2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
56 # include "http-ntlm.h"
69 extern char *version_string;
73 static char *create_authorization_line (const char *, const char *,
74 const char *, const char *,
75 const char *, bool *);
76 static char *basic_authentication_encode (const char *, const char *);
77 static bool known_authentication_scheme_p (const char *, const char *);
78 static void ensure_extension (struct http_stat *, const char *, int *);
79 static void load_cookies (void);
82 # define MIN(x, y) ((x) > (y) ? (y) : (x))
86 static bool cookies_loaded_p;
87 static struct cookie_jar *wget_cookie_jar;
89 #define TEXTHTML_S "text/html"
90 #define TEXTXHTML_S "application/xhtml+xml"
91 #define TEXTCSS_S "text/css"
93 /* Some status code validation macros: */
94 #define H_20X(x) (((x) >= 200) && ((x) < 300))
95 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
96 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
97 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
98 || (x) == HTTP_STATUS_SEE_OTHER \
99 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
101 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
102 /* Successful 2xx. */
103 #define HTTP_STATUS_OK 200
104 #define HTTP_STATUS_CREATED 201
105 #define HTTP_STATUS_ACCEPTED 202
106 #define HTTP_STATUS_NO_CONTENT 204
107 #define HTTP_STATUS_PARTIAL_CONTENTS 206
109 /* Redirection 3xx. */
110 #define HTTP_STATUS_MULTIPLE_CHOICES 300
111 #define HTTP_STATUS_MOVED_PERMANENTLY 301
112 #define HTTP_STATUS_MOVED_TEMPORARILY 302
113 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
114 #define HTTP_STATUS_NOT_MODIFIED 304
115 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
117 /* Client error 4xx. */
118 #define HTTP_STATUS_BAD_REQUEST 400
119 #define HTTP_STATUS_UNAUTHORIZED 401
120 #define HTTP_STATUS_FORBIDDEN 403
121 #define HTTP_STATUS_NOT_FOUND 404
122 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
124 /* Server errors 5xx. */
125 #define HTTP_STATUS_INTERNAL 500
126 #define HTTP_STATUS_NOT_IMPLEMENTED 501
127 #define HTTP_STATUS_BAD_GATEWAY 502
128 #define HTTP_STATUS_UNAVAILABLE 503
131 rel_none, rel_name, rel_value, rel_both
138 struct request_header {
140 enum rp release_policy;
142 int hcount, hcapacity;
145 /* Create a new, empty request. At least request_set_method must be
146 called before the request can be used. */
148 static struct request *
151 struct request *req = xnew0 (struct request);
153 req->headers = xnew_array (struct request_header, req->hcapacity);
157 /* Set the request's method and its arguments. METH should be a
158 literal string (or it should outlive the request) because it will
159 not be freed. ARG will be freed by request_free. */
162 request_set_method (struct request *req, const char *meth, char *arg)
168 /* Return the method string passed with the last call to
169 request_set_method. */
172 request_method (const struct request *req)
177 /* Free one header according to the release policy specified with
178 request_set_header. */
181 release_header (struct request_header *hdr)
183 switch (hdr->release_policy)
200 /* Set the request named NAME to VALUE. Specifically, this means that
201 a "NAME: VALUE\r\n" header line will be used in the request. If a
202 header with the same name previously existed in the request, its
203 value will be replaced by this one. A NULL value means do nothing.
205 RELEASE_POLICY determines whether NAME and VALUE should be released
206 (freed) with request_free. Allowed values are:
208 - rel_none - don't free NAME or VALUE
209 - rel_name - free NAME when done
210 - rel_value - free VALUE when done
211 - rel_both - free both NAME and VALUE when done
213 Setting release policy is useful when arguments come from different
214 sources. For example:
216 // Don't free literal strings!
217 request_set_header (req, "Pragma", "no-cache", rel_none);
219 // Don't free a global variable, we'll need it later.
220 request_set_header (req, "Referer", opt.referer, rel_none);
222 // Value freshly allocated, free it when done.
223 request_set_header (req, "Range",
224 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
229 request_set_header (struct request *req, char *name, char *value,
230 enum rp release_policy)
232 struct request_header *hdr;
237 /* A NULL value is a no-op; if freeing the name is requested,
238 free it now to avoid leaks. */
239 if (release_policy == rel_name || release_policy == rel_both)
244 for (i = 0; i < req->hcount; i++)
246 hdr = &req->headers[i];
247 if (0 == strcasecmp (name, hdr->name))
249 /* Replace existing header. */
250 release_header (hdr);
253 hdr->release_policy = release_policy;
258 /* Install new header. */
260 if (req->hcount >= req->hcapacity)
262 req->hcapacity <<= 1;
263 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
265 hdr = &req->headers[req->hcount++];
268 hdr->release_policy = release_policy;
271 /* Like request_set_header, but sets the whole header line, as
272 provided by the user using the `--header' option. For example,
273 request_set_user_header (req, "Foo: bar") works just like
274 request_set_header (req, "Foo", "bar"). */
277 request_set_user_header (struct request *req, const char *header)
280 const char *p = strchr (header, ':');
283 BOUNDED_TO_ALLOCA (header, p, name);
285 while (c_isspace (*p))
287 request_set_header (req, xstrdup (name), (char *) p, rel_name);
290 /* Remove the header with specified name from REQ. Returns true if
291 the header was actually removed, false otherwise. */
294 request_remove_header (struct request *req, char *name)
297 for (i = 0; i < req->hcount; i++)
299 struct request_header *hdr = &req->headers[i];
300 if (0 == strcasecmp (name, hdr->name))
302 release_header (hdr);
303 /* Move the remaining headers by one. */
304 if (i < req->hcount - 1)
305 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
313 #define APPEND(p, str) do { \
314 int A_len = strlen (str); \
315 memcpy (p, str, A_len); \
319 /* Construct the request and write it to FD using fd_write. */
322 request_send (const struct request *req, int fd)
324 char *request_string, *p;
325 int i, size, write_error;
327 /* Count the request size. */
330 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
331 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
333 for (i = 0; i < req->hcount; i++)
335 struct request_header *hdr = &req->headers[i];
336 /* NAME ": " VALUE "\r\n" */
337 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
343 p = request_string = alloca_array (char, size);
345 /* Generate the request. */
347 APPEND (p, req->method); *p++ = ' ';
348 APPEND (p, req->arg); *p++ = ' ';
349 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
351 for (i = 0; i < req->hcount; i++)
353 struct request_header *hdr = &req->headers[i];
354 APPEND (p, hdr->name);
355 *p++ = ':', *p++ = ' ';
356 APPEND (p, hdr->value);
357 *p++ = '\r', *p++ = '\n';
360 *p++ = '\r', *p++ = '\n', *p++ = '\0';
361 assert (p - request_string == size);
365 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
367 /* Send the request to the server. */
369 write_error = fd_write (fd, request_string, size - 1, -1);
371 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
376 /* Release the resources used by REQ. */
379 request_free (struct request *req)
382 xfree_null (req->arg);
383 for (i = 0; i < req->hcount; i++)
384 release_header (&req->headers[i]);
385 xfree_null (req->headers);
389 static struct hash_table *basic_authed_hosts;
391 /* Find out if this host has issued a Basic challenge yet; if so, give
392 * it the username, password. A temporary measure until we can get
393 * proper authentication in place. */
396 maybe_send_basic_creds (const char *hostname, const char *user,
397 const char *passwd, struct request *req)
399 bool do_challenge = false;
401 if (opt.auth_without_challenge)
403 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
406 else if (basic_authed_hosts
407 && hash_table_contains(basic_authed_hosts, hostname))
409 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
414 DEBUGP(("Host %s has not issued a general basic challenge.\n",
419 request_set_header (req, "Authorization",
420 basic_authentication_encode (user, passwd),
427 register_basic_auth_host (const char *hostname)
429 if (!basic_authed_hosts)
431 basic_authed_hosts = make_nocase_string_hash_table (1);
433 if (!hash_table_contains(basic_authed_hosts, hostname))
435 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
436 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
441 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
442 PROMISED_SIZE bytes are sent over the wire -- if the file is
443 longer, read only that much; if the file is shorter, report an error. */
446 post_file (int sock, const char *file_name, wgint promised_size)
448 static char chunk[8192];
453 DEBUGP (("[writing POST file %s ... ", file_name));
455 fp = fopen (file_name, "rb");
458 while (!feof (fp) && written < promised_size)
461 int length = fread (chunk, 1, sizeof (chunk), fp);
464 towrite = MIN (promised_size - written, length);
465 write_error = fd_write (sock, chunk, towrite, -1);
475 /* If we've written less than was promised, report a (probably
476 nonsensical) error rather than break the promise. */
477 if (written < promised_size)
483 assert (written == promised_size);
484 DEBUGP (("done]\n"));
488 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
489 If so, return the pointer to the position after the line, otherwise
490 return NULL. This is used as callback to fd_read_hunk. The data
491 between START and PEEKED has been read and cannot be "unread"; the
492 data after PEEKED has only been peeked. */
495 response_head_terminator (const char *start, const char *peeked, int peeklen)
499 /* If at first peek, verify whether HUNK starts with "HTTP". If
500 not, this is a HTTP/0.9 request and we must bail out without
502 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
505 /* Look for "\n[\r]\n", and return the following position if found.
