2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007 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 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
56 # include "http-ntlm.h"
69 extern char *version_string;
72 static char *create_authorization_line (const char *, const char *,
73 const char *, const char *,
74 const char *, bool *);
75 static char *basic_authentication_encode (const char *, const char *);
76 static bool known_authentication_scheme_p (const char *, const char *);
77 static void load_cookies (void);
80 # define MIN(x, y) ((x) > (y) ? (y) : (x))
84 static bool cookies_loaded_p;
85 static struct cookie_jar *wget_cookie_jar;
87 #define TEXTHTML_S "text/html"
88 #define TEXTXHTML_S "application/xhtml+xml"
90 /* Some status code validation macros: */
91 #define H_20X(x) (((x) >= 200) && ((x) < 300))
92 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
93 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
94 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
95 || (x) == HTTP_STATUS_SEE_OTHER \
96 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
98 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
100 #define HTTP_STATUS_OK 200
101 #define HTTP_STATUS_CREATED 201
102 #define HTTP_STATUS_ACCEPTED 202
103 #define HTTP_STATUS_NO_CONTENT 204
104 #define HTTP_STATUS_PARTIAL_CONTENTS 206
106 /* Redirection 3xx. */
107 #define HTTP_STATUS_MULTIPLE_CHOICES 300
108 #define HTTP_STATUS_MOVED_PERMANENTLY 301
109 #define HTTP_STATUS_MOVED_TEMPORARILY 302
110 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
111 #define HTTP_STATUS_NOT_MODIFIED 304
112 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
114 /* Client error 4xx. */
115 #define HTTP_STATUS_BAD_REQUEST 400
116 #define HTTP_STATUS_UNAUTHORIZED 401
117 #define HTTP_STATUS_FORBIDDEN 403
118 #define HTTP_STATUS_NOT_FOUND 404
119 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
121 /* Server errors 5xx. */
122 #define HTTP_STATUS_INTERNAL 500
123 #define HTTP_STATUS_NOT_IMPLEMENTED 501
124 #define HTTP_STATUS_BAD_GATEWAY 502
125 #define HTTP_STATUS_UNAVAILABLE 503
128 rel_none, rel_name, rel_value, rel_both
135 struct request_header {
137 enum rp release_policy;
139 int hcount, hcapacity;
142 /* Create a new, empty request. At least request_set_method must be
143 called before the request can be used. */
145 static struct request *
148 struct request *req = xnew0 (struct request);
150 req->headers = xnew_array (struct request_header, req->hcapacity);
154 /* Set the request's method and its arguments. METH should be a
155 literal string (or it should outlive the request) because it will
156 not be freed. ARG will be freed by request_free. */
159 request_set_method (struct request *req, const char *meth, char *arg)
165 /* Return the method string passed with the last call to
166 request_set_method. */
169 request_method (const struct request *req)
174 /* Free one header according to the release policy specified with
175 request_set_header. */
178 release_header (struct request_header *hdr)
180 switch (hdr->release_policy)
197 /* Set the request named NAME to VALUE. Specifically, this means that
198 a "NAME: VALUE\r\n" header line will be used in the request. If a
199 header with the same name previously existed in the request, its
200 value will be replaced by this one. A NULL value means do nothing.
202 RELEASE_POLICY determines whether NAME and VALUE should be released
203 (freed) with request_free. Allowed values are:
205 - rel_none - don't free NAME or VALUE
206 - rel_name - free NAME when done
207 - rel_value - free VALUE when done
208 - rel_both - free both NAME and VALUE when done
210 Setting release policy is useful when arguments come from different
211 sources. For example:
213 // Don't free literal strings!
214 request_set_header (req, "Pragma", "no-cache", rel_none);
216 // Don't free a global variable, we'll need it later.
217 request_set_header (req, "Referer", opt.referer, rel_none);
219 // Value freshly allocated, free it when done.
220 request_set_header (req, "Range",
221 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
226 request_set_header (struct request *req, char *name, char *value,
227 enum rp release_policy)
229 struct request_header *hdr;
234 /* A NULL value is a no-op; if freeing the name is requested,
235 free it now to avoid leaks. */
236 if (release_policy == rel_name || release_policy == rel_both)
241 for (i = 0; i < req->hcount; i++)
243 hdr = &req->headers[i];
244 if (0 == strcasecmp (name, hdr->name))
246 /* Replace existing header. */
247 release_header (hdr);
250 hdr->release_policy = release_policy;
255 /* Install new header. */
257 if (req->hcount >= req->hcapacity)
259 req->hcapacity <<= 1;
260 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
262 hdr = &req->headers[req->hcount++];
265 hdr->release_policy = release_policy;
268 /* Like request_set_header, but sets the whole header line, as
269 provided by the user using the `--header' option. For example,
270 request_set_user_header (req, "Foo: bar") works just like
271 request_set_header (req, "Foo", "bar"). */
274 request_set_user_header (struct request *req, const char *header)
277 const char *p = strchr (header, ':');
280 BOUNDED_TO_ALLOCA (header, p, name);
284 request_set_header (req, xstrdup (name), (char *) p, rel_name);
287 /* Remove the header with specified name from REQ. Returns true if
288 the header was actually removed, false otherwise. */
291 request_remove_header (struct request *req, char *name)
294 for (i = 0; i < req->hcount; i++)
296 struct request_header *hdr = &req->headers[i];
297 if (0 == strcasecmp (name, hdr->name))
299 release_header (hdr);
300 /* Move the remaining headers by one. */
301 if (i < req->hcount - 1)
302 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
310 #define APPEND(p, str) do { \
311 int A_len = strlen (str); \
312 memcpy (p, str, A_len); \
316 /* Construct the request and write it to FD using fd_write. */
319 request_send (const struct request *req, int fd)
321 char *request_string, *p;
322 int i, size, write_error;
324 /* Count the request size. */
327 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
328 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
330 for (i = 0; i < req->hcount; i++)
332 struct request_header *hdr = &req->headers[i];
333 /* NAME ": " VALUE "\r\n" */
334 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
340 p = request_string = alloca_array (char, size);
342 /* Generate the request. */
344 APPEND (p, req->method); *p++ = ' ';
345 APPEND (p, req->arg); *p++ = ' ';
346 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
348 for (i = 0; i < req->hcount; i++)
350 struct request_header *hdr = &req->headers[i];
351 APPEND (p, hdr->name);
352 *p++ = ':', *p++ = ' ';
353 APPEND (p, hdr->value);
354 *p++ = '\r', *p++ = '\n';
357 *p++ = '\r', *p++ = '\n', *p++ = '\0';
358 assert (p - request_string == size);
362 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
364 /* Send the request to the server. */
366 write_error = fd_write (fd, request_string, size - 1, -1);
368 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
373 /* Release the resources used by REQ. */
376 request_free (struct request *req)
379 xfree_null (req->arg);
380 for (i = 0; i < req->hcount; i++)
381 release_header (&req->headers[i]);
382 xfree_null (req->headers);
386 static struct hash_table *basic_authed_hosts;
388 /* Find out if this host has issued a Basic challenge yet; if so, give
389 * it the username, password. A temporary measure until we can get
390 * proper authentication in place. */
393 maybe_send_basic_creds (const char *hostname, const char *user,
394 const char *passwd, struct request *req)
396 int did_challenge = 0;
398 if (basic_authed_hosts
399 && hash_table_contains(basic_authed_hosts, hostname))
401 DEBUGP(("Found `%s' in basic_authed_hosts.\n", hostname));
402 request_set_header (req, "Authorization",
403 basic_authentication_encode (user, passwd),
409 DEBUGP(("Host `%s' has not issued a general basic challenge.\n",
412 return did_challenge;
416 register_basic_auth_host (const char *hostname)
418 if (!basic_authed_hosts)
420 basic_authed_hosts = make_nocase_string_hash_table (1);
422 if (!hash_table_contains(basic_authed_hosts, hostname))
424 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
425 DEBUGP(("Inserted `%s' into basic_authed_hosts\n", hostname));
430 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
431 PROMISED_SIZE bytes are sent over the wire -- if the file is
432 longer, read only that much; if the file is shorter, report an error. */
435 post_file (int sock, const char *file_name, wgint promised_size)
437 static char chunk[8192];
442 DEBUGP (("[writing POST file %s ... ", file_name));
444 fp = fopen (file_name, "rb");
447 while (!feof (fp) && written < promised_size)
450 int length = fread (chunk, 1, sizeof (chunk), fp);
453 towrite = MIN (promised_size - written, length);
454 write_error = fd_write (sock, chunk, towrite, -1);
464 /* If we've written less than was promised, report a (probably
465 nonsensical) error rather than break the promise. */
466 if (written < promised_size)
472 assert (written == promised_size);
473 DEBUGP (("done]\n"));
477 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
478 If so, return the pointer to the position after the line, otherwise
479 return NULL. This is used as callback to fd_read_hunk. The data
480 between START and PEEKED has been read and cannot be "unread"; the
481 data after PEEKED has only been peeked. */
484 response_head_terminator (const char *start, const char *peeked, int peeklen)
488 /* If at first peek, verify whether HUNK starts with "HTTP". If
489 not, this is a HTTP/0.9 request and we must bail out without
491 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
494 /* Look for "\n[\r]\n", and return the following position if found.
