range' without the "bytes"
853 if (0 == strncasecmp (hdr, "bytes", 5))
856 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
860 while (c_isspace (*hdr))
865 if (!c_isdigit (*hdr))
867 for (num = 0; c_isdigit (*hdr); hdr++)
868 num = 10 * num + (*hdr - '0');
869 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
871 *first_byte_ptr = num;
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
877 *last_byte_ptr = num;
882 for (num = 0; c_isdigit (*hdr); hdr++)
883 num = 10 * num + (*hdr - '0');
884 *entity_length_ptr = num;
888 /* Read the body of the request, but don't store it anywhere and don't
889 display a progress gauge. This is useful for reading the bodies of
890 administrative responses to which we will soon issue another
891 request. The response is not useful to the user, but reading it
892 allows us to continue using the same connection to the server.
894 If reading fails, false is returned, true otherwise. In debug
895 mode, the body is displayed for debugging purposes. */
898 skip_short_body (int fd, wgint contlen)
901 SKIP_SIZE = 512, /* size of the download buffer */
902 SKIP_THRESHOLD = 4096 /* the largest size we read */
904 char dlbuf[SKIP_SIZE + 1];
905 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
907 /* We shouldn't get here with unknown contlen. (This will change
908 with HTTP/1.1, which supports "chunked" transfer.) */
909 assert (contlen != -1);
911 /* If the body is too large, it makes more sense to simply close the
912 connection than to try to read the body. */
913 if (contlen > SKIP_THRESHOLD)
916 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
920 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
923 /* Don't normally report the error since this is an
924 optimization that should be invisible to the user. */
925 DEBUGP (("] aborting (%s).\n",
926 ret < 0 ? fd_errstr (fd) : "EOF received"));
930 /* Safe even if %.*s bogusly expects terminating \0 because
931 we've zero-terminated dlbuf above. */
932 DEBUGP (("%.*s", ret, dlbuf));
935 DEBUGP (("] done.\n"));
939 /* Extract a parameter from the string (typically an HTTP header) at
940 **SOURCE and advance SOURCE to the next parameter. Return false
941 when there are no more parameters to extract. The name of the
942 parameter is returned in NAME, and the value in VALUE. If the
943 parameter has no value, the token's value is zeroed out.
945 For example, if *SOURCE points to the string "attachment;
946 filename=\"foo bar\"", the first call to this function will return
947 the token named "attachment" and no value, and the second call will
948 return the token named "filename" and value "foo bar". The third
949 call will return false, indicating no more valid tokens. */
952 extract_param (const char **source, param_token *name, param_token *value,
955 const char *p = *source;
957 while (c_isspace (*p)) ++p;
961 return false; /* no error; nothing more to extract */
966 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
968 if (name->b == name->e)
969 return false; /* empty name: error */
970 while (c_isspace (*p)) ++p;
971 if (*p == separator || !*p) /* no value */
974 if (*p == separator) ++p;
979 return false; /* error */
981 /* *p is '=', extract value */
983 while (c_isspace (*p)) ++p;
984 if (*p == '"') /* quoted */
987 while (*p && *p != '"') ++p;
991 /* Currently at closing quote; find the end of param. */
992 while (c_isspace (*p)) ++p;
993 while (*p && *p != separator) ++p;
997 /* garbage after closed quote, e.g. foo="bar"baz */
1003 while (*p && *p != separator) ++p;
1005 while (value->e != value->b && c_isspace (value->e[-1]))
1007 if (*p == separator) ++p;
1014 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1016 /* Parse the contents of the `Content-Disposition' header, extracting
1017 the information useful to Wget. Content-Disposition is a header
1018 borrowed from MIME; when used in HTTP, it typically serves for
1019 specifying the desired file name of the resource. For example:
1021 Content-Disposition: attachment; filename="flora.jpg"
1023 Wget will skip the tokens it doesn't care about, such as
1024 "attachment" in the previous example; it will also skip other
1025 unrecognized params. If the header is syntactically correct and
1026 contains a file name, a copy of the file name is stored in
1027 *filename and true is returned. Otherwise, the function returns
1030 The file name is stripped of directory components and must not be
1034 parse_content_disposition (const char *hdr, char **filename)
1036 param_token name, value;
1037 while (extract_param (&hdr, &name, &value, ';'))
1038 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1040 /* Make the file name begin at the last slash or backslash. */
1041 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1042 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1043 if (last_slash && last_bs)
1044 value.b = 1 + MAX (last_slash, last_bs);
1045 else if (last_slash || last_bs)
1046 value.b = 1 + (last_slash ? last_slash : last_bs);
1047 if (value.b == value.e)
1049 /* Start with the directory prefix, if specified. */
1052 int prefix_length = strlen (opt.dir_prefix);
1053 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1058 total_length = prefix_length + (value.e - value.b);
1059 *filename = xmalloc (total_length + 1);
1060 strcpy (*filename, opt.dir_prefix);
1062 (*filename)[prefix_length - 1] = '/';
1063 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1064 (*filename)[total_length] = '\0';
1067 *filename = strdupdelim (value.b, value.e);
1073 /* Persistent connections. Currently, we cache the most recently used
1074 connection as persistent, provided that the HTTP server agrees to
1075 make it such. The persistence data is stored in the variables
1076 below. Ideally, it should be possible to cache an arbitrary fixed
1077 number of these connections. */
1079 /* Whether a persistent connection is active. */
1080 static bool pconn_active;
1083 /* The socket of the connection. */
1086 /* Host and port of the currently active persistent connection. */
1090 /* Whether a ssl handshake has occoured on this connection. */
1093 /* Whether the connection was authorized. This is only done by
1094 NTLM, which authorizes *connections* rather than individual
1095 requests. (That practice is peculiar for HTTP, but it is a
1096 useful optimization.) */
1100 /* NTLM data of the current connection. */
1101 struct ntlmdata ntlm;
1105 /* Mark the persistent connection as invalid and free the resources it
1106 uses. This is used by the CLOSE_* macros after they forcefully
1107 close a registered persistent connection. */
1110 invalidate_persistent (void)
1112 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1113 pconn_active = false;
1114 fd_close (pconn.socket);
1119 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1120 persistent. This will enable someone to use the same connection
1121 later. In the context of HTTP, this must be called only AFTER the
1122 response has been received and the server has promised that the
1123 connection will remain alive.
1125 If a previous connection was persistent, it is closed. */
1128 register_persistent (const char *host, int port, int fd, bool ssl)
1132 if (pconn.socket == fd)
1134 /* The connection FD is already registered. */
1139 /* The old persistent connection is still active; close it
1140 first. This situation arises whenever a persistent
1141 connection exists, but we then connect to a different
1142 host, and try to register a persistent connection to that
1144 invalidate_persistent ();
1148 pconn_active = true;
1150 pconn.host = xstrdup (host);
1153 pconn.authorized = false;
1155 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1158 /* Return true if a persistent connection is available for connecting
1162 persistent_available_p (const char *host, int port, bool ssl,
1163 bool *host_lookup_failed)
1165 /* First, check whether a persistent connection is active at all. */
1169 /* If we want SSL and the last connection wasn't or vice versa,
1170 don't use it. Checking for host and port is not enough because
1171 HTTP and HTTPS can apparently coexist on the same port. */
1172 if (ssl != pconn.ssl)
1175 /* If we're not connecting to the same port, we're not interested. */
1176 if (port != pconn.port)
1179 /* If the host is the same, we're in business. If not, there is
1180 still hope -- read below. */
1181 if (0 != strcasecmp (host, pconn.host))
1183 /* Check if pconn.socket is talking to HOST under another name.