506 Start two chars before the current to cover the possibility that
507 part of the terminator (e.g. "\n\r") arrived in the previous
509 p = peeked - start < 2 ? start : peeked - 2;
510 end = peeked + peeklen;
512 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
513 for (; p < end - 2; p++)
516 if (p[1] == '\r' && p[2] == '\n')
518 else if (p[1] == '\n')
521 /* p==end-2: check for \n\n directly preceding END. */
522 if (p[0] == '\n' && p[1] == '\n')
528 /* The maximum size of a single HTTP response we care to read. Rather
529 than being a limit of the reader implementation, this limit
530 prevents Wget from slurping all available memory upon encountering
531 malicious or buggy server output, thus protecting the user. Define
532 it to 0 to remove the limit. */
534 #define HTTP_RESPONSE_MAX_SIZE 65536
536 /* Read the HTTP request head from FD and return it. The error
537 conditions are the same as with fd_read_hunk.
539 To support HTTP/0.9 responses, this function tries to make sure
540 that the data begins with "HTTP". If this is not the case, no data
541 is read and an empty request is returned, so that the remaining
542 data can be treated as body. */
545 read_http_response_head (int fd)
547 return fd_read_hunk (fd, response_head_terminator, 512,
548 HTTP_RESPONSE_MAX_SIZE);
552 /* The response data. */
555 /* The array of pointers that indicate where each header starts.
556 For example, given this HTTP response:
563 The headers are located like this:
565 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
567 headers[0] headers[1] headers[2] headers[3]
569 I.e. headers[0] points to the beginning of the request,
570 headers[1] points to the end of the first header and the
571 beginning of the second one, etc. */
573 const char **headers;
576 /* Create a new response object from the text of the HTTP response,
577 available in HEAD. That text is automatically split into
578 constituent header lines for fast retrieval using
581 static struct response *
582 resp_new (const char *head)
587 struct response *resp = xnew0 (struct response);
592 /* Empty head means that we're dealing with a headerless
593 (HTTP/0.9) response. In that case, don't set HEADERS at
598 /* Split HEAD into header lines, so that resp_header_* functions
599 don't need to do this over and over again. */
605 DO_REALLOC (resp->headers, size, count + 1, const char *);
606 resp->headers[count++] = hdr;
608 /* Break upon encountering an empty line. */
609 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
612 /* Find the end of HDR, including continuations. */
615 const char *end = strchr (hdr, '\n');
621 while (*hdr == ' ' || *hdr == '\t');
623 DO_REALLOC (resp->headers, size, count + 1, const char *);
624 resp->headers[count] = NULL;
629 /* Locate the header named NAME in the request data, starting with
630 position START. This allows the code to loop through the request
631 data, filtering for all requests of a given name. Returns the
632 found position, or -1 for failure. The code that uses this
633 function typically looks like this:
635 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
636 ... do something with header ...
638 If you only care about one header, use resp_header_get instead of
642 resp_header_locate (const struct response *resp, const char *name, int start,
643 const char **begptr, const char **endptr)
646 const char **headers = resp->headers;
649 if (!headers || !headers[1])
652 name_len = strlen (name);
658 for (; headers[i + 1]; i++)
660 const char *b = headers[i];
661 const char *e = headers[i + 1];
663 && b[name_len] == ':'
664 && 0 == strncasecmp (b, name, name_len))
667 while (b < e && c_isspace (*b))
669 while (b < e && c_isspace (e[-1]))
679 /* Find and retrieve the header named NAME in the request data. If
680 found, set *BEGPTR to its starting, and *ENDPTR to its ending
681 position, and return true. Otherwise return false.
683 This function is used as a building block for resp_header_copy
684 and resp_header_strdup. */
687 resp_header_get (const struct response *resp, const char *name,
688 const char **begptr, const char **endptr)
690 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
694 /* Copy the response header named NAME to buffer BUF, no longer than
695 BUFSIZE (BUFSIZE includes the terminating 0). If the header
696 exists, true is returned, false otherwise. If there should be no
697 limit on the size of the header, use resp_header_strdup instead.
699 If BUFSIZE is 0, no data is copied, but the boolean indication of
700 whether the header is present is still returned. */
703 resp_header_copy (const struct response *resp, const char *name,
704 char *buf, int bufsize)
707 if (!resp_header_get (resp, name, &b, &e))
711 int len = MIN (e - b, bufsize - 1);
712 memcpy (buf, b, len);
718 /* Return the value of header named NAME in RESP, allocated with
719 malloc. If such a header does not exist in RESP, return NULL. */
722 resp_header_strdup (const struct response *resp, const char *name)
725 if (!resp_header_get (resp, name, &b, &e))
727 return strdupdelim (b, e);
730 /* Parse the HTTP status line, which is of format:
732 HTTP-Version SP Status-Code SP Reason-Phrase
734 The function returns the status-code, or -1 if the status line
735 appears malformed. The pointer to "reason-phrase" message is
736 returned in *MESSAGE. */
739 resp_status (const struct response *resp, char **message)
746 /* For a HTTP/0.9 response, assume status 200. */
748 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
752 p = resp->headers[0];
753 end = resp->headers[1];
759 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
763 /* Match the HTTP version. This is optional because Gnutella
764 servers have been reported to not specify HTTP version. */
765 if (p < end && *p == '/')
768 while (p < end && c_isdigit (*p))
770 if (p < end && *p == '.')
772 while (p < end && c_isdigit (*p))
776 while (p < end && c_isspace (*p))
778 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
781 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
786 while (p < end && c_isspace (*p))
788 while (p < end && c_isspace (end[-1]))
790 *message = strdupdelim (p, end);
796 /* Release the resources used by RESP. */
799 resp_free (struct response *resp)
801 xfree_null (resp->headers);
805 /* Print a single line of response, the characters [b, e). We tried
807 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
808 but that failed to escape the non-printable characters and, in fact,
809 caused crashes in UTF-8 locales. */
812 print_response_line(const char *prefix, const char *b, const char *e)
815 BOUNDED_TO_ALLOCA(b, e, copy);
816 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
817 quotearg_style (escape_quoting_style, copy));
820 /* Print the server response, line by line, omitting the trailing CRLF
821 from individual header lines, and prefixed with PREFIX. */
824 print_server_response (const struct response *resp, const char *prefix)
829 for (i = 0; resp->headers[i + 1]; i++)
831 const char *b = resp->headers[i];
832 const char *e = resp->headers[i + 1];
834 if (b < e && e[-1] == '\n')
836 if (b < e && e[-1] == '\r')
838 print_response_line(prefix, b, e);
842 /* Parse the `Content-Range' header and extract the information it
843 contains. Returns true if successful, false otherwise. */
845 parse_content_range (const char *hdr, wgint *first_byte_ptr,
846 wgint *last_byte_ptr, wgint *entity_length_ptr)
850 /* Ancient versions of Netscape proxy server, presumably predating
851 rfc2068, sent out `Content-Range' without the "bytes"
853 if (0 == strncasecmp (hdr, "bytes", 5))
856 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
860 while (c_isspace (*hdr))
865 if (!c_isdigit (*hdr))
867 for (num = 0; c_isdigit (*hdr); hdr++)
868 num = 10 * num + (*hdr - '0');
869 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
871 *first_byte_ptr = num;
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
877 *last_byte_ptr = num;
882 for (num = 0; c_isdigit (*hdr); hdr++)
883 num = 10 * num + (*hdr - '0');
884 *entity_length_ptr = num;
888 /* Read the body of the request, but don't store it anywhere and don't
889 display a progress gauge. This is useful for reading the bodies of
890 administrative responses to which we will soon issue another
891 request. The response is not useful to the user, but reading it
892 allows us to continue using the same connection to the server.
894 If reading fails, false is returned, true otherwise. In debug
895 mode, the body is displayed for debugging purposes. */
898 skip_short_body (int fd, wgint contlen)
901 SKIP_SIZE = 512, /* size of the download buffer */
902 SKIP_THRESHOLD = 4096 /* the largest size we read */
904 char dlbuf[SKIP_SIZE + 1];
905 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
907 /* We shouldn't get here with unknown contlen. (This will change
908 with HTTP/1.1, which supports "chunked" transfer.) */
909 assert (contlen != -1);
911 /* If the body is too large, it makes more sense to simply close the
912 connection than to try to read the body. */
913 if (contlen > SKIP_THRESHOLD)
916 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
920 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
923 /* Don't normally report the error since this is an
924 optimization that should be invisible to the user. */
925 DEBUGP (("] aborting (%s).\n",
926 ret < 0 ? fd_errstr (fd) : "EOF received"));
930 /* Safe even if %.*s bogusly expects terminating \0 because
931 we've zero-terminated dlbuf above. */
932 DEBUGP (("%.*s", ret, dlbuf));
935 DEBUGP (("] done.\n"));
939 /* Extract a parameter from the string (typically an HTTP header) at
940 **SOURCE and advance SOURCE to the next parameter. Return false
941 when there are no more parameters to extract. The name of the
942 parameter is returned in NAME, and the value in VALUE. If the
943 parameter has no value, the token's value is zeroed out.