495 Start two chars before the current to cover the possibility that
496 part of the terminator (e.g. "\n\r") arrived in the previous
498 p = peeked - start < 2 ? start : peeked - 2;
499 end = peeked + peeklen;
501 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
502 for (; p < end - 2; p++)
505 if (p[1] == '\r' && p[2] == '\n')
507 else if (p[1] == '\n')
510 /* p==end-2: check for \n\n directly preceding END. */
511 if (p[0] == '\n' && p[1] == '\n')
517 /* The maximum size of a single HTTP response we care to read. Rather
518 than being a limit of the reader implementation, this limit
519 prevents Wget from slurping all available memory upon encountering
520 malicious or buggy server output, thus protecting the user. Define
521 it to 0 to remove the limit. */
523 #define HTTP_RESPONSE_MAX_SIZE 65536
525 /* Read the HTTP request head from FD and return it. The error
526 conditions are the same as with fd_read_hunk.
528 To support HTTP/0.9 responses, this function tries to make sure
529 that the data begins with "HTTP". If this is not the case, no data
530 is read and an empty request is returned, so that the remaining
531 data can be treated as body. */
534 read_http_response_head (int fd)
536 return fd_read_hunk (fd, response_head_terminator, 512,
537 HTTP_RESPONSE_MAX_SIZE);
541 /* The response data. */
544 /* The array of pointers that indicate where each header starts.
545 For example, given this HTTP response:
552 The headers are located like this:
554 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
556 headers[0] headers[1] headers[2] headers[3]
558 I.e. headers[0] points to the beginning of the request,
559 headers[1] points to the end of the first header and the
560 beginning of the second one, etc. */
562 const char **headers;
565 /* Create a new response object from the text of the HTTP response,
566 available in HEAD. That text is automatically split into
567 constituent header lines for fast retrieval using
570 static struct response *
571 resp_new (const char *head)
576 struct response *resp = xnew0 (struct response);
581 /* Empty head means that we're dealing with a headerless
582 (HTTP/0.9) response. In that case, don't set HEADERS at
587 /* Split HEAD into header lines, so that resp_header_* functions
588 don't need to do this over and over again. */
594 DO_REALLOC (resp->headers, size, count + 1, const char *);
595 resp->headers[count++] = hdr;
597 /* Break upon encountering an empty line. */
598 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
601 /* Find the end of HDR, including continuations. */
604 const char *end = strchr (hdr, '\n');
610 while (*hdr == ' ' || *hdr == '\t');
612 DO_REALLOC (resp->headers, size, count + 1, const char *);
613 resp->headers[count] = NULL;
618 /* Locate the header named NAME in the request data, starting with
619 position START. This allows the code to loop through the request
620 data, filtering for all requests of a given name. Returns the
621 found position, or -1 for failure. The code that uses this
622 function typically looks like this:
624 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
625 ... do something with header ...
627 If you only care about one header, use resp_header_get instead of
631 resp_header_locate (const struct response *resp, const char *name, int start,
632 const char **begptr, const char **endptr)
635 const char **headers = resp->headers;
638 if (!headers || !headers[1])
641 name_len = strlen (name);
647 for (; headers[i + 1]; i++)
649 const char *b = headers[i];
650 const char *e = headers[i + 1];
652 && b[name_len] == ':'
653 && 0 == strncasecmp (b, name, name_len))
656 while (b < e && ISSPACE (*b))
658 while (b < e && ISSPACE (e[-1]))
668 /* Find and retrieve the header named NAME in the request data. If
669 found, set *BEGPTR to its starting, and *ENDPTR to its ending
670 position, and return true. Otherwise return false.
672 This function is used as a building block for resp_header_copy
673 and resp_header_strdup. */
676 resp_header_get (const struct response *resp, const char *name,
677 const char **begptr, const char **endptr)
679 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
683 /* Copy the response header named NAME to buffer BUF, no longer than
684 BUFSIZE (BUFSIZE includes the terminating 0). If the header
685 exists, true is returned, false otherwise. If there should be no
686 limit on the size of the header, use resp_header_strdup instead.
688 If BUFSIZE is 0, no data is copied, but the boolean indication of
689 whether the header is present is still returned. */
692 resp_header_copy (const struct response *resp, const char *name,
693 char *buf, int bufsize)
696 if (!resp_header_get (resp, name, &b, &e))
700 int len = MIN (e - b, bufsize - 1);
701 memcpy (buf, b, len);
707 /* Return the value of header named NAME in RESP, allocated with
708 malloc. If such a header does not exist in RESP, return NULL. */
711 resp_header_strdup (const struct response *resp, const char *name)
714 if (!resp_header_get (resp, name, &b, &e))
716 return strdupdelim (b, e);
719 /* Parse the HTTP status line, which is of format:
721 HTTP-Version SP Status-Code SP Reason-Phrase
723 The function returns the status-code, or -1 if the status line
724 appears malformed. The pointer to "reason-phrase" message is
725 returned in *MESSAGE. */
728 resp_status (const struct response *resp, char **message)
735 /* For a HTTP/0.9 response, assume status 200. */
737 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
741 p = resp->headers[0];
742 end = resp->headers[1];
748 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
752 /* Match the HTTP version. This is optional because Gnutella
753 servers have been reported to not specify HTTP version. */
754 if (p < end && *p == '/')
757 while (p < end && ISDIGIT (*p))
759 if (p < end && *p == '.')
761 while (p < end && ISDIGIT (*p))
765 while (p < end && ISSPACE (*p))
767 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
770 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
775 while (p < end && ISSPACE (*p))
777 while (p < end && ISSPACE (end[-1]))
779 *message = strdupdelim (p, end);
785 /* Release the resources used by RESP. */
788 resp_free (struct response *resp)
790 xfree_null (resp->headers);
794 /* Print a single line of response, the characters [b, e). We tried
796 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
797 but that failed to escape the non-printable characters and, in fact,
798 caused crashes in UTF-8 locales. */
801 print_response_line(const char *prefix, const char *b, const char *e)
804 BOUNDED_TO_ALLOCA(b, e, copy);
805 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
808 /* Print the server response, line by line, omitting the trailing CRLF
809 from individual header lines, and prefixed with PREFIX. */
812 print_server_response (const struct response *resp, const char *prefix)
817 for (i = 0; resp->headers[i + 1]; i++)
819 const char *b = resp->headers[i];
820 const char *e = resp->headers[i + 1];
822 if (b < e && e[-1] == '\n')
824 if (b < e && e[-1] == '\r')
826 print_response_line(prefix, b, e);
830 /* Parse the `Content-Range' header and extract the information it
831 contains. Returns true if successful, false otherwise. */
833 parse_content_range (const char *hdr, wgint *first_byte_ptr,
834 wgint *last_byte_ptr, wgint *entity_length_ptr)
838 /* Ancient versions of Netscape proxy server, presumably predating
839 rfc2068, sent out `Content-Range' without the "bytes"
841 if (0 == strncasecmp (hdr, "bytes", 5))
844 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
848 while (ISSPACE (*hdr))
855 for (num = 0; ISDIGIT (*hdr); hdr++)
856 num = 10 * num + (*hdr - '0');
857 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
859 *first_byte_ptr = num;
861 for (num = 0; ISDIGIT (*hdr); hdr++)
862 num = 10 * num + (*hdr - '0');
863 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
865 *last_byte_ptr = num;
867 for (num = 0; ISDIGIT (*hdr); hdr++)
868 num = 10 * num + (*hdr - '0');
869 *entity_length_ptr = num;
873 /* Read the body of the request, but don't store it anywhere and don't
874 display a progress gauge. This is useful for reading the bodies of
875 administrative responses to which we will soon issue another
876 request. The response is not useful to the user, but reading it
877 allows us to continue using the same connection to the server.
879 If reading fails, false is returned, true otherwise. In debug
880 mode, the body is displayed for debugging purposes. */
883 skip_short_body (int fd, wgint contlen)
886 SKIP_SIZE = 512, /* size of the download buffer */
887 SKIP_THRESHOLD = 4096 /* the largest size we read */
889 char dlbuf[SKIP_SIZE + 1];
890 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
892 /* We shouldn't get here with unknown contlen. (This will change
893 with HTTP/1.1, which supports "chunked" transfer.) */
894 assert (contlen != -1);
896 /* If the body is too large, it makes more sense to simply close the
897 connection than to try to read the body. */
898 if (contlen > SKIP_THRESHOLD)
901 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
905 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
908 /* Don't normally report the error since this is an
909 optimization that should be invisible to the user. */
910 DEBUGP (("] aborting (%s).\n",
911 ret < 0 ? fd_errstr (fd) : "EOF received"));
915 /* Safe even if %.*s bogusly expects terminating \0 because
916 we've zero-terminated dlbuf above. */
917 DEBUGP (("%.*s", ret, dlbuf));
920 DEBUGP (("] done.\n"));
924 /* Extract a parameter from the string (typically an HTTP header) at
925 **SOURCE and advance SOURCE to the next parameter. Return false
926 when there are no more parameters to extract. The name of the
927 parameter is returned in NAME, and the value in VALUE. If the
928 parameter has no value, the token's value is zeroed out.
