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. */
57 # include "http-ntlm.h"
70 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 load_cookies (void);
81 # define MIN(x, y) ((x) > (y) ? (y) : (x))
85 static bool cookies_loaded_p;
86 static struct cookie_jar *wget_cookie_jar;
88 #define TEXTHTML_S "text/html"
89 #define TEXTXHTML_S "application/xhtml+xml"
91 /* Some status code validation macros: */
92 #define H_20X(x) (((x) >= 200) && ((x) < 300))
93 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
94 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
95 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
96 || (x) == HTTP_STATUS_SEE_OTHER \
97 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
99 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
100 /* Successful 2xx. */
101 #define HTTP_STATUS_OK 200
102 #define HTTP_STATUS_CREATED 201
103 #define HTTP_STATUS_ACCEPTED 202
104 #define HTTP_STATUS_NO_CONTENT 204
105 #define HTTP_STATUS_PARTIAL_CONTENTS 206
107 /* Redirection 3xx. */
108 #define HTTP_STATUS_MULTIPLE_CHOICES 300
109 #define HTTP_STATUS_MOVED_PERMANENTLY 301
110 #define HTTP_STATUS_MOVED_TEMPORARILY 302
111 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
112 #define HTTP_STATUS_NOT_MODIFIED 304
113 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
115 /* Client error 4xx. */
116 #define HTTP_STATUS_BAD_REQUEST 400
117 #define HTTP_STATUS_UNAUTHORIZED 401
118 #define HTTP_STATUS_FORBIDDEN 403
119 #define HTTP_STATUS_NOT_FOUND 404
120 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
122 /* Server errors 5xx. */
123 #define HTTP_STATUS_INTERNAL 500
124 #define HTTP_STATUS_NOT_IMPLEMENTED 501
125 #define HTTP_STATUS_BAD_GATEWAY 502
126 #define HTTP_STATUS_UNAVAILABLE 503
129 rel_none, rel_name, rel_value, rel_both
136 struct request_header {
138 enum rp release_policy;
140 int hcount, hcapacity;
143 /* Create a new, empty request. At least request_set_method must be
144 called before the request can be used. */
146 static struct request *
149 struct request *req = xnew0 (struct request);
151 req->headers = xnew_array (struct request_header, req->hcapacity);
155 /* Set the request's method and its arguments. METH should be a
156 literal string (or it should outlive the request) because it will
157 not be freed. ARG will be freed by request_free. */
160 request_set_method (struct request *req, const char *meth, char *arg)
166 /* Return the method string passed with the last call to
167 request_set_method. */
170 request_method (const struct request *req)
175 /* Free one header according to the release policy specified with
176 request_set_header. */
179 release_header (struct request_header *hdr)
181 switch (hdr->release_policy)
198 /* Set the request named NAME to VALUE. Specifically, this means that
199 a "NAME: VALUE\r\n" header line will be used in the request. If a
200 header with the same name previously existed in the request, its
201 value will be replaced by this one. A NULL value means do nothing.
203 RELEASE_POLICY determines whether NAME and VALUE should be released
204 (freed) with request_free. Allowed values are:
206 - rel_none - don't free NAME or VALUE
207 - rel_name - free NAME when done
208 - rel_value - free VALUE when done
209 - rel_both - free both NAME and VALUE when done
211 Setting release policy is useful when arguments come from different
212 sources. For example:
214 // Don't free literal strings!
215 request_set_header (req, "Pragma", "no-cache", rel_none);
217 // Don't free a global variable, we'll need it later.
218 request_set_header (req, "Referer", opt.referer, rel_none);
220 // Value freshly allocated, free it when done.
221 request_set_header (req, "Range",
222 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
227 request_set_header (struct request *req, char *name, char *value,
228 enum rp release_policy)
230 struct request_header *hdr;
235 /* A NULL value is a no-op; if freeing the name is requested,
236 free it now to avoid leaks. */
237 if (release_policy == rel_name || release_policy == rel_both)
242 for (i = 0; i < req->hcount; i++)
244 hdr = &req->headers[i];
245 if (0 == strcasecmp (name, hdr->name))
247 /* Replace existing header. */
248 release_header (hdr);
251 hdr->release_policy = release_policy;
256 /* Install new header. */
258 if (req->hcount >= req->hcapacity)
260 req->hcapacity <<= 1;
261 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
263 hdr = &req->headers[req->hcount++];
266 hdr->release_policy = release_policy;
269 /* Like request_set_header, but sets the whole header line, as
270 provided by the user using the `--header' option. For example,
271 request_set_user_header (req, "Foo: bar") works just like
272 request_set_header (req, "Foo", "bar"). */
275 request_set_user_header (struct request *req, const char *header)
278 const char *p = strchr (header, ':');
281 BOUNDED_TO_ALLOCA (header, p, name);
283 while (c_isspace (*p))
285 request_set_header (req, xstrdup (name), (char *) p, rel_name);
288 /* Remove the header with specified name from REQ. Returns true if
289 the header was actually removed, false otherwise. */
292 request_remove_header (struct request *req, char *name)
295 for (i = 0; i < req->hcount; i++)
297 struct request_header *hdr = &req->headers[i];
298 if (0 == strcasecmp (name, hdr->name))
300 release_header (hdr);
301 /* Move the remaining headers by one. */
302 if (i < req->hcount - 1)
303 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
311 #define APPEND(p, str) do { \
312 int A_len = strlen (str); \
313 memcpy (p, str, A_len); \
317 /* Construct the request and write it to FD using fd_write. */
320 request_send (const struct request *req, int fd)
322 char *request_string, *p;
323 int i, size, write_error;
325 /* Count the request size. */
328 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
329 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
331 for (i = 0; i < req->hcount; i++)
333 struct request_header *hdr = &req->headers[i];
334 /* NAME ": " VALUE "\r\n" */
335 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
341 p = request_string = alloca_array (char, size);
343 /* Generate the request. */
345 APPEND (p, req->method); *p++ = ' ';
346 APPEND (p, req->arg); *p++ = ' ';
347 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
349 for (i = 0; i < req->hcount; i++)
351 struct request_header *hdr = &req->headers[i];
352 APPEND (p, hdr->name);
353 *p++ = ':', *p++ = ' ';
354 APPEND (p, hdr->value);
355 *p++ = '\r', *p++ = '\n';
358 *p++ = '\r', *p++ = '\n', *p++ = '\0';
359 assert (p - request_string == size);
363 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
365 /* Send the request to the server. */
367 write_error = fd_write (fd, request_string, size - 1, -1);
369 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
374 /* Release the resources used by REQ. */
377 request_free (struct request *req)
380 xfree_null (req->arg);
381 for (i = 0; i < req->hcount; i++)
382 release_header (&req->headers[i]);
383 xfree_null (req->headers);
387 static struct hash_table *basic_authed_hosts;
389 /* Find out if this host has issued a Basic challenge yet; if so, give
390 * it the username, password. A temporary measure until we can get
391 * proper authentication in place. */
394 maybe_send_basic_creds (const char *hostname, const char *user,
395 const char *passwd, struct request *req)
397 bool do_challenge = false;
399 if (opt.auth_without_challenge)
401 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
404 else if (basic_authed_hosts
405 && hash_table_contains(basic_authed_hosts, hostname))
407 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
412 DEBUGP(("Host %s has not issued a general basic challenge.\n",
417 request_set_header (req, "Authorization",
418 basic_authentication_encode (user, passwd),
425 register_basic_auth_host (const char *hostname)
427 if (!basic_authed_hosts)
429 basic_authed_hosts = make_nocase_string_hash_table (1);
431 if (!hash_table_contains(basic_authed_hosts, hostname))
433 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
434 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
439 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
440 PROMISED_SIZE bytes are sent over the wire -- if the file is
441 longer, read only that much; if the file is shorter, report an error. */
444 post_file (int sock, const char *file_name, wgint promised_size)
446 static char chunk[8192];
451 DEBUGP (("[writing POST file %s ... ", file_name));
453 fp = fopen (file_name, "rb");
456 while (!feof (fp) && written < promised_size)
459 int length = fread (chunk, 1, sizeof (chunk), fp);
462 towrite = MIN (promised_size - written, length);
463 write_error = fd_write (sock, chunk, towrite, -1);
473 /* If we've written less than was promised, report a (probably
474 nonsensical) error rather than break the promise. */
475 if (written < promised_size)
481 assert (written == promised_size);
482 DEBUGP (("done]\n"));
486 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
487 If so, return the pointer to the position after the line, otherwise
488 return NULL. This is used as callback to fd_read_hunk. The data
489 between START and PEEKED has been read and cannot be "unread"; the
490 data after PEEKED has only been peeked. */
493 response_head_terminator (const char *start, const char *peeked, int peeklen)
497 /* If at first peek, verify whether HUNK starts with "HTTP". If
498 not, this is a HTTP/0.9 request and we must bail out without
500 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
503 /* Look for "\n[\r]\n", and return the following position if found.