1184 This happens often when both sites are virtual hosts
1185 distinguished only by name and served by the same network
1186 interface, and hence the same web server (possibly set up by
1187 the ISP and serving many different web sites). This
1188 admittedly unconventional optimization does not contradict
1189 HTTP and works well with popular server software. */
1193 struct address_list *al;
1196 /* Don't try to talk to two different SSL sites over the same
1197 secure connection! (Besides, it's not clear that
1198 name-based virtual hosting is even possible with SSL.) */
1201 /* If pconn.socket's peer is one of the IP addresses HOST
1202 resolves to, pconn.socket is for all intents and purposes
1203 already talking to HOST. */
1205 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1207 /* Can't get the peer's address -- something must be very
1208 wrong with the connection. */
1209 invalidate_persistent ();
1212 al = lookup_host (host, 0);
1215 *host_lookup_failed = true;
1219 found = address_list_contains (al, &ip);
1220 address_list_release (al);
1225 /* The persistent connection's peer address was found among the
1226 addresses HOST resolved to; therefore, pconn.sock is in fact
1227 already talking to HOST -- no need to reconnect. */
1230 /* Finally, check whether the connection is still open. This is
1231 important because most servers implement liberal (short) timeout
1232 on persistent connections. Wget can of course always reconnect
1233 if the connection doesn't work out, but it's nicer to know in
1234 advance. This test is a logical followup of the first test, but
1235 is "expensive" and therefore placed at the end of the list.
1237 (Current implementation of test_socket_open has a nice side
1238 effect that it treats sockets with pending data as "closed".
1239 This is exactly what we want: if a broken server sends message
1240 body in response to HEAD, or if it sends more than conent-length
1241 data, we won't reuse the corrupted connection.) */
1243 if (!test_socket_open (pconn.socket))
1245 /* Oops, the socket is no longer open. Now that we know that,
1246 let's invalidate the persistent connection before returning
1248 invalidate_persistent ();
1255 /* The idea behind these two CLOSE macros is to distinguish between
1256 two cases: one when the job we've been doing is finished, and we
1257 want to close the connection and leave, and two when something is
1258 seriously wrong and we're closing the connection as part of
1261 In case of keep_alive, CLOSE_FINISH should leave the connection
1262 open, while CLOSE_INVALIDATE should still close it.
1264 Note that the semantics of the flag `keep_alive' is "this
1265 connection *will* be reused (the server has promised not to close
1266 the connection once we're done)", while the semantics of
1267 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1268 active, registered connection". */
1270 #define CLOSE_FINISH(fd) do { \
1273 if (pconn_active && (fd) == pconn.socket) \
1274 invalidate_persistent (); \
1283 #define CLOSE_INVALIDATE(fd) do { \
1284 if (pconn_active && (fd) == pconn.socket) \
1285 invalidate_persistent (); \
1293 wgint len; /* received length */
1294 wgint contlen; /* expected length */
1295 wgint restval; /* the restart value */
1296 int res; /* the result of last read */
1297 char *rderrmsg; /* error message from read error */
1298 char *newloc; /* new location (redirection) */
1299 char *remote_time; /* remote time-stamp string */
1300 char *error; /* textual HTTP error */
1301 int statcode; /* status code */
1302 char *message; /* status message */
1303 wgint rd_size; /* amount of data read from socket */
1304 double dltime; /* time it took to download the data */
1305 const char *referer; /* value of the referer header. */
1306 char *local_file; /* local file name. */
1307 bool existence_checked; /* true if we already checked for a file's
1308 existence after having begun to download
1309 (needed in gethttp for when connection is
1310 interrupted/restarted. */
1311 bool timestamp_checked; /* true if pre-download time-stamping checks
1312 * have already been performed */
1313 char *orig_file_name; /* name of file to compare for time-stamping
1314 * (might be != local_file if -K is set) */
1315 wgint orig_file_size; /* size of file to compare for time-stamping */
1316 time_t orig_file_tstamp; /* time-stamp of file to compare for
1321 free_hstat (struct http_stat *hs)
1323 xfree_null (hs->newloc);
1324 xfree_null (hs->remote_time);
1325 xfree_null (hs->error);
1326 xfree_null (hs->rderrmsg);
1327 xfree_null (hs->local_file);
1328 xfree_null (hs->orig_file_name);
1329 xfree_null (hs->message);
1331 /* Guard against being called twice. */
1333 hs->remote_time = NULL;
1337 #define BEGINS_WITH(line, string_constant) \
1338 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1339 && (c_isspace (line[sizeof (string_constant) - 1]) \
1340 || !line[sizeof (string_constant) - 1]))
1342 #define SET_USER_AGENT(req) do { \
1343 if (!opt.useragent) \
1344 request_set_header (req, "User-Agent", \
1345 aprintf ("Wget/%s", version_string), rel_value); \
1346 else if (*opt.useragent) \
1347 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1350 /* The flags that allow clobbering the file (opening with "wb").
1351 Defined here to avoid repetition later. #### This will require
1353 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1354 || opt.dirstruct || opt.output_document)
1356 /* Retrieve a document through HTTP protocol. It recognizes status
1357 code, and correctly handles redirections. It closes the network
1358 socket. If it receives an error from the functions below it, it
1359 will print it if there is enough information to do so (almost
1360 always), returning the error to the caller (i.e. http_loop).
1362 Various HTTP parameters are stored to hs.
1364 If PROXY is non-NULL, the connection will be made to the proxy
1365 server, and u->url will be requested. */
1367 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1370 struct request *req;
1373 char *user, *passwd;
1377 wgint contlen, contrange;
1384 /* Set to 1 when the authorization has already been sent and should
1385 not be tried again. */
1386 bool auth_finished = false;
1388 /* Set to 1 when just globally-set Basic authorization has been sent;
1389 * should prevent further Basic negotiations, but not other
1391 bool basic_auth_finished = false;
1393 /* Whether NTLM authentication is used for this request. */
1394 bool ntlm_seen = false;
1396 /* Whether our connection to the remote host is through SSL. */
1397 bool using_ssl = false;
1399 /* Whether a HEAD request will be issued (as opposed to GET or
1401 bool head_only = !!(*dt & HEAD_ONLY);
1404 struct response *resp;
1408 /* Whether this connection will be kept alive after the HTTP request
1412 /* Whether keep-alive should be inhibited.