930 For example, if *SOURCE points to the string "attachment;
931 filename=\"foo bar\"", the first call to this function will return
932 the token named "attachment" and no value, and the second call will
933 return the token named "filename" and value "foo bar". The third
934 call will return false, indicating no more valid tokens. */
937 extract_param (const char **source, param_token *name, param_token *value,
940 const char *p = *source;
942 while (ISSPACE (*p)) ++p;
946 return false; /* no error; nothing more to extract */
951 while (*p && !ISSPACE (*p) && *p != '=' && *p != separator) ++p;
953 if (name->b == name->e)
954 return false; /* empty name: error */
955 while (ISSPACE (*p)) ++p;
956 if (*p == separator || !*p) /* no value */
959 if (*p == separator) ++p;
964 return false; /* error */
966 /* *p is '=', extract value */
968 while (ISSPACE (*p)) ++p;
969 if (*p == '"') /* quoted */
972 while (*p && *p != '"') ++p;
976 /* Currently at closing quote; find the end of param. */
977 while (ISSPACE (*p)) ++p;
978 while (*p && *p != separator) ++p;
982 /* garbage after closed quote, e.g. foo="bar"baz */
988 while (*p && *p != separator) ++p;
990 while (value->e != value->b && ISSPACE (value->e[-1]))
992 if (*p == separator) ++p;
999 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1001 /* Parse the contents of the `Content-Disposition' header, extracting
1002 the information useful to Wget. Content-Disposition is a header
1003 borrowed from MIME; when used in HTTP, it typically serves for
1004 specifying the desired file name of the resource. For example:
1006 Content-Disposition: attachment; filename="flora.jpg"
1008 Wget will skip the tokens it doesn't care about, such as
1009 "attachment" in the previous example; it will also skip other
1010 unrecognized params. If the header is syntactically correct and
1011 contains a file name, a copy of the file name is stored in
1012 *filename and true is returned. Otherwise, the function returns
1015 The file name is stripped of directory components and must not be
1019 parse_content_disposition (const char *hdr, char **filename)
1021 param_token name, value;
1022 while (extract_param (&hdr, &name, &value, ';'))
1023 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1025 /* Make the file name begin at the last slash or backslash. */
1026 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1027 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1028 if (last_slash && last_bs)
1029 value.b = 1 + MAX (last_slash, last_bs);
1030 else if (last_slash || last_bs)
1031 value.b = 1 + (last_slash ? last_slash : last_bs);
1032 if (value.b == value.e)
1034 /* Start with the directory prefix, if specified. */
1037 int prefix_length = strlen (opt.dir_prefix);
1038 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1043 total_length = prefix_length + (value.e - value.b);
1044 *filename = xmalloc (total_length + 1);
1045 strcpy (*filename, opt.dir_prefix);
1047 (*filename)[prefix_length - 1] = '/';
1048 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1049 (*filename)[total_length] = '\0';
1052 *filename = strdupdelim (value.b, value.e);
1058 /* Persistent connections. Currently, we cache the most recently used
1059 connection as persistent, provided that the HTTP server agrees to
1060 make it such. The persistence data is stored in the variables
1061 below. Ideally, it should be possible to cache an arbitrary fixed
1062 number of these connections. */
1064 /* Whether a persistent connection is active. */
1065 static bool pconn_active;
1068 /* The socket of the connection. */
1071 /* Host and port of the currently active persistent connection. */
1075 /* Whether a ssl handshake has occoured on this connection. */
1078 /* Whether the connection was authorized. This is only done by
1079 NTLM, which authorizes *connections* rather than individual
1080 requests. (That practice is peculiar for HTTP, but it is a
1081 useful optimization.) */
1085 /* NTLM data of the current connection. */
1086 struct ntlmdata ntlm;
1090 /* Mark the persistent connection as invalid and free the resources it
1091 uses. This is used by the CLOSE_* macros after they forcefully
1092 close a registered persistent connection. */
1095 invalidate_persistent (void)
1097 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1098 pconn_active = false;
1099 fd_close (pconn.socket);
1104 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1105 persistent. This will enable someone to use the same connection
1106 later. In the context of HTTP, this must be called only AFTER the
1107 response has been received and the server has promised that the
1108 connection will remain alive.
1110 If a previous connection was persistent, it is closed. */
1113 register_persistent (const char *host, int port, int fd, bool ssl)
1117 if (pconn.socket == fd)
1119 /* The connection FD is already registered. */
1124 /* The old persistent connection is still active; close it
1125 first. This situation arises whenever a persistent
1126 connection exists, but we then connect to a different
1127 host, and try to register a persistent connection to that
1129 invalidate_persistent ();
1133 pconn_active = true;
1135 pconn.host = xstrdup (host);
1138 pconn.authorized = false;
1140 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1143 /* Return true if a persistent connection is available for connecting
1147 persistent_available_p (const char *host, int port, bool ssl,
1148 bool *host_lookup_failed)
1150 /* First, check whether a persistent connection is active at all. */
1154 /* If we want SSL and the last connection wasn't or vice versa,
1155 don't use it. Checking for host and port is not enough because
1156 HTTP and HTTPS can apparently coexist on the same port. */
1157 if (ssl != pconn.ssl)
1160 /* If we're not connecting to the same port, we're not interested. */
1161 if (port != pconn.port)
1164 /* If the host is the same, we're in business. If not, there is
1165 still hope -- read below. */
1166 if (0 != strcasecmp (host, pconn.host))
1168 /* Check if pconn.socket is talking to HOST under another name.
1169 This happens often when both sites are virtual hosts
1170 distinguished only by name and served by the same network
1171 interface, and hence the same web server (possibly set up by
1172 the ISP and serving many different web sites). This
1173 admittedly unconventional optimization does not contradict
1174 HTTP and works well with popular server software. */
1178 struct address_list *al;
1181 /* Don't try to talk to two different SSL sites over the same
1182 secure connection! (Besides, it's not clear that
1183 name-based virtual hosting is even possible with SSL.) */
1186 /* If pconn.socket's peer is one of the IP addresses HOST
1187 resolves to, pconn.socket is for all intents and purposes
1188 already talking to HOST. */
1190 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1192 /* Can't get the peer's address -- something must be very
1193 wrong with the connection. */
1194 invalidate_persistent ();
1197 al = lookup_host (host, 0);
1200 *host_lookup_failed = true;
1204 found = address_list_contains (al, &ip);
1205 address_list_release (al);
1210 /* The persistent connection's peer address was found among the
1211 addresses HOST resolved to; therefore, pconn.sock is in fact
1212 already talking to HOST -- no need to reconnect. */
1215 /* Finally, check whether the connection is still open. This is
1216 important because most servers implement liberal (short) timeout
1217 on persistent connections. Wget can of course always reconnect
1218 if the connection doesn't work out, but it's nicer to know in
1219 advance. This test is a logical followup of the first test, but
1220 is "expensive" and therefore placed at the end of the list.
1222 (Current implementation of test_socket_open has a nice side
1223 effect that it treats sockets with pending data as "closed".
1224 This is exactly what we want: if a broken server sends message
1225 body in response to HEAD, or if it sends more than conent-length
1226 data, we won't reuse the corrupted connection.) */
1228 if (!test_socket_open (pconn.socket))
1230 /* Oops, the socket is no longer open. Now that we know that,
1231 let's invalidate the persistent connection before returning
1233 invalidate_persistent ();
1240 /* The idea behind these two CLOSE macros is to distinguish between
1241 two cases: one when the job we've been doing is finished, and we
1242 want to close the connection and leave, and two when something is
1243 seriously wrong and we're closing the connection as part of
1246 In case of keep_alive, CLOSE_FINISH should leave the connection
1247 open, while CLOSE_INVALIDATE should still close it.