504 Start two chars before the current to cover the possibility that
505 part of the terminator (e.g. "\n\r") arrived in the previous
507 p = peeked - start < 2 ? start : peeked - 2;
508 end = peeked + peeklen;
510 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
511 for (; p < end - 2; p++)
514 if (p[1] == '\r' && p[2] == '\n')
516 else if (p[1] == '\n')
519 /* p==end-2: check for \n\n directly preceding END. */
520 if (p[0] == '\n' && p[1] == '\n')
526 /* The maximum size of a single HTTP response we care to read. Rather
527 than being a limit of the reader implementation, this limit
528 prevents Wget from slurping all available memory upon encountering
529 malicious or buggy server output, thus protecting the user. Define
530 it to 0 to remove the limit. */
532 #define HTTP_RESPONSE_MAX_SIZE 65536
534 /* Read the HTTP request head from FD and return it. The error
535 conditions are the same as with fd_read_hunk.
537 To support HTTP/0.9 responses, this function tries to make sure
538 that the data begins with "HTTP". If this is not the case, no data
539 is read and an empty request is returned, so that the remaining
540 data can be treated as body. */
543 read_http_response_head (int fd)
545 return fd_read_hunk (fd, response_head_terminator, 512,
546 HTTP_RESPONSE_MAX_SIZE);
550 /* The response data. */
553 /* The array of pointers that indicate where each header starts.
554 For example, given this HTTP response:
561 The headers are located like this:
563 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
565 headers[0] headers[1] headers[2] headers[3]
567 I.e. headers[0] points to the beginning of the request,
568 headers[1] points to the end of the first header and the
569 beginning of the second one, etc. */
571 const char **headers;
574 /* Create a new response object from the text of the HTTP response,
575 available in HEAD. That text is automatically split into
576 constituent header lines for fast retrieval using
579 static struct response *
580 resp_new (const char *head)
585 struct response *resp = xnew0 (struct response);
590 /* Empty head means that we're dealing with a headerless
591 (HTTP/0.9) response. In that case, don't set HEADERS at
596 /* Split HEAD into header lines, so that resp_header_* functions
597 don't need to do this over and over again. */
603 DO_REALLOC (resp->headers, size, count + 1, const char *);
604 resp->headers[count++] = hdr;
606 /* Break upon encountering an empty line. */
607 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
610 /* Find the end of HDR, including continuations. */
613 const char *end = strchr (hdr, '\n');
619 while (*hdr == ' ' || *hdr == '\t');
621 DO_REALLOC (resp->headers, size, count + 1, const char *);
622 resp->headers[count] = NULL;
627 /* Locate the header named NAME in the request data, starting with
628 position START. This allows the code to loop through the request
629 data, filtering for all requests of a given name. Returns the
630 found position, or -1 for failure. The code that uses this
631 function typically looks like this:
633 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
634 ... do something with header ...
636 If you only care about one header, use resp_header_get instead of
640 resp_header_locate (const struct response *resp, const char *name, int start,
641 const char **begptr, const char **endptr)
644 const char **headers = resp->headers;
647 if (!headers || !headers[1])
650 name_len = strlen (name);
656 for (; headers[i + 1]; i++)
658 const char *b = headers[i];
659 const char *e = headers[i + 1];
661 && b[name_len] == ':'
662 && 0 == strncasecmp (b, name, name_len))
665 while (b < e && c_isspace (*b))
667 while (b < e && c_isspace (e[-1]))
677 /* Find and retrieve the header named NAME in the request data. If
678 found, set *BEGPTR to its starting, and *ENDPTR to its ending
679 position, and return true. Otherwise return false.
681 This function is used as a building block for resp_header_copy
682 and resp_header_strdup. */
685 resp_header_get (const struct response *resp, const char *name,
686 const char **begptr, const char **endptr)
688 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
692 /* Copy the response header named NAME to buffer BUF, no longer than
693 BUFSIZE (BUFSIZE includes the terminating 0). If the header
694 exists, true is returned, false otherwise. If there should be no
695 limit on the size of the header, use resp_header_strdup instead.
697 If BUFSIZE is 0, no data is copied, but the boolean indication of
698 whether the header is present is still returned. */
701 resp_header_copy (const struct response *resp, const char *name,
702 char *buf, int bufsize)
705 if (!resp_header_get (resp, name, &b, &e))
709 int len = MIN (e - b, bufsize - 1);
710 memcpy (buf, b, len);
716 /* Return the value of header named NAME in RESP, allocated with
717 malloc. If such a header does not exist in RESP, return NULL. */
720 resp_header_strdup (const struct response *resp, const char *name)
723 if (!resp_header_get (resp, name, &b, &e))
725 return strdupdelim (b, e);
728 /* Parse the HTTP status line, which is of format:
730 HTTP-Version SP Status-Code SP Reason-Phrase
732 The function returns the status-code, or -1 if the status line
733 appears malformed. The pointer to "reason-phrase" message is
734 returned in *MESSAGE. */
737 resp_status (const struct response *resp, char **message)
744 /* For a HTTP/0.9 response, assume status 200. */
746 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
750 p = resp->headers[0];
751 end = resp->headers[1];
757 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
761 /* Match the HTTP version. This is optional because Gnutella
762 servers have been reported to not specify HTTP version. */
763 if (p < end && *p == '/')
766 while (p < end && c_isdigit (*p))
768 if (p < end && *p == '.')
770 while (p < end && c_isdigit (*p))
774 while (p < end && c_isspace (*p))
776 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
779 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
784 while (p < end && c_isspace (*p))
786 while (p < end && c_isspace (end[-1]))
788 *message = strdupdelim (p, end);
794 /* Release the resources used by RESP. */
797 resp_free (struct response *resp)
799 xfree_null (resp->headers);
803 /* Print a single line of response, the characters [b, e). We tried
805 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
806 but that failed to escape the non-printable characters and, in fact,
807 caused crashes in UTF-8 locales. */
810 print_response_line(const char *prefix, const char *b, const char *e)
813 BOUNDED_TO_ALLOCA(b, e, copy);
814 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
815 quotearg_style (escape_quoting_style, copy));
818 /* Print the server response, line by line, omitting the trailing CRLF
819 from individual header lines, and prefixed with PREFIX. */
822 print_server_response (const struct response *resp, const char *prefix)
827 for (i = 0; resp->headers[i + 1]; i++)
829 const char *b = resp->headers[i];
830 const char *e = resp->headers[i + 1];
832 if (b < e && e[-1] == '\n')
834 if (b < e && e[-1] == '\r')
836 print_response_line(prefix, b, e);
840 /* Parse the `Content-Range' header and extract the information it
841 contains. Returns true if successful, false otherwise. */
843 parse_content_range (const char *hdr, wgint *first_byte_ptr,
844 wgint *last_byte_ptr, wgint *entity_length_ptr)
848 /* Ancient versions of Netscape proxy server, presumably predating
849 rfc2068, sent out `Content-Range' without the "bytes"
851 if (0 == strncasecmp (hdr, "bytes", 5))
854 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
858 while (c_isspace (*hdr))
863 if (!c_isdigit (*hdr))
865 for (num = 0; c_isdigit (*hdr); hdr++)
866 num = 10 * num + (*hdr - '0');
867 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
869 *first_byte_ptr = num;
871 for (num = 0; c_isdigit (*hdr); hdr++)
872 num = 10 * num + (*hdr - '0');
873 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
875 *last_byte_ptr = num;
880 for (num = 0; c_isdigit (*hdr); hdr++)
881 num = 10 * num + (*hdr - '0');
882 *entity_length_ptr = num;
886 /* Read the body of the request, but don't store it anywhere and don't
887 display a progress gauge. This is useful for reading the bodies of
888 administrative responses to which we will soon issue another
889 request. The response is not useful to the user, but reading it
890 allows us to continue using the same connection to the server.
892 If reading fails, false is returned, true otherwise. In debug
893 mode, the body is displayed for debugging purposes. */
896 skip_short_body (int fd, wgint contlen)
899 SKIP_SIZE = 512, /* size of the download buffer */
900 SKIP_THRESHOLD = 4096 /* the largest size we read */
902 char dlbuf[SKIP_SIZE + 1];
903 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
905 /* We shouldn't get here with unknown contlen. (This will change
906 with HTTP/1.1, which supports "chunked" transfer.) */
907 assert (contlen != -1);
909 /* If the body is too large, it makes more sense to simply close the
910 connection than to try to read the body. */
911 if (contlen > SKIP_THRESHOLD)
914 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
918 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
921 /* Don't normally report the error since this is an
922 optimization that should be invisible to the user. */
923 DEBUGP (("] aborting (%s).\n",
924 ret < 0 ? fd_errstr (fd) : "EOF received"));
928 /* Safe even if %.*s bogusly expects terminating \0 because
929 we've zero-terminated dlbuf above. */
930 DEBUGP (("%.*s", ret, dlbuf));
933 DEBUGP (("] done.\n"));
937 /* Extract a parameter from the string (typically an HTTP header) at
938 **SOURCE and advance SOURCE to the next parameter. Return false
939 when there are no more parameters to extract. The name of the
940 parameter is returned in NAME, and the value in VALUE. If the
941 parameter has no value, the token's value is zeroed out.