945 For example, if *SOURCE points to the string "attachment;
946 filename=\"foo bar\"", the first call to this function will return
947 the token named "attachment" and no value, and the second call will
948 return the token named "filename" and value "foo bar". The third
949 call will return false, indicating no more valid tokens. */
952 extract_param (const char **source, param_token *name, param_token *value,
955 const char *p = *source;
957 while (c_isspace (*p)) ++p;
961 return false; /* no error; nothing more to extract */
966 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
968 if (name->b == name->e)
969 return false; /* empty name: error */
970 while (c_isspace (*p)) ++p;
971 if (*p == separator || !*p) /* no value */
974 if (*p == separator) ++p;
979 return false; /* error */
981 /* *p is '=', extract value */
983 while (c_isspace (*p)) ++p;
984 if (*p == '"') /* quoted */
987 while (*p && *p != '"') ++p;
991 /* Currently at closing quote; find the end of param. */
992 while (c_isspace (*p)) ++p;
993 while (*p && *p != separator) ++p;
997 /* garbage after closed quote, e.g. foo="bar"baz */
1003 while (*p && *p != separator) ++p;
1005 while (value->e != value->b && c_isspace (value->e[-1]))
1007 if (*p == separator) ++p;
1014 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1016 /* Parse the contents of the `Content-Disposition' header, extracting
1017 the information useful to Wget. Content-Disposition is a header
1018 borrowed from MIME; when used in HTTP, it typically serves for
1019 specifying the desired file name of the resource. For example:
1021 Content-Disposition: attachment; filename="flora.jpg"
1023 Wget will skip the tokens it doesn't care about, such as
1024 "attachment" in the previous example; it will also skip other
1025 unrecognized params. If the header is syntactically correct and
1026 contains a file name, a copy of the file name is stored in
1027 *filename and true is returned. Otherwise, the function returns
1030 The file name is stripped of directory components and must not be
1034 parse_content_disposition (const char *hdr, char **filename)
1036 param_token name, value;
1037 while (extract_param (&hdr, &name, &value, ';'))
1038 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1040 /* Make the file name begin at the last slash or backslash. */
1041 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1042 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1043 if (last_slash && last_bs)
1044 value.b = 1 + MAX (last_slash, last_bs);
1045 else if (last_slash || last_bs)
1046 value.b = 1 + (last_slash ? last_slash : last_bs);
1047 if (value.b == value.e)
1049 /* Start with the directory prefix, if specified. */
1052 int prefix_length = strlen (opt.dir_prefix);
1053 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1058 total_length = prefix_length + (value.e - value.b);
1059 *filename = xmalloc (total_length + 1);
1060 strcpy (*filename, opt.dir_prefix);
1062 (*filename)[prefix_length - 1] = '/';
1063 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1064 (*filename)[total_length] = '\0';
1067 *filename = strdupdelim (value.b, value.e);
1073 /* Persistent connections. Currently, we cache the most recently used
1074 connection as persistent, provided that the HTTP server agrees to
1075 make it such. The persistence data is stored in the variables
1076 below. Ideally, it should be possible to cache an arbitrary fixed
1077 number of these connections. */
1079 /* Whether a persistent connection is active. */
1080 static bool pconn_active;
1083 /* The socket of the connection. */
1086 /* Host and port of the currently active persistent connection. */
1090 /* Whether a ssl handshake has occoured on this connection. */
1093 /* Whether the connection was authorized. This is only done by
1094 NTLM, which authorizes *connections* rather than individual
1095 requests. (That practice is peculiar for HTTP, but it is a
1096 useful optimization.) */
1100 /* NTLM data of the current connection. */
1101 struct ntlmdata ntlm;
1105 /* Mark the persistent connection as invalid and free the resources it
1106 uses. This is used by the CLOSE_* macros after they forcefully
1107 close a registered persistent connection. */
1110 invalidate_persistent (void)
1112 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1113 pconn_active = false;
1114 fd_close (pconn.socket);
1119 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1120 persistent. This will enable someone to use the same connection
1121 later. In the context of HTTP, this must be called only AFTER the
1122 response has been received and the server has promised that the
1123 connection will remain alive.
1125 If a previous connection was persistent, it is closed. */
1128 register_persistent (const char *host, int port, int fd, bool ssl)
1132 if (pconn.socket == fd)
1134 /* The connection FD is already registered. */
1139 /* The old persistent connection is still active; close it
1140 first. This situation arises whenever a persistent
1141 connection exists, but we then connect to a different
1142 host, and try to register a persistent connection to that
1144 invalidate_persistent ();
1148 pconn_active = true;
1150 pconn.host = xstrdup (host);
1153 pconn.authorized = false;
1155 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1158 /* Return true if a persistent connection is available for connecting
1162 persistent_available_p (const char *host, int port, bool ssl,
1163 bool *host_lookup_failed)
1165 /* First, check whether a persistent connection is active at all. */
1169 /* If we want SSL and the last connection wasn't or vice versa,
1170 don't use it. Checking for host and port is not enough because
1171 HTTP and HTTPS can apparently coexist on the same port. */
1172 if (ssl != pconn.ssl)
1175 /* If we're not connecting to the same port, we're not interested. */
1176 if (port != pconn.port)
1179 /* If the host is the same, we're in business. If not, there is
1180 still hope -- read below. */
1181 if (0 != strcasecmp (host, pconn.host))
1183 /* Check if pconn.socket is talking to HOST under another name.
1184 This happens often when both sites are virtual hosts
1185 distinguished only by name and served by the same network
1186 interface, and hence the same web server (possibly set up by
1187 the ISP and serving many different web sites). This
1188 admittedly unconventional optimization does not contradict
1189 HTTP and works well with popular server software. */
1193 struct address_list *al;
1196 /* Don't try to talk to two different SSL sites over the same
1197 secure connection! (Besides, it's not clear that
1198 name-based virtual hosting is even possible with SSL.) */
1201 /* If pconn.socket's peer is one of the IP addresses HOST
1202 resolves to, pconn.socket is for all intents and purposes
1203 already talking to HOST. */
1205 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1207 /* Can't get the peer's address -- something must be very
1208 wrong with the connection. */
1209 invalidate_persistent ();
1212 al = lookup_host (host, 0);
1215 *host_lookup_failed = true;
1219 found = address_list_contains (al, &ip);
1220 address_list_release (al);
1225 /* The persistent connection's peer address was found among the
1226 addresses HOST resolved to; therefore, pconn.sock is in fact
1227 already talking to HOST -- no need to reconnect. */
1230 /* Finally, check whether the connection is still open. This is
1231 important because most servers implement liberal (short) timeout
1232 on persistent connections. Wget can of course always reconnect
1233 if the connection doesn't work out, but it's nicer to know in
1234 advance. This test is a logical followup of the first test, but
1235 is "expensive" and therefore placed at the end of the list.
1237 (Current implementation of test_socket_open has a nice side
1238 effect that it treats sockets with pending data as "closed".
1239 This is exactly what we want: if a broken server sends message
1240 body in response to HEAD, or if it sends more than conent-length
1241 data, we won't reuse the corrupted connection.) */
1243 if (!test_socket_open (pconn.socket))
1245 /* Oops, the socket is no longer open. Now that we know that,
1246 let's invalidate the persistent connection before returning
1248 invalidate_persistent ();
1255 /* The idea behind these two CLOSE macros is to distinguish between
1256 two cases: one when the job we've been doing is finished, and we
1257 want to close the connection and leave, and two when something is
1258 seriously wrong and we're closing the connection as part of
1261 In case of keep_alive, CLOSE_FINISH should leave the connection
1262 open, while CLOSE_INVALIDATE should still close it.
1264 Note that the semantics of the flag `keep_alive' is "this
1265 connection *will* be reused (the server has promised not to close
1266 the connection once we're done)", while the semantics of
1267 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1268 active, registered connection". */
1270 #define CLOSE_FINISH(fd) do { \
1273 if (pconn_active && (fd) == pconn.socket) \
1274 invalidate_persistent (); \
1283 #define CLOSE_INVALIDATE(fd) do { \
1284 if (pconn_active && (fd) == pconn.socket) \
1285 invalidate_persistent (); \
1293 wgint len; /* received length */
1294 wgint contlen; /* expected length */
1295 wgint restval; /* the restart value */
1296 int res; /* the result of last read */
1297 char *rderrmsg; /* error message from read error */
1298 char *newloc; /* new location (redirection) */
1299 char *remote_time; /* remote time-stamp string */
1300 char *error; /* textual HTTP error */
1301 int statcode; /* status code */
1302 char *message; /* status message */
1303 wgint rd_size; /* amount of data read from socket */
1304 double dltime; /* time it took to download the data */
1305 const char *referer; /* value of the referer header. */
1306 char *local_file; /* local file name. */
1307 bool existence_checked; /* true if we already checked for a file's
1308 existence after having begun to download
1309 (needed in gethttp for when connection is
1310 interrupted/restarted. */
1311 bool timestamp_checked; /* true if pre-download time-stamping checks
1312 * have already been performed */
1313 char *orig_file_name; /* name of file to compare for time-stamping
1314 * (might be != local_file if -K is set) */
1315 wgint orig_file_size; /* size of file to compare for time-stamping */
1316 time_t orig_file_tstamp; /* time-stamp of file to compare for
1321 free_hstat (struct http_stat *hs)
1323 xfree_null (hs->newloc);
1324 xfree_null (hs->remote_time);
1325 xfree_null (hs->error);
1326 xfree_null (hs->rderrmsg);
1327 xfree_null (hs->local_file);
1328 xfree_null (hs->orig_file_name);
1329 xfree_null (hs->message);
1331 /* Guard against being called twice. */
1333 hs->remote_time = NULL;
1337 #define BEGINS_WITH(line, string_constant) \
1338 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1339 && (c_isspace (line[sizeof (string_constant) - 1]) \
1340 || !line[sizeof (string_constant) - 1]))
1342 #define SET_USER_AGENT(req) do { \
1343 if (!opt.useragent) \
1344 request_set_header (req, "User-Agent", \
1345 aprintf ("Wget/%s", version_string), rel_value); \
1346 else if (*opt.useragent) \
1347 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1350 /* The flags that allow clobbering the file (opening with "wb").