1249 Note that the semantics of the flag `keep_alive' is "this
1250 connection *will* be reused (the server has promised not to close
1251 the connection once we're done)", while the semantics of
1252 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1253 active, registered connection". */
1255 #define CLOSE_FINISH(fd) do { \
1258 if (pconn_active && (fd) == pconn.socket) \
1259 invalidate_persistent (); \
1268 #define CLOSE_INVALIDATE(fd) do { \
1269 if (pconn_active && (fd) == pconn.socket) \
1270 invalidate_persistent (); \
1278 wgint len; /* received length */
1279 wgint contlen; /* expected length */
1280 wgint restval; /* the restart value */
1281 int res; /* the result of last read */
1282 char *rderrmsg; /* error message from read error */
1283 char *newloc; /* new location (redirection) */
1284 char *remote_time; /* remote time-stamp string */
1285 char *error; /* textual HTTP error */
1286 int statcode; /* status code */
1287 wgint rd_size; /* amount of data read from socket */
1288 double dltime; /* time it took to download the data */
1289 const char *referer; /* value of the referer header. */
1290 char *local_file; /* local file name. */
1291 bool timestamp_checked; /* true if pre-download time-stamping checks
1292 * have already been performed */
1293 char *orig_file_name; /* name of file to compare for time-stamping
1294 * (might be != local_file if -K is set) */
1295 wgint orig_file_size; /* size of file to compare for time-stamping */
1296 time_t orig_file_tstamp; /* time-stamp of file to compare for
1301 free_hstat (struct http_stat *hs)
1303 xfree_null (hs->newloc);
1304 xfree_null (hs->remote_time);
1305 xfree_null (hs->error);
1306 xfree_null (hs->rderrmsg);
1307 xfree_null (hs->local_file);
1308 xfree_null (hs->orig_file_name);
1310 /* Guard against being called twice. */
1312 hs->remote_time = NULL;
1316 #define BEGINS_WITH(line, string_constant) \
1317 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1318 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1319 || !line[sizeof (string_constant) - 1]))
1321 #define SET_USER_AGENT(req) do { \
1322 if (!opt.useragent) \
1323 request_set_header (req, "User-Agent", \
1324 aprintf ("Wget/%s", version_string), rel_value); \
1325 else if (*opt.useragent) \
1326 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1329 /* The flags that allow clobbering the file (opening with "wb").
1330 Defined here to avoid repetition later. #### This will require
1332 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1333 || opt.dirstruct || opt.output_document)
1335 /* Retrieve a document through HTTP protocol. It recognizes status
1336 code, and correctly handles redirections. It closes the network
1337 socket. If it receives an error from the functions below it, it
1338 will print it if there is enough information to do so (almost
1339 always), returning the error to the caller (i.e. http_loop).
1341 Various HTTP parameters are stored to hs.
1343 If PROXY is non-NULL, the connection will be made to the proxy
1344 server, and u->url will be requested. */
1346 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1348 struct request *req;
1351 char *user, *passwd;
1355 wgint contlen, contrange;
1362 /* Set to 1 when the authorization has already been sent and should
1363 not be tried again. */
1364 bool auth_finished = false;
1366 /* Set to 1 when just globally-set Basic authorization has been sent;
1367 * should prevent further Basic negotiations, but not other
1369 bool basic_auth_finished = false;
1371 /* Whether NTLM authentication is used for this request. */
1372 bool ntlm_seen = false;
1374 /* Whether our connection to the remote host is through SSL. */
1375 bool using_ssl = false;
1377 /* Whether a HEAD request will be issued (as opposed to GET or
1379 bool head_only = !!(*dt & HEAD_ONLY);
1382 struct response *resp;
1386 /* Whether this connection will be kept alive after the HTTP request
1390 /* Whether keep-alive should be inhibited.
1392 RFC 2068 requests that 1.0 clients not send keep-alive requests
1393 to proxies. This is because many 1.0 proxies do not interpret
1394 the Connection header and transfer it to the remote server,
1395 causing it to not close the connection and leave both the proxy
1396 and the client hanging. */
1397 bool inhibit_keep_alive =
1398 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1400 /* Headers sent when using POST. */
1401 wgint post_data_size = 0;
1403 bool host_lookup_failed = false;
1406 if (u->scheme == SCHEME_HTTPS)
1408 /* Initialize the SSL context. After this has once been done,
1409 it becomes a no-op. */
1412 scheme_disable (SCHEME_HTTPS);
1413 logprintf (LOG_NOTQUIET,
1414 _("Disabling SSL due to encountered errors.\n"));
1415 return SSLINITFAILED;
1418 #endif /* HAVE_SSL */
1420 /* Initialize certain elements of struct http_stat. */
1424 hs->rderrmsg = NULL;
1426 hs->remote_time = NULL;
1431 /* Prepare the request to send. */
1433 req = request_new ();
1436 const char *meth = "GET";
1439 else if (opt.post_file_name || opt.post_data)
1441 /* Use the full path, i.e. one that includes the leading slash and
1442 the query string. E.g. if u->path is "foo/bar" and u->query is
1443 "param=value", full_path will be "/foo/bar?param=value". */
1446 /* When using SSL over proxy, CONNECT establishes a direct
1447 connection to the HTTPS server. Therefore use the same
1448 argument as when talking to the server directly. */
1449 && u->scheme != SCHEME_HTTPS
1452 meth_arg = xstrdup (u->url);
1454 meth_arg = url_full_path (u);
1455 request_set_method (req, meth, meth_arg);
1458 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1459 if (*dt & SEND_NOCACHE)
1460 request_set_header (req, "Pragma", "no-cache", rel_none);
1462 request_set_header (req, "Range",
1463 aprintf ("bytes=%s-",
1464 number_to_static_string (hs->restval)),
1466 SET_USER_AGENT (req);
1467 request_set_header (req, "Accept", "*/*", rel_none);
1469 /* Find the username and password for authentication. */
1472 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1473 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1474 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1477 && !u->user) /* We only do "site-wide" authentication with "global"
1478 user/password values; URL user/password info overrides. */
1480 /* If this is a host for which we've already received a Basic
1481 * challenge, we'll go ahead and send Basic authentication creds. */
1482 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1488 char *proxy_user, *proxy_passwd;
1489 /* For normal username and password, URL components override
1490 command-line/wgetrc parameters. With proxy
1491 authentication, it's the reverse, because proxy URLs are
1492 normally the "permanent" ones, so command-line args
1493 should take precedence. */
1494 if (opt.proxy_user && opt.proxy_passwd)
1496 proxy_user = opt.proxy_user;
1497 proxy_passwd = opt.proxy_passwd;
1501 proxy_user = proxy->user;
1502 proxy_passwd = proxy->passwd;
1504 /* #### This does not appear right. Can't the proxy request,
1505 say, `Digest' authentication? */
1506 if (proxy_user && proxy_passwd)
1507 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1509 /* If we're using a proxy, we will be connecting to the proxy
1513 /* Proxy authorization over SSL is handled below. */
1515 if (u->scheme != SCHEME_HTTPS)
1517 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1520 /* Generate the Host header, HOST:PORT. Take into account that:
1522 - Broken server-side software often doesn't recognize the PORT
1523 argument, so we must generate "Host: www.server.com" instead of
1524 "Host: www.server.com:80" (and likewise for https port).