943 For example, if *SOURCE points to the string "attachment;
944 filename=\"foo bar\"", the first call to this function will return
945 the token named "attachment" and no value, and the second call will
946 return the token named "filename" and value "foo bar". The third
947 call will return false, indicating no more valid tokens. */
950 extract_param (const char **source, param_token *name, param_token *value,
953 const char *p = *source;
955 while (c_isspace (*p)) ++p;
959 return false; /* no error; nothing more to extract */
964 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
966 if (name->b == name->e)
967 return false; /* empty name: error */
968 while (c_isspace (*p)) ++p;
969 if (*p == separator || !*p) /* no value */
972 if (*p == separator) ++p;
977 return false; /* error */
979 /* *p is '=', extract value */
981 while (c_isspace (*p)) ++p;
982 if (*p == '"') /* quoted */
985 while (*p && *p != '"') ++p;
989 /* Currently at closing quote; find the end of param. */
990 while (c_isspace (*p)) ++p;
991 while (*p && *p != separator) ++p;
995 /* garbage after closed quote, e.g. foo="bar"baz */
1001 while (*p && *p != separator) ++p;
1003 while (value->e != value->b && c_isspace (value->e[-1]))
1005 if (*p == separator) ++p;
1012 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1014 /* Parse the contents of the `Content-Disposition' header, extracting
1015 the information useful to Wget. Content-Disposition is a header
1016 borrowed from MIME; when used in HTTP, it typically serves for
1017 specifying the desired file name of the resource. For example:
1019 Content-Disposition: attachment; filename="flora.jpg"
1021 Wget will skip the tokens it doesn't care about, such as
1022 "attachment" in the previous example; it will also skip other
1023 unrecognized params. If the header is syntactically correct and
1024 contains a file name, a copy of the file name is stored in
1025 *filename and true is returned. Otherwise, the function returns
1028 The file name is stripped of directory components and must not be
1032 parse_content_disposition (const char *hdr, char **filename)
1034 param_token name, value;
1035 while (extract_param (&hdr, &name, &value, ';'))
1036 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1038 /* Make the file name begin at the last slash or backslash. */
1039 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1040 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1041 if (last_slash && last_bs)
1042 value.b = 1 + MAX (last_slash, last_bs);
1043 else if (last_slash || last_bs)
1044 value.b = 1 + (last_slash ? last_slash : last_bs);
1045 if (value.b == value.e)
1047 /* Start with the directory prefix, if specified. */
1050 int prefix_length = strlen (opt.dir_prefix);
1051 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1056 total_length = prefix_length + (value.e - value.b);
1057 *filename = xmalloc (total_length + 1);
1058 strcpy (*filename, opt.dir_prefix);
1060 (*filename)[prefix_length - 1] = '/';
1061 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1062 (*filename)[total_length] = '\0';
1065 *filename = strdupdelim (value.b, value.e);
1071 /* Persistent connections. Currently, we cache the most recently used
1072 connection as persistent, provided that the HTTP server agrees to
1073 make it such. The persistence data is stored in the variables
1074 below. Ideally, it should be possible to cache an arbitrary fixed
1075 number of these connections. */
1077 /* Whether a persistent connection is active. */
1078 static bool pconn_active;
1081 /* The socket of the connection. */
1084 /* Host and port of the currently active persistent connection. */
1088 /* Whether a ssl handshake has occoured on this connection. */
1091 /* Whether the connection was authorized. This is only done by
1092 NTLM, which authorizes *connections* rather than individual
1093 requests. (That practice is peculiar for HTTP, but it is a
1094 useful optimization.) */
1098 /* NTLM data of the current connection. */
1099 struct ntlmdata ntlm;
1103 /* Mark the persistent connection as invalid and free the resources it
1104 uses. This is used by the CLOSE_* macros after they forcefully
1105 close a registered persistent connection. */
1108 invalidate_persistent (void)
1110 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1111 pconn_active = false;
1112 fd_close (pconn.socket);
1117 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1118 persistent. This will enable someone to use the same connection
1119 later. In the context of HTTP, this must be called only AFTER the
1120 response has been received and the server has promised that the
1121 connection will remain alive.
1123 If a previous connection was persistent, it is closed. */
1126 register_persistent (const char *host, int port, int fd, bool ssl)
1130 if (pconn.socket == fd)
1132 /* The connection FD is already registered. */
1137 /* The old persistent connection is still active; close it
1138 first. This situation arises whenever a persistent
1139 connection exists, but we then connect to a different
1140 host, and try to register a persistent connection to that
1142 invalidate_persistent ();
1146 pconn_active = true;
1148 pconn.host = xstrdup (host);
1151 pconn.authorized = false;
1153 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1156 /* Return true if a persistent connection is available for connecting
1160 persistent_available_p (const char *host, int port, bool ssl,
1161 bool *host_lookup_failed)
1163 /* First, check whether a persistent connection is active at all. */
1167 /* If we want SSL and the last connection wasn't or vice versa,
1168 don't use it. Checking for host and port is not enough because
1169 HTTP and HTTPS can apparently coexist on the same port. */
1170 if (ssl != pconn.ssl)
1173 /* If we're not connecting to the same port, we're not interested. */
1174 if (port != pconn.port)
1177 /* If the host is the same, we're in business. If not, there is
1178 still hope -- read below. */
1179 if (0 != strcasecmp (host, pconn.host))
1181 /* Check if pconn.socket is talking to HOST under another name.
1182 This happens often when both sites are virtual hosts
1183 distinguished only by name and served by the same network
1184 interface, and hence the same web server (possibly set up by
1185 the ISP and serving many different web sites). This
1186 admittedly unconventional optimization does not contradict
1187 HTTP and works well with popular server software. */
1191 struct address_list *al;
1194 /* Don't try to talk to two different SSL sites over the same
1195 secure connection! (Besides, it's not clear that
1196 name-based virtual hosting is even possible with SSL.) */
1199 /* If pconn.socket's peer is one of the IP addresses HOST
1200 resolves to, pconn.socket is for all intents and purposes
1201 already talking to HOST. */
1203 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1205 /* Can't get the peer's address -- something must be very
1206 wrong with the connection. */
1207 invalidate_persistent ();
1210 al = lookup_host (host, 0);
1213 *host_lookup_failed = true;
1217 found = address_list_contains (al, &ip);
1218 address_list_release (al);
1223 /* The persistent connection's peer address was found among the
1224 addresses HOST resolved to; therefore, pconn.sock is in fact
1225 already talking to HOST -- no need to reconnect. */
1228 /* Finally, check whether the connection is still open. This is
1229 important because most servers implement liberal (short) timeout
1230 on persistent connections. Wget can of course always reconnect
1231 if the connection doesn't work out, but it's nicer to know in
1232 advance. This test is a logical followup of the first test, but
1233 is "expensive" and therefore placed at the end of the list.
1235 (Current implementation of test_socket_open has a nice side
1236 effect that it treats sockets with pending data as "closed".
1237 This is exactly what we want: if a broken server sends message
1238 body in response to HEAD, or if it sends more than conent-length
1239 data, we won't reuse the corrupted connection.) */
1241 if (!test_socket_open (pconn.socket))
1243 /* Oops, the socket is no longer open. Now that we know that,
1244 let's invalidate the persistent connection before returning
1246 invalidate_persistent ();
1253 /* The idea behind these two CLOSE macros is to distinguish between
1254 two cases: one when the job we've been doing is finished, and we
1255 want to close the connection and leave, and two when something is
1256 seriously wrong and we're closing the connection as part of
1259 In case of keep_alive, CLOSE_FINISH should leave the connection
1260 open, while CLOSE_INVALIDATE should still close it.
1262 Note that the semantics of the flag `keep_alive' is "this
1263 connection *will* be reused (the server has promised not to close
1264 the connection once we're done)", while the semantics of
1265 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1266 active, registered connection". */
1268 #define CLOSE_FINISH(fd) do { \
1271 if (pconn_active && (fd) == pconn.socket) \
1272 invalidate_persistent (); \
1281 #define CLOSE_INVALIDATE(fd) do { \
1282 if (pconn_active && (fd) == pconn.socket) \
1283 invalidate_persistent (); \
1291 wgint len; /* received length */
1292 wgint contlen; /* expected length */
1293 wgint restval; /* the restart value */
1294 int res; /* the result of last read */
1295 char *rderrmsg; /* error message from read error */
1296 char *newloc; /* new location (redirection) */
1297 char *remote_time; /* remote time-stamp string */
1298 char *error; /* textual HTTP error */
1299 int statcode; /* status code */
1300 wgint rd_size; /* amount of data read from socket */
1301 double dltime; /* time it took to download the data */
1302 const char *referer; /* value of the referer header. */
1303 char *local_file; /* local file name. */
1304 bool existence_checked; /* true if we already checked for a file's
1305 existence after having begun to download
1306 (needed in gethttp for when connection is
1307 interrupted/restarted. */
1308 bool timestamp_checked; /* true if pre-download time-stamping checks
1309 * have already been performed */
1310 char *orig_file_name; /* name of file to compare for time-stamping
1311 * (might be != local_file if -K is set) */
1312 wgint orig_file_size; /* size of file to compare for time-stamping */
1313 time_t orig_file_tstamp; /* time-stamp of file to compare for
1318 free_hstat (struct http_stat *hs)
1320 xfree_null (hs->newloc);
1321 xfree_null (hs->remote_time);
1322 xfree_null (hs->error);
1323 xfree_null (hs->rderrmsg);
1324 xfree_null (hs->local_file);
1325 xfree_null (hs->orig_file_name);
1327 /* Guard against being called twice. */
1329 hs->remote_time = NULL;
1333 #define BEGINS_WITH(line, string_constant) \
1334 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1335 && (c_isspace (line[sizeof (string_constant) - 1]) \
1336 || !line[sizeof (string_constant) - 1]))
1338 #define SET_USER_AGENT(req) do { \
1339 if (!opt.useragent) \
1340 request_set_header (req, "User-Agent", \
1341 aprintf ("Wget/%s", version_string), rel_value); \
1342 else if (*opt.useragent) \
1343 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1346 /* The flags that allow clobbering the file (opening with "wb").
