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. */
31 #define USE_GNULIB_ALLOC
58 # include "http-ntlm.h"
71 extern char *version_string;
74 static char *create_authorization_line (const char *, const char *,
75 const char *, const char *,
76 const char *, bool *);
77 static char *basic_authentication_encode (const char *, const char *);
78 static bool known_authentication_scheme_p (const char *, const char *);
79 static void 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"
92 /* Some status code validation macros: */
93 #define H_20X(x) (((x) >= 200) && ((x) < 300))
94 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
95 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
96 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
97 || (x) == HTTP_STATUS_SEE_OTHER \
98 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
100 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
101 /* Successful 2xx. */
102 #define HTTP_STATUS_OK 200
103 #define HTTP_STATUS_CREATED 201
104 #define HTTP_STATUS_ACCEPTED 202
105 #define HTTP_STATUS_NO_CONTENT 204
106 #define HTTP_STATUS_PARTIAL_CONTENTS 206
108 /* Redirection 3xx. */
109 #define HTTP_STATUS_MULTIPLE_CHOICES 300
110 #define HTTP_STATUS_MOVED_PERMANENTLY 301
111 #define HTTP_STATUS_MOVED_TEMPORARILY 302
112 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
113 #define HTTP_STATUS_NOT_MODIFIED 304
114 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
116 /* Client error 4xx. */
117 #define HTTP_STATUS_BAD_REQUEST 400
118 #define HTTP_STATUS_UNAUTHORIZED 401
119 #define HTTP_STATUS_FORBIDDEN 403
120 #define HTTP_STATUS_NOT_FOUND 404
121 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
123 /* Server errors 5xx. */
124 #define HTTP_STATUS_INTERNAL 500
125 #define HTTP_STATUS_NOT_IMPLEMENTED 501
126 #define HTTP_STATUS_BAD_GATEWAY 502
127 #define HTTP_STATUS_UNAVAILABLE 503
130 rel_none, rel_name, rel_value, rel_both
137 struct request_header {
139 enum rp release_policy;
141 int hcount, hcapacity;
144 /* Create a new, empty request. At least request_set_method must be
145 called before the request can be used. */
147 static struct request *
150 struct request *req = xnew0 (struct request);
152 req->headers = xnew_array (struct request_header, req->hcapacity);
156 /* Set the request's method and its arguments. METH should be a
157 literal string (or it should outlive the request) because it will
158 not be freed. ARG will be freed by request_free. */
161 request_set_method (struct request *req, const char *meth, char *arg)
167 /* Return the method string passed with the last call to
168 request_set_method. */
171 request_method (const struct request *req)
176 /* Free one header according to the release policy specified with
177 request_set_header. */
180 release_header (struct request_header *hdr)
182 switch (hdr->release_policy)
199 /* Set the request named NAME to VALUE. Specifically, this means that
200 a "NAME: VALUE\r\n" header line will be used in the request. If a
201 header with the same name previously existed in the request, its
202 value will be replaced by this one. A NULL value means do nothing.
204 RELEASE_POLICY determines whether NAME and VALUE should be released
205 (freed) with request_free. Allowed values are:
207 - rel_none - don't free NAME or VALUE
208 - rel_name - free NAME when done
209 - rel_value - free VALUE when done
210 - rel_both - free both NAME and VALUE when done
212 Setting release policy is useful when arguments come from different
213 sources. For example:
215 // Don't free literal strings!
216 request_set_header (req, "Pragma", "no-cache", rel_none);
218 // Don't free a global variable, we'll need it later.
219 request_set_header (req, "Referer", opt.referer, rel_none);
221 // Value freshly allocated, free it when done.
222 request_set_header (req, "Range",
223 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
228 request_set_header (struct request *req, char *name, char *value,
229 enum rp release_policy)
231 struct request_header *hdr;
236 /* A NULL value is a no-op; if freeing the name is requested,
237 free it now to avoid leaks. */
238 if (release_policy == rel_name || release_policy == rel_both)
243 for (i = 0; i < req->hcount; i++)
245 hdr = &req->headers[i];
246 if (0 == strcasecmp (name, hdr->name))
248 /* Replace existing header. */
249 release_header (hdr);
252 hdr->release_policy = release_policy;
257 /* Install new header. */
259 if (req->hcount >= req->hcapacity)
261 req->hcapacity <<= 1;
262 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
264 hdr = &req->headers[req->hcount++];
267 hdr->release_policy = release_policy;
270 /* Like request_set_header, but sets the whole header line, as
271 provided by the user using the `--header' option. For example,
272 request_set_user_header (req, "Foo: bar") works just like
273 request_set_header (req, "Foo", "bar"). */
276 request_set_user_header (struct request *req, const char *header)
279 const char *p = strchr (header, ':');
282 BOUNDED_TO_ALLOCA (header, p, name);
284 while (c_isspace (*p))
286 request_set_header (req, xstrdup (name), (char *) p, rel_name);
289 /* Remove the header with specified name from REQ. Returns true if
290 the header was actually removed, false otherwise. */
293 request_remove_header (struct request *req, char *name)
296 for (i = 0; i < req->hcount; i++)
298 struct request_header *hdr = &req->headers[i];
299 if (0 == strcasecmp (name, hdr->name))
301 release_header (hdr);
302 /* Move the remaining headers by one. */
303 if (i < req->hcount - 1)
304 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
312 #define APPEND(p, str) do { \
313 int A_len = strlen (str); \
314 memcpy (p, str, A_len); \
318 /* Construct the request and write it to FD using fd_write. */
321 request_send (const struct request *req, int fd)
323 char *request_string, *p;
324 int i, size, write_error;
326 /* Count the request size. */
329 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
330 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
332 for (i = 0; i < req->hcount; i++)
334 struct request_header *hdr = &req->headers[i];
335 /* NAME ": " VALUE "\r\n" */
336 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
342 p = request_string = alloca_array (char, size);
344 /* Generate the request. */
346 APPEND (p, req->method); *p++ = ' ';
347 APPEND (p, req->arg); *p++ = ' ';
348 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
350 for (i = 0; i < req->hcount; i++)
352 struct request_header *hdr = &req->headers[i];
353 APPEND (p, hdr->name);
354 *p++ = ':', *p++ = ' ';
355 APPEND (p, hdr->value);
356 *p++ = '\r', *p++ = '\n';
359 *p++ = '\r', *p++ = '\n', *p++ = '\0';
360 assert (p - request_string == size);
364 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
366 /* Send the request to the server. */
368 write_error = fd_write (fd, request_string, size - 1, -1);
370 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
375 /* Release the resources used by REQ. */
378 request_free (struct request *req)
381 xfree_null (req->arg);
382 for (i = 0; i < req->hcount; i++)
383 release_header (&req->headers[i]);
384 xfree_null (req->headers);
388 static struct hash_table *basic_authed_hosts;
390 /* Find out if this host has issued a Basic challenge yet; if so, give
391 * it the username, password. A temporary measure until we can get
392 * proper authentication in place. */
395 maybe_send_basic_creds (const char *hostname, const char *user,
396 const char *passwd, struct request *req)
398 bool do_challenge = false;
400 if (opt.auth_without_challenge)
402 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
405 else if (basic_authed_hosts
406 && hash_table_contains(basic_authed_hosts, hostname))
408 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
413 DEBUGP(("Host %s has not issued a general basic challenge.\n",
418 request_set_header (req, "Authorization",
419 basic_authentication_encode (user, passwd),
426 register_basic_auth_host (const char *hostname)
428 if (!basic_authed_hosts)
430 basic_authed_hosts = make_nocase_string_hash_table (1);
432 if (!hash_table_contains(basic_authed_hosts, hostname))
434 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
435 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
440 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
441 PROMISED_SIZE bytes are sent over the wire -- if the file is
442 longer, read only that much; if the file is shorter, report an error. */
445 post_file (int sock, const char *file_name, wgint promised_size)
447 static char chunk[8192];
452 DEBUGP (("[writing POST file %s ... ", file_name));
454 fp = fopen (file_name, "rb");
457 while (!feof (fp) && written < promised_size)
460 int length = fread (chunk, 1, sizeof (chunk), fp);
463 towrite = MIN (promised_size - written, length);
464 write_error = fd_write (sock, chunk, towrite, -1);
474 /* If we've written less than was promised, report a (probably
475 nonsensical) error rather than break the promise. */
476 if (written < promised_size)
482 assert (written == promised_size);
483 DEBUGP (("done]\n"));
487 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
488 If so, return the pointer to the position after the line, otherwise
489 return NULL. This is used as callback to fd_read_hunk. The data
490 between START and PEEKED has been read and cannot be "unread"; the
491 data after PEEKED has only been peeked. */
494 response_head_terminator (const char *start, const char *peeked, int peeklen)
498 /* If at first peek, verify whether HUNK starts with "HTTP". If
499 not, this is a HTTP/0.9 request and we must bail out without
501 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
504 /* Look for "\n[\r]\n", and return the following position if found.