1351 Defined here to avoid repetition later. #### This will require
1353 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1354 || opt.dirstruct || opt.output_document)
1356 /* Retrieve a document through HTTP protocol. It recognizes status
1357 code, and correctly handles redirections. It closes the network
1358 socket. If it receives an error from the functions below it, it
1359 will print it if there is enough information to do so (almost
1360 always), returning the error to the caller (i.e. http_loop).
1362 Various HTTP parameters are stored to hs.
1364 If PROXY is non-NULL, the connection will be made to the proxy
1365 server, and u->url will be requested. */
1367 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1370 struct request *req;
1373 char *user, *passwd;
1377 wgint contlen, contrange;
1384 /* Set to 1 when the authorization has already been sent and should
1385 not be tried again. */
1386 bool auth_finished = false;
1388 /* Set to 1 when just globally-set Basic authorization has been sent;
1389 * should prevent further Basic negotiations, but not other
1391 bool basic_auth_finished = false;
1393 /* Whether NTLM authentication is used for this request. */
1394 bool ntlm_seen = false;
1396 /* Whether our connection to the remote host is through SSL. */
1397 bool using_ssl = false;
1399 /* Whether a HEAD request will be issued (as opposed to GET or
1401 bool head_only = !!(*dt & HEAD_ONLY);
1404 struct response *resp;
1408 /* Whether this connection will be kept alive after the HTTP request
1412 /* Whether keep-alive should be inhibited.
1414 RFC 2068 requests that 1.0 clients not send keep-alive requests
1415 to proxies. This is because many 1.0 proxies do not interpret
1416 the Connection header and transfer it to the remote server,
1417 causing it to not close the connection and leave both the proxy
1418 and the client hanging. */
1419 bool inhibit_keep_alive =
1420 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1422 /* Headers sent when using POST. */
1423 wgint post_data_size = 0;
1425 bool host_lookup_failed = false;
1428 if (u->scheme == SCHEME_HTTPS)
1430 /* Initialize the SSL context. After this has once been done,
1431 it becomes a no-op. */
1434 scheme_disable (SCHEME_HTTPS);
1435 logprintf (LOG_NOTQUIET,
1436 _("Disabling SSL due to encountered errors.\n"));
1437 return SSLINITFAILED;
1440 #endif /* HAVE_SSL */
1442 /* Initialize certain elements of struct http_stat. */
1446 hs->rderrmsg = NULL;
1448 hs->remote_time = NULL;
1454 /* Prepare the request to send. */
1456 req = request_new ();
1459 const char *meth = "GET";
1462 else if (opt.post_file_name || opt.post_data)
1464 /* Use the full path, i.e. one that includes the leading slash and
1465 the query string. E.g. if u->path is "foo/bar" and u->query is
1466 "param=value", full_path will be "/foo/bar?param=value". */
1469 /* When using SSL over proxy, CONNECT establishes a direct
1470 connection to the HTTPS server. Therefore use the same
1471 argument as when talking to the server directly. */
1472 && u->scheme != SCHEME_HTTPS
1475 meth_arg = xstrdup (u->url);
1477 meth_arg = url_full_path (u);
1478 request_set_method (req, meth, meth_arg);
1481 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1482 if (*dt & SEND_NOCACHE)
1483 request_set_header (req, "Pragma", "no-cache", rel_none);
1485 request_set_header (req, "Range",
1486 aprintf ("bytes=%s-",
1487 number_to_static_string (hs->restval)),
1489 SET_USER_AGENT (req);
1490 request_set_header (req, "Accept", "*/*", rel_none);
1492 /* Find the username and password for authentication. */
1495 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1496 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1497 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1499 /* We only do "site-wide" authentication with "global" user/password
1500 * values unless --auth-no-challange has been requested; URL user/password
1501 * info overrides. */
1502 if (user && passwd && (!u->user || opt.auth_without_challenge))
1504 /* If this is a host for which we've already received a Basic
1505 * challenge, we'll go ahead and send Basic authentication creds. */
1506 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1509 /* Generate the Host header, HOST:PORT. Take into account that:
1511 - Broken server-side software often doesn't recognize the PORT
1512 argument, so we must generate "Host: www.server.com" instead of
1513 "Host: www.server.com:80" (and likewise for https port).
1515 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1516 becomes ambiguous and needs to be rewritten as "Host:
1517 [3ffe:8100:200:2::2]:1234". */
1519 /* Formats arranged for hfmt[add_port][add_squares]. */
1520 static const char *hfmt[][2] = {
1521 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1523 int add_port = u->port != scheme_default_port (u->scheme);
1524 int add_squares = strchr (u->host, ':') != NULL;
1525 request_set_header (req, "Host",
1526 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1530 if (!inhibit_keep_alive)
1531 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1534 request_set_header (req, "Cookie",
1535 cookie_header (wget_cookie_jar,
1536 u->host, u->port, u->path,
1538 u->scheme == SCHEME_HTTPS
1545 if (opt.post_data || opt.post_file_name)
1547 request_set_header (req, "Content-Type",
1548 "application/x-www-form-urlencoded", rel_none);
1550 post_data_size = strlen (opt.post_data);
1553 post_data_size = file_size (opt.post_file_name);
1554 if (post_data_size == -1)
1556 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1557 quote (opt.post_file_name), strerror (errno));
1561 request_set_header (req, "Content-Length",
1562 xstrdup (number_to_static_string (post_data_size)),
1566 /* Add the user headers. */
1567 if (opt.user_headers)
1570 for (i = 0; opt.user_headers[i]; i++)
1571 request_set_user_header (req, opt.user_headers[i]);
1575 /* We need to come back here when the initial attempt to retrieve
1576 without authorization header fails. (Expected to happen at least
1577 for the Digest authorization scheme.) */
1582 char *proxy_user, *proxy_passwd;
1583 /* For normal username and password, URL components override
1584 command-line/wgetrc parameters. With proxy
1585 authentication, it's the reverse, because proxy URLs are
1586 normally the "permanent" ones, so command-line args
1587 should take precedence. */
1588 if (opt.proxy_user && opt.proxy_passwd)
1590 proxy_user = opt.proxy_user;
1591 proxy_passwd = opt.proxy_passwd;
1595 proxy_user = proxy->user;
1596 proxy_passwd = proxy->passwd;
1598 /* #### This does not appear right. Can't the proxy request,
1599 say, `Digest' authentication? */
1600 if (proxy_user && proxy_passwd)
1601 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1603 /* If we're using a proxy, we will be connecting to the proxy
1607 /* Proxy authorization over SSL is handled below. */
1609 if (u->scheme != SCHEME_HTTPS)
1611 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1616 /* Establish the connection. */
1618 if (!inhibit_keep_alive)
1620 /* Look for a persistent connection to target host, unless a
1621 proxy is used. The exception is when SSL is in use, in which
1622 case the proxy is nothing but a passthrough to the target
1623 host, registered as a connection to the latter. */
1624 struct url *relevant = conn;
1626 if (u->scheme == SCHEME_HTTPS)
1630 if (persistent_available_p (relevant->host, relevant->port,
1632 relevant->scheme == SCHEME_HTTPS,
1636 &host_lookup_failed))
1638 sock = pconn.socket;
1639 using_ssl = pconn.ssl;
1640 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1641 quotearg_style (escape_quoting_style, pconn.host),
1643 DEBUGP (("Reusing fd %d.\n", sock));
1644 if (pconn.authorized)
1645 /* If the connection is already authorized, the "Basic"
1646 authorization added by code above is unnecessary and
1648 request_remove_header (req, "Authorization");
1650 else if (host_lookup_failed)
1653 logprintf(LOG_NOTQUIET,
1654 _("%s: unable to resolve host address %s\n"),
1655 exec_name, quote (relevant->host));
1662 sock = connect_to_host (conn->host, conn->port);
1671 return (retryable_socket_connect_error (errno)
1672 ? CONERROR : CONIMPOSSIBLE);
1676 if (proxy && u->scheme == SCHEME_HTTPS)
1678 /* When requesting SSL URLs through proxies, use the
1679 CONNECT method to request passthrough. */
1680 struct request *connreq = request_new ();
1681 request_set_method (connreq, "CONNECT",
1682 aprintf ("%s:%d", u->host, u->port));
1683 SET_USER_AGENT (connreq);
1686 request_set_header (connreq, "Proxy-Authorization",
1687 proxyauth, rel_value);
1688 /* Now that PROXYAUTH is part of the CONNECT request,
1689 zero it out so we don't send proxy authorization with
1690 the regular request below. */
1693 /* Examples in rfc2817 use the Host header in CONNECT
1694 requests. I don't see how that gains anything, given
1695 that the contents of Host would be exactly the same as
1696 the contents of CONNECT. */
1698 write_error = request_send (connreq, sock);
1699 request_free (connreq);
1700 if (write_error < 0)
1702 CLOSE_INVALIDATE (sock);
1706 head = read_http_response_head (sock);
1709 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1711 CLOSE_INVALIDATE (sock);
1720 DEBUGP (("proxy responded with: [%s]\n", head));
1722 resp = resp_new (head);
1723 statcode = resp_status (resp, &message);
1724 hs->message = xstrdup (message);
1727 if (statcode != 200)
1730 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1731 message ? quotearg_style (escape_quoting_style, message) : "?");
1732 xfree_null (message);
1735 xfree_null (message);
1737 /* SOCK is now *really* connected to u->host, so update CONN
1738 to reflect this. That way register_persistent will
1739 register SOCK as being connected to u->host:u->port. */
1743 if (conn->scheme == SCHEME_HTTPS)
1745 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1752 #endif /* HAVE_SSL */
1755 /* Send the request to server. */
1756 write_error = request_send (req, sock);
1758 if (write_error >= 0)
1762 DEBUGP (("[POST data: %s]\n", opt.post_data));
1763 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1765 else if (opt.post_file_name && post_data_size != 0)
1766 write_error = post_file (sock, opt.post_file_name, post_data_size);
1769 if (write_error < 0)
1771 CLOSE_INVALIDATE (sock);
1775 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1776 proxy ? "Proxy" : "HTTP");
1781 head = read_http_response_head (sock);
1786 logputs (LOG_NOTQUIET, _("No data received.\n"));
1787 CLOSE_INVALIDATE (sock);
1793 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1795 CLOSE_INVALIDATE (sock);
1800 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1802 resp = resp_new (head);
1804 /* Check for status line. */
1806 statcode = resp_status (resp, &message);
1807 hs->message = xstrdup (message);
1808 if (!opt.server_response)
1809 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1810 message ? quotearg_style (escape_quoting_style, message) : "");
1813 logprintf (LOG_VERBOSE, "\n");
1814 print_server_response (resp, " ");
1817 /* Determine the local filename if needed. Notice that if -O is used
1818 * hstat.local_file is set by http_loop to the argument of -O. */
1819 if (!hs->local_file)
1821 /* Honor Content-Disposition whether possible. */
1822 if (!opt.content_disposition
1823 || !resp_header_copy (resp, "Content-Disposition",
1824 hdrval, sizeof (hdrval))
1825 || !parse_content_disposition (hdrval, &hs->local_file))
1827 /* The Content-Disposition header is missing or broken.