1414 RFC 2068 requests that 1.0 clients not send keep-alive requests
1415 to proxies. This is because many 1.0 proxies do not interpret
1416 the Connection header and transfer it to the remote server,
1417 causing it to not close the connection and leave both the proxy
1418 and the client hanging. */
1419 bool inhibit_keep_alive =
1420 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1422 /* Headers sent when using POST. */
1423 wgint post_data_size = 0;
1425 bool host_lookup_failed = false;
1428 if (u->scheme == SCHEME_HTTPS)
1430 /* Initialize the SSL context. After this has once been done,
1431 it becomes a no-op. */
1434 scheme_disable (SCHEME_HTTPS);
1435 logprintf (LOG_NOTQUIET,
1436 _("Disabling SSL due to encountered errors.\n"));
1437 return SSLINITFAILED;
1440 #endif /* HAVE_SSL */
1442 /* Initialize certain elements of struct http_stat. */
1446 hs->rderrmsg = NULL;
1448 hs->remote_time = NULL;
1454 /* Prepare the request to send. */
1456 req = request_new ();
1459 const char *meth = "GET";
1462 else if (opt.post_file_name || opt.post_data)
1464 /* Use the full path, i.e. one that includes the leading slash and
1465 the query string. E.g. if u->path is "foo/bar" and u->query is
1466 "param=value", full_path will be "/foo/bar?param=value". */
1469 /* When using SSL over proxy, CONNECT establishes a direct
1470 connection to the HTTPS server. Therefore use the same
1471 argument as when talking to the server directly. */
1472 && u->scheme != SCHEME_HTTPS
1475 meth_arg = xstrdup (u->url);
1477 meth_arg = url_full_path (u);
1478 request_set_method (req, meth, meth_arg);
1481 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1482 if (*dt & SEND_NOCACHE)
1483 request_set_header (req, "Pragma", "no-cache", rel_none);
1485 request_set_header (req, "Range",
1486 aprintf ("bytes=%s-",
1487 number_to_static_string (hs->restval)),
1489 SET_USER_AGENT (req);
1490 request_set_header (req, "Accept", "*/*", rel_none);
1492 /* Find the username and password for authentication. */
1495 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1496 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1497 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1500 && !u->user) /* We only do "site-wide" authentication with "global"
1501 user/password values; URL user/password info overrides. */
1503 /* If this is a host for which we've already received a Basic
1504 * challenge, we'll go ahead and send Basic authentication creds. */
1505 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1508 /* Generate the Host header, HOST:PORT. Take into account that:
1510 - Broken server-side software often doesn't recognize the PORT
1511 argument, so we must generate "Host: www.server.com" instead of
1512 "Host: www.server.com:80" (and likewise for https port).
1514 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1515 becomes ambiguous and needs to be rewritten as "Host:
1516 [3ffe:8100:200:2::2]:1234". */
1518 /* Formats arranged for hfmt[add_port][add_squares]. */
1519 static const char *hfmt[][2] = {
1520 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1522 int add_port = u->port != scheme_default_port (u->scheme);
1523 int add_squares = strchr (u->host, ':') != NULL;
1524 request_set_header (req, "Host",
1525 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1529 if (!inhibit_keep_alive)
1530 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1533 request_set_header (req, "Cookie",
1534 cookie_header (wget_cookie_jar,
1535 u->host, u->port, u->path,
1537 u->scheme == SCHEME_HTTPS
1544 if (opt.post_data || opt.post_file_name)
1546 request_set_header (req, "Content-Type",
1547 "application/x-www-form-urlencoded", rel_none);
1549 post_data_size = strlen (opt.post_data);
1552 post_data_size = file_size (opt.post_file_name);
1553 if (post_data_size == -1)
1555 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1556 quote (opt.post_file_name), strerror (errno));
1560 request_set_header (req, "Content-Length",
1561 xstrdup (number_to_static_string (post_data_size)),
1565 /* Add the user headers. */
1566 if (opt.user_headers)
1569 for (i = 0; opt.user_headers[i]; i++)
1570 request_set_user_header (req, opt.user_headers[i]);
1574 /* We need to come back here when the initial attempt to retrieve
1575 without authorization header fails. (Expected to happen at least
1576 for the Digest authorization scheme.) */
1581 char *proxy_user, *proxy_passwd;
1582 /* For normal username and password, URL components override
1583 command-line/wgetrc parameters. With proxy
1584 authentication, it's the reverse, because proxy URLs are
1585 normally the "permanent" ones, so command-line args
1586 should take precedence. */
1587 if (opt.proxy_user && opt.proxy_passwd)
1589 proxy_user = opt.proxy_user;
1590 proxy_passwd = opt.proxy_passwd;
1594 proxy_user = proxy->user;
1595 proxy_passwd = proxy->passwd;
1597 /* #### This does not appear right. Can't the proxy request,
1598 say, `Digest' authentication? */
1599 if (proxy_user && proxy_passwd)
1600 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1602 /* If we're using a proxy, we will be connecting to the proxy
1606 /* Proxy authorization over SSL is handled below. */
1608 if (u->scheme != SCHEME_HTTPS)
1610 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1615 /* Establish the connection. */
1617 if (!inhibit_keep_alive)
1619 /* Look for a persistent connection to target host, unless a
1620 proxy is used. The exception is when SSL is in use, in which
1621 case the proxy is nothing but a passthrough to the target
1622 host, registered as a connection to the latter. */
1623 struct url *relevant = conn;
1625 if (u->scheme == SCHEME_HTTPS)
1629 if (persistent_available_p (relevant->host, relevant->port,
1631 relevant->scheme == SCHEME_HTTPS,
1635 &host_lookup_failed))
1637 sock = pconn.socket;
1638 using_ssl = pconn.ssl;
1639 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1640 quotearg_style (escape_quoting_style, pconn.host),
1642 DEBUGP (("Reusing fd %d.\n", sock));
1643 if (pconn.authorized)
1644 /* If the connection is already authorized, the "Basic"
1645 authorization added by code above is unnecessary and
1647 request_remove_header (req, "Authorization");
1649 else if (host_lookup_failed)
1652 logprintf(LOG_NOTQUIET,
1653 _("%s: unable to resolve host address %s\n"),
1654 exec_name, quote (relevant->host));
1661 sock = connect_to_host (conn->host, conn->port);
1670 return (retryable_socket_connect_error (errno)
1671 ? CONERROR : CONIMPOSSIBLE);
1675 if (proxy && u->scheme == SCHEME_HTTPS)
1677 /* When requesting SSL URLs through proxies, use the
1678 CONNECT method to request passthrough. */
1679 struct request *connreq = request_new ();
1680 request_set_method (connreq, "CONNECT",
1681 aprintf ("%s:%d", u->host, u->port));
1682 SET_USER_AGENT (connreq);
1685 request_set_header (connreq, "Proxy-Authorization",
1686 proxyauth, rel_value);
1687 /* Now that PROXYAUTH is part of the CONNECT request,
1688 zero it out so we don't send proxy authorization with
1689 the regular request below. */
1692 /* Examples in rfc2817 use the Host header in CONNECT
1693 requests. I don't see how that gains anything, given
1694 that the contents of Host would be exactly the same as
1695 the contents of CONNECT. */
1697 write_error = request_send (connreq, sock);
1698 request_free (connreq);
1699 if (write_error < 0)
1701 CLOSE_INVALIDATE (sock);
1705 head = read_http_response_head (sock);
1708 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1710 CLOSE_INVALIDATE (sock);
1719 DEBUGP (("proxy responded with: [%s]\n", head));
1721 resp = resp_new (head);
1722 statcode = resp_status (resp, &message);
1723 hs->message = xstrdup (message);
1726 if (statcode != 200)
1729 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1730 message ? quotearg_style (escape_quoting_style, message) : "?");
1731 xfree_null (message);
1734 xfree_null (message);
1736 /* SOCK is now *really* connected to u->host, so update CONN
1737 to reflect this. That way register_persistent will
1738 register SOCK as being connected to u->host:u->port. */
1742 if (conn->scheme == SCHEME_HTTPS)
1744 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1751 #endif /* HAVE_SSL */
1754 /* Send the request to server. */
1755 write_error = request_send (req, sock);
1757 if (write_error >= 0)
1761 DEBUGP (("[POST data: %s]\n", opt.post_data));
1762 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1764 else if (opt.post_file_name && post_data_size != 0)
1765 write_error = post_file (sock, opt.post_file_name, post_data_size);
1768 if (write_error < 0)
1770 CLOSE_INVALIDATE (sock);
1774 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1775 proxy ? "Proxy" : "HTTP");
1780 head = read_http_response_head (sock);
1785 logputs (LOG_NOTQUIET, _("No data received.\n"));
1786 CLOSE_INVALIDATE (sock);
1792 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1794 CLOSE_INVALIDATE (sock);
1799 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1801 resp = resp_new (head);
1803 /* Check for status line. */
1805 statcode = resp_status (resp, &message);
1806 hs->message = xstrdup (message);
1807 if (!opt.server_response)
1808 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1809 message ? quotearg_style (escape_quoting_style, message) : "");
1812 logprintf (LOG_VERBOSE, "\n");
1813 print_server_response (resp, " ");
1816 /* Determine the local filename if needed. Notice that if -O is used
1817 * hstat.local_file is set by http_loop to the argument of -O. */
1818 if (!hs->local_file)
1820 /* Honor Content-Disposition whether possible. */
1821 if (!opt.content_disposition
1822 || !resp_header_copy (resp, "Content-Disposition",
1823 hdrval, sizeof (hdrval))
1824 || !parse_content_disposition (hdrval, &hs->local_file))
1826 /* The Content-Disposition header is missing or broken.