505 Start two chars before the current to cover the possibility that
506 part of the terminator (e.g. "\n\r") arrived in the previous
508 p = peeked - start < 2 ? start : peeked - 2;
509 end = peeked + peeklen;
511 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
512 for (; p < end - 2; p++)
515 if (p[1] == '\r' && p[2] == '\n')
517 else if (p[1] == '\n')
520 /* p==end-2: check for \n\n directly preceding END. */
521 if (p[0] == '\n' && p[1] == '\n')
527 /* The maximum size of a single HTTP response we care to read. Rather
528 than being a limit of the reader implementation, this limit
529 prevents Wget from slurping all available memory upon encountering
530 malicious or buggy server output, thus protecting the user. Define
531 it to 0 to remove the limit. */
533 #define HTTP_RESPONSE_MAX_SIZE 65536
535 /* Read the HTTP request head from FD and return it. The error
536 conditions are the same as with fd_read_hunk.
538 To support HTTP/0.9 responses, this function tries to make sure
539 that the data begins with "HTTP". If this is not the case, no data
540 is read and an empty request is returned, so that the remaining
541 data can be treated as body. */
544 read_http_response_head (int fd)
546 return fd_read_hunk (fd, response_head_terminator, 512,
547 HTTP_RESPONSE_MAX_SIZE);
551 /* The response data. */
554 /* The array of pointers that indicate where each header starts.
555 For example, given this HTTP response:
562 The headers are located like this:
564 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
566 headers[0] headers[1] headers[2] headers[3]
568 I.e. headers[0] points to the beginning of the request,
569 headers[1] points to the end of the first header and the
570 beginning of the second one, etc. */
572 const char **headers;
575 /* Create a new response object from the text of the HTTP response,
576 available in HEAD. That text is automatically split into
577 constituent header lines for fast retrieval using
580 static struct response *
581 resp_new (const char *head)
586 struct response *resp = xnew0 (struct response);
591 /* Empty head means that we're dealing with a headerless
592 (HTTP/0.9) response. In that case, don't set HEADERS at
597 /* Split HEAD into header lines, so that resp_header_* functions
598 don't need to do this over and over again. */
604 DO_REALLOC (resp->headers, size, count + 1, const char *);
605 resp->headers[count++] = hdr;
607 /* Break upon encountering an empty line. */
608 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
611 /* Find the end of HDR, including continuations. */
614 const char *end = strchr (hdr, '\n');
620 while (*hdr == ' ' || *hdr == '\t');
622 DO_REALLOC (resp->headers, size, count + 1, const char *);
623 resp->headers[count] = NULL;
628 /* Locate the header named NAME in the request data, starting with
629 position START. This allows the code to loop through the request
630 data, filtering for all requests of a given name. Returns the
631 found position, or -1 for failure. The code that uses this
632 function typically looks like this:
634 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
635 ... do something with header ...
637 If you only care about one header, use resp_header_get instead of
641 resp_header_locate (const struct response *resp, const char *name, int start,
642 const char **begptr, const char **endptr)
645 const char **headers = resp->headers;
648 if (!headers || !headers[1])
651 name_len = strlen (name);
657 for (; headers[i + 1]; i++)
659 const char *b = headers[i];
660 const char *e = headers[i + 1];
662 && b[name_len] == ':'
663 && 0 == strncasecmp (b, name, name_len))
666 while (b < e && c_isspace (*b))
668 while (b < e && c_isspace (e[-1]))
678 /* Find and retrieve the header named NAME in the request data. If
679 found, set *BEGPTR to its starting, and *ENDPTR to its ending
680 position, and return true. Otherwise return false.
682 This function is used as a building block for resp_header_copy
683 and resp_header_strdup. */
686 resp_header_get (const struct response *resp, const char *name,
687 const char **begptr, const char **endptr)
689 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
693 /* Copy the response header named NAME to buffer BUF, no longer than
694 BUFSIZE (BUFSIZE includes the terminating 0). If the header
695 exists, true is returned, false otherwise. If there should be no
696 limit on the size of the header, use resp_header_strdup instead.
698 If BUFSIZE is 0, no data is copied, but the boolean indication of
699 whether the header is present is still returned. */
702 resp_header_copy (const struct response *resp, const char *name,
703 char *buf, int bufsize)
706 if (!resp_header_get (resp, name, &b, &e))
710 int len = MIN (e - b, bufsize - 1);
711 memcpy (buf, b, len);
717 /* Return the value of header named NAME in RESP, allocated with
718 malloc. If such a header does not exist in RESP, return NULL. */
721 resp_header_strdup (const struct response *resp, const char *name)
724 if (!resp_header_get (resp, name, &b, &e))
726 return strdupdelim (b, e);
729 /* Parse the HTTP status line, which is of format:
731 HTTP-Version SP Status-Code SP Reason-Phrase
733 The function returns the status-code, or -1 if the status line
734 appears malformed. The pointer to "reason-phrase" message is
735 returned in *MESSAGE. */
738 resp_status (const struct response *resp, char **message)
745 /* For a HTTP/0.9 response, assume status 200. */
747 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
751 p = resp->headers[0];
752 end = resp->headers[1];
758 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
762 /* Match the HTTP version. This is optional because Gnutella
763 servers have been reported to not specify HTTP version. */
764 if (p < end && *p == '/')
767 while (p < end && c_isdigit (*p))
769 if (p < end && *p == '.')
771 while (p < end && c_isdigit (*p))
775 while (p < end && c_isspace (*p))
777 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
780 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
785 while (p < end && c_isspace (*p))
787 while (p < end && c_isspace (end[-1]))
789 *message = strdupdelim (p, end);
795 /* Release the resources used by RESP. */
798 resp_free (struct response *resp)
800 xfree_null (resp->headers);
804 /* Print a single line of response, the characters [b, e). We tried
806 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
807 but that failed to escape the non-printable characters and, in fact,
808 caused crashes in UTF-8 locales. */
811 print_response_line(const char *prefix, const char *b, const char *e)
814 BOUNDED_TO_ALLOCA(b, e, copy);
815 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
818 /* Print the server response, line by line, omitting the trailing CRLF
819 from individual header lines, and prefixed with PREFIX. */
822 print_server_response (const struct response *resp, const char *prefix)
827 for (i = 0; resp->headers[i + 1]; i++)
829 const char *b = resp->headers[i];
830 const char *e = resp->headers[i + 1];
832 if (b < e && e[-1] == '\n')
834 if (b < e && e[-1] == '\r')
836 print_response_line(prefix, b, e);
840 /* Parse the `Content-Range' header and extract the information it
841 contains. Returns true if successful, false otherwise. */
843 parse_content_range (const char *hdr, wgint *first_byte_ptr,
844 wgint *last_byte_ptr, wgint *entity_length_ptr)
848 /* Ancient versions of Netscape proxy server, presumably predating
849 rfc2068, sent out `Content-Range' without the "bytes"
851 if (0 == strncasecmp (hdr, "bytes", 5))
854 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
858 while (c_isspace (*hdr))
863 if (!c_isdigit (*hdr))
865 for (num = 0; c_isdigit (*hdr); hdr++)
866 num = 10 * num + (*hdr - '0');
867 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
869 *first_byte_ptr = num;
871 for (num = 0; c_isdigit (*hdr); hdr++)
872 num = 10 * num + (*hdr - '0');
873 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
875 *last_byte_ptr = num;
880 for (num = 0; c_isdigit (*hdr); hdr++)
881 num = 10 * num + (*hdr - '0');
882 *entity_length_ptr = num;
886 /* Read the body of the request, but don't store it anywhere and don't
887 display a progress gauge. This is useful for reading the bodies of
888 administrative responses to which we will soon issue another
889 request. The response is not useful to the user, but reading it
890 allows us to continue using the same connection to the server.
892 If reading fails, false is returned, true otherwise. In debug
893 mode, the body is displayed for debugging purposes. */
896 skip_short_body (int fd, wgint contlen)
899 SKIP_SIZE = 512, /* size of the download buffer */
900 SKIP_THRESHOLD = 4096 /* the largest size we read */
902 char dlbuf[SKIP_SIZE + 1];
903 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
905 /* We shouldn't get here with unknown contlen. (This will change
906 with HTTP/1.1, which supports "chunked" transfer.) */
907 assert (contlen != -1);
909 /* If the body is too large, it makes more sense to simply close the
910 connection than to try to read the body. */
911 if (contlen > SKIP_THRESHOLD)
914 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
918 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
921 /* Don't normally report the error since this is an
922 optimization that should be invisible to the user. */
923 DEBUGP (("] aborting (%s).\n",
924 ret < 0 ? fd_errstr (fd) : "EOF received"));
928 /* Safe even if %.*s bogusly expects terminating \0 because
929 we've zero-terminated dlbuf above. */
930 DEBUGP (("%.*s", ret, dlbuf));
933 DEBUGP (("] done.\n"));
937 /* Extract a parameter from the string (typically an HTTP header) at
938 **SOURCE and advance SOURCE to the next parameter. Return false
939 when there are no more parameters to extract. The name of the
940 parameter is returned in NAME, and the value in VALUE. If the
941 parameter has no value, the token's value is zeroed out.