1347 Defined here to avoid repetition later. #### This will require
1349 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1350 || opt.dirstruct || opt.output_document)
1352 /* Retrieve a document through HTTP protocol. It recognizes status
1353 code, and correctly handles redirections. It closes the network
1354 socket. If it receives an error from the functions below it, it
1355 will print it if there is enough information to do so (almost
1356 always), returning the error to the caller (i.e. http_loop).
1358 Various HTTP parameters are stored to hs.
1360 If PROXY is non-NULL, the connection will be made to the proxy
1361 server, and u->url will be requested. */
1363 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1365 struct request *req;
1368 char *user, *passwd;
1372 wgint contlen, contrange;
1379 /* Set to 1 when the authorization has already been sent and should
1380 not be tried again. */
1381 bool auth_finished = false;
1383 /* Set to 1 when just globally-set Basic authorization has been sent;
1384 * should prevent further Basic negotiations, but not other
1386 bool basic_auth_finished = false;
1388 /* Whether NTLM authentication is used for this request. */
1389 bool ntlm_seen = false;
1391 /* Whether our connection to the remote host is through SSL. */
1392 bool using_ssl = false;
1394 /* Whether a HEAD request will be issued (as opposed to GET or
1396 bool head_only = !!(*dt & HEAD_ONLY);
1399 struct response *resp;
1403 /* Whether this connection will be kept alive after the HTTP request
1407 /* Whether keep-alive should be inhibited.
1409 RFC 2068 requests that 1.0 clients not send keep-alive requests
1410 to proxies. This is because many 1.0 proxies do not interpret
1411 the Connection header and transfer it to the remote server,
1412 causing it to not close the connection and leave both the proxy
1413 and the client hanging. */
1414 bool inhibit_keep_alive =
1415 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1417 /* Headers sent when using POST. */
1418 wgint post_data_size = 0;
1420 bool host_lookup_failed = false;
1423 if (u->scheme == SCHEME_HTTPS)
1425 /* Initialize the SSL context. After this has once been done,
1426 it becomes a no-op. */
1429 scheme_disable (SCHEME_HTTPS);
1430 logprintf (LOG_NOTQUIET,
1431 _("Disabling SSL due to encountered errors.\n"));
1432 return SSLINITFAILED;
1435 #endif /* HAVE_SSL */
1437 /* Initialize certain elements of struct http_stat. */
1441 hs->rderrmsg = NULL;
1443 hs->remote_time = NULL;
1448 /* Prepare the request to send. */
1450 req = request_new ();
1453 const char *meth = "GET";
1456 else if (opt.post_file_name || opt.post_data)
1458 /* Use the full path, i.e. one that includes the leading slash and
1459 the query string. E.g. if u->path is "foo/bar" and u->query is
1460 "param=value", full_path will be "/foo/bar?param=value". */
1463 /* When using SSL over proxy, CONNECT establishes a direct
1464 connection to the HTTPS server. Therefore use the same
1465 argument as when talking to the server directly. */
1466 && u->scheme != SCHEME_HTTPS
1469 meth_arg = xstrdup (u->url);
1471 meth_arg = url_full_path (u);
1472 request_set_method (req, meth, meth_arg);
1475 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1476 if (*dt & SEND_NOCACHE)
1477 request_set_header (req, "Pragma", "no-cache", rel_none);
1479 request_set_header (req, "Range",
1480 aprintf ("bytes=%s-",
1481 number_to_static_string (hs->restval)),
1483 SET_USER_AGENT (req);
1484 request_set_header (req, "Accept", "*/*", rel_none);
1486 /* Find the username and password for authentication. */
1489 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1490 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1491 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1494 && !u->user) /* We only do "site-wide" authentication with "global"
1495 user/password values; URL user/password info overrides. */
1497 /* If this is a host for which we've already received a Basic
1498 * challenge, we'll go ahead and send Basic authentication creds. */
1499 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1502 /* Generate the Host header, HOST:PORT. Take into account that:
1504 - Broken server-side software often doesn't recognize the PORT
1505 argument, so we must generate "Host: www.server.com" instead of
1506 "Host: www.server.com:80" (and likewise for https port).
1508 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1509 becomes ambiguous and needs to be rewritten as "Host:
1510 [3ffe:8100:200:2::2]:1234". */
1512 /* Formats arranged for hfmt[add_port][add_squares]. */
1513 static const char *hfmt[][2] = {
1514 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1516 int add_port = u->port != scheme_default_port (u->scheme);
1517 int add_squares = strchr (u->host, ':') != NULL;
1518 request_set_header (req, "Host",
1519 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1523 if (!inhibit_keep_alive)
1524 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1527 request_set_header (req, "Cookie",
1528 cookie_header (wget_cookie_jar,
1529 u->host, u->port, u->path,
1531 u->scheme == SCHEME_HTTPS
1538 if (opt.post_data || opt.post_file_name)
1540 request_set_header (req, "Content-Type",
1541 "application/x-www-form-urlencoded", rel_none);
1543 post_data_size = strlen (opt.post_data);
1546 post_data_size = file_size (opt.post_file_name);
1547 if (post_data_size == -1)
1549 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1550 quote (opt.post_file_name), strerror (errno));
1554 request_set_header (req, "Content-Length",
1555 xstrdup (number_to_static_string (post_data_size)),
1559 /* Add the user headers. */
1560 if (opt.user_headers)
1563 for (i = 0; opt.user_headers[i]; i++)
1564 request_set_user_header (req, opt.user_headers[i]);
1568 /* We need to come back here when the initial attempt to retrieve
1569 without authorization header fails. (Expected to happen at least
1570 for the Digest authorization scheme.) */
1575 char *proxy_user, *proxy_passwd;
1576 /* For normal username and password, URL components override
1577 command-line/wgetrc parameters. With proxy
1578 authentication, it's the reverse, because proxy URLs are
1579 normally the "permanent" ones, so command-line args
1580 should take precedence. */
1581 if (opt.proxy_user && opt.proxy_passwd)
1583 proxy_user = opt.proxy_user;
1584 proxy_passwd = opt.proxy_passwd;
1588 proxy_user = proxy->user;
1589 proxy_passwd = proxy->passwd;
1591 /* #### This does not appear right. Can't the proxy request,
1592 say, `Digest' authentication? */
1593 if (proxy_user && proxy_passwd)
1594 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1596 /* If we're using a proxy, we will be connecting to the proxy
1600 /* Proxy authorization over SSL is handled below. */
1602 if (u->scheme != SCHEME_HTTPS)
1604 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1609 /* Establish the connection. */
1611 if (!inhibit_keep_alive)
1613 /* Look for a persistent connection to target host, unless a
1614 proxy is used. The exception is when SSL is in use, in which
1615 case the proxy is nothing but a passthrough to the target
1616 host, registered as a connection to the latter. */
1617 struct url *relevant = conn;
1619 if (u->scheme == SCHEME_HTTPS)
1623 if (persistent_available_p (relevant->host, relevant->port,
1625 relevant->scheme == SCHEME_HTTPS,
1629 &host_lookup_failed))
1631 sock = pconn.socket;
1632 using_ssl = pconn.ssl;
1633 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1634 quotearg_style (escape_quoting_style, pconn.host),
1636 DEBUGP (("Reusing fd %d.\n", sock));
1637 if (pconn.authorized)
1638 /* If the connection is already authorized, the "Basic"
1639 authorization added by code above is unnecessary and
1641 request_remove_header (req, "Authorization");
1643 else if (host_lookup_failed)
1646 logprintf(LOG_NOTQUIET,
1647 _("%s: unable to resolve host address %s\n"),
1648 exec_name, quote (relevant->host));
1655 sock = connect_to_host (conn->host, conn->port);
1664 return (retryable_socket_connect_error (errno)
1665 ? CONERROR : CONIMPOSSIBLE);
1669 if (proxy && u->scheme == SCHEME_HTTPS)
1671 /* When requesting SSL URLs through proxies, use the
1672 CONNECT method to request passthrough. */
1673 struct request *connreq = request_new ();
1674 request_set_method (connreq, "CONNECT",
1675 aprintf ("%s:%d", u->host, u->port));
1676 SET_USER_AGENT (connreq);
1679 request_set_header (connreq, "Proxy-Authorization",
1680 proxyauth, rel_value);
1681 /* Now that PROXYAUTH is part of the CONNECT request,
1682 zero it out so we don't send proxy authorization with
1683 the regular request below. */
1686 /* Examples in rfc2817 use the Host header in CONNECT
1687 requests. I don't see how that gains anything, given
1688 that the contents of Host would be exactly the same as
1689 the contents of CONNECT. */
1691 write_error = request_send (connreq, sock);
1692 request_free (connreq);
1693 if (write_error < 0)
1695 CLOSE_INVALIDATE (sock);
1699 head = read_http_response_head (sock);
1702 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1704 CLOSE_INVALIDATE (sock);
1713 DEBUGP (("proxy responded with: [%s]\n", head));
1715 resp = resp_new (head);
1716 statcode = resp_status (resp, &message);
1719 if (statcode != 200)
1722 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1723 message ? quotearg_style (escape_quoting_style, message) : "?");
1724 xfree_null (message);
1727 xfree_null (message);
1729 /* SOCK is now *really* connected to u->host, so update CONN
1730 to reflect this. That way register_persistent will
1731 register SOCK as being connected to u->host:u->port. */
1735 if (conn->scheme == SCHEME_HTTPS)
1737 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1744 #endif /* HAVE_SSL */
1747 /* Send the request to server. */
1748 write_error = request_send (req, sock);
1750 if (write_error >= 0)
1754 DEBUGP (("[POST data: %s]\n", opt.post_data));
1755 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1757 else if (opt.post_file_name && post_data_size != 0)
1758 write_error = post_file (sock, opt.post_file_name, post_data_size);
1761 if (write_error < 0)
1763 CLOSE_INVALIDATE (sock);
1767 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1768 proxy ? "Proxy" : "HTTP");
1773 head = read_http_response_head (sock);
1778 logputs (LOG_NOTQUIET, _("No data received.\n"));
1779 CLOSE_INVALIDATE (sock);
1785 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1787 CLOSE_INVALIDATE (sock);
1792 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1794 resp = resp_new (head);
1796 /* Check for status line. */
1798 statcode = resp_status (resp, &message);
1799 if (!opt.server_response)
1800 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1801 message ? quotearg_style (escape_quoting_style, message) : "");
