2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008, 2009 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"
71 #endif /* def __VMS */
73 extern char *version_string;
77 static char *create_authorization_line (const char *, const char *,
78 const char *, const char *,
79 const char *, bool *);
80 static char *basic_authentication_encode (const char *, const char *);
81 static bool known_authentication_scheme_p (const char *, const char *);
82 static void ensure_extension (struct http_stat *, const char *, int *);
83 static void load_cookies (void);
86 # define MIN(x, y) ((x) > (y) ? (y) : (x))
90 static bool cookies_loaded_p;
91 static struct cookie_jar *wget_cookie_jar;
93 #define TEXTHTML_S "text/html"
94 #define TEXTXHTML_S "application/xhtml+xml"
95 #define TEXTCSS_S "text/css"
97 /* Some status code validation macros: */
98 #define H_20X(x) (((x) >= 200) && ((x) < 300))
99 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
100 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
101 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
102 || (x) == HTTP_STATUS_SEE_OTHER \
103 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
105 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
106 /* Successful 2xx. */
107 #define HTTP_STATUS_OK 200
108 #define HTTP_STATUS_CREATED 201
109 #define HTTP_STATUS_ACCEPTED 202
110 #define HTTP_STATUS_NO_CONTENT 204
111 #define HTTP_STATUS_PARTIAL_CONTENTS 206
113 /* Redirection 3xx. */
114 #define HTTP_STATUS_MULTIPLE_CHOICES 300
115 #define HTTP_STATUS_MOVED_PERMANENTLY 301
116 #define HTTP_STATUS_MOVED_TEMPORARILY 302
117 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
118 #define HTTP_STATUS_NOT_MODIFIED 304
119 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
121 /* Client error 4xx. */
122 #define HTTP_STATUS_BAD_REQUEST 400
123 #define HTTP_STATUS_UNAUTHORIZED 401
124 #define HTTP_STATUS_FORBIDDEN 403
125 #define HTTP_STATUS_NOT_FOUND 404
126 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
128 /* Server errors 5xx. */
129 #define HTTP_STATUS_INTERNAL 500
130 #define HTTP_STATUS_NOT_IMPLEMENTED 501
131 #define HTTP_STATUS_BAD_GATEWAY 502
132 #define HTTP_STATUS_UNAVAILABLE 503
135 rel_none, rel_name, rel_value, rel_both
142 struct request_header {
144 enum rp release_policy;
146 int hcount, hcapacity;
151 /* Create a new, empty request. At least request_set_method must be
152 called before the request can be used. */
154 static struct request *
157 struct request *req = xnew0 (struct request);
159 req->headers = xnew_array (struct request_header, req->hcapacity);
163 /* Set the request's method and its arguments. METH should be a
164 literal string (or it should outlive the request) because it will
165 not be freed. ARG will be freed by request_free. */
168 request_set_method (struct request *req, const char *meth, char *arg)
174 /* Return the method string passed with the last call to
175 request_set_method. */
178 request_method (const struct request *req)
183 /* Free one header according to the release policy specified with
184 request_set_header. */
187 release_header (struct request_header *hdr)
189 switch (hdr->release_policy)
206 /* Set the request named NAME to VALUE. Specifically, this means that
207 a "NAME: VALUE\r\n" header line will be used in the request. If a
208 header with the same name previously existed in the request, its
209 value will be replaced by this one. A NULL value means do nothing.
211 RELEASE_POLICY determines whether NAME and VALUE should be released
212 (freed) with request_free. Allowed values are:
214 - rel_none - don't free NAME or VALUE
215 - rel_name - free NAME when done
216 - rel_value - free VALUE when done
217 - rel_both - free both NAME and VALUE when done
219 Setting release policy is useful when arguments come from different
220 sources. For example:
222 // Don't free literal strings!
223 request_set_header (req, "Pragma", "no-cache", rel_none);
225 // Don't free a global variable, we'll need it later.
226 request_set_header (req, "Referer", opt.referer, rel_none);
228 // Value freshly allocated, free it when done.