943 For example, if *SOURCE points to the string "attachment;
944 filename=\"foo bar\"", the first call to this function will return
945 the token named "attachment" and no value, and the second call will
946 return the token named "filename" and value "foo bar". The third
947 call will return false, indicating no more valid tokens. */
950 extract_param (const char **source, param_token *name, param_token *value,
953 const char *p = *source;
955 while (c_isspace (*p)) ++p;
959 return false; /* no error; nothing more to extract */
964 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
966 if (name->b == name->e)
967 return false; /* empty name: error */
968 while (c_isspace (*p)) ++p;
969 if (*p == separator || !*p) /* no value */
972 if (*p == separator) ++p;
977 return false; /* error */
979 /* *p is '=', extract value */
981 while (c_isspace (*p)) ++p;
982 if (*p == '"') /* quoted */
985 while (*p && *p != '"') ++p;
989 /* Currently at closing quote; find the end of param. */
990 while (c_isspace (*p)) ++p;
991 while (*p && *p != separator) ++p;
995 /* garbage after closed quote, e.g. foo="bar"baz */
1001 while (*p && *p != separator) ++p;
1003 while (value->e != value->b && c_isspace (value->e[-1]))
1005 if (*p == separator) ++p;
1012 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1014 /* Parse the contents of the `Content-Disposition' header, extracting
1015 the information useful to Wget. Content-Disposition is a header
1016 borrowed from MIME; when used in HTTP, it typically serves for
1017 specifying the desired file name of the resource. For example:
1019 Content-Disposition: attachment; filename="flora.jpg"
1021 Wget will skip the tokens it doesn't care about, such as
1022 "attachment" in the previous example; it will also skip other
1023 unrecognized params. If the header is syntactically correct and
1024 contains a file name, a copy of the file name is stored in
1025 *filename and true is returned. Otherwise, the function returns
1028 The file name is stripped of directory components and must not be
1032 parse_content_disposition (const char *hdr, char **filename)
1034 param_token name, value;
1035 while (extract_param (&hdr, &name, &value, ';'))
1036 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1038 /* Make the file name begin at the last slash or backslash. */
1039 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1040 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1041 if (last_slash && last_bs)
1042 value.b = 1 + MAX (last_slash, last_bs);
1043 else if (last_slash || last_bs)
1044 value.b = 1 + (last_slash ? last_slash : last_bs);
1045 if (value.b == value.e)
1047 /* Start with the directory prefix, if specified. */
1050 int prefix_length = strlen (opt.dir_prefix);
1051 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1056 total_length = prefix_length + (value.e - value.b);
1057 *filename = xmalloc (total_length + 1);
1058 strcpy (*filename, opt.dir_prefix);
1060 (*filename)[prefix_length - 1] = '/';
1061 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1062 (*filename)[total_length] = '\0';
1065 *filename = strdupdelim (value.b, value.e);
1071 /* Persistent connections. Currently, we cache the most recently used
1072 connection as persistent, provided that the HTTP server agrees to
1073 make it such. The persistence data is stored in the variables
1074 below. Ideally, it should be possible to cache an arbitrary fixed
1075 number of these connections. */
1077 /* Whether a persistent connection is active. */
1078 static bool pconn_active;
1081 /* The socket of the connection. */
1084 /* Host and port of the currently active persistent connection. */
1088 /* Whether a ssl handshake has occoured on this connection. */
1091 /* Whether the connection was authorized. This is only done by
1092 NTLM, which authorizes *connections* rather than individual
1093 requests. (That practice is peculiar for HTTP, but it is a
1094 useful optimization.) */
1098 /* NTLM data of the current connection. */
1099 struct ntlmdata ntlm;
1103 /* Mark the persistent connection as invalid and free the resources it
1104 uses. This is used by the CLOSE_* macros after they forcefully
1105 close a registered persistent connection. */
1108 invalidate_persistent (void)
1110 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1111 pconn_active = false;
1112 fd_close (pconn.socket);
1117 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1118 persistent. This will enable someone to use the same connection
1119 later. In the context of HTTP, this must be called only AFTER the
1120 response has been received and the server has promised that the
1121 connection will remain alive.
1123 If a previous connection was persistent, it is closed. */
1126 register_persistent (const char *host, int port, int fd, bool ssl)
1130 if (pconn.socket == fd)
1132 /* The connection FD is already registered. */
1137 /* The old persistent connection is still active; close it
1138 first. This situation arises whenever a persistent
1139 connection exists, but we then connect to a different
1140 host, and try to register a persistent connection to that
1142 invalidate_persistent ();
1146 pconn_active = true;
1148 pconn.host = xstrdup (host);
1151 pconn.authorized = false;
1153 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1156 /* Return true if a persistent connection is available for connecting
1160 persistent_available_p (const char *host, int port, bool ssl,
1161 bool *host_lookup_failed)
1163 /* First, check whether a persistent connection is active at all. */
1167 /* If we want SSL and the last connection wasn't or vice versa,
1168 don't use it. Checking for host and port is not enough because
1169 HTTP and HTTPS can apparently coexist on the same port. */
1170 if (ssl != pconn.ssl)
1173 /* If we're not connecting to the same port, we're not interested. */
1174 if (port != pconn.port)
1177 /* If the host is the same, we're in business. If not, there is
1178 still hope -- read below. */
1179 if (0 != strcasecmp (host, pconn.host))
1181 /* Check if pconn.socket is talking to HOST under another name.
1182 This happens often when both sites are virtual hosts
1183 distinguished only by name and served by the same network
1184 interface, and hence the same web server (possibly set up by
1185 the ISP and serving many different web sites). This
1186 admittedly unconventional optimization does not contradict
1187 HTTP and works well with popular server software. */
1191 struct address_list *al;
1194 /* Don't try to talk to two different SSL sites over the same
1195 secure connection! (Besides, it's not clear that
1196 name-based virtual hosting is even possible with SSL.) */
1199 /* If pconn.socket's peer is one of the IP addresses HOST
1200 resolves to, pconn.socket is for all intents and purposes
1201 already talking to HOST. */
1203 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1205 /* Can't get the peer's address -- something must be very
1206 wrong with the connection. */
1207 invalidate_persistent ();
1210 al = lookup_host (host, 0);
1213 *host_lookup_failed = true;
1217 found = address_list_contains (al, &ip);
1218 address_list_release (al);
1223 /* The persistent connection's peer address was found among the
1224 addresses HOST resolved to; therefore, pconn.sock is in fact
1225 already talking to HOST -- no need to reconnect. */
1228 /* Finally, check whether the connection is still open. This is
1229 important because most servers implement liberal (short) timeout
1230 on persistent connections. Wget can of course always reconnect
1231 if the connection doesn't work out, but it's nicer to know in
1232 advance. This test is a logical followup of the first test, but
1233 is "expensive" and therefore placed at the end of the list.
1235 (Current implementation of test_socket_open has a nice side
1236 effect that it treats sockets with pending data as "closed".
1237 This is exactly what we want: if a broken server sends message
1238 body in response to HEAD, or if it sends more than conent-length
1239 data, we won't reuse the corrupted connection.) */
1241 if (!test_socket_open (pconn.socket))
1243 /* Oops, the socket is no longer open. Now that we know that,
1244 let's invalidate the persistent connection before returning
1246 invalidate_persistent ();
1253 /* The idea behind these two CLOSE macros is to distinguish between
1254 two cases: one when the job we've been doing is finished, and we
1255 want to close the connection and leave, and two when something is
1256 seriously wrong and we're closing the connection as part of
1259 In case of keep_alive, CLOSE_FINISH should leave the connection
1260 open, while CLOSE_INVALIDATE should still close it.
1262 Note that the semantics of the flag `keep_alive' is "this
1263 connection *will* be reused (the server has promised not to close
1264 the connection once we're done)", while the semantics of
1265 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1266 active, registered connection". */
1268 #define CLOSE_FINISH(fd) do { \
1271 if (pconn_active && (fd) == pconn.socket) \
1272 invalidate_persistent (); \
1281 #define CLOSE_INVALIDATE(fd) do { \
1282 if (pconn_active && (fd) == pconn.socket) \
1283 invalidate_persistent (); \
1291 wgint len; /* received length */
1292 wgint contlen; /* expected length */
1293 wgint restval; /* the restart value */
1294 int res; /* the result of last read */
1295 char *rderrmsg; /* error message from read error */
1296 char *newloc; /* new location (redirection) */
1297 char *remote_time; /* remote time-stamp string */
1298 char *error; /* textual HTTP error */
1299 int statcode; /* status code */
1300 wgint rd_size; /* amount of data read from socket */
1301 double dltime; /* time it took to download the data */
1302 const char *referer; /* value of the referer header. */
1303 char *local_file; /* local file name. */
1304 bool existence_checked; /* true if we already checked for a file's
1305 existence after having begun to download
1306 (needed in gethttp for when connection is
1307 interrupted/restarted. */
1308 bool timestamp_checked; /* true if pre-download time-stamping checks
1309 * have already been performed */
1310 char *orig_file_name; /* name of file to compare for time-stamping
1311 * (might be != local_file if -K is set) */
1312 wgint orig_file_size; /* size of file to compare for time-stamping */
1313 time_t orig_file_tstamp; /* time-stamp of file to compare for
1318 free_hstat (struct http_stat *hs)
1320 xfree_null (hs->newloc);
1321 xfree_null (hs->remote_time);
1322 xfree_null (hs->error);
1323 xfree_null (hs->rderrmsg);
1324 xfree_null (hs->local_file);
1325 xfree_null (hs->orig_file_name);
1327 /* Guard against being called twice. */
1329 hs->remote_time = NULL;
1333 #define BEGINS_WITH(line, string_constant) \
1334 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1335 && (c_isspace (line[sizeof (string_constant) - 1]) \
1336 || !line[sizeof (string_constant) - 1]))
1338 #define SET_USER_AGENT(req) do { \
1339 if (!opt.useragent) \
1340 request_set_header (req, "User-Agent", \
1341 aprintf ("Wget/%s", version_string), rel_value); \
1342 else if (*opt.useragent) \
1343 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1346 /* The flags that allow clobbering the file (opening with "wb").