1804 logprintf (LOG_VERBOSE, "\n");
1805 print_server_response (resp, " ");
1808 /* Determine the local filename if needed. Notice that if -O is used
1809 * hstat.local_file is set by http_loop to the argument of -O. */
1810 if (!hs->local_file)
1812 /* Honor Content-Disposition whether possible. */
1813 if (!opt.content_disposition
1814 || !resp_header_copy (resp, "Content-Disposition",
1815 hdrval, sizeof (hdrval))
1816 || !parse_content_disposition (hdrval, &hs->local_file))
1818 /* The Content-Disposition header is missing or broken.
1819 * Choose unique file name according to given URL. */
1820 hs->local_file = url_file_name (u);
1824 /* TODO: perform this check only once. */
1825 if (!hs->existence_checked && file_exists_p (hs->local_file))
1827 if (opt.noclobber && !opt.output_document)
1829 /* If opt.noclobber is turned on and file already exists, do not
1830 retrieve the file. But if the output_document was given, then this
1831 test was already done and the file didn't exist. Hence the !opt.output_document */
1832 logprintf (LOG_VERBOSE, _("\
1833 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1834 /* If the file is there, we suppose it's retrieved OK. */
1837 /* #### Bogusness alert. */
1838 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1839 if (has_html_suffix_p (hs->local_file))
1842 return RETRUNNEEDED;
1844 else if (!ALLOW_CLOBBER)
1846 char *unique = unique_name (hs->local_file, true);
1847 if (unique != hs->local_file)
1848 xfree (hs->local_file);
1849 hs->local_file = unique;
1852 hs->existence_checked = true;
1854 /* Support timestamping */
1855 /* TODO: move this code out of gethttp. */
1856 if (opt.timestamping && !hs->timestamp_checked)
1858 size_t filename_len = strlen (hs->local_file);
1859 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1860 bool local_dot_orig_file_exists = false;
1861 char *local_filename = NULL;
1864 if (opt.backup_converted)
1865 /* If -K is specified, we'll act on the assumption that it was specified
1866 last time these files were downloaded as well, and instead of just
1867 comparing local file X against server file X, we'll compare local
1868 file X.orig (if extant, else X) against server file X. If -K
1869 _wasn't_ specified last time, or the server contains files called
1870 *.orig, -N will be back to not operating correctly with -k. */
1872 /* Would a single s[n]printf() call be faster? --dan
1874 Definitely not. sprintf() is horribly slow. It's a
1875 different question whether the difference between the two
1876 affects a program. Usually I'd say "no", but at one
1877 point I profiled Wget, and found that a measurable and
1878 non-negligible amount of time was lost calling sprintf()
1879 in url.c. Replacing sprintf with inline calls to
1880 strcpy() and number_to_string() made a difference.
1882 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1883 memcpy (filename_plus_orig_suffix + filename_len,
1884 ".orig", sizeof (".orig"));
1886 /* Try to stat() the .orig file. */
1887 if (stat (filename_plus_orig_suffix, &st) == 0)
1889 local_dot_orig_file_exists = true;
1890 local_filename = filename_plus_orig_suffix;
1894 if (!local_dot_orig_file_exists)
1895 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1896 if (stat (hs->local_file, &st) == 0)
1897 local_filename = hs->local_file;
1899 if (local_filename != NULL)
1900 /* There was a local file, so we'll check later to see if the version
1901 the server has is the same version we already have, allowing us to
1904 hs->orig_file_name = xstrdup (local_filename);
1905 hs->orig_file_size = st.st_size;
1906 hs->orig_file_tstamp = st.st_mtime;
1908 /* Modification time granularity is 2 seconds for Windows, so
1909 increase local time by 1 second for later comparison. */
1910 ++hs->orig_file_tstamp;
1915 if (!opt.ignore_length
1916 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1920 parsed = str_to_wgint (hdrval, NULL, 10);
1921 if (parsed == WGINT_MAX && errno == ERANGE)
1924 #### If Content-Length is out of range, it most likely
1925 means that the file is larger than 2G and that we're
1926 compiled without LFS. In that case we should probably
1927 refuse to even attempt to download the file. */
1930 else if (parsed < 0)
1932 /* Negative Content-Length; nonsensical, so we can't
1933 assume any information about the content to receive. */
1940 /* Check for keep-alive related responses. */
1941 if (!inhibit_keep_alive && contlen != -1)
1943 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1945 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1947 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1952 /* The server has promised that it will not close the connection
1953 when we're done. This means that we can register it. */
1954 register_persistent (conn->host, conn->port, sock, using_ssl);
1956 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1958 /* Authorization is required. */
1959 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1960 CLOSE_FINISH (sock);
1962 CLOSE_INVALIDATE (sock);
1963 pconn.authorized = false;
1964 if (!auth_finished && (user && passwd))
1966 /* IIS sends multiple copies of WWW-Authenticate, one with
1967 the value "negotiate", and other(s) with data. Loop over
1968 all the occurrences and pick the one we recognize. */
1970 const char *wabeg, *waend;
1971 char *www_authenticate = NULL;
1973 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1974 &wabeg, &waend)) != -1;
1976 if (known_authentication_scheme_p (wabeg, waend))
1978 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1982 if (!www_authenticate)
1984 /* If the authentication header is missing or
1985 unrecognized, there's no sense in retrying. */
1986 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1988 else if (!basic_auth_finished
1989 || !BEGINS_WITH (www_authenticate, "Basic"))
1992 pth = url_full_path (u);
1993 request_set_header (req, "Authorization",
1994 create_authorization_line (www_authenticate,
1996 request_method (req),
2000 if (BEGINS_WITH (www_authenticate, "NTLM"))
2002 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2004 /* Need to register this host as using basic auth,
2005 * so we automatically send creds next time. */
2006 register_basic_auth_host (u->host);
2009 goto retry_with_auth;
2013 /* We already did Basic auth, and it failed. Gotta
2017 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2021 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2023 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2025 pconn.authorized = true;
2029 hs->statcode = statcode;
2031 hs->error = xstrdup (_("Malformed status line"));
2033 hs->error = xstrdup (_("(no description)"));
2035 hs->error = xstrdup (message);
2036 xfree_null (message);
2038 type = resp_header_strdup (resp, "Content-Type");
2041 char *tmp = strchr (type, ';');
2044 /* sXXXav: only needed if IRI support is enabled */
2045 char *tmp2 = tmp + 1;
2047 while (tmp > type && c_isspace (tmp[-1]))
2051 /* Try to get remote encoding if needed */
2052 if (opt.enable_iri && !opt.encoding_remote)
2053 /* xxx = */ parse_charset (tmp2);
2056 hs->newloc = resp_header_strdup (resp, "Location");
2057 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2059 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2063 const char *scbeg, *scend;
2064 /* The jar should have been created by now. */
2065 assert (wget_cookie_jar != NULL);
2067 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2068 &scbeg, &scend)) != -1;
2071 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2072 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2073 u->path, set_cookie);
2077 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2079 wgint first_byte_pos, last_byte_pos, entity_length;
2080 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2083 contrange = first_byte_pos;
2084 contlen = last_byte_pos - first_byte_pos + 1;
2089 /* 20x responses are counted among successful by default. */
2090 if (H_20X (statcode))
2093 /* Return if redirected. */
2094 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2096 /* RFC2068 says that in case of the 300 (multiple choices)
2097 response, the server can output a preferred URL through
2098 `Location' header; otherwise, the request should be treated
2099 like GET. So, if the location is set, it will be a
2100 redirection; otherwise, just proceed normally. */
2101 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2105 logprintf (LOG_VERBOSE,
2106 _("Location: %s%s\n"),
2107 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2108 hs->newloc ? _(" [following]") : "");
2109 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2110 CLOSE_FINISH (sock);
2112 CLOSE_INVALIDATE (sock);
2118 /* If content-type is not given, assume text/html. This is because
2119 of the multitude of broken CGI's that "forget" to generate the
2122 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2123 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2128 if (opt.html_extension && (*dt & TEXTHTML))
2129 /* -E / --html-extension / html_extension = on was specified, and this is a
2130 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2131 already the file's suffix, tack on ".html". */
2133 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2135 if (last_period_in_local_filename == NULL
2136 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2137 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2139 int local_filename_len = strlen (hs->local_file);
2140 /* Resize the local file, allowing for ".html" preceded by
2141 optional ".NUMBER". */
2142 hs->local_file = xrealloc (hs->local_file,
2143 local_filename_len + 24 + sizeof (".html"));
2144 strcpy(hs->local_file + local_filename_len, ".html");
2145 /* If clobbering is not allowed and the file, as named,
2146 exists, tack on ".NUMBER.html" instead. */
2147 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2151 sprintf (hs->local_file + local_filename_len,
2152 ".%d.html", ext_num++);
2153 while (file_exists_p (hs->local_file));
2155 *dt |= ADDED_HTML_EXTENSION;
2159 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2161 /* If `-c' is in use and the file has been fully downloaded (or
2162 the remote file has shrunk), Wget effectively requests bytes
2163 after the end of file and the server response with 416. */
2164 logputs (LOG_VERBOSE, _("\
2165 \n The file is already fully retrieved; nothing to do.\n\n"));
2166 /* In case the caller inspects. */
2169 /* Mark as successfully retrieved. */
2172 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2173 might be more bytes in the body. */
2174 return RETRUNNEEDED;
2176 if ((contrange != 0 && contrange != hs->restval)
2177 || (H_PARTIAL (statcode) && !contrange))
2179 /* The Range request was somehow misunderstood by the server.