229 request_set_header (req, "Range",
230 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
235 request_set_header (struct request *req, char *name, char *value,
236 enum rp release_policy)
238 struct request_header *hdr;
243 /* A NULL value is a no-op; if freeing the name is requested,
244 free it now to avoid leaks. */
245 if (release_policy == rel_name || release_policy == rel_both)
250 for (i = 0; i < req->hcount; i++)
252 hdr = &req->headers[i];
253 if (0 == strcasecmp (name, hdr->name))
255 /* Replace existing header. */
256 release_header (hdr);
259 hdr->release_policy = release_policy;
264 /* Install new header. */
266 if (req->hcount >= req->hcapacity)
268 req->hcapacity <<= 1;
269 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
271 hdr = &req->headers[req->hcount++];
274 hdr->release_policy = release_policy;
277 /* Like request_set_header, but sets the whole header line, as
278 provided by the user using the `--header' option. For example,
279 request_set_user_header (req, "Foo: bar") works just like
280 request_set_header (req, "Foo", "bar"). */
283 request_set_user_header (struct request *req, const char *header)
286 const char *p = strchr (header, ':');
289 BOUNDED_TO_ALLOCA (header, p, name);
291 while (c_isspace (*p))
293 request_set_header (req, xstrdup (name), (char *) p, rel_name);
296 /* Remove the header with specified name from REQ. Returns true if
297 the header was actually removed, false otherwise. */
300 request_remove_header (struct request *req, char *name)
303 for (i = 0; i < req->hcount; i++)
305 struct request_header *hdr = &req->headers[i];
306 if (0 == strcasecmp (name, hdr->name))
308 release_header (hdr);
309 /* Move the remaining headers by one. */
310 if (i < req->hcount - 1)
311 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
319 #define APPEND(p, str) do { \
320 int A_len = strlen (str); \
321 memcpy (p, str, A_len); \
325 /* Construct the request and write it to FD using fd_write. */
328 request_send (const struct request *req, int fd)
330 char *request_string, *p;
331 int i, size, write_error;
333 /* Count the request size. */
336 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
337 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
339 for (i = 0; i < req->hcount; i++)
341 struct request_header *hdr = &req->headers[i];
342 /* NAME ": " VALUE "\r\n" */
343 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
349 p = request_string = alloca_array (char, size);
351 /* Generate the request. */
353 APPEND (p, req->method); *p++ = ' ';
354 APPEND (p, req->arg); *p++ = ' ';
355 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
357 for (i = 0; i < req->hcount; i++)
359 struct request_header *hdr = &req->headers[i];
360 APPEND (p, hdr->name);
361 *p++ = ':', *p++ = ' ';
362 APPEND (p, hdr->value);
363 *p++ = '\r', *p++ = '\n';
366 *p++ = '\r', *p++ = '\n', *p++ = '\0';
367 assert (p - request_string == size);
371 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
373 /* Send the request to the server. */
375 write_error = fd_write (fd, request_string, size - 1, -1);
377 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
382 /* Release the resources used by REQ. */
385 request_free (struct request *req)
388 xfree_null (req->arg);
389 for (i = 0; i < req->hcount; i++)
390 release_header (&req->headers[i]);
391 xfree_null (req->headers);
395 static struct hash_table *basic_authed_hosts;
397 /* Find out if this host has issued a Basic challenge yet; if so, give
398 * it the username, password. A temporary measure until we can get
399 * proper authentication in place. */
402 maybe_send_basic_creds (const char *hostname, const char *user,
403 const char *passwd, struct request *req)
405 bool do_challenge = false;
407 if (opt.auth_without_challenge)
409 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
412 else if (basic_authed_hosts
413 && hash_table_contains(basic_authed_hosts, hostname))
415 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
420 DEBUGP(("Host %s has not issued a general basic challenge.\n",
425 request_set_header (req, "Authorization",
426 basic_authentication_encode (user, passwd),
433 register_basic_auth_host (const char *hostname)
435 if (!basic_authed_hosts)
437 basic_authed_hosts = make_nocase_string_hash_table (1);
439 if (!hash_table_contains(basic_authed_hosts, hostname))
441 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
442 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
447 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
448 PROMISED_SIZE bytes are sent over the wire -- if the file is
449 longer, read only that much; if the file is shorter, report an error. */
452 post_file (int sock, const char *file_name, wgint promised_size)
454 static char chunk[8192];
459 DEBUGP (("[writing POST file %s ... ", file_name));
461 fp = fopen (file_name, "rb");
464 while (!feof (fp) && written < promised_size)
467 int length = fread (chunk, 1, sizeof (chunk), fp);
470 towrite = MIN (promised_size - written, length);
471 write_error = fd_write (sock, chunk, towrite, -1);
481 /* If we've written less than was promised, report a (probably
482 nonsensical) error rather than break the promise. */
483 if (written < promised_size)
489 assert (written == promised_size);
490 DEBUGP (("done]\n"));
494 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
495 If so, return the pointer to the position after the line, otherwise
496 return NULL. This is used as callback to fd_read_hunk. The data
497 between START and PEEKED has been read and cannot be "unread"; the
498 data after PEEKED has only been peeked. */
501 response_head_terminator (const char *start, const char *peeked, int peeklen)
505 /* If at first peek, verify whether HUNK starts with "HTTP". If
506 not, this is a HTTP/0.9 request and we must bail out without
508 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
511 /* Look for "\n[\r]\n", and return the following position if found.