1347 Defined here to avoid repetition later. #### This will require
1349 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1350 || opt.dirstruct || opt.output_document)
1352 /* Retrieve a document through HTTP protocol. It recognizes status
1353 code, and correctly handles redirections. It closes the network
1354 socket. If it receives an error from the functions below it, it
1355 will print it if there is enough information to do so (almost
1356 always), returning the error to the caller (i.e. http_loop).
1358 Various HTTP parameters are stored to hs.
1360 If PROXY is non-NULL, the connection will be made to the proxy
1361 server, and u->url will be requested. */
1363 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1365 struct request *req;
1368 char *user, *passwd;
1372 wgint contlen, contrange;
1379 /* Set to 1 when the authorization has already been sent and should
1380 not be tried again. */
1381 bool auth_finished = false;
1383 /* Set to 1 when just globally-set Basic authorization has been sent;
1384 * should prevent further Basic negotiations, but not other
1386 bool basic_auth_finished = false;
1388 /* Whether NTLM authentication is used for this request. */
1389 bool ntlm_seen = false;
1391 /* Whether our connection to the remote host is through SSL. */
1392 bool using_ssl = false;
1394 /* Whether a HEAD request will be issued (as opposed to GET or
1396 bool head_only = !!(*dt & HEAD_ONLY);
1399 struct response *resp;
1403 /* Whether this connection will be kept alive after the HTTP request
1407 /* Whether keep-alive should be inhibited.
1409 RFC 2068 requests that 1.0 clients not send keep-alive requests
1410 to proxies. This is because many 1.0 proxies do not interpret
1411 the Connection header and transfer it to the remote server,
1412 causing it to not close the connection and leave both the proxy
1413 and the client hanging. */
1414 bool inhibit_keep_alive =
1415 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1417 /* Headers sent when using POST. */
1418 wgint post_data_size = 0;
1420 bool host_lookup_failed = false;
1423 if (u->scheme == SCHEME_HTTPS)
1425 /* Initialize the SSL context. After this has once been done,
1426 it becomes a no-op. */
1429 scheme_disable (SCHEME_HTTPS);
1430 logprintf (LOG_NOTQUIET,
1431 _("Disabling SSL due to encountered errors.\n"));
1432 return SSLINITFAILED;
1435 #endif /* HAVE_SSL */
1437 /* Initialize certain elements of struct http_stat. */
1441 hs->rderrmsg = NULL;
1443 hs->remote_time = NULL;
1448 /* Prepare the request to send. */
1450 req = request_new ();
1453 const char *meth = "GET";
1456 else if (opt.post_file_name || opt.post_data)
1458 /* Use the full path, i.e. one that includes the leading slash and
1459 the query string. E.g. if u->path is "foo/bar" and u->query is
1460 "param=value", full_path will be "/foo/bar?param=value". */
1463 /* When using SSL over proxy, CONNECT establishes a direct
1464 connection to the HTTPS server. Therefore use the same
1465 argument as when talking to the server directly. */
1466 && u->scheme != SCHEME_HTTPS
1469 meth_arg = xstrdup (u->url);
1471 meth_arg = url_full_path (u);
1472 request_set_method (req, meth, meth_arg);
1475 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1476 if (*dt & SEND_NOCACHE)
1477 request_set_header (req, "Pragma", "no-cache", rel_none);
1479 request_set_header (req, "Range",
1480 aprintf ("bytes=%s-",
1481 number_to_static_string (hs->restval)),
1483 SET_USER_AGENT (req);
1484 request_set_header (req, "Accept", "*/*", rel_none);
1486 /* Find the username and password for authentication. */
1489 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1490 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1491 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1494 && !u->user) /* We only do "site-wide" authentication with "global"
1495 user/password values; URL user/password info overrides. */
1497 /* If this is a host for which we've already received a Basic
1498 * challenge, we'll go ahead and send Basic authentication creds. */
1499 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1505 char *proxy_user, *proxy_passwd;
1506 /* For normal username and password, URL components override
1507 command-line/wgetrc parameters. With proxy
1508 authentication, it's the reverse, because proxy URLs are
1509 normally the "permanent" ones, so command-line args
1510 should take precedence. */
1511 if (opt.proxy_user && opt.proxy_passwd)
1513 proxy_user = opt.proxy_user;
1514 proxy_passwd = opt.proxy_passwd;
1518 proxy_user = proxy->user;
1519 proxy_passwd = proxy->passwd;
1521 /* #### This does not appear right. Can't the proxy request,
1522 say, `Digest' authentication? */
1523 if (proxy_user && proxy_passwd)
1524 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1526 /* If we're using a proxy, we will be connecting to the proxy
1530 /* Proxy authorization over SSL is handled below. */
1532 if (u->scheme != SCHEME_HTTPS)
1534 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1537 /* Generate the Host header, HOST:PORT. Take into account that:
1539 - Broken server-side software often doesn't recognize the PORT
1540 argument, so we must generate "Host: www.server.com" instead of
1541 "Host: www.server.com:80" (and likewise for https port).
1543 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1544 becomes ambiguous and needs to be rewritten as "Host:
1545 [3ffe:8100:200:2::2]:1234". */
1547 /* Formats arranged for hfmt[add_port][add_squares]. */
1548 static const char *hfmt[][2] = {
1549 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1551 int add_port = u->port != scheme_default_port (u->scheme);
1552 int add_squares = strchr (u->host, ':') != NULL;
1553 request_set_header (req, "Host",
1554 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1558 if (!inhibit_keep_alive)
1559 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1562 request_set_header (req, "Cookie",
1563 cookie_header (wget_cookie_jar,
1564 u->host, u->port, u->path,
1566 u->scheme == SCHEME_HTTPS
1573 if (opt.post_data || opt.post_file_name)
1575 request_set_header (req, "Content-Type",
1576 "application/x-www-form-urlencoded", rel_none);
1578 post_data_size = strlen (opt.post_data);
1581 post_data_size = file_size (opt.post_file_name);
1582 if (post_data_size == -1)
1584 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1585 quote (opt.post_file_name), strerror (errno));
1589 request_set_header (req, "Content-Length",
1590 xstrdup (number_to_static_string (post_data_size)),
1594 /* Add the user headers. */
1595 if (opt.user_headers)
1598 for (i = 0; opt.user_headers[i]; i++)
1599 request_set_user_header (req, opt.user_headers[i]);
1603 /* We need to come back here when the initial attempt to retrieve
1604 without authorization header fails. (Expected to happen at least
1605 for the Digest authorization scheme.) */
1609 /* Establish the connection. */
1611 if (!inhibit_keep_alive)
1613 /* Look for a persistent connection to target host, unless a
1614 proxy is used. The exception is when SSL is in use, in which
1615 case the proxy is nothing but a passthrough to the target
1616 host, registered as a connection to the latter. */
1617 struct url *relevant = conn;
1619 if (u->scheme == SCHEME_HTTPS)
1623 if (persistent_available_p (relevant->host, relevant->port,
1625 relevant->scheme == SCHEME_HTTPS,
1629 &host_lookup_failed))
1631 sock = pconn.socket;
1632 using_ssl = pconn.ssl;
1633 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1634 escnonprint (pconn.host), pconn.port);
1635 DEBUGP (("Reusing fd %d.\n", sock));
1636 if (pconn.authorized)
1637 /* If the connection is already authorized, the "Basic"
1638 authorization added by code above is unnecessary and
1640 request_remove_header (req, "Authorization");
1642 else if (host_lookup_failed)
1645 logprintf(LOG_NOTQUIET,
1646 _("%s: unable to resolve host address %s\n"),
1647 exec_name, quote (relevant->host));
1654 sock = connect_to_host (conn->host, conn->port);
1663 return (retryable_socket_connect_error (errno)
1664 ? CONERROR : CONIMPOSSIBLE);
1668 if (proxy && u->scheme == SCHEME_HTTPS)
1670 /* When requesting SSL URLs through proxies, use the
1671 CONNECT method to request passthrough. */
1672 struct request *connreq = request_new ();
1673 request_set_method (connreq, "CONNECT",
1674 aprintf ("%s:%d", u->host, u->port));
1675 SET_USER_AGENT (connreq);
1678 request_set_header (connreq, "Proxy-Authorization",
1679 proxyauth, rel_value);
1680 /* Now that PROXYAUTH is part of the CONNECT request,
1681 zero it out so we don't send proxy authorization with
1682 the regular request below. */
1685 /* Examples in rfc2817 use the Host header in CONNECT
1686 requests. I don't see how that gains anything, given
1687 that the contents of Host would be exactly the same as
1688 the contents of CONNECT. */
1690 write_error = request_send (connreq, sock);
1691 request_free (connreq);
1692 if (write_error < 0)
1694 CLOSE_INVALIDATE (sock);
1698 head = read_http_response_head (sock);
1701 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1703 CLOSE_INVALIDATE (sock);
1712 DEBUGP (("proxy responded with: [%s]\n", head));
1714 resp = resp_new (head);
1715 statcode = resp_status (resp, &message);
1718 if (statcode != 200)
1721 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1722 message ? escnonprint (message) : "?");
1723 xfree_null (message);
1726 xfree_null (message);
1728 /* SOCK is now *really* connected to u->host, so update CONN
1729 to reflect this. That way register_persistent will
1730 register SOCK as being connected to u->host:u->port. */
1734 if (conn->scheme == SCHEME_HTTPS)
1736 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1743 #endif /* HAVE_SSL */
1746 /* Send the request to server. */
1747 write_error = request_send (req, sock);
1749 if (write_error >= 0)
1753 DEBUGP (("[POST data: %s]\n", opt.post_data));
1754 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1756 else if (opt.post_file_name && post_data_size != 0)
1757 write_error = post_file (sock, opt.post_file_name, post_data_size);
1760 if (write_error < 0)
1762 CLOSE_INVALIDATE (sock);
1766 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1767 proxy ? "Proxy" : "HTTP");
1772 head = read_http_response_head (sock);
1777 logputs (LOG_NOTQUIET, _("No data received.\n"));
1778 CLOSE_INVALIDATE (sock);
1784 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1786 CLOSE_INVALIDATE (sock);
1791 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1793 resp = resp_new (head);
1795 /* Check for status line. */