1828 * Choose unique file name according to given URL. */
1829 hs->local_file = url_file_name (u);
1833 /* TODO: perform this check only once. */
1834 if (!hs->existence_checked && file_exists_p (hs->local_file))
1836 if (opt.noclobber && !opt.output_document)
1838 /* If opt.noclobber is turned on and file already exists, do not
1839 retrieve the file. But if the output_document was given, then this
1840 test was already done and the file didn't exist. Hence the !opt.output_document */
1841 logprintf (LOG_VERBOSE, _("\
1842 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1843 /* If the file is there, we suppose it's retrieved OK. */
1846 /* #### Bogusness alert. */
1847 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1848 if (has_html_suffix_p (hs->local_file))
1851 return RETRUNNEEDED;
1853 else if (!ALLOW_CLOBBER)
1855 char *unique = unique_name (hs->local_file, true);
1856 if (unique != hs->local_file)
1857 xfree (hs->local_file);
1858 hs->local_file = unique;
1861 hs->existence_checked = true;
1863 /* Support timestamping */
1864 /* TODO: move this code out of gethttp. */
1865 if (opt.timestamping && !hs->timestamp_checked)
1867 size_t filename_len = strlen (hs->local_file);
1868 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1869 bool local_dot_orig_file_exists = false;
1870 char *local_filename = NULL;
1873 if (opt.backup_converted)
1874 /* If -K is specified, we'll act on the assumption that it was specified
1875 last time these files were downloaded as well, and instead of just
1876 comparing local file X against server file X, we'll compare local
1877 file X.orig (if extant, else X) against server file X. If -K
1878 _wasn't_ specified last time, or the server contains files called
1879 *.orig, -N will be back to not operating correctly with -k. */
1881 /* Would a single s[n]printf() call be faster? --dan
1883 Definitely not. sprintf() is horribly slow. It's a
1884 different question whether the difference between the two
1885 affects a program. Usually I'd say "no", but at one
1886 point I profiled Wget, and found that a measurable and
1887 non-negligible amount of time was lost calling sprintf()
1888 in url.c. Replacing sprintf with inline calls to
1889 strcpy() and number_to_string() made a difference.
1891 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1892 memcpy (filename_plus_orig_suffix + filename_len,
1893 ".orig", sizeof (".orig"));
1895 /* Try to stat() the .orig file. */
1896 if (stat (filename_plus_orig_suffix, &st) == 0)
1898 local_dot_orig_file_exists = true;
1899 local_filename = filename_plus_orig_suffix;
1903 if (!local_dot_orig_file_exists)
1904 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1905 if (stat (hs->local_file, &st) == 0)
1906 local_filename = hs->local_file;
1908 if (local_filename != NULL)
1909 /* There was a local file, so we'll check later to see if the version
1910 the server has is the same version we already have, allowing us to
1913 hs->orig_file_name = xstrdup (local_filename);
1914 hs->orig_file_size = st.st_size;
1915 hs->orig_file_tstamp = st.st_mtime;
1917 /* Modification time granularity is 2 seconds for Windows, so
1918 increase local time by 1 second for later comparison. */
1919 ++hs->orig_file_tstamp;
1924 if (!opt.ignore_length
1925 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1929 parsed = str_to_wgint (hdrval, NULL, 10);
1930 if (parsed == WGINT_MAX && errno == ERANGE)
1933 #### If Content-Length is out of range, it most likely
1934 means that the file is larger than 2G and that we're
1935 compiled without LFS. In that case we should probably
1936 refuse to even attempt to download the file. */
1939 else if (parsed < 0)
1941 /* Negative Content-Length; nonsensical, so we can't
1942 assume any information about the content to receive. */
1949 /* Check for keep-alive related responses. */
1950 if (!inhibit_keep_alive && contlen != -1)
1952 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1954 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1956 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1961 /* The server has promised that it will not close the connection
1962 when we're done. This means that we can register it. */
1963 register_persistent (conn->host, conn->port, sock, using_ssl);
1965 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1967 /* Authorization is required. */
1968 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1969 CLOSE_FINISH (sock);
1971 CLOSE_INVALIDATE (sock);
1972 pconn.authorized = false;
1973 if (!auth_finished && (user && passwd))
1975 /* IIS sends multiple copies of WWW-Authenticate, one with
1976 the value "negotiate", and other(s) with data. Loop over
1977 all the occurrences and pick the one we recognize. */
1979 const char *wabeg, *waend;
1980 char *www_authenticate = NULL;
1982 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1983 &wabeg, &waend)) != -1;
1985 if (known_authentication_scheme_p (wabeg, waend))
1987 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1991 if (!www_authenticate)
1993 /* If the authentication header is missing or
1994 unrecognized, there's no sense in retrying. */
1995 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1997 else if (!basic_auth_finished
1998 || !BEGINS_WITH (www_authenticate, "Basic"))
2001 pth = url_full_path (u);
2002 request_set_header (req, "Authorization",
2003 create_authorization_line (www_authenticate,
2005 request_method (req),
2009 if (BEGINS_WITH (www_authenticate, "NTLM"))
2011 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2013 /* Need to register this host as using basic auth,
2014 * so we automatically send creds next time. */
2015 register_basic_auth_host (u->host);
2018 goto retry_with_auth;
2022 /* We already did Basic auth, and it failed. Gotta
2026 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2030 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2032 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2034 pconn.authorized = true;
2038 hs->statcode = statcode;
2040 hs->error = xstrdup (_("Malformed status line"));
2042 hs->error = xstrdup (_("(no description)"));
2044 hs->error = xstrdup (message);
2045 xfree_null (message);
2047 type = resp_header_strdup (resp, "Content-Type");
2050 char *tmp = strchr (type, ';');
2053 /* sXXXav: only needed if IRI support is enabled */
2054 char *tmp2 = tmp + 1;
2056 while (tmp > type && c_isspace (tmp[-1]))
2060 /* Try to get remote encoding if needed */
2061 if (opt.enable_iri && !opt.encoding_remote)
2063 tmp = parse_charset (tmp2);
2065 set_content_encoding (iri, tmp);
2069 hs->newloc = resp_header_strdup (resp, "Location");
2070 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2072 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2076 const char *scbeg, *scend;
2077 /* The jar should have been created by now. */
2078 assert (wget_cookie_jar != NULL);
2080 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2081 &scbeg, &scend)) != -1;
2084 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2085 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2086 u->path, set_cookie);
2090 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2092 wgint first_byte_pos, last_byte_pos, entity_length;
2093 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2096 contrange = first_byte_pos;
2097 contlen = last_byte_pos - first_byte_pos + 1;
2102 /* 20x responses are counted among successful by default. */
2103 if (H_20X (statcode))
2106 /* Return if redirected. */
2107 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2109 /* RFC2068 says that in case of the 300 (multiple choices)
2110 response, the server can output a preferred URL through
2111 `Location' header; otherwise, the request should be treated
2112 like GET. So, if the location is set, it will be a
2113 redirection; otherwise, just proceed normally. */
2114 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2118 logprintf (LOG_VERBOSE,
2119 _("Location: %s%s\n"),
2120 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2121 hs->newloc ? _(" [following]") : "");
2122 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2123 CLOSE_FINISH (sock);
2125 CLOSE_INVALIDATE (sock);
2131 /* If content-type is not given, assume text/html. This is because
2132 of the multitude of broken CGI's that "forget" to generate the
2135 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2136 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2142 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2147 if (opt.html_extension)
2150 /* -E / --html-extension / html_extension = on was specified,
2151 and this is a text/html file. If some case-insensitive
2152 variation on ".htm[l]" isn't already the file's suffix,
2155 ensure_extension (hs, ".html", dt);
2157 else if (*dt & TEXTCSS)
2159 ensure_extension (hs, ".css", dt);
2163 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2164 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2165 && contrange == 0 && hs->restval >= contlen)
2168 /* If `-c' is in use and the file has been fully downloaded (or
2169 the remote file has shrunk), Wget effectively requests bytes
2170 after the end of file and the server response with 416
2171 (or 200 with a <= Content-Length. */
2172 logputs (LOG_VERBOSE, _("\
2173 \n The file is already fully retrieved; nothing to do.\n\n"));
2174 /* In case the caller inspects. */
2177 /* Mark as successfully retrieved. */
2180 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2181 might be more bytes in the body. */
2182 return RETRUNNEEDED;
2184 if ((contrange != 0 && contrange != hs->restval)
2185 || (H_PARTIAL (statcode) && !contrange))
2187 /* The Range request was somehow misunderstood by the server.