512 Start two chars before the current to cover the possibility that
513 part of the terminator (e.g. "\n\r") arrived in the previous
515 p = peeked - start < 2 ? start : peeked - 2;
516 end = peeked + peeklen;
518 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
519 for (; p < end - 2; p++)
522 if (p[1] == '\r' && p[2] == '\n')
524 else if (p[1] == '\n')
527 /* p==end-2: check for \n\n directly preceding END. */
528 if (p[0] == '\n' && p[1] == '\n')
534 /* The maximum size of a single HTTP response we care to read. Rather
535 than being a limit of the reader implementation, this limit
536 prevents Wget from slurping all available memory upon encountering
537 malicious or buggy server output, thus protecting the user. Define
538 it to 0 to remove the limit. */
540 #define HTTP_RESPONSE_MAX_SIZE 65536
542 /* Read the HTTP request head from FD and return it. The error
543 conditions are the same as with fd_read_hunk.
545 To support HTTP/0.9 responses, this function tries to make sure
546 that the data begins with "HTTP". If this is not the case, no data
547 is read and an empty request is returned, so that the remaining
548 data can be treated as body. */
551 read_http_response_head (int fd)
553 return fd_read_hunk (fd, response_head_terminator, 512,
554 HTTP_RESPONSE_MAX_SIZE);
558 /* The response data. */
561 /* The array of pointers that indicate where each header starts.
562 For example, given this HTTP response:
569 The headers are located like this:
571 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
573 headers[0] headers[1] headers[2] headers[3]
575 I.e. headers[0] points to the beginning of the request,
576 headers[1] points to the end of the first header and the
577 beginning of the second one, etc. */
579 const char **headers;
582 /* Create a new response object from the text of the HTTP response,
583 available in HEAD. That text is automatically split into
584 constituent header lines for fast retrieval using
587 static struct response *
588 resp_new (const char *head)
593 struct response *resp = xnew0 (struct response);
598 /* Empty head means that we're dealing with a headerless
599 (HTTP/0.9) response. In that case, don't set HEADERS at
604 /* Split HEAD into header lines, so that resp_header_* functions
605 don't need to do this over and over again. */
611 DO_REALLOC (resp->headers, size, count + 1, const char *);
612 resp->headers[count++] = hdr;
614 /* Break upon encountering an empty line. */
615 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
618 /* Find the end of HDR, including continuations. */
621 const char *end = strchr (hdr, '\n');
627 while (*hdr == ' ' || *hdr == '\t');
629 DO_REALLOC (resp->headers, size, count + 1, const char *);
630 resp->headers[count] = NULL;
635 /* Locate the header named NAME in the request data, starting with
636 position START. This allows the code to loop through the request
637 data, filtering for all requests of a given name. Returns the
638 found position, or -1 for failure. The code that uses this
639 function typically looks like this:
641 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
642 ... do something with header ...
644 If you only care about one header, use resp_header_get instead of
648 resp_header_locate (const struct response *resp, const char *name, int start,
649 const char **begptr, const char **endptr)
652 const char **headers = resp->headers;
655 if (!headers || !headers[1])
658 name_len = strlen (name);
664 for (; headers[i + 1]; i++)
666 const char *b = headers[i];
667 const char *e = headers[i + 1];
669 && b[name_len] == ':'
670 && 0 == strncasecmp (b, name, name_len))
673 while (b < e && c_isspace (*b))
675 while (b < e && c_isspace (e[-1]))
685 /* Find and retrieve the header named NAME in the request data. If
686 found, set *BEGPTR to its starting, and *ENDPTR to its ending
687 position, and return true. Otherwise return false.
689 This function is used as a building block for resp_header_copy
690 and resp_header_strdup. */
693 resp_header_get (const struct response *resp, const char *name,
694 const char **begptr, const char **endptr)
696 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
700 /* Copy the response header named NAME to buffer BUF, no longer than
701 BUFSIZE (BUFSIZE includes the terminating 0). If the header
702 exists, true is returned, false otherwise. If there should be no
703 limit on the size of the header, use resp_header_strdup instead.