1526 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1527 becomes ambiguous and needs to be rewritten as "Host:
1528 [3ffe:8100:200:2::2]:1234". */
1530 /* Formats arranged for hfmt[add_port][add_squares]. */
1531 static const char *hfmt[][2] = {
1532 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1534 int add_port = u->port != scheme_default_port (u->scheme);
1535 int add_squares = strchr (u->host, ':') != NULL;
1536 request_set_header (req, "Host",
1537 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1541 if (!inhibit_keep_alive)
1542 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1545 request_set_header (req, "Cookie",
1546 cookie_header (wget_cookie_jar,
1547 u->host, u->port, u->path,
1549 u->scheme == SCHEME_HTTPS
1556 if (opt.post_data || opt.post_file_name)
1558 request_set_header (req, "Content-Type",
1559 "application/x-www-form-urlencoded", rel_none);
1561 post_data_size = strlen (opt.post_data);
1564 post_data_size = file_size (opt.post_file_name);
1565 if (post_data_size == -1)
1567 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1568 opt.post_file_name, strerror (errno));
1572 request_set_header (req, "Content-Length",
1573 xstrdup (number_to_static_string (post_data_size)),
1577 /* Add the user headers. */
1578 if (opt.user_headers)
1581 for (i = 0; opt.user_headers[i]; i++)
1582 request_set_user_header (req, opt.user_headers[i]);
1586 /* We need to come back here when the initial attempt to retrieve
1587 without authorization header fails. (Expected to happen at least
1588 for the Digest authorization scheme.) */
1592 /* Establish the connection. */
1594 if (!inhibit_keep_alive)
1596 /* Look for a persistent connection to target host, unless a
1597 proxy is used. The exception is when SSL is in use, in which
1598 case the proxy is nothing but a passthrough to the target
1599 host, registered as a connection to the latter. */
1600 struct url *relevant = conn;
1602 if (u->scheme == SCHEME_HTTPS)
1606 if (persistent_available_p (relevant->host, relevant->port,
1608 relevant->scheme == SCHEME_HTTPS,
1612 &host_lookup_failed))
1614 sock = pconn.socket;
1615 using_ssl = pconn.ssl;
1616 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1617 escnonprint (pconn.host), pconn.port);
1618 DEBUGP (("Reusing fd %d.\n", sock));
1619 if (pconn.authorized)
1620 /* If the connection is already authorized, the "Basic"
1621 authorization added by code above is unnecessary and
1623 request_remove_header (req, "Authorization");
1625 else if (host_lookup_failed)
1628 logprintf(LOG_NOTQUIET,
1629 _("%s: unable to resolve host address `%s'\n"),
1630 exec_name, relevant->host);
1637 sock = connect_to_host (conn->host, conn->port);
1646 return (retryable_socket_connect_error (errno)
1647 ? CONERROR : CONIMPOSSIBLE);
1651 if (proxy && u->scheme == SCHEME_HTTPS)
1653 /* When requesting SSL URLs through proxies, use the
1654 CONNECT method to request passthrough. */
1655 struct request *connreq = request_new ();
1656 request_set_method (connreq, "CONNECT",
1657 aprintf ("%s:%d", u->host, u->port));
1658 SET_USER_AGENT (connreq);
1661 request_set_header (connreq, "Proxy-Authorization",
1662 proxyauth, rel_value);
1663 /* Now that PROXYAUTH is part of the CONNECT request,
1664 zero it out so we don't send proxy authorization with
1665 the regular request below. */
1668 /* Examples in rfc2817 use the Host header in CONNECT
1669 requests. I don't see how that gains anything, given
1670 that the contents of Host would be exactly the same as
1671 the contents of CONNECT. */
1673 write_error = request_send (connreq, sock);
1674 request_free (connreq);
1675 if (write_error < 0)
1677 CLOSE_INVALIDATE (sock);
1681 head = read_http_response_head (sock);
1684 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1686 CLOSE_INVALIDATE (sock);
1695 DEBUGP (("proxy responded with: [%s]\n", head));
1697 resp = resp_new (head);
1698 statcode = resp_status (resp, &message);
1701 if (statcode != 200)
1704 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1705 message ? escnonprint (message) : "?");
1706 xfree_null (message);
1709 xfree_null (message);
1711 /* SOCK is now *really* connected to u->host, so update CONN
1712 to reflect this. That way register_persistent will
1713 register SOCK as being connected to u->host:u->port. */
1717 if (conn->scheme == SCHEME_HTTPS)
1719 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1726 #endif /* HAVE_SSL */
1729 /* Send the request to server. */
1730 write_error = request_send (req, sock);
1732 if (write_error >= 0)
1736 DEBUGP (("[POST data: %s]\n", opt.post_data));
1737 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1739 else if (opt.post_file_name && post_data_size != 0)
1740 write_error = post_file (sock, opt.post_file_name, post_data_size);
1743 if (write_error < 0)
1745 CLOSE_INVALIDATE (sock);
1749 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1750 proxy ? "Proxy" : "HTTP");
1755 head = read_http_response_head (sock);
1760 logputs (LOG_NOTQUIET, _("No data received.\n"));
1761 CLOSE_INVALIDATE (sock);
1767 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1769 CLOSE_INVALIDATE (sock);
1774 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1776 resp = resp_new (head);
1778 /* Check for status line. */
1780 statcode = resp_status (resp, &message);
1781 if (!opt.server_response)
1782 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1783 message ? escnonprint (message) : "");
1786 logprintf (LOG_VERBOSE, "\n");
1787 print_server_response (resp, " ");
1790 /* Determine the local filename if needed. Notice that if -O is used
1791 * hstat.local_file is set by http_loop to the argument of -O. */
1792 if (!hs->local_file)
1794 /* Honor Content-Disposition whether possible. */
1795 if (!opt.content_disposition
1796 || !resp_header_copy (resp, "Content-Disposition",
1797 hdrval, sizeof (hdrval))
1798 || !parse_content_disposition (hdrval, &hs->local_file))
1800 /* The Content-Disposition header is missing or broken.
1801 * Choose unique file name according to given URL. */
1802 hs->local_file = url_file_name (u);
1806 /* TODO: perform this check only once. */
1807 if (file_exists_p (hs->local_file))
1811 /* If opt.noclobber is turned on and file already exists, do not
1812 retrieve the file */
1813 logprintf (LOG_VERBOSE, _("\
1814 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1815 /* If the file is there, we suppose it's retrieved OK. */
1818 /* #### Bogusness alert. */
1819 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1820 if (has_html_suffix_p (hs->local_file))
1823 return RETRUNNEEDED;
1825 else if (!ALLOW_CLOBBER)
1827 char *unique = unique_name (hs->local_file, true);
1828 if (unique != hs->local_file)
1829 xfree (hs->local_file);
1830 hs->local_file = unique;
1834 /* Support timestamping */
1835 /* TODO: move this code out of gethttp. */
1836 if (opt.timestamping && !hs->timestamp_checked)
1838 size_t filename_len = strlen (hs->local_file);
1839 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1840 bool local_dot_orig_file_exists = false;
1841 char *local_filename = NULL;
1844 if (opt.backup_converted)
1845 /* If -K is specified, we'll act on the assumption that it was specified
1846 last time these files were downloaded as well, and instead of just
1847 comparing local file X against server file X, we'll compare local
1848 file X.orig (if extant, else X) against server file X. If -K
1849 _wasn't_ specified last time, or the server contains files called
1850 *.orig, -N will be back to not operating correctly with -k. */
1852 /* Would a single s[n]printf() call be faster? --dan
1854 Definitely not. sprintf() is horribly slow. It's a
1855 different question whether the difference between the two
1856 affects a program. Usually I'd say "no", but at one
1857 point I profiled Wget, and found that a measurable and
1858 non-negligible amount of time was lost calling sprintf()
1859 in url.c. Replacing sprintf with inline calls to
1860 strcpy() and number_to_string() made a difference.
1862 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1863 memcpy (filename_plus_orig_suffix + filename_len,
1864 ".orig", sizeof (".orig"));
1866 /* Try to stat() the .orig file. */
1867 if (stat (filename_plus_orig_suffix, &st) == 0)
1869 local_dot_orig_file_exists = true;
1870 local_filename = filename_plus_orig_suffix;
1874 if (!local_dot_orig_file_exists)
1875 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1876 if (stat (hs->local_file, &st) == 0)
1877 local_filename = hs->local_file;
1879 if (local_filename != NULL)
1880 /* There was a local file, so we'll check later to see if the version
1881 the server has is the same version we already have, allowing us to
1884 hs->orig_file_name = xstrdup (local_filename);
1885 hs->orig_file_size = st.st_size;
1886 hs->orig_file_tstamp = st.st_mtime;
1888 /* Modification time granularity is 2 seconds for Windows, so
1889 increase local time by 1 second for later comparison. */
1890 ++hs->orig_file_tstamp;
1895 if (!opt.ignore_length
1896 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1900 parsed = str_to_wgint (hdrval, NULL, 10);
1901 if (parsed == WGINT_MAX && errno == ERANGE)
1904 #### If Content-Length is out of range, it most likely
1905 means that the file is larger than 2G and that we're
1906 compiled without LFS. In that case we should probably
1907 refuse to even attempt to download the file. */
1910 else if (parsed < 0)
1912 /* Negative Content-Length; nonsensical, so we can't
1913 assume any information about the content to receive. */
1920 /* Check for keep-alive related responses. */
1921 if (!inhibit_keep_alive && contlen != -1)
1923 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1925 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1927 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1932 /* The server has promised that it will not close the connection
1933 when we're done. This means that we can register it. */
1934 register_persistent (conn->host, conn->port, sock, using_ssl);
1936 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1938 /* Authorization is required. */
1939 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1940 CLOSE_FINISH (sock);
1942 CLOSE_INVALIDATE (sock);
1943 pconn.authorized = false;
1944 if (!auth_finished && (user && passwd))
1946 /* IIS sends multiple copies of WWW-Authenticate, one with
1947 the value "negotiate", and other(s) with data. Loop over
1948 all the occurrences and pick the one we recognize. */
1950 const char *wabeg, *waend;
1951 char *www_authenticate = NULL;
1953 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1954 &wabeg, &waend)) != -1;
1956 if (known_authentication_scheme_p (wabeg, waend))
1958 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1962 if (!www_authenticate)
1964 /* If the authentication header is missing or
1965 unrecognized, there's no sense in retrying. */
1966 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1968 else if (!basic_auth_finished
1969 || !BEGINS_WITH (www_authenticate, "Basic"))
1972 pth = url_full_path (u);
1973 request_set_header (req, "Authorization",
1974 create_authorization_line (www_authenticate,