2182 CLOSE_INVALIDATE (sock);
2188 hs->contlen = contlen + contrange;
2194 /* No need to print this output if the body won't be
2195 downloaded at all, or if the original server response is
2197 logputs (LOG_VERBOSE, _("Length: "));
2200 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2201 if (contlen + contrange >= 1024)
2202 logprintf (LOG_VERBOSE, " (%s)",
2203 human_readable (contlen + contrange));
2206 if (contlen >= 1024)
2207 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2208 number_to_static_string (contlen),
2209 human_readable (contlen));
2211 logprintf (LOG_VERBOSE, _(", %s remaining"),
2212 number_to_static_string (contlen));
2216 logputs (LOG_VERBOSE,
2217 opt.ignore_length ? _("ignored") : _("unspecified"));
2219 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2221 logputs (LOG_VERBOSE, "\n");
2225 type = NULL; /* We don't need it any more. */
2227 /* Return if we have no intention of further downloading. */
2228 if (!(*dt & RETROKF) || head_only)
2230 /* In case the caller cares to look... */
2235 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2236 servers not to send body in response to a HEAD request, and
2237 those that do will likely be caught by test_socket_open.
2238 If not, they can be worked around using
2239 `--no-http-keep-alive'. */
2240 CLOSE_FINISH (sock);
2241 else if (keep_alive && skip_short_body (sock, contlen))
2242 /* Successfully skipped the body; also keep using the socket. */
2243 CLOSE_FINISH (sock);
2245 CLOSE_INVALIDATE (sock);
2246 return RETRFINISHED;
2249 /* Open the local file. */
2252 mkalldirs (hs->local_file);
2254 rotate_backups (hs->local_file);
2256 fp = fopen (hs->local_file, "ab");
2257 else if (ALLOW_CLOBBER)
2258 fp = fopen (hs->local_file, "wb");
2261 fp = fopen_excl (hs->local_file, true);
2262 if (!fp && errno == EEXIST)
2264 /* We cannot just invent a new name and use it (which is
2265 what functions like unique_create typically do)
2266 because we told the user we'd use this name.
2267 Instead, return and retry the download. */
2268 logprintf (LOG_NOTQUIET,
2269 _("%s has sprung into existence.\n"),
2271 CLOSE_INVALIDATE (sock);
2272 return FOPEN_EXCL_ERR;
2277 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2278 CLOSE_INVALIDATE (sock);
2285 /* Print fetch message, if opt.verbose. */
2288 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2289 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2292 /* This confuses the timestamping code that checks for file size.
2293 #### The timestamping code should be smarter about file size. */
2294 if (opt.save_headers && hs->restval == 0)
2295 fwrite (head, 1, strlen (head), fp);
2297 /* Now we no longer need to store the response header. */
2300 /* Download the request body. */
2303 /* If content-length is present, read that much; otherwise, read
2304 until EOF. The HTTP spec doesn't require the server to
2305 actually close the connection when it's done sending data. */
2306 flags |= rb_read_exactly;
2307 if (hs->restval > 0 && contrange == 0)
2308 /* If the server ignored our range request, instruct fd_read_body
2309 to skip the first RESTVAL bytes of body. */
2310 flags |= rb_skip_startpos;
2311 hs->len = hs->restval;
2313 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2314 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2318 CLOSE_FINISH (sock);
2322 hs->rderrmsg = xstrdup (fd_errstr (sock));
2323 CLOSE_INVALIDATE (sock);
2330 return RETRFINISHED;
2333 /* The genuine HTTP loop! This is the part where the retrieval is
2334 retried, and retried, and retried, and... */
2336 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2337 int *dt, struct url *proxy)
2340 bool got_head = false; /* used for time-stamping and filename detection */
2341 bool time_came_from_head = false;
2342 bool got_name = false;
2345 uerr_t err, ret = TRYLIMEXC;
2346 time_t tmr = -1; /* remote time-stamp */
2347 struct http_stat hstat; /* HTTP status */
2349 bool send_head_first = true;
2351 /* Assert that no value for *LOCAL_FILE was passed. */
2352 assert (local_file == NULL || *local_file == NULL);
2354 /* Set LOCAL_FILE parameter. */
2355 if (local_file && opt.output_document)
2356 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2358 /* Reset NEWLOC parameter. */
2361 /* This used to be done in main(), but it's a better idea to do it
2362 here so that we don't go through the hoops if we're just using
2367 /* Warn on (likely bogus) wildcard usage in HTTP. */
2368 if (opt.ftp_glob && has_wildcards_p (u->path))
2369 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2371 /* Setup hstat struct. */
2373 hstat.referer = referer;
2375 if (opt.output_document)
2377 hstat.local_file = xstrdup (opt.output_document);
2380 else if (!opt.content_disposition)
2382 hstat.local_file = url_file_name (u);
2386 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2387 * screaming for some refactoring. */
2388 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2390 /* If opt.noclobber is turned on and file already exists, do not
2391 retrieve the file. But if the output_document was given, then this
2392 test was already done and the file didn't exist. Hence the !opt.output_document */
2393 logprintf (LOG_VERBOSE, _("\
2394 File %s already there; not retrieving.\n\n"),
2395 quote (hstat.local_file));
2396 /* If the file is there, we suppose it's retrieved OK. */
2399 /* #### Bogusness alert. */
2400 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2401 if (has_html_suffix_p (hstat.local_file))
2408 /* Reset the counter. */
2411 /* Reset the document type. */
2414 /* Skip preliminary HEAD request if we're not in spider mode AND
2415 * if -O was given or HTTP Content-Disposition support is disabled. */
2417 && (got_name || !opt.content_disposition))
2418 send_head_first = false;
2420 /* Send preliminary HEAD request if -N is given and we have an existing
2421 * destination file. */
2422 if (opt.timestamping
2423 && !opt.content_disposition
2424 && file_exists_p (url_file_name (u)))
2425 send_head_first = true;
2430 /* Increment the pass counter. */
2432 sleep_between_retrievals (count);
2434 /* Get the current time string. */
2435 tms = datetime_str (time (NULL));
2437 if (opt.spider && !got_head)
2438 logprintf (LOG_VERBOSE, _("\
2439 Spider mode enabled. Check if remote file exists.\n"));
2441 /* Print fetch message, if opt.verbose. */
2444 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2449 sprintf (tmp, _("(try:%2d)"), count);
2450 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2455 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2460 ws_changetitle (hurl);
2465 /* Default document type is empty. However, if spider mode is
2466 on or time-stamping is employed, HEAD_ONLY commands is
2467 encoded within *dt. */
2468 if (send_head_first && !got_head)
2473 /* Decide whether or not to restart. */
2476 && stat (hstat.local_file, &st) == 0
2477 && S_ISREG (st.st_mode))
2478 /* When -c is used, continue from on-disk size. (Can't use
2479 hstat.len even if count>1 because we don't want a failed
2480 first attempt to clobber existing data.) */
2481 hstat.restval = st.st_size;
2483 /* otherwise, continue where the previous try left off */
2484 hstat.restval = hstat.len;
2488 /* Decide whether to send the no-cache directive. We send it in
2490 a) we're using a proxy, and we're past our first retrieval.