705 If BUFSIZE is 0, no data is copied, but the boolean indication of
706 whether the header is present is still returned. */
709 resp_header_copy (const struct response *resp, const char *name,
710 char *buf, int bufsize)
713 if (!resp_header_get (resp, name, &b, &e))
717 int len = MIN (e - b, bufsize - 1);
718 memcpy (buf, b, len);
724 /* Return the value of header named NAME in RESP, allocated with
725 malloc. If such a header does not exist in RESP, return NULL. */
728 resp_header_strdup (const struct response *resp, const char *name)
731 if (!resp_header_get (resp, name, &b, &e))
733 return strdupdelim (b, e);
736 /* Parse the HTTP status line, which is of format:
738 HTTP-Version SP Status-Code SP Reason-Phrase
740 The function returns the status-code, or -1 if the status line
741 appears malformed. The pointer to "reason-phrase" message is
742 returned in *MESSAGE. */
745 resp_status (const struct response *resp, char **message)
752 /* For a HTTP/0.9 response, assume status 200. */
754 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
758 p = resp->headers[0];
759 end = resp->headers[1];
765 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
769 /* Match the HTTP version. This is optional because Gnutella
770 servers have been reported to not specify HTTP version. */
771 if (p < end && *p == '/')
774 while (p < end && c_isdigit (*p))
776 if (p < end && *p == '.')
778 while (p < end && c_isdigit (*p))
782 while (p < end && c_isspace (*p))
784 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
787 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
792 while (p < end && c_isspace (*p))
794 while (p < end && c_isspace (end[-1]))
796 *message = strdupdelim (p, end);
802 /* Release the resources used by RESP. */
805 resp_free (struct response *resp)
807 xfree_null (resp->headers);
811 /* Print a single line of response, the characters [b, e). We tried
813 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
814 but that failed to escape the non-printable characters and, in fact,
815 caused crashes in UTF-8 locales. */
818 print_response_line(const char *prefix, const char *b, const char *e)
821 BOUNDED_TO_ALLOCA(b, e, copy);
822 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
823 quotearg_style (escape_quoting_style, copy));
826 /* Print the server response, line by line, omitting the trailing CRLF
827 from individual header lines, and prefixed with PREFIX. */
830 print_server_response (const struct response *resp, const char *prefix)
835 for (i = 0; resp->headers[i + 1]; i++)
837 const char *b = resp->headers[i];
838 const char *e = resp->headers[i + 1];
840 if (b < e && e[-1] == '\n')
842 if (b < e && e[-1] == '\r')
844 print_response_line(prefix, b, e);
848 /* Parse the `Content-Range' header and extract the information it
849 contains. Returns true if successful, false otherwise. */
851 parse_content_range (const char *hdr, wgint *first_byte_ptr,
852 wgint *last_byte_ptr, wgint *entity_length_ptr)
856 /* Ancient versions of Netscape proxy server, presumably predating
857 rfc2068, sent out `Content-Range' without the "bytes"
859 if (0 == strncasecmp (hdr, "bytes", 5))
862 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
866 while (c_isspace (*hdr))
871 if (!c_isdigit (*hdr))
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
877 *first_byte_ptr = num;
879 for (num = 0; c_isdigit (*hdr); hdr++)
880 num = 10 * num + (*hdr - '0');
881 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
883 *last_byte_ptr = num;
888 for (num = 0; c_isdigit (*hdr); hdr++)
889 num = 10 * num + (*hdr - '0');
890 *entity_length_ptr = num;
894 /* Read the body of the request, but don't store it anywhere and don't
895 display a progress gauge. This is useful for reading the bodies of
896 administrative responses to which we will soon issue another
897 request. The response is not useful to the user, but reading it
898 allows us to continue using the same connection to the server.
900 If reading fails, false is returned, true otherwise. In debug
901 mode, the body is displayed for debugging purposes. */
904 skip_short_body (int fd, wgint contlen)
907 SKIP_SIZE = 512, /* size of the download buffer */
908 SKIP_THRESHOLD = 4096 /* the largest size we read */
910 char dlbuf[SKIP_SIZE + 1];
911 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
913 /* We shouldn't get here with unknown contlen. (This will change
914 with HTTP/1.1, which supports "chunked" transfer.) */
915 assert (contlen != -1);
917 /* If the body is too large, it makes more sense to simply close the
918 connection than to try to read the body. */
919 if (contlen > SKIP_THRESHOLD)
922 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
926 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
929 /* Don't normally report the error since this is an
930 optimization that should be invisible to the user. */
931 DEBUGP (("] aborting (%s).\n",
932 ret < 0 ? fd_errstr (fd) : "EOF received"));
936 /* Safe even if %.*s bogusly expects terminating \0 because
937 we've zero-terminated dlbuf above. */
938 DEBUGP (("%.*s", ret, dlbuf));
941 DEBUGP (("] done.\n"));
945 /* Extract a parameter from the string (typically an HTTP header) at
946 **SOURCE and advance SOURCE to the next parameter. Return false
947 when there are no more parameters to extract. The name of the
948 parameter is returned in NAME, and the value in VALUE. If the
949 parameter has no value, the token's value is zeroed out.