1827 * Choose unique file name according to given URL. */
1828 hs->local_file = url_file_name (u);
1832 /* TODO: perform this check only once. */
1833 if (!hs->existence_checked && file_exists_p (hs->local_file))
1835 if (opt.noclobber && !opt.output_document)
1837 /* If opt.noclobber is turned on and file already exists, do not
1838 retrieve the file. But if the output_document was given, then this
1839 test was already done and the file didn't exist. Hence the !opt.output_document */
1840 logprintf (LOG_VERBOSE, _("\
1841 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1842 /* If the file is there, we suppose it's retrieved OK. */
1845 /* #### Bogusness alert. */
1846 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1847 if (has_html_suffix_p (hs->local_file))
1850 return RETRUNNEEDED;
1852 else if (!ALLOW_CLOBBER)
1854 char *unique = unique_name (hs->local_file, true);
1855 if (unique != hs->local_file)
1856 xfree (hs->local_file);
1857 hs->local_file = unique;
1860 hs->existence_checked = true;
1862 /* Support timestamping */
1863 /* TODO: move this code out of gethttp. */
1864 if (opt.timestamping && !hs->timestamp_checked)
1866 size_t filename_len = strlen (hs->local_file);
1867 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1868 bool local_dot_orig_file_exists = false;
1869 char *local_filename = NULL;
1872 if (opt.backup_converted)
1873 /* If -K is specified, we'll act on the assumption that it was specified
1874 last time these files were downloaded as well, and instead of just
1875 comparing local file X against server file X, we'll compare local
1876 file X.orig (if extant, else X) against server file X. If -K
1877 _wasn't_ specified last time, or the server contains files called
1878 *.orig, -N will be back to not operating correctly with -k. */
1880 /* Would a single s[n]printf() call be faster? --dan
1882 Definitely not. sprintf() is horribly slow. It's a
1883 different question whether the difference between the two
1884 affects a program. Usually I'd say "no", but at one
1885 point I profiled Wget, and found that a measurable and
1886 non-negligible amount of time was lost calling sprintf()
1887 in url.c. Replacing sprintf with inline calls to
1888 strcpy() and number_to_string() made a difference.
1890 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1891 memcpy (filename_plus_orig_suffix + filename_len,
1892 ".orig", sizeof (".orig"));
1894 /* Try to stat() the .orig file. */
1895 if (stat (filename_plus_orig_suffix, &st) == 0)
1897 local_dot_orig_file_exists = true;
1898 local_filename = filename_plus_orig_suffix;
1902 if (!local_dot_orig_file_exists)
1903 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1904 if (stat (hs->local_file, &st) == 0)
1905 local_filename = hs->local_file;
1907 if (local_filename != NULL)
1908 /* There was a local file, so we'll check later to see if the version
1909 the server has is the same version we already have, allowing us to
1912 hs->orig_file_name = xstrdup (local_filename);
1913 hs->orig_file_size = st.st_size;
1914 hs->orig_file_tstamp = st.st_mtime;
1916 /* Modification time granularity is 2 seconds for Windows, so
1917 increase local time by 1 second for later comparison. */
1918 ++hs->orig_file_tstamp;
1923 if (!opt.ignore_length
1924 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1928 parsed = str_to_wgint (hdrval, NULL, 10);
1929 if (parsed == WGINT_MAX && errno == ERANGE)
1932 #### If Content-Length is out of range, it most likely
1933 means that the file is larger than 2G and that we're
1934 compiled without LFS. In that case we should probably
1935 refuse to even attempt to download the file. */
1938 else if (parsed < 0)
1940 /* Negative Content-Length; nonsensical, so we can't
1941 assume any information about the content to receive. */
1948 /* Check for keep-alive related responses. */
1949 if (!inhibit_keep_alive && contlen != -1)
1951 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1953 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1955 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1960 /* The server has promised that it will not close the connection
1961 when we're done. This means that we can register it. */
1962 register_persistent (conn->host, conn->port, sock, using_ssl);
1964 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1966 /* Authorization is required. */
1967 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1968 CLOSE_FINISH (sock);
1970 CLOSE_INVALIDATE (sock);
1971 pconn.authorized = false;
1972 if (!auth_finished && (user && passwd))
1974 /* IIS sends multiple copies of WWW-Authenticate, one with
1975 the value "negotiate", and other(s) with data. Loop over
1976 all the occurrences and pick the one we recognize. */
1978 const char *wabeg, *waend;
1979 char *www_authenticate = NULL;
1981 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1982 &wabeg, &waend)) != -1;
1984 if (known_authentication_scheme_p (wabeg, waend))
1986 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1990 if (!www_authenticate)
1992 /* If the authentication header is missing or
1993 unrecognized, there's no sense in retrying. */
1994 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1996 else if (!basic_auth_finished
1997 || !BEGINS_WITH (www_authenticate, "Basic"))
2000 pth = url_full_path (u);
2001 request_set_header (req, "Authorization",
2002 create_authorization_line (www_authenticate,
2004 request_method (req),
2008 if (BEGINS_WITH (www_authenticate, "NTLM"))
2010 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2012 /* Need to register this host as using basic auth,
2013 * so we automatically send creds next time. */
2014 register_basic_auth_host (u->host);
2017 goto retry_with_auth;
2021 /* We already did Basic auth, and it failed. Gotta
2025 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2029 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2031 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2033 pconn.authorized = true;
2037 hs->statcode = statcode;