1976 request_method (req),
1980 if (BEGINS_WITH (www_authenticate, "NTLM"))
1982 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1984 /* Need to register this host as using basic auth,
1985 * so we automatically send creds next time. */
1986 register_basic_auth_host (u->host);
1989 goto retry_with_auth;
1993 /* We already did Basic auth, and it failed. Gotta
1997 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2001 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2003 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2005 pconn.authorized = true;
2009 hs->statcode = statcode;
2011 hs->error = xstrdup (_("Malformed status line"));
2013 hs->error = xstrdup (_("(no description)"));
2015 hs->error = xstrdup (message);
2016 xfree_null (message);
2018 type = resp_header_strdup (resp, "Content-Type");
2021 char *tmp = strchr (type, ';');
2024 while (tmp > type && ISSPACE (tmp[-1]))
2029 hs->newloc = resp_header_strdup (resp, "Location");
2030 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2032 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2036 const char *scbeg, *scend;
2037 /* The jar should have been created by now. */
2038 assert (wget_cookie_jar != NULL);
2040 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2041 &scbeg, &scend)) != -1;
2044 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2045 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2046 u->path, set_cookie);
2050 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2052 wgint first_byte_pos, last_byte_pos, entity_length;
2053 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2055 contrange = first_byte_pos;
2059 /* 20x responses are counted among successful by default. */
2060 if (H_20X (statcode))
2063 /* Return if redirected. */
2064 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2066 /* RFC2068 says that in case of the 300 (multiple choices)
2067 response, the server can output a preferred URL through
2068 `Location' header; otherwise, the request should be treated
2069 like GET. So, if the location is set, it will be a
2070 redirection; otherwise, just proceed normally. */
2071 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2075 logprintf (LOG_VERBOSE,
2076 _("Location: %s%s\n"),
2077 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2078 hs->newloc ? _(" [following]") : "");
2079 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2080 CLOSE_FINISH (sock);
2082 CLOSE_INVALIDATE (sock);
2088 /* If content-type is not given, assume text/html. This is because
2089 of the multitude of broken CGI's that "forget" to generate the
2092 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2093 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2098 if (opt.html_extension && (*dt & TEXTHTML))
2099 /* -E / --html-extension / html_extension = on was specified, and this is a
2100 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2101 already the file's suffix, tack on ".html". */
2103 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2105 if (last_period_in_local_filename == NULL
2106 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2107 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2109 int local_filename_len = strlen (hs->local_file);
2110 /* Resize the local file, allowing for ".html" preceded by
2111 optional ".NUMBER". */
2112 hs->local_file = xrealloc (hs->local_file,
2113 local_filename_len + 24 + sizeof (".html"));
2114 strcpy(hs->local_file + local_filename_len, ".html");
2115 /* If clobbering is not allowed and the file, as named,
2116 exists, tack on ".NUMBER.html" instead. */
2117 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2121 sprintf (hs->local_file + local_filename_len,
2122 ".%d.html", ext_num++);
2123 while (file_exists_p (hs->local_file));
2125 *dt |= ADDED_HTML_EXTENSION;
2129 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2131 /* If `-c' is in use and the file has been fully downloaded (or
2132 the remote file has shrunk), Wget effectively requests bytes
2133 after the end of file and the server response with 416. */
2134 logputs (LOG_VERBOSE, _("\
2135 \n The file is already fully retrieved; nothing to do.\n\n"));
2136 /* In case the caller inspects. */
2139 /* Mark as successfully retrieved. */
2142 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2143 might be more bytes in the body. */
2144 return RETRUNNEEDED;
2146 if ((contrange != 0 && contrange != hs->restval)
2147 || (H_PARTIAL (statcode) && !contrange))
2149 /* The Range request was somehow misunderstood by the server.
2152 CLOSE_INVALIDATE (sock);
2155 hs->contlen = contlen + contrange;
2161 /* No need to print this output if the body won't be
2162 downloaded at all, or if the original server response is
2164 logputs (LOG_VERBOSE, _("Length: "));
2167 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2168 if (contlen + contrange >= 1024)
2169 logprintf (LOG_VERBOSE, " (%s)",
2170 human_readable (contlen + contrange));
2173 if (contlen >= 1024)
2174 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2175 number_to_static_string (contlen),
2176 human_readable (contlen));
2178 logprintf (LOG_VERBOSE, _(", %s remaining"),
2179 number_to_static_string (contlen));
2183 logputs (LOG_VERBOSE,
2184 opt.ignore_length ? _("ignored") : _("unspecified"));
2186 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2188 logputs (LOG_VERBOSE, "\n");
2192 type = NULL; /* We don't need it any more. */
2194 /* Return if we have no intention of further downloading. */
2195 if (!(*dt & RETROKF) || head_only)
2197 /* In case the caller cares to look... */
2202 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2203 servers not to send body in response to a HEAD request, and
2204 those that do will likely be caught by test_socket_open.
2205 If not, they can be worked around using
2206 `--no-http-keep-alive'. */
2207 CLOSE_FINISH (sock);
2208 else if (keep_alive && skip_short_body (sock, contlen))
2209 /* Successfully skipped the body; also keep using the socket. */
2210 CLOSE_FINISH (sock);
2212 CLOSE_INVALIDATE (sock);
2213 return RETRFINISHED;
2216 /* Open the local file. */
2219 mkalldirs (hs->local_file);
2221 rotate_backups (hs->local_file);
2223 fp = fopen (hs->local_file, "ab");
2224 else if (ALLOW_CLOBBER)
2225 fp = fopen (hs->local_file, "wb");
2228 fp = fopen_excl (hs->local_file, true);
2229 if (!fp && errno == EEXIST)
2231 /* We cannot just invent a new name and use it (which is
2232 what functions like unique_create typically do)
2233 because we told the user we'd use this name.
2234 Instead, return and retry the download. */
2235 logprintf (LOG_NOTQUIET,
2236 _("%s has sprung into existence.\n"),
2238 CLOSE_INVALIDATE (sock);
2239 return FOPEN_EXCL_ERR;
2244 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2245 CLOSE_INVALIDATE (sock);
2252 /* Print fetch message, if opt.verbose. */
2255 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2256 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2259 /* This confuses the timestamping code that checks for file size.
2260 #### The timestamping code should be smarter about file size. */
2261 if (opt.save_headers && hs->restval == 0)
2262 fwrite (head, 1, strlen (head), fp);
2264 /* Now we no longer need to store the response header. */
2267 /* Download the request body. */
2270 /* If content-length is present, read that much; otherwise, read
2271 until EOF. The HTTP spec doesn't require the server to
2272 actually close the connection when it's done sending data. */
2273 flags |= rb_read_exactly;
2274 if (hs->restval > 0 && contrange == 0)
2275 /* If the server ignored our range request, instruct fd_read_body
2276 to skip the first RESTVAL bytes of body. */
2277 flags |= rb_skip_startpos;
2278 hs->len = hs->restval;
2280 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2281 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2285 CLOSE_FINISH (sock);
2289 hs->rderrmsg = xstrdup (fd_errstr (sock));
2290 CLOSE_INVALIDATE (sock);
2297 return RETRFINISHED;
2300 /* The genuine HTTP loop! This is the part where the retrieval is
2301 retried, and retried, and retried, and... */
2303 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2304 int *dt, struct url *proxy)
2307 bool got_head = false; /* used for time-stamping and filename detection */
2308 bool time_came_from_head = false;
2309 bool got_name = false;
2312 uerr_t err, ret = TRYLIMEXC;
2313 time_t tmr = -1; /* remote time-stamp */
2314 struct http_stat hstat; /* HTTP status */
2316 bool send_head_first = true;
2318 /* Assert that no value for *LOCAL_FILE was passed. */
2319 assert (local_file == NULL || *local_file == NULL);
2321 /* Set LOCAL_FILE parameter. */
2322 if (local_file && opt.output_document)
2323 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2325 /* Reset NEWLOC parameter. */
2328 /* This used to be done in main(), but it's a better idea to do it
2329 here so that we don't go through the hoops if we're just using
2334 /* Warn on (likely bogus) wildcard usage in HTTP. */
2335 if (opt.ftp_glob && has_wildcards_p (u->path))
2336 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2338 /* Setup hstat struct. */
2340 hstat.referer = referer;
2342 if (opt.output_document)
2344 hstat.local_file = xstrdup (opt.output_document);
2347 else if (!opt.content_disposition)
2349 hstat.local_file = url_file_name (u);
2353 /* Reset the counter. */
2356 /* Reset the document type. */
2359 /* Skip preliminary HEAD request if we're not in spider mode AND
2360 * if -O was given or HTTP Content-Disposition support is disabled. */
2362 && (got_name || !opt.content_disposition))
2363 send_head_first = false;
2365 /* Send preliminary HEAD request if -N is given and we have an existing
2366 * destination file. */
2367 if (opt.timestamping
2368 && !opt.content_disposition
2369 && file_exists_p (url_file_name (u)))
2370 send_head_first = true;
2375 /* Increment the pass counter. */
2377 sleep_between_retrievals (count);
2379 /* Get the current time string. */
2380 tms = datetime_str (time (NULL));
2382 if (opt.spider && !got_head)
2383 logprintf (LOG_VERBOSE, _("\
2384 Spider mode enabled. Check if remote file exists.\n"));
2386 /* Print fetch message, if opt.verbose. */
2389 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2394 sprintf (tmp, _("(try:%2d)"), count);
2395 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2400 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2405 ws_changetitle (hurl);
2410 /* Default document type is empty. However, if spider mode is
2411 on or time-stamping is employed, HEAD_ONLY commands is
2412 encoded within *dt. */
2413 if (send_head_first && !got_head)
2418 /* Decide whether or not to restart. */
2421 && stat (hstat.local_file, &st) == 0
2422 && S_ISREG (st.st_mode))
2423 /* When -c is used, continue from on-disk size. (Can't use
2424 hstat.len even if count>1 because we don't want a failed
2425 first attempt to clobber existing data.) */
2426 hstat.restval = st.st_size;
2428 /* otherwise, continue where the previous try left off */
2429 hstat.restval = hstat.len;
2433 /* Decide whether to send the no-cache directive. We send it in
2435 a) we're using a proxy, and we're past our first retrieval.