951 For example, if *SOURCE points to the string "attachment;
952 filename=\"foo bar\"", the first call to this function will return
953 the token named "attachment" and no value, and the second call will
954 return the token named "filename" and value "foo bar". The third
955 call will return false, indicating no more valid tokens. */
958 extract_param (const char **source, param_token *name, param_token *value,
961 const char *p = *source;
963 while (c_isspace (*p)) ++p;
967 return false; /* no error; nothing more to extract */
972 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
974 if (name->b == name->e)
975 return false; /* empty name: error */
976 while (c_isspace (*p)) ++p;
977 if (*p == separator || !*p) /* no value */
980 if (*p == separator) ++p;
985 return false; /* error */
987 /* *p is '=', extract value */
989 while (c_isspace (*p)) ++p;
990 if (*p == '"') /* quoted */
993 while (*p && *p != '"') ++p;
997 /* Currently at closing quote; find the end of param. */
998 while (c_isspace (*p)) ++p;
999 while (*p && *p != separator) ++p;
1000 if (*p == separator)
1003 /* garbage after closed quote, e.g. foo="bar"baz */
1009 while (*p && *p != separator) ++p;
1011 while (value->e != value->b && c_isspace (value->e[-1]))
1013 if (*p == separator) ++p;
1020 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1022 /* Parse the contents of the `Content-Disposition' header, extracting
1023 the information useful to Wget. Content-Disposition is a header
1024 borrowed from MIME; when used in HTTP, it typically serves for
1025 specifying the desired file name of the resource. For example:
1027 Content-Disposition: attachment; filename="flora.jpg"
1029 Wget will skip the tokens it doesn't care about, such as
1030 "attachment" in the previous example; it will also skip other
1031 unrecognized params. If the header is syntactically correct and
1032 contains a file name, a copy of the file name is stored in
1033 *filename and true is returned. Otherwise, the function returns
1036 The file name is stripped of directory components and must not be
1040 parse_content_disposition (const char *hdr, char **filename)
1042 param_token name, value;
1043 while (extract_param (&hdr, &name, &value, ';'))
1044 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1046 /* Make the file name begin at the last slash or backslash. */
1047 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1048 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1049 if (last_slash && last_bs)
1050 value.b = 1 + MAX (last_slash, last_bs);
1051 else if (last_slash || last_bs)
1052 value.b = 1 + (last_slash ? last_slash : last_bs);
1053 if (value.b == value.e)
1055 /* Start with the directory prefix, if specified. */
1058 int prefix_length = strlen (opt.dir_prefix);
1059 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1064 total_length = prefix_length + (value.e - value.b);
1065 *filename = xmalloc (total_length + 1);
1066 strcpy (*filename, opt.dir_prefix);
1068 (*filename)[prefix_length - 1] = '/';
1069 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1070 (*filename)[total_length] = '\0';
1073 *filename = strdupdelim (value.b, value.e);
1079 /* Persistent connections. Currently, we cache the most recently used
1080 connection as persistent, provided that the HTTP server agrees to
1081 make it such. The persistence data is stored in the variables
1082 below. Ideally, it should be possible to cache an arbitrary fixed
1083 number of these connections. */
1085 /* Whether a persistent connection is active. */
1086 static bool pconn_active;
1089 /* The socket of the connection. */
1092 /* Host and port of the currently active persistent connection. */
1096 /* Whether a ssl handshake has occoured on this connection. */
1099 /* Whether the connection was authorized. This is only done by
1100 NTLM, which authorizes *connections* rather than individual
1101 requests. (That practice is peculiar for HTTP, but it is a
1102 useful optimization.) */
1106 /* NTLM data of the current connection. */
1107 struct ntlmdata ntlm;
1111 /* Mark the persistent connection as invalid and free the resources it
1112 uses. This is used by the CLOSE_* macros after they forcefully
1113 close a registered persistent connection. */
1116 invalidate_persistent (void)
1118 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1119 pconn_active = false;
1120 fd_close (pconn.socket);
1125 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1126 persistent. This will enable someone to use the same connection
1127 later. In the context of HTTP, this must be called only AFTER the
1128 response has been received and the server has promised that the
1129 connection will remain alive.