2190 CLOSE_INVALIDATE (sock);
2196 hs->contlen = contlen + contrange;
2202 /* No need to print this output if the body won't be
2203 downloaded at all, or if the original server response is
2205 logputs (LOG_VERBOSE, _("Length: "));
2208 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2209 if (contlen + contrange >= 1024)
2210 logprintf (LOG_VERBOSE, " (%s)",
2211 human_readable (contlen + contrange));
2214 if (contlen >= 1024)
2215 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2216 number_to_static_string (contlen),
2217 human_readable (contlen));
2219 logprintf (LOG_VERBOSE, _(", %s remaining"),
2220 number_to_static_string (contlen));
2224 logputs (LOG_VERBOSE,
2225 opt.ignore_length ? _("ignored") : _("unspecified"));
2227 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2229 logputs (LOG_VERBOSE, "\n");
2233 type = NULL; /* We don't need it any more. */
2235 /* Return if we have no intention of further downloading. */
2236 if (!(*dt & RETROKF) || head_only)
2238 /* In case the caller cares to look... */
2243 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2244 servers not to send body in response to a HEAD request, and
2245 those that do will likely be caught by test_socket_open.
2246 If not, they can be worked around using
2247 `--no-http-keep-alive'. */
2248 CLOSE_FINISH (sock);
2249 else if (keep_alive && skip_short_body (sock, contlen))
2250 /* Successfully skipped the body; also keep using the socket. */
2251 CLOSE_FINISH (sock);
2253 CLOSE_INVALIDATE (sock);
2254 return RETRFINISHED;
2257 /* Open the local file. */
2260 mkalldirs (hs->local_file);
2262 rotate_backups (hs->local_file);
2264 fp = fopen (hs->local_file, "ab");
2265 else if (ALLOW_CLOBBER)
2266 fp = fopen (hs->local_file, "wb");
2269 fp = fopen_excl (hs->local_file, true);
2270 if (!fp && errno == EEXIST)
2272 /* We cannot just invent a new name and use it (which is
2273 what functions like unique_create typically do)
2274 because we told the user we'd use this name.
2275 Instead, return and retry the download. */
2276 logprintf (LOG_NOTQUIET,
2277 _("%s has sprung into existence.\n"),
2279 CLOSE_INVALIDATE (sock);
2280 return FOPEN_EXCL_ERR;
2285 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2286 CLOSE_INVALIDATE (sock);
2293 /* Print fetch message, if opt.verbose. */
2296 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2297 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2300 /* This confuses the timestamping code that checks for file size.
2301 #### The timestamping code should be smarter about file size. */
2302 if (opt.save_headers && hs->restval == 0)
2303 fwrite (head, 1, strlen (head), fp);
2305 /* Now we no longer need to store the response header. */
2308 /* Download the request body. */
2311 /* If content-length is present, read that much; otherwise, read
2312 until EOF. The HTTP spec doesn't require the server to
2313 actually close the connection when it's done sending data. */
2314 flags |= rb_read_exactly;
2315 if (hs->restval > 0 && contrange == 0)
2316 /* If the server ignored our range request, instruct fd_read_body
2317 to skip the first RESTVAL bytes of body. */
2318 flags |= rb_skip_startpos;
2319 hs->len = hs->restval;
2321 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2322 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2326 CLOSE_FINISH (sock);
2330 hs->rderrmsg = xstrdup (fd_errstr (sock));
2331 CLOSE_INVALIDATE (sock);
2338 return RETRFINISHED;
2341 /* The genuine HTTP loop! This is the part where the retrieval is
2342 retried, and retried, and retried, and... */
2344 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2345 int *dt, struct url *proxy, struct iri *iri)
2348 bool got_head = false; /* used for time-stamping and filename detection */
2349 bool time_came_from_head = false;
2350 bool got_name = false;
2353 uerr_t err, ret = TRYLIMEXC;
2354 time_t tmr = -1; /* remote time-stamp */
2355 struct http_stat hstat; /* HTTP status */
2357 bool send_head_first = true;
2359 /* Assert that no value for *LOCAL_FILE was passed. */
2360 assert (local_file == NULL || *local_file == NULL);
2362 /* Set LOCAL_FILE parameter. */
2363 if (local_file && opt.output_document)
2364 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2366 /* Reset NEWLOC parameter. */
2369 /* This used to be done in main(), but it's a better idea to do it
2370 here so that we don't go through the hoops if we're just using
2375 /* Warn on (likely bogus) wildcard usage in HTTP. */
2376 if (opt.ftp_glob && has_wildcards_p (u->path))
2377 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2379 /* Setup hstat struct. */
2381 hstat.referer = referer;
2383 if (opt.output_document)
2385 hstat.local_file = xstrdup (opt.output_document);
2388 else if (!opt.content_disposition)
2390 hstat.local_file = url_file_name (u);
2394 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2395 * screaming for some refactoring. */
2396 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2398 /* If opt.noclobber is turned on and file already exists, do not
2399 retrieve the file. But if the output_document was given, then this
2400 test was already done and the file didn't exist. Hence the !opt.output_document */
2401 logprintf (LOG_VERBOSE, _("\
2402 File %s already there; not retrieving.\n\n"),
2403 quote (hstat.local_file));
2404 /* If the file is there, we suppose it's retrieved OK. */
2407 /* #### Bogusness alert. */
2408 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2409 if (has_html_suffix_p (hstat.local_file))
2416 /* Reset the counter. */
2419 /* Reset the document type. */
2422 /* Skip preliminary HEAD request if we're not in spider mode AND
2423 * if -O was given or HTTP Content-Disposition support is disabled. */
2425 && (got_name || !opt.content_disposition))
2426 send_head_first = false;
2428 /* Send preliminary HEAD request if -N is given and we have an existing
2429 * destination file. */
2430 if (opt.timestamping
2431 && !opt.content_disposition
2432 && file_exists_p (url_file_name (u)))
2433 send_head_first = true;
2438 /* Increment the pass counter. */
2440 sleep_between_retrievals (count);
2442 /* Get the current time string. */
2443 tms = datetime_str (time (NULL));
2445 if (opt.spider && !got_head)
2446 logprintf (LOG_VERBOSE, _("\
2447 Spider mode enabled. Check if remote file exists.\n"));
2449 /* Print fetch message, if opt.verbose. */
2452 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2457 sprintf (tmp, _("(try:%2d)"), count);
2458 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2463 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2468 ws_changetitle (hurl);
2473 /* Default document type is empty. However, if spider mode is
2474 on or time-stamping is employed, HEAD_ONLY commands is
2475 encoded within *dt. */
2476 if (send_head_first && !got_head)
2481 /* Decide whether or not to restart. */
2484 && stat (hstat.local_file, &st) == 0
2485 && S_ISREG (st.st_mode))
2486 /* When -c is used, continue from on-disk size. (Can't use
2487 hstat.len even if count>1 because we don't want a failed
2488 first attempt to clobber existing data.) */
2489 hstat.restval = st.st_size;
2491 /* otherwise, continue where the previous try left off */
2492 hstat.restval = hstat.len;
2496 /* Decide whether to send the no-cache directive. We send it in
2498 a) we're using a proxy, and we're past our first retrieval.
2499 Some proxies are notorious for caching incomplete data, so
2500 we require a fresh get.