2436 Some proxies are notorious for caching incomplete data, so
2437 we require a fresh get.
2438 b) caching is explicitly inhibited. */
2439 if ((proxy && count > 1) /* a */
2440 || !opt.allow_cache) /* b */
2441 *dt |= SEND_NOCACHE;
2443 *dt &= ~SEND_NOCACHE;
2445 /* Try fetching the document, or at least its head. */
2446 err = gethttp (u, &hstat, dt, proxy);
2449 tms = datetime_str (time (NULL));
2451 /* Get the new location (with or without the redirection). */
2453 *newloc = xstrdup (hstat.newloc);
2457 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2458 case CONERROR: case READERR: case WRITEFAILED:
2459 case RANGEERR: case FOPEN_EXCL_ERR:
2460 /* Non-fatal errors continue executing the loop, which will
2461 bring them to "while" statement at the end, to judge
2462 whether the number of tries was exceeded. */
2463 printwhat (count, opt.ntry);
2465 case FWRITEERR: case FOPENERR:
2466 /* Another fatal error. */
2467 logputs (LOG_VERBOSE, "\n");
2468 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2469 hstat.local_file, strerror (errno));
2470 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2471 case SSLINITFAILED: case CONTNOTSUPPORTED:
2472 /* Fatal errors just return from the function. */
2476 /* Another fatal error. */
2477 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2481 /* Return the new location to the caller. */
2484 logprintf (LOG_NOTQUIET,
2485 _("ERROR: Redirection (%d) without location.\n"),
2495 /* The file was already fully retrieved. */
2499 /* Deal with you later. */
2502 /* All possibilities should have been exhausted. */
2506 if (!(*dt & RETROKF))
2511 /* #### Ugly ugly ugly! */
2512 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2513 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2516 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2518 && (hstat.statcode == 500 || hstat.statcode == 501))
2523 /* Maybe we should always keep track of broken links, not just in
2525 else if (opt.spider)
2527 /* #### Again: ugly ugly ugly! */
2529 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2530 nonexisting_url (hurl);
2531 logprintf (LOG_NOTQUIET, _("\
2532 Remote file does not exist -- broken link!!!\n"));
2536 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2537 tms, hstat.statcode, escnonprint (hstat.error));
2539 logputs (LOG_VERBOSE, "\n");
2545 /* Did we get the time-stamp? */
2548 got_head = true; /* no more time-stamping */
2550 if (opt.timestamping && !hstat.remote_time)
2552 logputs (LOG_NOTQUIET, _("\
2553 Last-modified header missing -- time-stamps turned off.\n"));
2555 else if (hstat.remote_time)
2557 /* Convert the date-string into struct tm. */
2558 tmr = http_atotm (hstat.remote_time);
2559 if (tmr == (time_t) (-1))
2560 logputs (LOG_VERBOSE, _("\
2561 Last-modified header invalid -- time-stamp ignored.\n"));
2562 if (*dt & HEAD_ONLY)
2563 time_came_from_head = true;
2566 if (send_head_first)
2568 /* The time-stamping section. */
2569 if (opt.timestamping)
2571 if (hstat.orig_file_name) /* Perform the following
2572 checks only if the file
2574 download already exists. */
2576 if (hstat.remote_time &&
2577 tmr != (time_t) (-1))
2579 /* Now time-stamping can be used validly.
2580 Time-stamping means that if the sizes of
2581 the local and remote file match, and local
2582 file is newer than the remote file, it will
2583 not be retrieved. Otherwise, the normal
2584 download procedure is resumed. */
2585 if (hstat.orig_file_tstamp >= tmr)
2587 if (hstat.contlen == -1
2588 || hstat.orig_file_size == hstat.contlen)
2590 logprintf (LOG_VERBOSE, _("\
2591 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2592 hstat.orig_file_name);
2598 logprintf (LOG_VERBOSE, _("\
2599 The sizes do not match (local %s) -- retrieving.\n"),
2600 number_to_static_string (hstat.orig_file_size));
2604 logputs (LOG_VERBOSE,
2605 _("Remote file is newer, retrieving.\n"));
2607 logputs (LOG_VERBOSE, "\n");
2611 /* free_hstat (&hstat); */
2612 hstat.timestamp_checked = true;
2621 logputs (LOG_VERBOSE, _("\
2622 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2626 logprintf (LOG_VERBOSE, _("\
2627 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2628 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2634 logprintf (LOG_VERBOSE, _("\
2635 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2636 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2643 count = 0; /* the retrieve count for HEAD is reset */
2645 } /* send_head_first */
2648 if ((tmr != (time_t) (-1))
2649 && ((hstat.len == hstat.contlen) ||
2650 ((hstat.res == 0) && (hstat.contlen == -1))))
2652 /* #### This code repeats in http.c and ftp.c. Move it to a
2654 const char *fl = NULL;
2655 if (opt.output_document)
2657 if (output_stream_regular)
2658 fl = opt.output_document;
2661 fl = hstat.local_file;
2665 /* Reparse time header, in case it's changed. */
2666 if (time_came_from_head
2667 && hstat.remote_time && hstat.remote_time[0])
2669 newtmr = http_atotm (hstat.remote_time);
2676 /* End of time-stamping section. */
2678 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2679 total_download_time += hstat.dltime;
2681 if (hstat.len == hstat.contlen)
2685 logprintf (LOG_VERBOSE,
2686 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2687 tms, tmrate, hstat.local_file,
2688 number_to_static_string (hstat.len),
2689 number_to_static_string (hstat.contlen));
2690 logprintf (LOG_NONVERBOSE,
2691 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2693 number_to_static_string (hstat.len),
2694 number_to_static_string (hstat.contlen),
2695 hstat.local_file, count);
2698 total_downloaded_bytes += hstat.len;
2700 /* Remember that we downloaded the file for later ".orig" code. */
2701 if (*dt & ADDED_HTML_EXTENSION)
2702 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2704 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2709 else if (hstat.res == 0) /* No read error */
2711 if (hstat.contlen == -1) /* We don't know how much we were supposed
2712 to get, so assume we succeeded. */
2716 logprintf (LOG_VERBOSE,
2717 _("%s (%s) - `%s' saved [%s]\n\n"),
2718 tms, tmrate, hstat.local_file,
2719 number_to_static_string (hstat.len));
2720 logprintf (LOG_NONVERBOSE,
2721 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2722 tms, u->url, number_to_static_string (hstat.len),
2723 hstat.local_file, count);
2726 total_downloaded_bytes += hstat.len;
2728 /* Remember that we downloaded the file for later ".orig" code. */
2729 if (*dt & ADDED_HTML_EXTENSION)
2730 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2732 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2737 else if (hstat.len < hstat.contlen) /* meaning we lost the
2738 connection too soon */
2740 logprintf (LOG_VERBOSE,
2741 _("%s (%s) - Connection closed at byte %s. "),
2742 tms, tmrate, number_to_static_string (hstat.len));
2743 printwhat (count, opt.ntry);
2747 /* Getting here would mean reading more data than
2748 requested with content-length, which we never do. */
2751 else /* from now on hstat.res can only be -1 */
2753 if (hstat.contlen == -1)
2755 logprintf (LOG_VERBOSE,
2756 _("%s (%s) - Read error at byte %s (%s)."),
2757 tms, tmrate, number_to_static_string (hstat.len),
2759 printwhat (count, opt.ntry);
2762 else /* hstat.res == -1 and contlen is given */
2764 logprintf (LOG_VERBOSE,
2765 _("%s (%s) - Read error at byte %s/%s (%s). "),
2767 number_to_static_string (hstat.len),
2768 number_to_static_string (hstat.contlen),
2770 printwhat (count, opt.ntry);
2776 while (!opt.ntry || (count < opt.ntry));
2780 *local_file = xstrdup (hstat.local_file);
2781 free_hstat (&hstat);
2786 /* Check whether the result of strptime() indicates success.
2787 strptime() returns the pointer to how far it got to in the string.
2788 The processing has been successful if the string is at `GMT' or
2789 `+X', or at the end of the string.
2791 In extended regexp parlance, the function returns 1 if P matches
2792 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2793 can return) is considered a failure and 0 is returned. */
2795 check_end (const char *p)
2799 while (ISSPACE (*p))
2802 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2803 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2809 /* Convert the textual specification of time in TIME_STRING to the
2810 number of seconds since the Epoch.