1131 If a previous connection was persistent, it is closed. */
1134 register_persistent (const char *host, int port, int fd, bool ssl)
1138 if (pconn.socket == fd)
1140 /* The connection FD is already registered. */
1145 /* The old persistent connection is still active; close it
1146 first. This situation arises whenever a persistent
1147 connection exists, but we then connect to a different
1148 host, and try to register a persistent connection to that
1150 invalidate_persistent ();
1154 pconn_active = true;
1156 pconn.host = xstrdup (host);
1159 pconn.authorized = false;
1161 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1164 /* Return true if a persistent connection is available for connecting
1168 persistent_available_p (const char *host, int port, bool ssl,
1169 bool *host_lookup_failed)
1171 /* First, check whether a persistent connection is active at all. */
1175 /* If we want SSL and the last connection wasn't or vice versa,
1176 don't use it. Checking for host and port is not enough because
1177 HTTP and HTTPS can apparently coexist on the same port. */
1178 if (ssl != pconn.ssl)
1181 /* If we're not connecting to the same port, we're not interested. */
1182 if (port != pconn.port)
1185 /* If the host is the same, we're in business. If not, there is
1186 still hope -- read below. */
1187 if (0 != strcasecmp (host, pconn.host))
1189 /* Check if pconn.socket is talking to HOST under another name.
1190 This happens often when both sites are virtual hosts
1191 distinguished only by name and served by the same network
1192 interface, and hence the same web server (possibly set up by
1193 the ISP and serving many different web sites). This
1194 admittedly unconventional optimization does not contradict
1195 HTTP and works well with popular server software. */
1199 struct address_list *al;
1202 /* Don't try to talk to two different SSL sites over the same
1203 secure connection! (Besides, it's not clear that
1204 name-based virtual hosting is even possible with SSL.) */
1207 /* If pconn.socket's peer is one of the IP addresses HOST
1208 resolves to, pconn.socket is for all intents and purposes
1209 already talking to HOST. */
1211 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1213 /* Can't get the peer's address -- something must be very
1214 wrong with the connection. */
1215 invalidate_persistent ();
1218 al = lookup_host (host, 0);
1221 *host_lookup_failed = true;
1225 found = address_list_contains (al, &ip);
1226 address_list_release (al);
1231 /* The persistent connection's peer address was found among the
1232 addresses HOST resolved to; therefore, pconn.sock is in fact
1233 already talking to HOST -- no need to reconnect. */
1236 /* Finally, check whether the connection is still open. This is
1237 important because most servers implement liberal (short) timeout
1238 on persistent connections. Wget can of course always reconnect
1239 if the connection doesn't work out, but it's nicer to know in
1240 advance. This test is a logical followup of the first test, but
1241 is "expensive" and therefore placed at the end of the list.
1243 (Current implementation of test_socket_open has a nice side
1244 effect that it treats sockets with pending data as "closed".
1245 This is exactly what we want: if a broken server sends message
1246 body in response to HEAD, or if it sends more than conent-length
1247 data, we won't reuse the corrupted connection.) */
1249 if (!test_socket_open (pconn.socket))
1251 /* Oops, the socket is no longer open. Now that we know that,
1252 let's invalidate the persistent connection before returning
1254 invalidate_persistent ();
1261 /* The idea behind these two CLOSE macros is to distinguish between
1262 two cases: one when the job we've been doing is finished, and we
1263 want to close the connection and leave, and two when something is
1264 seriously wrong and we're closing the connection as part of
1267 In case of keep_alive, CLOSE_FINISH should leave the connection
1268 open, while CLOSE_INVALIDATE should still close it.
1270 Note that the semantics of the flag `keep_alive' is "this
1271 connection *will* be reused (the server has promised not to close
1272 the connection once we're done)", while the semantics of
1273 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1274 active, registered connection". */
1276 #define CLOSE_FINISH(fd) do { \
1279 if (pconn_active && (fd) == pconn.socket) \
1280 invalidate_persistent (); \
1289 #define CLOSE_INVALIDATE(fd) do { \
1290 if (pconn_active && (fd) == pconn.socket) \
1291 invalidate_persistent (); \
1299 wgint len; /* received length */
1300 wgint contlen; /* expected length */
1301 wgint restval; /* the restart value */
1302 int res; /* the result of last read */
1303 char *rderrmsg; /* error message from read error */
1304 char *newloc; /* new location (redirection) */
1305 char *remote_time; /* remote time-stamp string */
1306 char *error; /* textual HTTP error */
1307 int statcode; /* status code */
1308 char *message; /* status message */
1309 wgint rd_size; /* amount of data read from socket */
1310 double dltime; /* time it took to download the data */
1311 const char *referer; /* value of the referer header. */
1312 char *local_file; /* local file name. */
1313 bool existence_checked; /* true if we already checked for a file's
1314 existence after having begun to download
1315 (needed in gethttp for when connection is
1316 interrupted/restarted. */
1317 bool timestamp_checked; /* true if pre-download time-stamping checks
1318 * have already been performed */
1319 char *orig_file_name; /* name of file to compare for time-stamping
1320 * (might be != local_file if -K is set) */
1321 wgint orig_file_size; /* size of file to compare for time-stamping */