2039 hs->error = xstrdup (_("Malformed status line"));
2041 hs->error = xstrdup (_("(no description)"));
2043 hs->error = xstrdup (message);
2044 xfree_null (message);
2046 type = resp_header_strdup (resp, "Content-Type");
2049 char *tmp = strchr (type, ';');
2052 /* sXXXav: only needed if IRI support is enabled */
2053 char *tmp2 = tmp + 1;
2055 while (tmp > type && c_isspace (tmp[-1]))
2059 /* Try to get remote encoding if needed */
2060 if (opt.enable_iri && !opt.encoding_remote)
2062 tmp = parse_charset (tmp2);
2064 set_content_encoding (iri, tmp);
2068 hs->newloc = resp_header_strdup (resp, "Location");
2069 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2071 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2075 const char *scbeg, *scend;
2076 /* The jar should have been created by now. */
2077 assert (wget_cookie_jar != NULL);
2079 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2080 &scbeg, &scend)) != -1;
2083 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2084 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2085 u->path, set_cookie);
2089 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2091 wgint first_byte_pos, last_byte_pos, entity_length;
2092 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2095 contrange = first_byte_pos;
2096 contlen = last_byte_pos - first_byte_pos + 1;
2101 /* 20x responses are counted among successful by default. */
2102 if (H_20X (statcode))
2105 /* Return if redirected. */
2106 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2108 /* RFC2068 says that in case of the 300 (multiple choices)
2109 response, the server can output a preferred URL through
2110 `Location' header; otherwise, the request should be treated
2111 like GET. So, if the location is set, it will be a
2112 redirection; otherwise, just proceed normally. */
2113 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2117 logprintf (LOG_VERBOSE,
2118 _("Location: %s%s\n"),
2119 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2120 hs->newloc ? _(" [following]") : "");
2121 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2122 CLOSE_FINISH (sock);
2124 CLOSE_INVALIDATE (sock);
2130 /* If content-type is not given, assume text/html. This is because
2131 of the multitude of broken CGI's that "forget" to generate the
2134 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2135 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2141 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2146 if (opt.html_extension)
2149 /* -E / --html-extension / html_extension = on was specified,
2150 and this is a text/html file. If some case-insensitive
2151 variation on ".htm[l]" isn't already the file's suffix,
2154 ensure_extension (hs, ".html", dt);
2156 else if (*dt & TEXTCSS)
2158 ensure_extension (hs, ".css", dt);
2162 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2164 /* If `-c' is in use and the file has been fully downloaded (or
2165 the remote file has shrunk), Wget effectively requests bytes
2166 after the end of file and the server response with 416. */
2167 logputs (LOG_VERBOSE, _("\
2168 \n The file is already fully retrieved; nothing to do.\n\n"));
2169 /* In case the caller inspects. */
2172 /* Mark as successfully retrieved. */
2175 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2176 might be more bytes in the body. */
2177 return RETRUNNEEDED;
2179 if ((contrange != 0 && contrange != hs->restval)
2180 || (H_PARTIAL (statcode) && !contrange))
2182 /* The Range request was somehow misunderstood by the server.
2185 CLOSE_INVALIDATE (sock);
2191 hs->contlen = contlen + contrange;
2197 /* No need to print this output if the body won't be
2198 downloaded at all, or if the original server response is
2200 logputs (LOG_VERBOSE, _("Length: "));
2203 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2204 if (contlen + contrange >= 1024)
2205 logprintf (LOG_VERBOSE, " (%s)",
2206 human_readable (contlen + contrange));
2209 if (contlen >= 1024)
2210 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2211 number_to_static_string (contlen),
2212 human_readable (contlen));
2214 logprintf (LOG_VERBOSE, _(", %s remaining"),
2215 number_to_static_string (contlen));
2219 logputs (LOG_VERBOSE,
2220 opt.ignore_length ? _("ignored") : _("unspecified"));
2222 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2224 logputs (LOG_VERBOSE, "\n");
2228 type = NULL; /* We don't need it any more. */
2230 /* Return if we have no intention of further downloading. */
2231 if (!(*dt & RETROKF) || head_only)
2233 /* In case the caller cares to look... */
2238 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2239 servers not to send body in response to a HEAD request, and
2240 those that do will likely be caught by test_socket_open.
2241 If not, they can be worked around using
2242 `--no-http-keep-alive'. */
2243 CLOSE_FINISH (sock);
2244 else if (keep_alive && skip_short_body (sock, contlen))
2245 /* Successfully skipped the body; also keep using the socket. */
2246 CLOSE_FINISH (sock);
2248 CLOSE_INVALIDATE (sock);
2249 return RETRFINISHED;
2252 /* Open the local file. */
2255 mkalldirs (hs->local_file);
2257 rotate_backups (hs->local_file);
2259 fp = fopen (hs->local_file, "ab");
2260 else if (ALLOW_CLOBBER)
2261 fp = fopen (hs->local_file, "wb");
2264 fp = fopen_excl (hs->local_file, true);
2265 if (!fp && errno == EEXIST)
2267 /* We cannot just invent a new name and use it (which is
2268 what functions like unique_create typically do)
2269 because we told the user we'd use this name.
2270 Instead, return and retry the download. */
2271 logprintf (LOG_NOTQUIET,
2272 _("%s has sprung into existence.\n"),
2274 CLOSE_INVALIDATE (sock);
2275 return FOPEN_EXCL_ERR;
2280 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2281 CLOSE_INVALIDATE (sock);
2288 /* Print fetch message, if opt.verbose. */
2291 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2292 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2295 /* This confuses the timestamping code that checks for file size.