2812 TIME_STRING can be in any of the three formats RFC2616 allows the
2813 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2814 as well as the time format used in the Set-Cookie header.
2815 Timezones are ignored, and should be GMT.
2817 Return the computed time_t representation, or -1 if the conversion
2820 This function uses strptime with various string formats for parsing
2821 TIME_STRING. This results in a parser that is not as lenient in
2822 interpreting TIME_STRING as I would like it to be. Being based on
2823 strptime, it always allows shortened months, one-digit days, etc.,
2824 but due to the multitude of formats in which time can be
2825 represented, an ideal HTTP time parser would be even more
2826 forgiving. It should completely ignore things like week days and
2827 concentrate only on the various forms of representing years,
2828 months, days, hours, minutes, and seconds. For example, it would
2829 be nice if it accepted ISO 8601 out of the box.
2831 I've investigated free and PD code for this purpose, but none was
2832 usable. getdate was big and unwieldy, and had potential copyright
2833 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2834 distributed with phttpd, is excellent, but we cannot use it because
2835 it is not assigned to the FSF. So I stuck it with strptime. */
2838 http_atotm (const char *time_string)
2840 /* NOTE: Solaris strptime man page claims that %n and %t match white
2841 space, but that's not universally available. Instead, we simply
2842 use ` ' to mean "skip all WS", which works under all strptime
2843 implementations I've tested. */
2845 static const char *time_formats[] = {
2846 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2847 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2848 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2849 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2850 (used in Set-Cookie, defined in the
2851 Netscape cookie specification.) */
2853 const char *oldlocale;
2855 time_t ret = (time_t) -1;
2857 /* Solaris strptime fails to recognize English month names in
2858 non-English locales, which we work around by temporarily setting
2859 locale to C before invoking strptime. */
2860 oldlocale = setlocale (LC_TIME, NULL);
2861 setlocale (LC_TIME, "C");
2863 for (i = 0; i < countof (time_formats); i++)
2867 /* Some versions of strptime use the existing contents of struct
2868 tm to recalculate the date according to format. Zero it out
2869 to prevent stack garbage from influencing strptime. */
2872 if (check_end (strptime (time_string, time_formats[i], &t)))
2879 /* Restore the previous locale. */
2880 setlocale (LC_TIME, oldlocale);
2885 /* Authorization support: We support three authorization schemes:
2887 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2889 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2890 consisting of answering to the server's challenge with the proper
2893 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2894 Stenberg for libcurl. Like digest, NTLM is based on a
2895 challenge-response mechanism, but unlike digest, it is non-standard
2896 (authenticates TCP connections rather than requests), undocumented
2897 and Microsoft-specific. */
2899 /* Create the authentication header contents for the `Basic' scheme.
2900 This is done by encoding the string "USER:PASS" to base64 and
2901 prepending the string "Basic " in front of it. */
2904 basic_authentication_encode (const char *user, const char *passwd)
2907 int len1 = strlen (user) + 1 + strlen (passwd);
2909 t1 = (char *)alloca (len1 + 1);
2910 sprintf (t1, "%s:%s", user, passwd);
2912 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2913 base64_encode (t1, len1, t2);
2915 return concat_strings ("Basic ", t2, (char *) 0);
2918 #define SKIP_WS(x) do { \
2919 while (ISSPACE (*(x))) \
2923 #ifdef ENABLE_DIGEST
2924 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2925 an array of 16 bytes containing the hash keys, and BUF should be a
2926 buffer of 33 writable characters (32 for hex digits plus one for
2927 zero termination). */
2929 dump_hash (char *buf, const unsigned char *hash)
2933 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2935 *buf++ = XNUM_TO_digit (*hash >> 4);
2936 *buf++ = XNUM_TO_digit (*hash & 0xf);
2941 /* Take the line apart to find the challenge, and compose a digest
2942 authorization header. See RFC2069 section 2.1.2. */
2944 digest_authentication_encode (const char *au, const char *user,
2945 const char *passwd, const char *method,
2948 static char *realm, *opaque, *nonce;
2953 { "realm", &realm },
2954 { "opaque", &opaque },
2958 param_token name, value;
2960 realm = opaque = nonce = NULL;
2962 au += 6; /* skip over `Digest' */
2963 while (extract_param (&au, &name, &value, ','))
2966 for (i = 0; i < countof (options); i++)
2967 if (name.e - name.b == strlen (options[i].name)
2968 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2970 *options[i].variable = strdupdelim (value.b, value.e);
2974 if (!realm || !nonce || !user || !passwd || !path || !method)
2977 xfree_null (opaque);
2982 /* Calculate the digest value. */
2984 ALLOCA_MD5_CONTEXT (ctx);
2985 unsigned char hash[MD5_HASHLEN];
2986 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2987 char response_digest[MD5_HASHLEN * 2 + 1];
2989 /* A1BUF = H(user ":" realm ":" password) */
2991 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2992 gen_md5_update ((unsigned char *)":", 1, ctx);
2993 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2994 gen_md5_update ((unsigned char *)":", 1, ctx);
2995 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2996 gen_md5_finish (ctx, hash);
2997 dump_hash (a1buf, hash);
2999 /* A2BUF = H(method ":" path) */
3001 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3002 gen_md5_update ((unsigned char *)":", 1, ctx);
3003 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3004 gen_md5_finish (ctx, hash);
3005 dump_hash (a2buf, hash);
3007 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3009 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3010 gen_md5_update ((unsigned char *)":", 1, ctx);
3011 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3012 gen_md5_update ((unsigned char *)":", 1, ctx);
3013 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3014 gen_md5_finish (ctx, hash);
3015 dump_hash (response_digest, hash);
3017 res = xmalloc (strlen (user)
3022 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3023 + (opaque ? strlen (opaque) : 0)
3025 sprintf (res, "Digest \
3026 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3027 user, realm, nonce, path, response_digest);
3030 char *p = res + strlen (res);
3031 strcat (p, ", opaque=\"");
3038 #endif /* ENABLE_DIGEST */
3040 /* Computing the size of a string literal must take into account that
3041 value returned by sizeof includes the terminating \0. */
3042 #define STRSIZE(literal) (sizeof (literal) - 1)
3044 /* Whether chars in [b, e) begin with the literal string provided as
3045 first argument and are followed by whitespace or terminating \0.
3046 The comparison is case-insensitive. */
3047 #define STARTS(literal, b, e) \
3048 ((e) - (b) >= STRSIZE (literal) \
3049 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3050 && ((e) - (b) == STRSIZE (literal) \
3051 || ISSPACE (b[STRSIZE (literal)])))
3054 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3056 return STARTS ("Basic", hdrbeg, hdrend)
3057 #ifdef ENABLE_DIGEST
3058 || STARTS ("Digest", hdrbeg, hdrend)
3061 || STARTS ("NTLM", hdrbeg, hdrend)
3068 /* Create the HTTP authorization request header. When the
3069 `WWW-Authenticate' response header is seen, according to the
3070 authorization scheme specified in that header (`Basic' and `Digest'
3071 are supported by the current implementation), produce an
3072 appropriate HTTP authorization request header. */
3074 create_authorization_line (const char *au, const char *user,
3075 const char *passwd, const char *method,
3076 const char *path, bool *finished)
3078 /* We are called only with known schemes, so we can dispatch on the
3080 switch (TOUPPER (*au))
3082 case 'B': /* Basic */
3084 return basic_authentication_encode (user, passwd);
3085 #ifdef ENABLE_DIGEST
3086 case 'D': /* Digest */
3088 return digest_authentication_encode (au, user, passwd, method, path);
3091 case 'N': /* NTLM */
3092 if (!ntlm_input (&pconn.ntlm, au))
3097 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3100 /* We shouldn't get here -- this function should be only called
3101 with values approved by known_authentication_scheme_p. */
3109 if (!wget_cookie_jar)
3110 wget_cookie_jar = cookie_jar_new ();
3111 if (opt.cookies_input && !cookies_loaded_p)
3113 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3114 cookies_loaded_p = true;
3121 if (wget_cookie_jar)
3122 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3128 xfree_null (pconn.host);
3129 if (wget_cookie_jar)
3130 cookie_jar_delete (wget_cookie_jar);
3137 test_parse_content_disposition()
3142 char *opt_dir_prefix;
3146 { "filename=\"file.ext\"", NULL, "file.ext", true },
3147 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3148 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3149 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3150 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3151 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3152 { "attachment", NULL, NULL, false },
3153 { "attachment", "somedir", NULL, false },
3156 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3161 opt.dir_prefix = test_array[i].opt_dir_prefix;
3162 res = parse_content_disposition (test_array[i].hdrval, &filename);
3164 mu_assert ("test_parse_content_disposition: wrong result",
3165 res == test_array[i].result
3167 || 0 == strcmp (test_array[i].filename, filename)));
3173 #endif /* TESTING */
3176 * vim: et sts=2 sw=2 cino+={s