1322 time_t orig_file_tstamp; /* time-stamp of file to compare for
1327 free_hstat (struct http_stat *hs)
1329 xfree_null (hs->newloc);
1330 xfree_null (hs->remote_time);
1331 xfree_null (hs->error);
1332 xfree_null (hs->rderrmsg);
1333 xfree_null (hs->local_file);
1334 xfree_null (hs->orig_file_name);
1335 xfree_null (hs->message);
1337 /* Guard against being called twice. */
1339 hs->remote_time = NULL;
1343 #define BEGINS_WITH(line, string_constant) \
1344 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1345 && (c_isspace (line[sizeof (string_constant) - 1]) \
1346 || !line[sizeof (string_constant) - 1]))
1349 #define SET_USER_AGENT(req) do { \
1350 if (!opt.useragent) \
1351 request_set_header (req, "User-Agent", \
1352 aprintf ("Wget/%s (VMS %s %s)", \
1353 version_string, vms_arch(), vms_vers()), \
1355 else if (*opt.useragent) \
1356 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1358 #else /* def __VMS */
1359 #define SET_USER_AGENT(req) do { \
1360 if (!opt.useragent) \
1361 request_set_header (req, "User-Agent", \
1362 aprintf ("Wget/%s (%s)", \
1363 version_string, OS_TYPE), \
1365 else if (*opt.useragent) \
1366 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1368 #endif /* def __VMS [else] */
1370 /* The flags that allow clobbering the file (opening with "wb").
1371 Defined here to avoid repetition later. #### This will require
1373 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1374 || opt.dirstruct || opt.output_document)
1376 /* Retrieve a document through HTTP protocol. It recognizes status
1377 code, and correctly handles redirections. It closes the network
1378 socket. If it receives an error from the functions below it, it
1379 will print it if there is enough information to do so (almost
1380 always), returning the error to the caller (i.e. http_loop).
1382 Various HTTP parameters are stored to hs.
1384 If PROXY is non-NULL, the connection will be made to the proxy
1385 server, and u->url will be requested. */
1387 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1390 struct request *req;
1393 char *user, *passwd;
1397 wgint contlen, contrange;
1404 /* Set to 1 when the authorization has already been sent and should
1405 not be tried again. */
1406 bool auth_finished = false;
1408 /* Set to 1 when just globally-set Basic authorization has been sent;
1409 * should prevent further Basic negotiations, but not other
1411 bool basic_auth_finished = false;
1413 /* Whether NTLM authentication is used for this request. */
1414 bool ntlm_seen = false;
1416 /* Whether our connection to the remote host is through SSL. */
1417 bool using_ssl = false;
1419 /* Whether a HEAD request will be issued (as opposed to GET or
1421 bool head_only = !!(*dt & HEAD_ONLY);
1424 struct response *resp;
1428 /* Whether this connection will be kept alive after the HTTP request
1432 /* Whether keep-alive should be inhibited.
1434 RFC 2068 requests that 1.0 clients not send keep-alive requests
1435 to proxies. This is because many 1.0 proxies do not interpret
1436 the Connection header and transfer it to the remote server,
1437 causing it to not close the connection and leave both the proxy
1438 and the client hanging. */
1439 bool inhibit_keep_alive =
1440 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1442 /* Headers sent when using POST. */
1443 wgint post_data_size = 0;
1445 bool host_lookup_failed = false;
1448 if (u->scheme == SCHEME_HTTPS)
1450 /* Initialize the SSL context. After this has once been done,
1451 it becomes a no-op. */
1454 scheme_disable (SCHEME_HTTPS);
1455 logprintf (LOG_NOTQUIET,
1456 _("Disabling SSL due to encountered errors.\n"));
1457 return SSLINITFAILED;
1460 #endif /* HAVE_SSL */
1462 /* Initialize certain elements of struct http_stat. */
1466 hs->rderrmsg = NULL;
1468 hs->remote_time = NULL;
1474 /* Prepare the request to send. */
1476 req = request_new ();
1479 const char *meth = "GET";
1482 else if (opt.post_file_name || opt.post_data)
1484 /* Use the full path, i.e. one that includes the leading slash and
1485 the query string. E.g. if u->path is "foo/bar" and u->query is
1486 "param=value", full_path will be "/foo/bar?param=value". */
1489 /* When using SSL over proxy, CONNECT establishes a direct
1490 connection to the HTTPS server. Therefore use the same
1491 argument as when talking to the server directly. */
1492 && u->scheme != SCHEME_HTTPS
1495 meth_arg = xstrdup (u->url);
1497 meth_arg = url_full_path (u);
1498 request_set_method (req, meth, meth_arg);
1501 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1502 if (*dt & SEND_NOCACHE)
1503 request_set_header (req, "Pragma", "no-cache", rel_none);
1505 request_set_header (req, "Range",
1506 aprintf ("bytes=%s-",
1507 number_to_static_string (hs->restval)),
1509 SET_USER_AGENT (req);
1510 request_set_header (req, "Accept", "*/*", rel_none);
1512 /* Find the username and password for authentication. */
1515 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1516 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1517 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1519 /* We only do "site-wide" authentication with "global" user/password
1520 * values unless --auth-no-challange has been requested; URL user/password
1521 * info overrides. */
1522 if (user && passwd && (!u->user || opt.auth_without_challenge))
1524 /* If this is a host for which we've already received a Basic
1525 * challenge, we'll go ahead and send Basic authentication creds. */
1526 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1529 /* Generate the Host header, HOST:PORT. Take into account that:
1531 - Broken server-side software often doesn't recognize the PORT
1532 argument, so we must generate "Host: www.server.com" instead of
1533 "Host: www.server.com:80" (and likewise for https port).