1797 statcode = resp_status (resp, &message);
1798 if (!opt.server_response)
1799 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1800 message ? escnonprint (message) : "");
1803 logprintf (LOG_VERBOSE, "\n");
1804 print_server_response (resp, " ");
1807 /* Determine the local filename if needed. Notice that if -O is used
1808 * hstat.local_file is set by http_loop to the argument of -O. */
1809 if (!hs->local_file)
1811 /* Honor Content-Disposition whether possible. */
1812 if (!opt.content_disposition
1813 || !resp_header_copy (resp, "Content-Disposition",
1814 hdrval, sizeof (hdrval))
1815 || !parse_content_disposition (hdrval, &hs->local_file))
1817 /* The Content-Disposition header is missing or broken.
1818 * Choose unique file name according to given URL. */
1819 hs->local_file = url_file_name (u);
1823 /* TODO: perform this check only once. */
1824 if (!hs->existence_checked && file_exists_p (hs->local_file))
1828 /* If opt.noclobber is turned on and file already exists, do not
1829 retrieve the file */
1830 logprintf (LOG_VERBOSE, _("\
1831 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1832 /* If the file is there, we suppose it's retrieved OK. */
1835 /* #### Bogusness alert. */
1836 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1837 if (has_html_suffix_p (hs->local_file))
1840 return RETRUNNEEDED;
1842 else if (!ALLOW_CLOBBER)
1844 char *unique = unique_name (hs->local_file, true);
1845 if (unique != hs->local_file)
1846 xfree (hs->local_file);
1847 hs->local_file = unique;
1850 hs->existence_checked = true;
1852 /* Support timestamping */
1853 /* TODO: move this code out of gethttp. */
1854 if (opt.timestamping && !hs->timestamp_checked)
1856 size_t filename_len = strlen (hs->local_file);
1857 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1858 bool local_dot_orig_file_exists = false;
1859 char *local_filename = NULL;
1862 if (opt.backup_converted)
1863 /* If -K is specified, we'll act on the assumption that it was specified
1864 last time these files were downloaded as well, and instead of just
1865 comparing local file X against server file X, we'll compare local
1866 file X.orig (if extant, else X) against server file X. If -K
1867 _wasn't_ specified last time, or the server contains files called
1868 *.orig, -N will be back to not operating correctly with -k. */
1870 /* Would a single s[n]printf() call be faster? --dan
1872 Definitely not. sprintf() is horribly slow. It's a
1873 different question whether the difference between the two
1874 affects a program. Usually I'd say "no", but at one
1875 point I profiled Wget, and found that a measurable and
1876 non-negligible amount of time was lost calling sprintf()
1877 in url.c. Replacing sprintf with inline calls to
1878 strcpy() and number_to_string() made a difference.
1880 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1881 memcpy (filename_plus_orig_suffix + filename_len,
1882 ".orig", sizeof (".orig"));
1884 /* Try to stat() the .orig file. */
1885 if (stat (filename_plus_orig_suffix, &st) == 0)
1887 local_dot_orig_file_exists = true;
1888 local_filename = filename_plus_orig_suffix;
1892 if (!local_dot_orig_file_exists)
1893 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1894 if (stat (hs->local_file, &st) == 0)
1895 local_filename = hs->local_file;
1897 if (local_filename != NULL)
1898 /* There was a local file, so we'll check later to see if the version
1899 the server has is the same version we already have, allowing us to
1902 hs->orig_file_name = xstrdup (local_filename);
1903 hs->orig_file_size = st.st_size;
1904 hs->orig_file_tstamp = st.st_mtime;
1906 /* Modification time granularity is 2 seconds for Windows, so
1907 increase local time by 1 second for later comparison. */
1908 ++hs->orig_file_tstamp;
1913 if (!opt.ignore_length
1914 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1918 parsed = str_to_wgint (hdrval, NULL, 10);
1919 if (parsed == WGINT_MAX && errno == ERANGE)
1922 #### If Content-Length is out of range, it most likely
1923 means that the file is larger than 2G and that we're
1924 compiled without LFS. In that case we should probably
1925 refuse to even attempt to download the file. */
1928 else if (parsed < 0)
1930 /* Negative Content-Length; nonsensical, so we can't
1931 assume any information about the content to receive. */
1938 /* Check for keep-alive related responses. */
1939 if (!inhibit_keep_alive && contlen != -1)
1941 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1943 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1945 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1950 /* The server has promised that it will not close the connection
1951 when we're done. This means that we can register it. */
1952 register_persistent (conn->host, conn->port, sock, using_ssl);
1954 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1956 /* Authorization is required. */
1957 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1958 CLOSE_FINISH (sock);
1960 CLOSE_INVALIDATE (sock);
1961 pconn.authorized = false;
1962 if (!auth_finished && (user && passwd))
1964 /* IIS sends multiple copies of WWW-Authenticate, one with
1965 the value "negotiate", and other(s) with data. Loop over
1966 all the occurrences and pick the one we recognize. */
1968 const char *wabeg, *waend;
1969 char *www_authenticate = NULL;
1971 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1972 &wabeg, &waend)) != -1;
1974 if (known_authentication_scheme_p (wabeg, waend))
1976 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1980 if (!www_authenticate)
1982 /* If the authentication header is missing or
1983 unrecognized, there's no sense in retrying. */
1984 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1986 else if (!basic_auth_finished
1987 || !BEGINS_WITH (www_authenticate, "Basic"))
1990 pth = url_full_path (u);
1991 request_set_header (req, "Authorization",
1992 create_authorization_line (www_authenticate,
1994 request_method (req),
1998 if (BEGINS_WITH (www_authenticate, "NTLM"))
2000 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2002 /* Need to register this host as using basic auth,
2003 * so we automatically send creds next time. */
2004 register_basic_auth_host (u->host);
2007 goto retry_with_auth;
2011 /* We already did Basic auth, and it failed. Gotta
2015 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2019 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2021 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2023 pconn.authorized = true;
2027 hs->statcode = statcode;
2029 hs->error = xstrdup (_("Malformed status line"));
2031 hs->error = xstrdup (_("(no description)"));
2033 hs->error = xstrdup (message);
2034 xfree_null (message);
2036 type = resp_header_strdup (resp, "Content-Type");
2039 char *tmp = strchr (type, ';');
2042 while (tmp > type && c_isspace (tmp[-1]))
2047 hs->newloc = resp_header_strdup (resp, "Location");
2048 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2050 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2054 const char *scbeg, *scend;
2055 /* The jar should have been created by now. */
2056 assert (wget_cookie_jar != NULL);
2058 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2059 &scbeg, &scend)) != -1;
2062 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2063 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2064 u->path, set_cookie);
2068 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2070 wgint first_byte_pos, last_byte_pos, entity_length;
2071 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2074 contrange = first_byte_pos;
2075 contlen = last_byte_pos - first_byte_pos + 1;
2080 /* 20x responses are counted among successful by default. */
2081 if (H_20X (statcode))
2084 /* Return if redirected. */
2085 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2087 /* RFC2068 says that in case of the 300 (multiple choices)
2088 response, the server can output a preferred URL through
2089 `Location' header; otherwise, the request should be treated
2090 like GET. So, if the location is set, it will be a
2091 redirection; otherwise, just proceed normally. */
2092 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2096 logprintf (LOG_VERBOSE,
2097 _("Location: %s%s\n"),
2098 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2099 hs->newloc ? _(" [following]") : "");
2100 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2101 CLOSE_FINISH (sock);
2103 CLOSE_INVALIDATE (sock);
2109 /* If content-type is not given, assume text/html. This is because
2110 of the multitude of broken CGI's that "forget" to generate the
2113 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2114 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2119 if (opt.html_extension && (*dt & TEXTHTML))
2120 /* -E / --html-extension / html_extension = on was specified, and this is a
2121 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2122 already the file's suffix, tack on ".html". */
2124 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2126 if (last_period_in_local_filename == NULL
2127 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2128 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2130 int local_filename_len = strlen (hs->local_file);
2131 /* Resize the local file, allowing for ".html" preceded by
2132 optional ".NUMBER". */
2133 hs->local_file = xrealloc (hs->local_file,
2134 local_filename_len + 24 + sizeof (".html"));
2135 strcpy(hs->local_file + local_filename_len, ".html");
2136 /* If clobbering is not allowed and the file, as named,
2137 exists, tack on ".NUMBER.html" instead. */
2138 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2142 sprintf (hs->local_file + local_filename_len,
2143 ".%d.html", ext_num++);
2144 while (file_exists_p (hs->local_file));
2146 *dt |= ADDED_HTML_EXTENSION;
2150 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2152 /* If `-c' is in use and the file has been fully downloaded (or
2153 the remote file has shrunk), Wget effectively requests bytes
2154 after the end of file and the server response with 416. */
2155 logputs (LOG_VERBOSE, _("\
2156 \n The file is already fully retrieved; nothing to do.\n\n"));
2157 /* In case the caller inspects. */
2160 /* Mark as successfully retrieved. */
2163 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2164 might be more bytes in the body. */
2165 return RETRUNNEEDED;
2167 if ((contrange != 0 && contrange != hs->restval)
2168 || (H_PARTIAL (statcode) && !contrange))
2170 /* The Range request was somehow misunderstood by the server.