2296 #### The timestamping code should be smarter about file size. */
2297 if (opt.save_headers && hs->restval == 0)
2298 fwrite (head, 1, strlen (head), fp);
2300 /* Now we no longer need to store the response header. */
2303 /* Download the request body. */
2306 /* If content-length is present, read that much; otherwise, read
2307 until EOF. The HTTP spec doesn't require the server to
2308 actually close the connection when it's done sending data. */
2309 flags |= rb_read_exactly;
2310 if (hs->restval > 0 && contrange == 0)
2311 /* If the server ignored our range request, instruct fd_read_body
2312 to skip the first RESTVAL bytes of body. */
2313 flags |= rb_skip_startpos;
2314 hs->len = hs->restval;
2316 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2317 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2321 CLOSE_FINISH (sock);
2325 hs->rderrmsg = xstrdup (fd_errstr (sock));
2326 CLOSE_INVALIDATE (sock);
2333 return RETRFINISHED;
2336 /* The genuine HTTP loop! This is the part where the retrieval is
2337 retried, and retried, and retried, and... */
2339 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2340 int *dt, struct url *proxy, struct iri *iri)
2343 bool got_head = false; /* used for time-stamping and filename detection */
2344 bool time_came_from_head = false;
2345 bool got_name = false;
2348 uerr_t err, ret = TRYLIMEXC;
2349 time_t tmr = -1; /* remote time-stamp */
2350 struct http_stat hstat; /* HTTP status */
2352 bool send_head_first = true;
2354 /* Assert that no value for *LOCAL_FILE was passed. */
2355 assert (local_file == NULL || *local_file == NULL);
2357 /* Set LOCAL_FILE parameter. */
2358 if (local_file && opt.output_document)
2359 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2361 /* Reset NEWLOC parameter. */
2364 /* This used to be done in main(), but it's a better idea to do it
2365 here so that we don't go through the hoops if we're just using
2370 /* Warn on (likely bogus) wildcard usage in HTTP. */
2371 if (opt.ftp_glob && has_wildcards_p (u->path))
2372 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2374 /* Setup hstat struct. */
2376 hstat.referer = referer;
2378 if (opt.output_document)
2380 hstat.local_file = xstrdup (opt.output_document);
2383 else if (!opt.content_disposition)
2385 hstat.local_file = url_file_name (u);
2389 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2390 * screaming for some refactoring. */
2391 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2393 /* If opt.noclobber is turned on and file already exists, do not
2394 retrieve the file. But if the output_document was given, then this
2395 test was already done and the file didn't exist. Hence the !opt.output_document */
2396 logprintf (LOG_VERBOSE, _("\
2397 File %s already there; not retrieving.\n\n"),
2398 quote (hstat.local_file));
2399 /* If the file is there, we suppose it's retrieved OK. */
2402 /* #### Bogusness alert. */
2403 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2404 if (has_html_suffix_p (hstat.local_file))
2411 /* Reset the counter. */
2414 /* Reset the document type. */
2417 /* Skip preliminary HEAD request if we're not in spider mode AND
2418 * if -O was given or HTTP Content-Disposition support is disabled. */
2420 && (got_name || !opt.content_disposition))
2421 send_head_first = false;
2423 /* Send preliminary HEAD request if -N is given and we have an existing
2424 * destination file. */
2425 if (opt.timestamping
2426 && !opt.content_disposition
2427 && file_exists_p (url_file_name (u)))
2428 send_head_first = true;
2433 /* Increment the pass counter. */
2435 sleep_between_retrievals (count);
2437 /* Get the current time string. */
2438 tms = datetime_str (time (NULL));
2440 if (opt.spider && !got_head)
2441 logprintf (LOG_VERBOSE, _("\
2442 Spider mode enabled. Check if remote file exists.\n"));
2444 /* Print fetch message, if opt.verbose. */
2447 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2452 sprintf (tmp, _("(try:%2d)"), count);
2453 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2458 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2463 ws_changetitle (hurl);
2468 /* Default document type is empty. However, if spider mode is
2469 on or time-stamping is employed, HEAD_ONLY commands is
2470 encoded within *dt. */
2471 if (send_head_first && !got_head)
2476 /* Decide whether or not to restart. */
2479 && stat (hstat.local_file, &st) == 0
2480 && S_ISREG (st.st_mode))
2481 /* When -c is used, continue from on-disk size. (Can't use
2482 hstat.len even if count>1 because we don't want a failed
2483 first attempt to clobber existing data.) */
2484 hstat.restval = st.st_size;
2486 /* otherwise, continue where the previous try left off */
2487 hstat.restval = hstat.len;
2491 /* Decide whether to send the no-cache directive. We send it in
2493 a) we're using a proxy, and we're past our first retrieval.
2494 Some proxies are notorious for caching incomplete data, so
2495 we require a fresh get.
2496 b) caching is explicitly inhibited. */
2497 if ((proxy && count > 1) /* a */
2498 || !opt.allow_cache) /* b */
2499 *dt |= SEND_NOCACHE;
2501 *dt &= ~SEND_NOCACHE;
2503 /* Try fetching the document, or at least its head. */
2504 err = gethttp (u, &hstat, dt, proxy, iri);
2507 tms = datetime_str (time (NULL));
2509 /* Get the new location (with or without the redirection). */
2511 *newloc = xstrdup (hstat.newloc);
2515 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2516 case CONERROR: case READERR: case WRITEFAILED:
2517 case RANGEERR: case FOPEN_EXCL_ERR:
2518 /* Non-fatal errors continue executing the loop, which will
2519 bring them to "while" statement at the end, to judge
2520 whether the number of tries was exceeded. */
2521 printwhat (count, opt.ntry);
2523 case FWRITEERR: case FOPENERR:
2524 /* Another fatal error. */
2525 logputs (LOG_VERBOSE, "\n");
2526 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2527 quote (hstat.local_file), strerror (errno));
2528 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2529 case SSLINITFAILED: case CONTNOTSUPPORTED:
2530 /* Fatal errors just return from the function. */
2534 /* Another fatal error. */
2535 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2539 /* Return the new location to the caller. */
2542 logprintf (LOG_NOTQUIET,
2543 _("ERROR: Redirection (%d) without location.\n"),
2553 /* The file was already fully retrieved. */
2557 /* Deal with you later. */
2560 /* All possibilities should have been exhausted. */
2564 if (!(*dt & RETROKF))
2569 /* #### Ugly ugly ugly! */
2570 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2571 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2574 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2576 && (hstat.statcode == 500 || hstat.statcode == 501))
2581 /* Maybe we should always keep track of broken links, not just in
2583 * Don't log error if it was UTF-8 encoded because we will try
2584 * once unencoded. */
2585 else if (opt.spider && !iri->utf8_encode)
2587 /* #### Again: ugly ugly ugly! */
2589 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2590 nonexisting_url (hurl);
2591 logprintf (LOG_NOTQUIET, _("\
2592 Remote file does not exist -- broken link!!!\n"));
2596 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2597 tms, hstat.statcode,
2598 quotearg_style (escape_quoting_style, hstat.error));
2600 logputs (LOG_VERBOSE, "\n");
2606 /* Did we get the time-stamp? */
2609 got_head = true; /* no more time-stamping */
2611 if (opt.timestamping && !hstat.remote_time)
2613 logputs (LOG_NOTQUIET, _("\
2614 Last-modified header missing -- time-stamps turned off.\n"));
2616 else if (hstat.remote_time)
2618 /* Convert the date-string into struct tm. */
2619 tmr = http_atotm (hstat.remote_time);
2620 if (tmr == (time_t) (-1))
2621 logputs (LOG_VERBOSE, _("\
2622 Last-modified header invalid -- time-stamp ignored.\n"));
2623 if (*dt & HEAD_ONLY)
2624 time_came_from_head = true;
2627 if (send_head_first)
2629 /* The time-stamping section. */
2630 if (opt.timestamping)
2632 if (hstat.orig_file_name) /* Perform the following
2633 checks only if the file
2635 download already exists. */
2637 if (hstat.remote_time &&
2638 tmr != (time_t) (-1))
2640 /* Now time-stamping can be used validly.