2491 Some proxies are notorious for caching incomplete data, so
2492 we require a fresh get.
2493 b) caching is explicitly inhibited. */
2494 if ((proxy && count > 1) /* a */
2495 || !opt.allow_cache) /* b */
2496 *dt |= SEND_NOCACHE;
2498 *dt &= ~SEND_NOCACHE;
2500 /* Try fetching the document, or at least its head. */
2501 err = gethttp (u, &hstat, dt, proxy);
2504 tms = datetime_str (time (NULL));
2506 /* Get the new location (with or without the redirection). */
2508 *newloc = xstrdup (hstat.newloc);
2512 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2513 case CONERROR: case READERR: case WRITEFAILED:
2514 case RANGEERR: case FOPEN_EXCL_ERR:
2515 /* Non-fatal errors continue executing the loop, which will
2516 bring them to "while" statement at the end, to judge
2517 whether the number of tries was exceeded. */
2518 printwhat (count, opt.ntry);
2520 case FWRITEERR: case FOPENERR:
2521 /* Another fatal error. */
2522 logputs (LOG_VERBOSE, "\n");
2523 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2524 quote (hstat.local_file), strerror (errno));
2525 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2526 case SSLINITFAILED: case CONTNOTSUPPORTED:
2527 /* Fatal errors just return from the function. */
2531 /* Another fatal error. */
2532 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2536 /* Return the new location to the caller. */
2539 logprintf (LOG_NOTQUIET,
2540 _("ERROR: Redirection (%d) without location.\n"),
2550 /* The file was already fully retrieved. */
2554 /* Deal with you later. */
2557 /* All possibilities should have been exhausted. */
2561 if (!(*dt & RETROKF))
2566 /* #### Ugly ugly ugly! */
2567 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2568 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2571 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2573 && (hstat.statcode == 500 || hstat.statcode == 501))
2578 /* Maybe we should always keep track of broken links, not just in
2580 else if (opt.spider)
2582 /* #### Again: ugly ugly ugly! */
2584 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2585 nonexisting_url (hurl);
2586 logprintf (LOG_NOTQUIET, _("\
2587 Remote file does not exist -- broken link!!!\n"));
2591 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2592 tms, hstat.statcode,
2593 quotearg_style (escape_quoting_style, hstat.error));
2595 logputs (LOG_VERBOSE, "\n");
2601 /* Did we get the time-stamp? */
2604 got_head = true; /* no more time-stamping */
2606 if (opt.timestamping && !hstat.remote_time)
2608 logputs (LOG_NOTQUIET, _("\
2609 Last-modified header missing -- time-stamps turned off.\n"));
2611 else if (hstat.remote_time)
2613 /* Convert the date-string into struct tm. */
2614 tmr = http_atotm (hstat.remote_time);
2615 if (tmr == (time_t) (-1))
2616 logputs (LOG_VERBOSE, _("\
2617 Last-modified header invalid -- time-stamp ignored.\n"));
2618 if (*dt & HEAD_ONLY)
2619 time_came_from_head = true;
2622 if (send_head_first)
2624 /* The time-stamping section. */
2625 if (opt.timestamping)
2627 if (hstat.orig_file_name) /* Perform the following
2628 checks only if the file
2630 download already exists. */
2632 if (hstat.remote_time &&
2633 tmr != (time_t) (-1))
2635 /* Now time-stamping can be used validly.
2636 Time-stamping means that if the sizes of
2637 the local and remote file match, and local
2638 file is newer than the remote file, it will
2639 not be retrieved. Otherwise, the normal
2640 download procedure is resumed. */
2641 if (hstat.orig_file_tstamp >= tmr)
2643 if (hstat.contlen == -1
2644 || hstat.orig_file_size == hstat.contlen)
2646 logprintf (LOG_VERBOSE, _("\
2647 Server file no newer than local file %s -- not retrieving.\n\n"),
2648 quote (hstat.orig_file_name));
2654 logprintf (LOG_VERBOSE, _("\
2655 The sizes do not match (local %s) -- retrieving.\n"),
2656 number_to_static_string (hstat.orig_file_size));
2660 logputs (LOG_VERBOSE,
2661 _("Remote file is newer, retrieving.\n"));
2663 logputs (LOG_VERBOSE, "\n");
2667 /* free_hstat (&hstat); */
2668 hstat.timestamp_checked = true;
2677 logputs (LOG_VERBOSE, _("\
2678 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2682 logprintf (LOG_VERBOSE, _("\
2683 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2684 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2692 logprintf (LOG_VERBOSE, _("\
2693 Remote file exists and could contain further links,\n\
2694 but recursion is disabled -- not retrieving.\n\n"));
2698 logprintf (LOG_VERBOSE, _("\
2699 Remote file exists.\n\n"));
2701 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2708 count = 0; /* the retrieve count for HEAD is reset */
2710 } /* send_head_first */
2713 if ((tmr != (time_t) (-1))
2714 && ((hstat.len == hstat.contlen) ||
2715 ((hstat.res == 0) && (hstat.contlen == -1))))
2717 /* #### This code repeats in http.c and ftp.c. Move it to a
2719 const char *fl = NULL;
2720 if (opt.output_document)
2722 if (output_stream_regular)
2723 fl = opt.output_document;
2726 fl = hstat.local_file;
2730 /* Reparse time header, in case it's changed. */
2731 if (time_came_from_head
2732 && hstat.remote_time && hstat.remote_time[0])
2734 newtmr = http_atotm (hstat.remote_time);
2741 /* End of time-stamping section. */
2743 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2744 total_download_time += hstat.dltime;
2746 if (hstat.len == hstat.contlen)
2750 logprintf (LOG_VERBOSE,
2751 _("%s (%s) - %s saved [%s/%s]\n\n"),
2752 tms, tmrate, quote (hstat.local_file),
2753 number_to_static_string (hstat.len),
2754 number_to_static_string (hstat.contlen));
2755 logprintf (LOG_NONVERBOSE,
2756 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2758 number_to_static_string (hstat.len),
2759 number_to_static_string (hstat.contlen),
2760 hstat.local_file, count);
2763 total_downloaded_bytes += hstat.len;
2765 /* Remember that we downloaded the file for later ".orig" code. */
2766 if (*dt & ADDED_HTML_EXTENSION)
2767 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2769 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2774 else if (hstat.res == 0) /* No read error */
2776 if (hstat.contlen == -1) /* We don't know how much we were supposed
2777 to get, so assume we succeeded. */
2781 logprintf (LOG_VERBOSE,
2782 _("%s (%s) - %s saved [%s]\n\n"),
2783 tms, tmrate, quote (hstat.local_file),
2784 number_to_static_string (hstat.len));
2785 logprintf (LOG_NONVERBOSE,
2786 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2787 tms, u->url, number_to_static_string (hstat.len),
2788 hstat.local_file, count);
2791 total_downloaded_bytes += hstat.len;
2793 /* Remember that we downloaded the file for later ".orig" code. */
2794 if (*dt & ADDED_HTML_EXTENSION)
2795 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2797 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2802 else if (hstat.len < hstat.contlen) /* meaning we lost the
2803 connection too soon */
2805 logprintf (LOG_VERBOSE,
2806 _("%s (%s) - Connection closed at byte %s. "),
2807 tms, tmrate, number_to_static_string (hstat.len));
2808 printwhat (count, opt.ntry);
2812 /* Getting here would mean reading more data than
2813 requested with content-length, which we never do. */
2816 else /* from now on hstat.res can only be -1 */
2818 if (hstat.contlen == -1)
2820 logprintf (LOG_VERBOSE,
2821 _("%s (%s) - Read error at byte %s (%s)."),
2822 tms, tmrate, number_to_static_string (hstat.len),
2824 printwhat (count, opt.ntry);
2827 else /* hstat.res == -1 and contlen is given */
2829 logprintf (LOG_VERBOSE,
2830 _("%s (%s) - Read error at byte %s/%s (%s). "),
2832 number_to_static_string (hstat.len),
2833 number_to_static_string (hstat.contlen),
2835 printwhat (count, opt.ntry);
2841 while (!opt.ntry || (count < opt.ntry));
2845 *local_file = xstrdup (hstat.local_file);
2846 free_hstat (&hstat);
2851 /* Check whether the result of strptime() indicates success.
2852 strptime() returns the pointer to how far it got to in the string.
2853 The processing has been successful if the string is at `GMT' or
2854 `+X', or at the end of the string.