2501 b) caching is explicitly inhibited. */
2502 if ((proxy && count > 1) /* a */
2503 || !opt.allow_cache) /* b */
2504 *dt |= SEND_NOCACHE;
2506 *dt &= ~SEND_NOCACHE;
2508 /* Try fetching the document, or at least its head. */
2509 err = gethttp (u, &hstat, dt, proxy, iri);
2512 tms = datetime_str (time (NULL));
2514 /* Get the new location (with or without the redirection). */
2516 *newloc = xstrdup (hstat.newloc);
2520 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2521 case CONERROR: case READERR: case WRITEFAILED:
2522 case RANGEERR: case FOPEN_EXCL_ERR:
2523 /* Non-fatal errors continue executing the loop, which will
2524 bring them to "while" statement at the end, to judge
2525 whether the number of tries was exceeded. */
2526 printwhat (count, opt.ntry);
2528 case FWRITEERR: case FOPENERR:
2529 /* Another fatal error. */
2530 logputs (LOG_VERBOSE, "\n");
2531 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2532 quote (hstat.local_file), strerror (errno));
2533 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2534 case SSLINITFAILED: case CONTNOTSUPPORTED:
2535 /* Fatal errors just return from the function. */
2539 /* Another fatal error. */
2540 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2544 /* Return the new location to the caller. */
2547 logprintf (LOG_NOTQUIET,
2548 _("ERROR: Redirection (%d) without location.\n"),
2558 /* The file was already fully retrieved. */
2562 /* Deal with you later. */
2565 /* All possibilities should have been exhausted. */
2569 if (!(*dt & RETROKF))
2574 /* #### Ugly ugly ugly! */
2575 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2576 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2579 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2581 && (hstat.statcode == 500 || hstat.statcode == 501))
2586 /* Maybe we should always keep track of broken links, not just in
2588 * Don't log error if it was UTF-8 encoded because we will try
2589 * once unencoded. */
2590 else if (opt.spider && !iri->utf8_encode)
2592 /* #### Again: ugly ugly ugly! */
2594 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2595 nonexisting_url (hurl);
2596 logprintf (LOG_NOTQUIET, _("\
2597 Remote file does not exist -- broken link!!!\n"));
2601 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2602 tms, hstat.statcode,
2603 quotearg_style (escape_quoting_style, hstat.error));
2605 logputs (LOG_VERBOSE, "\n");
2611 /* Did we get the time-stamp? */
2614 got_head = true; /* no more time-stamping */
2616 if (opt.timestamping && !hstat.remote_time)
2618 logputs (LOG_NOTQUIET, _("\
2619 Last-modified header missing -- time-stamps turned off.\n"));
2621 else if (hstat.remote_time)
2623 /* Convert the date-string into struct tm. */
2624 tmr = http_atotm (hstat.remote_time);
2625 if (tmr == (time_t) (-1))
2626 logputs (LOG_VERBOSE, _("\
2627 Last-modified header invalid -- time-stamp ignored.\n"));
2628 if (*dt & HEAD_ONLY)
2629 time_came_from_head = true;
2632 if (send_head_first)
2634 /* The time-stamping section. */
2635 if (opt.timestamping)
2637 if (hstat.orig_file_name) /* Perform the following
2638 checks only if the file
2640 download already exists. */
2642 if (hstat.remote_time &&
2643 tmr != (time_t) (-1))
2645 /* Now time-stamping can be used validly.
2646 Time-stamping means that if the sizes of
2647 the local and remote file match, and local
2648 file is newer than the remote file, it will
2649 not be retrieved. Otherwise, the normal
2650 download procedure is resumed. */
2651 if (hstat.orig_file_tstamp >= tmr)
2653 if (hstat.contlen == -1
2654 || hstat.orig_file_size == hstat.contlen)
2656 logprintf (LOG_VERBOSE, _("\
2657 Server file no newer than local file %s -- not retrieving.\n\n"),
2658 quote (hstat.orig_file_name));
2664 logprintf (LOG_VERBOSE, _("\
2665 The sizes do not match (local %s) -- retrieving.\n"),
2666 number_to_static_string (hstat.orig_file_size));
2670 logputs (LOG_VERBOSE,
2671 _("Remote file is newer, retrieving.\n"));
2673 logputs (LOG_VERBOSE, "\n");
2677 /* free_hstat (&hstat); */
2678 hstat.timestamp_checked = true;
2683 bool finished = true;
2688 logputs (LOG_VERBOSE, _("\
2689 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2694 logprintf (LOG_VERBOSE, _("\
2695 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2696 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2703 logprintf (LOG_VERBOSE, _("\
2704 Remote file exists and could contain further links,\n\
2705 but recursion is disabled -- not retrieving.\n\n"));
2709 logprintf (LOG_VERBOSE, _("\
2710 Remote file exists.\n\n"));
2712 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2717 logprintf (LOG_NONVERBOSE,
2718 _("%s URL:%s %2d %s\n"),
2719 tms, u->url, hstat.statcode,
2720 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2727 count = 0; /* the retrieve count for HEAD is reset */
2729 } /* send_head_first */
2732 if ((tmr != (time_t) (-1))
2733 && ((hstat.len == hstat.contlen) ||
2734 ((hstat.res == 0) && (hstat.contlen == -1))))
2736 /* #### This code repeats in http.c and ftp.c. Move it to a
2738 const char *fl = NULL;
2739 if (opt.output_document)
2741 if (output_stream_regular)
2742 fl = opt.output_document;
2745 fl = hstat.local_file;
2749 /* Reparse time header, in case it's changed. */
2750 if (time_came_from_head
2751 && hstat.remote_time && hstat.remote_time[0])
2753 newtmr = http_atotm (hstat.remote_time);
2760 /* End of time-stamping section. */
2762 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2763 total_download_time += hstat.dltime;
2765 if (hstat.len == hstat.contlen)
2769 logprintf (LOG_VERBOSE,
2770 _("%s (%s) - %s saved [%s/%s]\n\n"),
2771 tms, tmrate, quote (hstat.local_file),
2772 number_to_static_string (hstat.len),
2773 number_to_static_string (hstat.contlen));
2774 logprintf (LOG_NONVERBOSE,
2775 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2777 number_to_static_string (hstat.len),
2778 number_to_static_string (hstat.contlen),
2779 hstat.local_file, count);
2782 total_downloaded_bytes += hstat.len;
2784 /* Remember that we downloaded the file for later ".orig" code. */
2785 if (*dt & ADDED_HTML_EXTENSION)
2786 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2788 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2793 else if (hstat.res == 0) /* No read error */
2795 if (hstat.contlen == -1) /* We don't know how much we were supposed
2796 to get, so assume we succeeded. */
2800 logprintf (LOG_VERBOSE,
2801 _("%s (%s) - %s saved [%s]\n\n"),
2802 tms, tmrate, quote (hstat.local_file),
2803 number_to_static_string (hstat.len));
2804 logprintf (LOG_NONVERBOSE,
2805 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2806 tms, u->url, number_to_static_string (hstat.len),
2807 hstat.local_file, count);
2810 total_downloaded_bytes += hstat.len;
2812 /* Remember that we downloaded the file for later ".orig" code. */
2813 if (*dt & ADDED_HTML_EXTENSION)
2814 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2816 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2821 else if (hstat.len < hstat.contlen) /* meaning we lost the
2822 connection too soon */
2824 logprintf (LOG_VERBOSE,
2825 _("%s (%s) - Connection closed at byte %s. "),
2826 tms, tmrate, number_to_static_string (hstat.len));
2827 printwhat (count, opt.ntry);
2830 else if (hstat.len != hstat.restval)
2831 /* Getting here would mean reading more data than
2832 requested with content-length, which we never do. */
2836 /* Getting here probably means that the content-length was
2837 * _less_ than the original, local size. We should probably
2838 * truncate or re-read, or something. FIXME */
2843 else /* from now on hstat.res can only be -1 */
2845 if (hstat.contlen == -1)
2847 logprintf (LOG_VERBOSE,
2848 _("%s (%s) - Read error at byte %s (%s)."),
2849 tms, tmrate, number_to_static_string (hstat.len),
2851 printwhat (count, opt.ntry);
2854 else /* hstat.res == -1 and contlen is given */
2856 logprintf (LOG_VERBOSE,
2857 _("%s (%s) - Read error at byte %s/%s (%s). "),
2859 number_to_static_string (hstat.len),
2860 number_to_static_string (hstat.contlen),
2862 printwhat (count, opt.ntry);
2868 while (!opt.ntry || (count < opt.ntry));
2872 *local_file = xstrdup (hstat.local_file);
2873 free_hstat (&hstat);
2878 /* Check whether the result of strptime() indicates success.
2879 strptime() returns the pointer to how far it got to in the string.
2880 The processing has been successful if the string is at `GMT' or
2881 `+X', or at the end of the string.
2883 In extended regexp parlance, the function returns 1 if P matches
2884 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2885 can return) is considered a failure and 0 is returned. */
2887 check_end (const char *p)
2891 while (c_isspace (*p))
2894 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2895 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2901 /* Convert the textual specification of time in TIME_STRING to the
2902 number of seconds since the Epoch.
2904 TIME_STRING can be in any of the three formats RFC2616 allows the
2905 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2906 as well as the time format used in the Set-Cookie header.