2173 CLOSE_INVALIDATE (sock);
2179 hs->contlen = contlen + contrange;
2185 /* No need to print this output if the body won't be
2186 downloaded at all, or if the original server response is
2188 logputs (LOG_VERBOSE, _("Length: "));
2191 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2192 if (contlen + contrange >= 1024)
2193 logprintf (LOG_VERBOSE, " (%s)",
2194 human_readable (contlen + contrange));
2197 if (contlen >= 1024)
2198 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2199 number_to_static_string (contlen),
2200 human_readable (contlen));
2202 logprintf (LOG_VERBOSE, _(", %s remaining"),
2203 number_to_static_string (contlen));
2207 logputs (LOG_VERBOSE,
2208 opt.ignore_length ? _("ignored") : _("unspecified"));
2210 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2212 logputs (LOG_VERBOSE, "\n");
2216 type = NULL; /* We don't need it any more. */
2218 /* Return if we have no intention of further downloading. */
2219 if (!(*dt & RETROKF) || head_only)
2221 /* In case the caller cares to look... */
2226 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2227 servers not to send body in response to a HEAD request, and
2228 those that do will likely be caught by test_socket_open.
2229 If not, they can be worked around using
2230 `--no-http-keep-alive'. */
2231 CLOSE_FINISH (sock);
2232 else if (keep_alive && skip_short_body (sock, contlen))
2233 /* Successfully skipped the body; also keep using the socket. */
2234 CLOSE_FINISH (sock);
2236 CLOSE_INVALIDATE (sock);
2237 return RETRFINISHED;
2240 /* Open the local file. */
2243 mkalldirs (hs->local_file);
2245 rotate_backups (hs->local_file);
2247 fp = fopen (hs->local_file, "ab");
2248 else if (ALLOW_CLOBBER)
2249 fp = fopen (hs->local_file, "wb");
2252 fp = fopen_excl (hs->local_file, true);
2253 if (!fp && errno == EEXIST)
2255 /* We cannot just invent a new name and use it (which is
2256 what functions like unique_create typically do)
2257 because we told the user we'd use this name.
2258 Instead, return and retry the download. */
2259 logprintf (LOG_NOTQUIET,
2260 _("%s has sprung into existence.\n"),
2262 CLOSE_INVALIDATE (sock);
2263 return FOPEN_EXCL_ERR;
2268 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2269 CLOSE_INVALIDATE (sock);
2276 /* Print fetch message, if opt.verbose. */
2279 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2280 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2283 /* This confuses the timestamping code that checks for file size.
2284 #### The timestamping code should be smarter about file size. */
2285 if (opt.save_headers && hs->restval == 0)
2286 fwrite (head, 1, strlen (head), fp);
2288 /* Now we no longer need to store the response header. */
2291 /* Download the request body. */
2294 /* If content-length is present, read that much; otherwise, read
2295 until EOF. The HTTP spec doesn't require the server to
2296 actually close the connection when it's done sending data. */
2297 flags |= rb_read_exactly;
2298 if (hs->restval > 0 && contrange == 0)
2299 /* If the server ignored our range request, instruct fd_read_body
2300 to skip the first RESTVAL bytes of body. */
2301 flags |= rb_skip_startpos;
2302 hs->len = hs->restval;
2304 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2305 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2309 CLOSE_FINISH (sock);
2313 hs->rderrmsg = xstrdup (fd_errstr (sock));
2314 CLOSE_INVALIDATE (sock);
2321 return RETRFINISHED;
2324 /* The genuine HTTP loop! This is the part where the retrieval is
2325 retried, and retried, and retried, and... */
2327 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2328 int *dt, struct url *proxy)
2331 bool got_head = false; /* used for time-stamping and filename detection */
2332 bool time_came_from_head = false;
2333 bool got_name = false;
2336 uerr_t err, ret = TRYLIMEXC;
2337 time_t tmr = -1; /* remote time-stamp */
2338 struct http_stat hstat; /* HTTP status */
2340 bool send_head_first = true;
2342 /* Assert that no value for *LOCAL_FILE was passed. */
2343 assert (local_file == NULL || *local_file == NULL);
2345 /* Set LOCAL_FILE parameter. */
2346 if (local_file && opt.output_document)
2347 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2349 /* Reset NEWLOC parameter. */
2352 /* This used to be done in main(), but it's a better idea to do it
2353 here so that we don't go through the hoops if we're just using
2358 /* Warn on (likely bogus) wildcard usage in HTTP. */
2359 if (opt.ftp_glob && has_wildcards_p (u->path))
2360 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2362 /* Setup hstat struct. */
2364 hstat.referer = referer;
2366 if (opt.output_document)
2368 hstat.local_file = xstrdup (opt.output_document);
2371 else if (!opt.content_disposition)
2373 hstat.local_file = url_file_name (u);
2377 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2378 * screaming for some refactoring. */
2379 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber)
2381 /* If opt.noclobber is turned on and file already exists, do not
2382 retrieve the file */
2383 logprintf (LOG_VERBOSE, _("\
2384 File %s already there; not retrieving.\n\n"),
2385 quote (hstat.local_file));
2386 /* If the file is there, we suppose it's retrieved OK. */
2389 /* #### Bogusness alert. */
2390 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2391 if (has_html_suffix_p (hstat.local_file))
2394 return RETRUNNEEDED;
2397 /* Reset the counter. */
2400 /* Reset the document type. */
2403 /* Skip preliminary HEAD request if we're not in spider mode AND
2404 * if -O was given or HTTP Content-Disposition support is disabled. */
2406 && (got_name || !opt.content_disposition))
2407 send_head_first = false;
2409 /* Send preliminary HEAD request if -N is given and we have an existing
2410 * destination file. */
2411 if (opt.timestamping
2412 && !opt.content_disposition
2413 && file_exists_p (url_file_name (u)))
2414 send_head_first = true;
2419 /* Increment the pass counter. */
2421 sleep_between_retrievals (count);
2423 /* Get the current time string. */
2424 tms = datetime_str (time (NULL));
2426 if (opt.spider && !got_head)
2427 logprintf (LOG_VERBOSE, _("\
2428 Spider mode enabled. Check if remote file exists.\n"));
2430 /* Print fetch message, if opt.verbose. */
2433 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2438 sprintf (tmp, _("(try:%2d)"), count);
2439 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2444 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2449 ws_changetitle (hurl);
2454 /* Default document type is empty. However, if spider mode is
2455 on or time-stamping is employed, HEAD_ONLY commands is
2456 encoded within *dt. */
2457 if (send_head_first && !got_head)
2462 /* Decide whether or not to restart. */
2465 && stat (hstat.local_file, &st) == 0
2466 && S_ISREG (st.st_mode))
2467 /* When -c is used, continue from on-disk size. (Can't use
2468 hstat.len even if count>1 because we don't want a failed
2469 first attempt to clobber existing data.) */
2470 hstat.restval = st.st_size;
2472 /* otherwise, continue where the previous try left off */
2473 hstat.restval = hstat.len;
2477 /* Decide whether to send the no-cache directive. We send it in
2479 a) we're using a proxy, and we're past our first retrieval.