2856 In extended regexp parlance, the function returns 1 if P matches
2857 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2858 can return) is considered a failure and 0 is returned. */
2860 check_end (const char *p)
2864 while (c_isspace (*p))
2867 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2868 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2874 /* Convert the textual specification of time in TIME_STRING to the
2875 number of seconds since the Epoch.
2877 TIME_STRING can be in any of the three formats RFC2616 allows the
2878 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2879 as well as the time format used in the Set-Cookie header.
2880 Timezones are ignored, and should be GMT.
2882 Return the computed time_t representation, or -1 if the conversion
2885 This function uses strptime with various string formats for parsing
2886 TIME_STRING. This results in a parser that is not as lenient in
2887 interpreting TIME_STRING as I would like it to be. Being based on
2888 strptime, it always allows shortened months, one-digit days, etc.,
2889 but due to the multitude of formats in which time can be
2890 represented, an ideal HTTP time parser would be even more
2891 forgiving. It should completely ignore things like week days and
2892 concentrate only on the various forms of representing years,
2893 months, days, hours, minutes, and seconds. For example, it would
2894 be nice if it accepted ISO 8601 out of the box.
2896 I've investigated free and PD code for this purpose, but none was
2897 usable. getdate was big and unwieldy, and had potential copyright
2898 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2899 distributed with phttpd, is excellent, but we cannot use it because
2900 it is not assigned to the FSF. So I stuck it with strptime. */
2903 http_atotm (const char *time_string)
2905 /* NOTE: Solaris strptime man page claims that %n and %t match white
2906 space, but that's not universally available. Instead, we simply
2907 use ` ' to mean "skip all WS", which works under all strptime
2908 implementations I've tested. */
2910 static const char *time_formats[] = {
2911 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2912 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2913 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2914 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2915 (used in Set-Cookie, defined in the
2916 Netscape cookie specification.) */
2918 const char *oldlocale;
2920 time_t ret = (time_t) -1;
2922 /* Solaris strptime fails to recognize English month names in
2923 non-English locales, which we work around by temporarily setting
2924 locale to C before invoking strptime. */
2925 oldlocale = setlocale (LC_TIME, NULL);
2926 setlocale (LC_TIME, "C");
2928 for (i = 0; i < countof (time_formats); i++)
2932 /* Some versions of strptime use the existing contents of struct
2933 tm to recalculate the date according to format. Zero it out
2934 to prevent stack garbage from influencing strptime. */
2937 if (check_end (strptime (time_string, time_formats[i], &t)))
2944 /* Restore the previous locale. */
2945 setlocale (LC_TIME, oldlocale);
2950 /* Authorization support: We support three authorization schemes:
2952 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2954 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2955 consisting of answering to the server's challenge with the proper
2958 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2959 Stenberg for libcurl. Like digest, NTLM is based on a
2960 challenge-response mechanism, but unlike digest, it is non-standard
2961 (authenticates TCP connections rather than requests), undocumented
2962 and Microsoft-specific. */
2964 /* Create the authentication header contents for the `Basic' scheme.
2965 This is done by encoding the string "USER:PASS" to base64 and
2966 prepending the string "Basic " in front of it. */
2969 basic_authentication_encode (const char *user, const char *passwd)
2972 int len1 = strlen (user) + 1 + strlen (passwd);
2974 t1 = (char *)alloca (len1 + 1);
2975 sprintf (t1, "%s:%s", user, passwd);
2977 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2978 base64_encode (t1, len1, t2);
2980 return concat_strings ("Basic ", t2, (char *) 0);
2983 #define SKIP_WS(x) do { \
2984 while (c_isspace (*(x))) \
2988 #ifdef ENABLE_DIGEST
2989 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2990 an array of 16 bytes containing the hash keys, and BUF should be a
2991 buffer of 33 writable characters (32 for hex digits plus one for
2992 zero termination). */
2994 dump_hash (char *buf, const unsigned char *hash)
2998 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3000 *buf++ = XNUM_TO_digit (*hash >> 4);
3001 *buf++ = XNUM_TO_digit (*hash & 0xf);
3006 /* Take the line apart to find the challenge, and compose a digest
3007 authorization header. See RFC2069 section 2.1.2. */
3009 digest_authentication_encode (const char *au, const char *user,
3010 const char *passwd, const char *method,
3013 static char *realm, *opaque, *nonce;
3018 { "realm", &realm },
3019 { "opaque", &opaque },
3023 param_token name, value;
3025 realm = opaque = nonce = NULL;
3027 au += 6; /* skip over `Digest' */
3028 while (extract_param (&au, &name, &value, ','))
3031 size_t namelen = name.e - name.b;
3032 for (i = 0; i < countof (options); i++)
3033 if (namelen == strlen (options[i].name)
3034 && 0 == strncmp (name.b, options[i].name,
3037 *options[i].variable = strdupdelim (value.b, value.e);
3041 if (!realm || !nonce || !user || !passwd || !path || !method)
3044 xfree_null (opaque);
3049 /* Calculate the digest value. */
3051 ALLOCA_MD5_CONTEXT (ctx);
3052 unsigned char hash[MD5_HASHLEN];
3053 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3054 char response_digest[MD5_HASHLEN * 2 + 1];
3056 /* A1BUF = H(user ":" realm ":" password) */
3058 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3059 gen_md5_update ((unsigned char *)":", 1, ctx);
3060 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3061 gen_md5_update ((unsigned char *)":", 1, ctx);
3062 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3063 gen_md5_finish (ctx, hash);
3064 dump_hash (a1buf, hash);
3066 /* A2BUF = H(method ":" path) */
3068 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3069 gen_md5_update ((unsigned char *)":", 1, ctx);
3070 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3071 gen_md5_finish (ctx, hash);
3072 dump_hash (a2buf, hash);
3074 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3076 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3077 gen_md5_update ((unsigned char *)":", 1, ctx);
3078 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3079 gen_md5_update ((unsigned char *)":", 1, ctx);
3080 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3081 gen_md5_finish (ctx, hash);
3082 dump_hash (response_digest, hash);
3084 res = xmalloc (strlen (user)
3089 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3090 + (opaque ? strlen (opaque) : 0)
3092 sprintf (res, "Digest \
3093 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3094 user, realm, nonce, path, response_digest);
3097 char *p = res + strlen (res);
3098 strcat (p, ", opaque=\"");
3105 #endif /* ENABLE_DIGEST */
3107 /* Computing the size of a string literal must take into account that
3108 value returned by sizeof includes the terminating \0. */
3109 #define STRSIZE(literal) (sizeof (literal) - 1)
3111 /* Whether chars in [b, e) begin with the literal string provided as
3112 first argument and are followed by whitespace or terminating \0.
3113 The comparison is case-insensitive. */
3114 #define STARTS(literal, b, e) \
3116 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3117 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3118 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3119 || c_isspace (b[STRSIZE (literal)])))
3122 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3124 return STARTS ("Basic", hdrbeg, hdrend)
3125 #ifdef ENABLE_DIGEST
3126 || STARTS ("Digest", hdrbeg, hdrend)
3129 || STARTS ("NTLM", hdrbeg, hdrend)
3136 /* Create the HTTP authorization request header. When the
3137 `WWW-Authenticate' response header is seen, according to the
3138 authorization scheme specified in that header (`Basic' and `Digest'
3139 are supported by the current implementation), produce an
3140 appropriate HTTP authorization request header. */
3142 create_authorization_line (const char *au, const char *user,
3143 const char *passwd, const char *method,
3144 const char *path, bool *finished)
3146 /* We are called only with known schemes, so we can dispatch on the
3148 switch (c_toupper (*au))
3150 case 'B': /* Basic */
3152 return basic_authentication_encode (user, passwd);
3153 #ifdef ENABLE_DIGEST
3154 case 'D': /* Digest */
3156 return digest_authentication_encode (au, user, passwd, method, path);
3159 case 'N': /* NTLM */
3160 if (!ntlm_input (&pconn.ntlm, au))
3165 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3168 /* We shouldn't get here -- this function should be only called
3169 with values approved by known_authentication_scheme_p. */
3177 if (!wget_cookie_jar)
3178 wget_cookie_jar = cookie_jar_new ();
3179 if (opt.cookies_input && !cookies_loaded_p)
3181 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3182 cookies_loaded_p = true;
3189 if (wget_cookie_jar)
3190 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3196 xfree_null (pconn.host);
3197 if (wget_cookie_jar)
3198 cookie_jar_delete (wget_cookie_jar);
3205 test_parse_content_disposition()
3210 char *opt_dir_prefix;
3214 { "filename=\"file.ext\"", NULL, "file.ext", true },
3215 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3216 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3217 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3218 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3219 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3220 { "attachment", NULL, NULL, false },
3221 { "attachment", "somedir", NULL, false },
3224 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3229 opt.dir_prefix = test_array[i].opt_dir_prefix;
3230 res = parse_content_disposition (test_array[i].hdrval, &filename);
3232 mu_assert ("test_parse_content_disposition: wrong result",
3233 res == test_array[i].result
3235 || 0 == strcmp (test_array[i].filename, filename)));
3241 #endif /* TESTING */
3244 * vim: et sts=2 sw=2 cino+={s