2641 Time-stamping means that if the sizes of
2642 the local and remote file match, and local
2643 file is newer than the remote file, it will
2644 not be retrieved. Otherwise, the normal
2645 download procedure is resumed. */
2646 if (hstat.orig_file_tstamp >= tmr)
2648 if (hstat.contlen == -1
2649 || hstat.orig_file_size == hstat.contlen)
2651 logprintf (LOG_VERBOSE, _("\
2652 Server file no newer than local file %s -- not retrieving.\n\n"),
2653 quote (hstat.orig_file_name));
2659 logprintf (LOG_VERBOSE, _("\
2660 The sizes do not match (local %s) -- retrieving.\n"),
2661 number_to_static_string (hstat.orig_file_size));
2665 logputs (LOG_VERBOSE,
2666 _("Remote file is newer, retrieving.\n"));
2668 logputs (LOG_VERBOSE, "\n");
2672 /* free_hstat (&hstat); */
2673 hstat.timestamp_checked = true;
2678 bool finished = true;
2683 logputs (LOG_VERBOSE, _("\
2684 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2689 logprintf (LOG_VERBOSE, _("\
2690 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2691 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2698 logprintf (LOG_VERBOSE, _("\
2699 Remote file exists and could contain further links,\n\
2700 but recursion is disabled -- not retrieving.\n\n"));
2704 logprintf (LOG_VERBOSE, _("\
2705 Remote file exists.\n\n"));
2707 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2712 logprintf (LOG_NONVERBOSE,
2713 _("%s URL:%s %2d %s\n"),
2714 tms, u->url, hstat.statcode,
2715 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2722 count = 0; /* the retrieve count for HEAD is reset */
2724 } /* send_head_first */
2727 if ((tmr != (time_t) (-1))
2728 && ((hstat.len == hstat.contlen) ||
2729 ((hstat.res == 0) && (hstat.contlen == -1))))
2731 /* #### This code repeats in http.c and ftp.c. Move it to a
2733 const char *fl = NULL;
2734 if (opt.output_document)
2736 if (output_stream_regular)
2737 fl = opt.output_document;
2740 fl = hstat.local_file;
2744 /* Reparse time header, in case it's changed. */
2745 if (time_came_from_head
2746 && hstat.remote_time && hstat.remote_time[0])
2748 newtmr = http_atotm (hstat.remote_time);
2755 /* End of time-stamping section. */
2757 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2758 total_download_time += hstat.dltime;
2760 if (hstat.len == hstat.contlen)
2764 logprintf (LOG_VERBOSE,
2765 _("%s (%s) - %s saved [%s/%s]\n\n"),
2766 tms, tmrate, quote (hstat.local_file),
2767 number_to_static_string (hstat.len),
2768 number_to_static_string (hstat.contlen));
2769 logprintf (LOG_NONVERBOSE,
2770 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2772 number_to_static_string (hstat.len),
2773 number_to_static_string (hstat.contlen),
2774 hstat.local_file, count);
2777 total_downloaded_bytes += hstat.len;
2779 /* Remember that we downloaded the file for later ".orig" code. */
2780 if (*dt & ADDED_HTML_EXTENSION)
2781 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2783 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2788 else if (hstat.res == 0) /* No read error */
2790 if (hstat.contlen == -1) /* We don't know how much we were supposed
2791 to get, so assume we succeeded. */
2795 logprintf (LOG_VERBOSE,
2796 _("%s (%s) - %s saved [%s]\n\n"),
2797 tms, tmrate, quote (hstat.local_file),
2798 number_to_static_string (hstat.len));
2799 logprintf (LOG_NONVERBOSE,
2800 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2801 tms, u->url, number_to_static_string (hstat.len),
2802 hstat.local_file, count);
2805 total_downloaded_bytes += hstat.len;
2807 /* Remember that we downloaded the file for later ".orig" code. */
2808 if (*dt & ADDED_HTML_EXTENSION)
2809 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2811 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2816 else if (hstat.len < hstat.contlen) /* meaning we lost the
2817 connection too soon */
2819 logprintf (LOG_VERBOSE,
2820 _("%s (%s) - Connection closed at byte %s. "),
2821 tms, tmrate, number_to_static_string (hstat.len));
2822 printwhat (count, opt.ntry);
2825 else if (hstat.len != hstat.restval)
2826 /* Getting here would mean reading more data than
2827 requested with content-length, which we never do. */
2831 /* Getting here probably means that the content-length was
2832 * _less_ than the original, local size. We should probably
2833 * truncate or re-read, or something. FIXME */
2838 else /* from now on hstat.res can only be -1 */
2840 if (hstat.contlen == -1)
2842 logprintf (LOG_VERBOSE,
2843 _("%s (%s) - Read error at byte %s (%s)."),
2844 tms, tmrate, number_to_static_string (hstat.len),
2846 printwhat (count, opt.ntry);
2849 else /* hstat.res == -1 and contlen is given */
2851 logprintf (LOG_VERBOSE,
2852 _("%s (%s) - Read error at byte %s/%s (%s). "),
2854 number_to_static_string (hstat.len),
2855 number_to_static_string (hstat.contlen),
2857 printwhat (count, opt.ntry);
2863 while (!opt.ntry || (count < opt.ntry));
2867 *local_file = xstrdup (hstat.local_file);
2868 free_hstat (&hstat);
2873 /* Check whether the result of strptime() indicates success.
2874 strptime() returns the pointer to how far it got to in the string.
2875 The processing has been successful if the string is at `GMT' or
2876 `+X', or at the end of the string.
2878 In extended regexp parlance, the function returns 1 if P matches
2879 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2880 can return) is considered a failure and 0 is returned. */
2882 check_end (const char *p)
2886 while (c_isspace (*p))
2889 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2890 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2896 /* Convert the textual specification of time in TIME_STRING to the
2897 number of seconds since the Epoch.
2899 TIME_STRING can be in any of the three formats RFC2616 allows the
2900 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2901 as well as the time format used in the Set-Cookie header.
2902 Timezones are ignored, and should be GMT.
2904 Return the computed time_t representation, or -1 if the conversion
2907 This function uses strptime with various string formats for parsing
2908 TIME_STRING. This results in a parser that is not as lenient in
2909 interpreting TIME_STRING as I would like it to be. Being based on
2910 strptime, it always allows shortened months, one-digit days, etc.,
2911 but due to the multitude of formats in which time can be
2912 represented, an ideal HTTP time parser would be even more
2913 forgiving. It should completely ignore things like week days and
2914 concentrate only on the various forms of representing years,
2915 months, days, hours, minutes, and seconds. For example, it would
2916 be nice if it accepted ISO 8601 out of the box.
2918 I've investigated free and PD code for this purpose, but none was
2919 usable. getdate was big and unwieldy, and had potential copyright
2920 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2921 distributed with phttpd, is excellent, but we cannot use it because
2922 it is not assigned to the FSF. So I stuck it with strptime. */
2925 http_atotm (const char *time_string)
2927 /* NOTE: Solaris strptime man page claims that %n and %t match white
2928 space, but that's not universally available. Instead, we simply
2929 use ` ' to mean "skip all WS", which works under all strptime
2930 implementations I've tested. */
2932 static const char *time_formats[] = {
2933 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2934 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2935 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2936 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2937 (used in Set-Cookie, defined in the
2938 Netscape cookie specification.) */
2940 const char *oldlocale;
2942 time_t ret = (time_t) -1;
2944 /* Solaris strptime fails to recognize English month names in
2945 non-English locales, which we work around by temporarily setting
2946 locale to C before invoking strptime. */
2947 oldlocale = setlocale (LC_TIME, NULL);
2948 setlocale (LC_TIME, "C");
2950 for (i = 0; i < countof (time_formats); i++)
2954 /* Some versions of strptime use the existing contents of struct
2955 tm to recalculate the date according to format. Zero it out
2956 to prevent stack garbage from influencing strptime. */
2959 if (check_end (strptime (time_string, time_formats[i], &t)))
2966 /* Restore the previous locale. */
2967 setlocale (LC_TIME, oldlocale);
2972 /* Authorization support: We support three authorization schemes:
2974 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2976 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2977 consisting of answering to the server's challenge with the proper
2980 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2981 Stenberg for libcurl. Like digest, NTLM is based on a
2982 challenge-response mechanism, but unlike digest, it is non-standard
2983 (authenticates TCP connections rather than requests), undocumented
2984 and Microsoft-specific. */
2986 /* Create the authentication header contents for the `Basic' scheme.