2480 Some proxies are notorious for caching incomplete data, so
2481 we require a fresh get.
2482 b) caching is explicitly inhibited. */
2483 if ((proxy && count > 1) /* a */
2484 || !opt.allow_cache) /* b */
2485 *dt |= SEND_NOCACHE;
2487 *dt &= ~SEND_NOCACHE;
2489 /* Try fetching the document, or at least its head. */
2490 err = gethttp (u, &hstat, dt, proxy);
2493 tms = datetime_str (time (NULL));
2495 /* Get the new location (with or without the redirection). */
2497 *newloc = xstrdup (hstat.newloc);
2501 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2502 case CONERROR: case READERR: case WRITEFAILED:
2503 case RANGEERR: case FOPEN_EXCL_ERR:
2504 /* Non-fatal errors continue executing the loop, which will
2505 bring them to "while" statement at the end, to judge
2506 whether the number of tries was exceeded. */
2507 printwhat (count, opt.ntry);
2509 case FWRITEERR: case FOPENERR:
2510 /* Another fatal error. */
2511 logputs (LOG_VERBOSE, "\n");
2512 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2513 quote (hstat.local_file), strerror (errno));
2514 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2515 case SSLINITFAILED: case CONTNOTSUPPORTED:
2516 /* Fatal errors just return from the function. */
2520 /* Another fatal error. */
2521 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2525 /* Return the new location to the caller. */
2528 logprintf (LOG_NOTQUIET,
2529 _("ERROR: Redirection (%d) without location.\n"),
2539 /* The file was already fully retrieved. */
2543 /* Deal with you later. */
2546 /* All possibilities should have been exhausted. */
2550 if (!(*dt & RETROKF))
2555 /* #### Ugly ugly ugly! */
2556 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2557 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2560 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2562 && (hstat.statcode == 500 || hstat.statcode == 501))
2567 /* Maybe we should always keep track of broken links, not just in
2569 else if (opt.spider)
2571 /* #### Again: ugly ugly ugly! */
2573 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2574 nonexisting_url (hurl);
2575 logprintf (LOG_NOTQUIET, _("\
2576 Remote file does not exist -- broken link!!!\n"));
2580 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2581 tms, hstat.statcode, escnonprint (hstat.error));
2583 logputs (LOG_VERBOSE, "\n");
2589 /* Did we get the time-stamp? */
2592 got_head = true; /* no more time-stamping */
2594 if (opt.timestamping && !hstat.remote_time)
2596 logputs (LOG_NOTQUIET, _("\
2597 Last-modified header missing -- time-stamps turned off.\n"));
2599 else if (hstat.remote_time)
2601 /* Convert the date-string into struct tm. */
2602 tmr = http_atotm (hstat.remote_time);
2603 if (tmr == (time_t) (-1))
2604 logputs (LOG_VERBOSE, _("\
2605 Last-modified header invalid -- time-stamp ignored.\n"));
2606 if (*dt & HEAD_ONLY)
2607 time_came_from_head = true;
2610 if (send_head_first)
2612 /* The time-stamping section. */
2613 if (opt.timestamping)
2615 if (hstat.orig_file_name) /* Perform the following
2616 checks only if the file
2618 download already exists. */
2620 if (hstat.remote_time &&
2621 tmr != (time_t) (-1))
2623 /* Now time-stamping can be used validly.
2624 Time-stamping means that if the sizes of
2625 the local and remote file match, and local
2626 file is newer than the remote file, it will
2627 not be retrieved. Otherwise, the normal
2628 download procedure is resumed. */
2629 if (hstat.orig_file_tstamp >= tmr)
2631 if (hstat.contlen == -1
2632 || hstat.orig_file_size == hstat.contlen)
2634 logprintf (LOG_VERBOSE, _("\
2635 Server file no newer than local file %s -- not retrieving.\n\n"),
2636 quote (hstat.orig_file_name));
2642 logprintf (LOG_VERBOSE, _("\
2643 The sizes do not match (local %s) -- retrieving.\n"),
2644 number_to_static_string (hstat.orig_file_size));
2648 logputs (LOG_VERBOSE,
2649 _("Remote file is newer, retrieving.\n"));
2651 logputs (LOG_VERBOSE, "\n");
2655 /* free_hstat (&hstat); */
2656 hstat.timestamp_checked = true;
2665 logputs (LOG_VERBOSE, _("\
2666 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2670 logprintf (LOG_VERBOSE, _("\
2671 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2672 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2680 logprintf (LOG_VERBOSE, _("\
2681 Remote file exists and could contain further links,\n\
2682 but recursion is disabled -- not retrieving.\n\n"));
2686 logprintf (LOG_VERBOSE, _("\
2687 Remote file exists.\n\n"));
2689 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2696 count = 0; /* the retrieve count for HEAD is reset */
2698 } /* send_head_first */
2701 if ((tmr != (time_t) (-1))
2702 && ((hstat.len == hstat.contlen) ||
2703 ((hstat.res == 0) && (hstat.contlen == -1))))
2705 /* #### This code repeats in http.c and ftp.c. Move it to a
2707 const char *fl = NULL;
2708 if (opt.output_document)
2710 if (output_stream_regular)
2711 fl = opt.output_document;
2714 fl = hstat.local_file;
2718 /* Reparse time header, in case it's changed. */
2719 if (time_came_from_head
2720 && hstat.remote_time && hstat.remote_time[0])
2722 newtmr = http_atotm (hstat.remote_time);
2729 /* End of time-stamping section. */
2731 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2732 total_download_time += hstat.dltime;
2734 if (hstat.len == hstat.contlen)
2738 logprintf (LOG_VERBOSE,
2739 _("%s (%s) - %s saved [%s/%s]\n\n"),
2740 tms, tmrate, quote (hstat.local_file),
2741 number_to_static_string (hstat.len),
2742 number_to_static_string (hstat.contlen));
2743 logprintf (LOG_NONVERBOSE,
2744 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2746 number_to_static_string (hstat.len),
2747 number_to_static_string (hstat.contlen),
2748 hstat.local_file, count);
2751 total_downloaded_bytes += hstat.len;
2753 /* Remember that we downloaded the file for later ".orig" code. */
2754 if (*dt & ADDED_HTML_EXTENSION)
2755 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2757 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2762 else if (hstat.res == 0) /* No read error */
2764 if (hstat.contlen == -1) /* We don't know how much we were supposed
2765 to get, so assume we succeeded. */
2769 logprintf (LOG_VERBOSE,
2770 _("%s (%s) - %s saved [%s]\n\n"),
2771 tms, tmrate, quote (hstat.local_file),
2772 number_to_static_string (hstat.len));
2773 logprintf (LOG_NONVERBOSE,
2774 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2775 tms, u->url, number_to_static_string (hstat.len),
2776 hstat.local_file, count);
2779 total_downloaded_bytes += hstat.len;
2781 /* Remember that we downloaded the file for later ".orig" code. */
2782 if (*dt & ADDED_HTML_EXTENSION)
2783 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2785 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2790 else if (hstat.len < hstat.contlen) /* meaning we lost the
2791 connection too soon */
2793 logprintf (LOG_VERBOSE,
2794 _("%s (%s) - Connection closed at byte %s. "),
2795 tms, tmrate, number_to_static_string (hstat.len));
2796 printwhat (count, opt.ntry);
2800 /* Getting here would mean reading more data than
2801 requested with content-length, which we never do. */
2804 else /* from now on hstat.res can only be -1 */
2806 if (hstat.contlen == -1)
2808 logprintf (LOG_VERBOSE,
2809 _("%s (%s) - Read error at byte %s (%s)."),
2810 tms, tmrate, number_to_static_string (hstat.len),
2812 printwhat (count, opt.ntry);
2815 else /* hstat.res == -1 and contlen is given */
2817 logprintf (LOG_VERBOSE,
2818 _("%s (%s) - Read error at byte %s/%s (%s). "),
2820 number_to_static_string (hstat.len),
2821 number_to_static_string (hstat.contlen),
2823 printwhat (count, opt.ntry);
2829 while (!opt.ntry || (count < opt.ntry));
2833 *local_file = xstrdup (hstat.local_file);
2834 free_hstat (&hstat);
2839 /* Check whether the result of strptime() indicates success.
2840 strptime() returns the pointer to how far it got to in the string.
2841 The processing has been successful if the string is at `GMT' or
2842 `+X', or at the end of the string.