2907 Timezones are ignored, and should be GMT.
2909 Return the computed time_t representation, or -1 if the conversion
2912 This function uses strptime with various string formats for parsing
2913 TIME_STRING. This results in a parser that is not as lenient in
2914 interpreting TIME_STRING as I would like it to be. Being based on
2915 strptime, it always allows shortened months, one-digit days, etc.,
2916 but due to the multitude of formats in which time can be
2917 represented, an ideal HTTP time parser would be even more
2918 forgiving. It should completely ignore things like week days and
2919 concentrate only on the various forms of representing years,
2920 months, days, hours, minutes, and seconds. For example, it would
2921 be nice if it accepted ISO 8601 out of the box.
2923 I've investigated free and PD code for this purpose, but none was
2924 usable. getdate was big and unwieldy, and had potential copyright
2925 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2926 distributed with phttpd, is excellent, but we cannot use it because
2927 it is not assigned to the FSF. So I stuck it with strptime. */
2930 http_atotm (const char *time_string)
2932 /* NOTE: Solaris strptime man page claims that %n and %t match white
2933 space, but that's not universally available. Instead, we simply
2934 use ` ' to mean "skip all WS", which works under all strptime
2935 implementations I've tested. */
2937 static const char *time_formats[] = {
2938 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2939 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2940 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2941 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2942 (used in Set-Cookie, defined in the
2943 Netscape cookie specification.) */
2945 const char *oldlocale;
2947 time_t ret = (time_t) -1;
2949 /* Solaris strptime fails to recognize English month names in
2950 non-English locales, which we work around by temporarily setting
2951 locale to C before invoking strptime. */
2952 oldlocale = setlocale (LC_TIME, NULL);
2953 setlocale (LC_TIME, "C");
2955 for (i = 0; i < countof (time_formats); i++)
2959 /* Some versions of strptime use the existing contents of struct
2960 tm to recalculate the date according to format. Zero it out
2961 to prevent stack garbage from influencing strptime. */
2964 if (check_end (strptime (time_string, time_formats[i], &t)))
2971 /* Restore the previous locale. */
2972 setlocale (LC_TIME, oldlocale);
2977 /* Authorization support: We support three authorization schemes:
2979 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2981 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2982 consisting of answering to the server's challenge with the proper
2985 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2986 Stenberg for libcurl. Like digest, NTLM is based on a
2987 challenge-response mechanism, but unlike digest, it is non-standard
2988 (authenticates TCP connections rather than requests), undocumented
2989 and Microsoft-specific. */
2991 /* Create the authentication header contents for the `Basic' scheme.
2992 This is done by encoding the string "USER:PASS" to base64 and
2993 prepending the string "Basic " in front of it. */
2996 basic_authentication_encode (const char *user, const char *passwd)
2999 int len1 = strlen (user) + 1 + strlen (passwd);
3001 t1 = (char *)alloca (len1 + 1);
3002 sprintf (t1, "%s:%s", user, passwd);
3004 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3005 base64_encode (t1, len1, t2);
3007 return concat_strings ("Basic ", t2, (char *) 0);
3010 #define SKIP_WS(x) do { \
3011 while (c_isspace (*(x))) \
3015 #ifdef ENABLE_DIGEST
3016 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3017 an array of 16 bytes containing the hash keys, and BUF should be a
3018 buffer of 33 writable characters (32 for hex digits plus one for
3019 zero termination). */
3021 dump_hash (char *buf, const unsigned char *hash)
3025 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3027 *buf++ = XNUM_TO_digit (*hash >> 4);
3028 *buf++ = XNUM_TO_digit (*hash & 0xf);
3033 /* Take the line apart to find the challenge, and compose a digest
3034 authorization header. See RFC2069 section 2.1.2. */
3036 digest_authentication_encode (const char *au, const char *user,
3037 const char *passwd, const char *method,
3040 static char *realm, *opaque, *nonce;
3045 { "realm", &realm },
3046 { "opaque", &opaque },
3050 param_token name, value;
3052 realm = opaque = nonce = NULL;
3054 au += 6; /* skip over `Digest' */
3055 while (extract_param (&au, &name, &value, ','))
3058 size_t namelen = name.e - name.b;
3059 for (i = 0; i < countof (options); i++)
3060 if (namelen == strlen (options[i].name)
3061 && 0 == strncmp (name.b, options[i].name,
3064 *options[i].variable = strdupdelim (value.b, value.e);
3068 if (!realm || !nonce || !user || !passwd || !path || !method)
3071 xfree_null (opaque);
3076 /* Calculate the digest value. */
3078 ALLOCA_MD5_CONTEXT (ctx);
3079 unsigned char hash[MD5_HASHLEN];
3080 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3081 char response_digest[MD5_HASHLEN * 2 + 1];
3083 /* A1BUF = H(user ":" realm ":" password) */
3085 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3086 gen_md5_update ((unsigned char *)":", 1, ctx);
3087 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3088 gen_md5_update ((unsigned char *)":", 1, ctx);
3089 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3090 gen_md5_finish (ctx, hash);
3091 dump_hash (a1buf, hash);
3093 /* A2BUF = H(method ":" path) */
3095 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3096 gen_md5_update ((unsigned char *)":", 1, ctx);
3097 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3098 gen_md5_finish (ctx, hash);
3099 dump_hash (a2buf, hash);
3101 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3103 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3104 gen_md5_update ((unsigned char *)":", 1, ctx);
3105 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3106 gen_md5_update ((unsigned char *)":", 1, ctx);
3107 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3108 gen_md5_finish (ctx, hash);
3109 dump_hash (response_digest, hash);
3111 res = xmalloc (strlen (user)
3116 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3117 + (opaque ? strlen (opaque) : 0)
3119 sprintf (res, "Digest \
3120 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3121 user, realm, nonce, path, response_digest);
3124 char *p = res + strlen (res);
3125 strcat (p, ", opaque=\"");
3132 #endif /* ENABLE_DIGEST */
3134 /* Computing the size of a string literal must take into account that
3135 value returned by sizeof includes the terminating \0. */
3136 #define STRSIZE(literal) (sizeof (literal) - 1)
3138 /* Whether chars in [b, e) begin with the literal string provided as
3139 first argument and are followed by whitespace or terminating \0.
3140 The comparison is case-insensitive. */
3141 #define STARTS(literal, b, e) \
3143 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3144 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3145 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3146 || c_isspace (b[STRSIZE (literal)])))
3149 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3151 return STARTS ("Basic", hdrbeg, hdrend)
3152 #ifdef ENABLE_DIGEST
3153 || STARTS ("Digest", hdrbeg, hdrend)
3156 || STARTS ("NTLM", hdrbeg, hdrend)
3163 /* Create the HTTP authorization request header. When the
3164 `WWW-Authenticate' response header is seen, according to the
3165 authorization scheme specified in that header (`Basic' and `Digest'
3166 are supported by the current implementation), produce an
3167 appropriate HTTP authorization request header. */
3169 create_authorization_line (const char *au, const char *user,
3170 const char *passwd, const char *method,
3171 const char *path, bool *finished)
3173 /* We are called only with known schemes, so we can dispatch on the
3175 switch (c_toupper (*au))
3177 case 'B': /* Basic */
3179 return basic_authentication_encode (user, passwd);
3180 #ifdef ENABLE_DIGEST
3181 case 'D': /* Digest */
3183 return digest_authentication_encode (au, user, passwd, method, path);
3186 case 'N': /* NTLM */
3187 if (!ntlm_input (&pconn.ntlm, au))
3192 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3195 /* We shouldn't get here -- this function should be only called
3196 with values approved by known_authentication_scheme_p. */
3204 if (!wget_cookie_jar)
3205 wget_cookie_jar = cookie_jar_new ();
3206 if (opt.cookies_input && !cookies_loaded_p)
3208 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3209 cookies_loaded_p = true;
3216 if (wget_cookie_jar)
3217 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3223 xfree_null (pconn.host);
3224 if (wget_cookie_jar)
3225 cookie_jar_delete (wget_cookie_jar);
3229 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3231 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3233 int len = strlen (ext);
3236 strncpy (shortext, ext, len - 1);
3237 shortext[len - 2] = '\0';
3240 if (last_period_in_local_filename == NULL
3241 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3242 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3244 int local_filename_len = strlen (hs->local_file);
3245 /* Resize the local file, allowing for ".html" preceded by
3246 optional ".NUMBER". */
3247 hs->local_file = xrealloc (hs->local_file,
3248 local_filename_len + 24 + len);
3249 strcpy (hs->local_file + local_filename_len, ext);
3250 /* If clobbering is not allowed and the file, as named,
3251 exists, tack on ".NUMBER.html" instead. */
3252 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3256 sprintf (hs->local_file + local_filename_len,
3257 ".%d%s", ext_num++, ext);
3258 while (file_exists_p (hs->local_file));
3260 *dt |= ADDED_HTML_EXTENSION;
3268 test_parse_content_disposition()
3273 char *opt_dir_prefix;
3277 { "filename=\"file.ext\"", NULL, "file.ext", true },
3278 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3279 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3280 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3281 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3282 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3283 { "attachment", NULL, NULL, false },
3284 { "attachment", "somedir", NULL, false },
3287 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3292 opt.dir_prefix = test_array[i].opt_dir_prefix;
3293 res = parse_content_disposition (test_array[i].hdrval, &filename);
3295 mu_assert ("test_parse_content_disposition: wrong result",
3296 res == test_array[i].result
3298 || 0 == strcmp (test_array[i].filename, filename)));
3304 #endif /* TESTING */
3307 * vim: et sts=2 sw=2 cino+={s