2987 This is done by encoding the string "USER:PASS" to base64 and
2988 prepending the string "Basic " in front of it. */
2991 basic_authentication_encode (const char *user, const char *passwd)
2994 int len1 = strlen (user) + 1 + strlen (passwd);
2996 t1 = (char *)alloca (len1 + 1);
2997 sprintf (t1, "%s:%s", user, passwd);
2999 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3000 base64_encode (t1, len1, t2);
3002 return concat_strings ("Basic ", t2, (char *) 0);
3005 #define SKIP_WS(x) do { \
3006 while (c_isspace (*(x))) \
3010 #ifdef ENABLE_DIGEST
3011 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3012 an array of 16 bytes containing the hash keys, and BUF should be a
3013 buffer of 33 writable characters (32 for hex digits plus one for
3014 zero termination). */
3016 dump_hash (char *buf, const unsigned char *hash)
3020 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3022 *buf++ = XNUM_TO_digit (*hash >> 4);
3023 *buf++ = XNUM_TO_digit (*hash & 0xf);
3028 /* Take the line apart to find the challenge, and compose a digest
3029 authorization header. See RFC2069 section 2.1.2. */
3031 digest_authentication_encode (const char *au, const char *user,
3032 const char *passwd, const char *method,
3035 static char *realm, *opaque, *nonce;
3040 { "realm", &realm },
3041 { "opaque", &opaque },
3045 param_token name, value;
3047 realm = opaque = nonce = NULL;
3049 au += 6; /* skip over `Digest' */
3050 while (extract_param (&au, &name, &value, ','))
3053 size_t namelen = name.e - name.b;
3054 for (i = 0; i < countof (options); i++)
3055 if (namelen == strlen (options[i].name)
3056 && 0 == strncmp (name.b, options[i].name,
3059 *options[i].variable = strdupdelim (value.b, value.e);
3063 if (!realm || !nonce || !user || !passwd || !path || !method)
3066 xfree_null (opaque);
3071 /* Calculate the digest value. */
3073 ALLOCA_MD5_CONTEXT (ctx);
3074 unsigned char hash[MD5_HASHLEN];
3075 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3076 char response_digest[MD5_HASHLEN * 2 + 1];
3078 /* A1BUF = H(user ":" realm ":" password) */
3080 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3081 gen_md5_update ((unsigned char *)":", 1, ctx);
3082 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3083 gen_md5_update ((unsigned char *)":", 1, ctx);
3084 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3085 gen_md5_finish (ctx, hash);
3086 dump_hash (a1buf, hash);
3088 /* A2BUF = H(method ":" path) */
3090 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3091 gen_md5_update ((unsigned char *)":", 1, ctx);
3092 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3093 gen_md5_finish (ctx, hash);
3094 dump_hash (a2buf, hash);
3096 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3098 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3099 gen_md5_update ((unsigned char *)":", 1, ctx);
3100 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3101 gen_md5_update ((unsigned char *)":", 1, ctx);
3102 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3103 gen_md5_finish (ctx, hash);
3104 dump_hash (response_digest, hash);
3106 res = xmalloc (strlen (user)
3111 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3112 + (opaque ? strlen (opaque) : 0)
3114 sprintf (res, "Digest \
3115 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3116 user, realm, nonce, path, response_digest);
3119 char *p = res + strlen (res);
3120 strcat (p, ", opaque=\"");
3127 #endif /* ENABLE_DIGEST */
3129 /* Computing the size of a string literal must take into account that
3130 value returned by sizeof includes the terminating \0. */
3131 #define STRSIZE(literal) (sizeof (literal) - 1)
3133 /* Whether chars in [b, e) begin with the literal string provided as
3134 first argument and are followed by whitespace or terminating \0.
3135 The comparison is case-insensitive. */
3136 #define STARTS(literal, b, e) \
3138 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3139 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3140 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3141 || c_isspace (b[STRSIZE (literal)])))
3144 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3146 return STARTS ("Basic", hdrbeg, hdrend)
3147 #ifdef ENABLE_DIGEST
3148 || STARTS ("Digest", hdrbeg, hdrend)
3151 || STARTS ("NTLM", hdrbeg, hdrend)
3158 /* Create the HTTP authorization request header. When the
3159 `WWW-Authenticate' response header is seen, according to the
3160 authorization scheme specified in that header (`Basic' and `Digest'
3161 are supported by the current implementation), produce an
3162 appropriate HTTP authorization request header. */
3164 create_authorization_line (const char *au, const char *user,
3165 const char *passwd, const char *method,
3166 const char *path, bool *finished)
3168 /* We are called only with known schemes, so we can dispatch on the
3170 switch (c_toupper (*au))
3172 case 'B': /* Basic */
3174 return basic_authentication_encode (user, passwd);
3175 #ifdef ENABLE_DIGEST
3176 case 'D': /* Digest */
3178 return digest_authentication_encode (au, user, passwd, method, path);
3181 case 'N': /* NTLM */
3182 if (!ntlm_input (&pconn.ntlm, au))
3187 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3190 /* We shouldn't get here -- this function should be only called
3191 with values approved by known_authentication_scheme_p. */
3199 if (!wget_cookie_jar)
3200 wget_cookie_jar = cookie_jar_new ();
3201 if (opt.cookies_input && !cookies_loaded_p)
3203 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3204 cookies_loaded_p = true;
3211 if (wget_cookie_jar)
3212 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3218 xfree_null (pconn.host);
3219 if (wget_cookie_jar)
3220 cookie_jar_delete (wget_cookie_jar);
3224 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3226 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3228 int len = strlen (ext);
3231 strncpy (shortext, ext, len - 1);
3232 shortext[len - 2] = '\0';
3235 if (last_period_in_local_filename == NULL
3236 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3237 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3239 int local_filename_len = strlen (hs->local_file);
3240 /* Resize the local file, allowing for ".html" preceded by
3241 optional ".NUMBER". */
3242 hs->local_file = xrealloc (hs->local_file,
3243 local_filename_len + 24 + len);
3244 strcpy (hs->local_file + local_filename_len, ext);
3245 /* If clobbering is not allowed and the file, as named,
3246 exists, tack on ".NUMBER.html" instead. */
3247 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3251 sprintf (hs->local_file + local_filename_len,
3252 ".%d%s", ext_num++, ext);
3253 while (file_exists_p (hs->local_file));
3255 *dt |= ADDED_HTML_EXTENSION;
3263 test_parse_content_disposition()
3268 char *opt_dir_prefix;
3272 { "filename=\"file.ext\"", NULL, "file.ext", true },
3273 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3274 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3275 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3276 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3277 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3278 { "attachment", NULL, NULL, false },
3279 { "attachment", "somedir", NULL, false },
3282 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3287 opt.dir_prefix = test_array[i].opt_dir_prefix;
3288 res = parse_content_disposition (test_array[i].hdrval, &filename);
3290 mu_assert ("test_parse_content_disposition: wrong result",
3291 res == test_array[i].result
3293 || 0 == strcmp (test_array[i].filename, filename)));
3299 #endif /* TESTING */
3302 * vim: et sts=2 sw=2 cino+={s