2844 In extended regexp parlance, the function returns 1 if P matches
2845 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2846 can return) is considered a failure and 0 is returned. */
2848 check_end (const char *p)
2852 while (c_isspace (*p))
2855 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2856 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2862 /* Convert the textual specification of time in TIME_STRING to the
2863 number of seconds since the Epoch.
2865 TIME_STRING can be in any of the three formats RFC2616 allows the
2866 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2867 as well as the time format used in the Set-Cookie header.
2868 Timezones are ignored, and should be GMT.
2870 Return the computed time_t representation, or -1 if the conversion
2873 This function uses strptime with various string formats for parsing
2874 TIME_STRING. This results in a parser that is not as lenient in
2875 interpreting TIME_STRING as I would like it to be. Being based on
2876 strptime, it always allows shortened months, one-digit days, etc.,
2877 but due to the multitude of formats in which time can be
2878 represented, an ideal HTTP time parser would be even more
2879 forgiving. It should completely ignore things like week days and
2880 concentrate only on the various forms of representing years,
2881 months, days, hours, minutes, and seconds. For example, it would
2882 be nice if it accepted ISO 8601 out of the box.
2884 I've investigated free and PD code for this purpose, but none was
2885 usable. getdate was big and unwieldy, and had potential copyright
2886 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2887 distributed with phttpd, is excellent, but we cannot use it because
2888 it is not assigned to the FSF. So I stuck it with strptime. */
2891 http_atotm (const char *time_string)
2893 /* NOTE: Solaris strptime man page claims that %n and %t match white
2894 space, but that's not universally available. Instead, we simply
2895 use ` ' to mean "skip all WS", which works under all strptime
2896 implementations I've tested. */
2898 static const char *time_formats[] = {
2899 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2900 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2901 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2902 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2903 (used in Set-Cookie, defined in the
2904 Netscape cookie specification.) */
2906 const char *oldlocale;
2908 time_t ret = (time_t) -1;
2910 /* Solaris strptime fails to recognize English month names in
2911 non-English locales, which we work around by temporarily setting
2912 locale to C before invoking strptime. */
2913 oldlocale = setlocale (LC_TIME, NULL);
2914 setlocale (LC_TIME, "C");
2916 for (i = 0; i < countof (time_formats); i++)
2920 /* Some versions of strptime use the existing contents of struct
2921 tm to recalculate the date according to format. Zero it out
2922 to prevent stack garbage from influencing strptime. */
2925 if (check_end (strptime (time_string, time_formats[i], &t)))
2932 /* Restore the previous locale. */
2933 setlocale (LC_TIME, oldlocale);
2938 /* Authorization support: We support three authorization schemes:
2940 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2942 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2943 consisting of answering to the server's challenge with the proper
2946 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2947 Stenberg for libcurl. Like digest, NTLM is based on a
2948 challenge-response mechanism, but unlike digest, it is non-standard
2949 (authenticates TCP connections rather than requests), undocumented
2950 and Microsoft-specific. */
2952 /* Create the authentication header contents for the `Basic' scheme.
2953 This is done by encoding the string "USER:PASS" to base64 and
2954 prepending the string "Basic " in front of it. */
2957 basic_authentication_encode (const char *user, const char *passwd)
2960 int len1 = strlen (user) + 1 + strlen (passwd);
2962 t1 = (char *)alloca (len1 + 1);
2963 sprintf (t1, "%s:%s", user, passwd);
2965 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2966 base64_encode (t1, len1, t2);
2968 return concat_strings ("Basic ", t2, (char *) 0);
2971 #define SKIP_WS(x) do { \
2972 while (c_isspace (*(x))) \
2976 #ifdef ENABLE_DIGEST
2977 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2978 an array of 16 bytes containing the hash keys, and BUF should be a
2979 buffer of 33 writable characters (32 for hex digits plus one for
2980 zero termination). */
2982 dump_hash (char *buf, const unsigned char *hash)
2986 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2988 *buf++ = XNUM_TO_digit (*hash >> 4);
2989 *buf++ = XNUM_TO_digit (*hash & 0xf);
2994 /* Take the line apart to find the challenge, and compose a digest
2995 authorization header. See RFC2069 section 2.1.2. */
2997 digest_authentication_encode (const char *au, const char *user,
2998 const char *passwd, const char *method,
3001 static char *realm, *opaque, *nonce;
3006 { "realm", &realm },
3007 { "opaque", &opaque },
3011 param_token name, value;
3013 realm = opaque = nonce = NULL;
3015 au += 6; /* skip over `Digest' */
3016 while (extract_param (&au, &name, &value, ','))
3019 for (i = 0; i < countof (options); i++)
3020 if (name.e - name.b == strlen (options[i].name)
3021 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
3023 *options[i].variable = strdupdelim (value.b, value.e);
3027 if (!realm || !nonce || !user || !passwd || !path || !method)
3030 xfree_null (opaque);
3035 /* Calculate the digest value. */
3037 ALLOCA_MD5_CONTEXT (ctx);
3038 unsigned char hash[MD5_HASHLEN];
3039 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3040 char response_digest[MD5_HASHLEN * 2 + 1];
3042 /* A1BUF = H(user ":" realm ":" password) */
3044 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3045 gen_md5_update ((unsigned char *)":", 1, ctx);
3046 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3047 gen_md5_update ((unsigned char *)":", 1, ctx);
3048 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3049 gen_md5_finish (ctx, hash);
3050 dump_hash (a1buf, hash);
3052 /* A2BUF = H(method ":" path) */
3054 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3055 gen_md5_update ((unsigned char *)":", 1, ctx);
3056 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3057 gen_md5_finish (ctx, hash);
3058 dump_hash (a2buf, hash);
3060 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3062 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3063 gen_md5_update ((unsigned char *)":", 1, ctx);
3064 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3065 gen_md5_update ((unsigned char *)":", 1, ctx);
3066 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3067 gen_md5_finish (ctx, hash);
3068 dump_hash (response_digest, hash);
3070 res = xmalloc (strlen (user)
3075 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3076 + (opaque ? strlen (opaque) : 0)
3078 sprintf (res, "Digest \
3079 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3080 user, realm, nonce, path, response_digest);
3083 char *p = res + strlen (res);
3084 strcat (p, ", opaque=\"");
3091 #endif /* ENABLE_DIGEST */
3093 /* Computing the size of a string literal must take into account that
3094 value returned by sizeof includes the terminating \0. */
3095 #define STRSIZE(literal) (sizeof (literal) - 1)
3097 /* Whether chars in [b, e) begin with the literal string provided as
3098 first argument and are followed by whitespace or terminating \0.
3099 The comparison is case-insensitive. */
3100 #define STARTS(literal, b, e) \
3101 ((e) - (b) >= STRSIZE (literal) \
3102 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3103 && ((e) - (b) == STRSIZE (literal) \
3104 || c_isspace (b[STRSIZE (literal)])))
3107 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3109 return STARTS ("Basic", hdrbeg, hdrend)
3110 #ifdef ENABLE_DIGEST
3111 || STARTS ("Digest", hdrbeg, hdrend)
3114 || STARTS ("NTLM", hdrbeg, hdrend)
3121 /* Create the HTTP authorization request header. When the
3122 `WWW-Authenticate' response header is seen, according to the
3123 authorization scheme specified in that header (`Basic' and `Digest'
3124 are supported by the current implementation), produce an
3125 appropriate HTTP authorization request header. */
3127 create_authorization_line (const char *au, const char *user,
3128 const char *passwd, const char *method,
3129 const char *path, bool *finished)
3131 /* We are called only with known schemes, so we can dispatch on the
3133 switch (c_toupper (*au))
3135 case 'B': /* Basic */
3137 return basic_authentication_encode (user, passwd);
3138 #ifdef ENABLE_DIGEST
3139 case 'D': /* Digest */
3141 return digest_authentication_encode (au, user, passwd, method, path);
3144 case 'N': /* NTLM */
3145 if (!ntlm_input (&pconn.ntlm, au))
3150 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3153 /* We shouldn't get here -- this function should be only called
3154 with values approved by known_authentication_scheme_p. */
3162 if (!wget_cookie_jar)
3163 wget_cookie_jar = cookie_jar_new ();
3164 if (opt.cookies_input && !cookies_loaded_p)
3166 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3167 cookies_loaded_p = true;
3174 if (wget_cookie_jar)
3175 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3181 xfree_null (pconn.host);
3182 if (wget_cookie_jar)
3183 cookie_jar_delete (wget_cookie_jar);
3190 test_parse_content_disposition()
3195 char *opt_dir_prefix;
3199 { "filename=\"file.ext\"", NULL, "file.ext", true },
3200 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3201 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3202 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3203 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3204 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3205 { "attachment", NULL, NULL, false },
3206 { "attachment", "somedir", NULL, false },
3209 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3214 opt.dir_prefix = test_array[i].opt_dir_prefix;
3215 res = parse_content_disposition (test_array[i].hdrval, &filename);
3217 mu_assert ("test_parse_content_disposition: wrong result",
3218 res == test_array[i].result
3220 || 0 == strcmp (test_array[i].filename, filename)));
3226 #endif /* TESTING */
3229 * vim: et sts=2 sw=2 cino+={s