1535 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1536 becomes ambiguous and needs to be rewritten as "Host:
1537 [3ffe:8100:200:2::2]:1234". */
1539 /* Formats arranged for hfmt[add_port][add_squares]. */
1540 static const char *hfmt[][2] = {
1541 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1543 int add_port = u->port != scheme_default_port (u->scheme);
1544 int add_squares = strchr (u->host, ':') != NULL;
1545 request_set_header (req, "Host",
1546 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1550 if (!inhibit_keep_alive)
1551 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1554 request_set_header (req, "Cookie",
1555 cookie_header (wget_cookie_jar,
1556 u->host, u->port, u->path,
1558 u->scheme == SCHEME_HTTPS
1565 if (opt.post_data || opt.post_file_name)
1567 request_set_header (req, "Content-Type",
1568 "application/x-www-form-urlencoded", rel_none);
1570 post_data_size = strlen (opt.post_data);
1573 post_data_size = file_size (opt.post_file_name);
1574 if (post_data_size == -1)
1576 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1577 quote (opt.post_file_name), strerror (errno));
1581 request_set_header (req, "Content-Length",
1582 xstrdup (number_to_static_string (post_data_size)),
1586 /* Add the user headers. */
1587 if (opt.user_headers)
1590 for (i = 0; opt.user_headers[i]; i++)
1591 request_set_user_header (req, opt.user_headers[i]);
1595 /* We need to come back here when the initial attempt to retrieve
1596 without authorization header fails. (Expected to happen at least
1597 for the Digest authorization scheme.) */
1602 char *proxy_user, *proxy_passwd;
1603 /* For normal username and password, URL components override
1604 command-line/wgetrc parameters. With proxy
1605 authentication, it's the reverse, because proxy URLs are
1606 normally the "permanent" ones, so command-line args
1607 should take precedence. */
1608 if (opt.proxy_user && opt.proxy_passwd)
1610 proxy_user = opt.proxy_user;
1611 proxy_passwd = opt.proxy_passwd;
1615 proxy_user = proxy->user;
1616 proxy_passwd = proxy->passwd;
1618 /* #### This does not appear right. Can't the proxy request,
1619 say, `Digest' authentication? */
1620 if (proxy_user && proxy_passwd)
1621 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1623 /* If we're using a proxy, we will be connecting to the proxy
1627 /* Proxy authorization over SSL is handled below. */
1629 if (u->scheme != SCHEME_HTTPS)
1631 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1636 /* Establish the connection. */
1638 if (!inhibit_keep_alive)
1640 /* Look for a persistent connection to target host, unless a
1641 proxy is used. The exception is when SSL is in use, in which
1642 case the proxy is nothing but a passthrough to the target
1643 host, registered as a connection to the latter. */
1644 struct url *relevant = conn;
1646 if (u->scheme == SCHEME_HTTPS)
1650 if (persistent_available_p (relevant->host, relevant->port,
1652 relevant->scheme == SCHEME_HTTPS,
1656 &host_lookup_failed))
1658 sock = pconn.socket;
1659 using_ssl = pconn.ssl;
1660 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1661 quotearg_style (escape_quoting_style, pconn.host),
1663 DEBUGP (("Reusing fd %d.\n", sock));
1664 if (pconn.authorized)
1665 /* If the connection is already authorized, the "Basic"
1666 authorization added by code above is unnecessary and
1668 request_remove_header (req, "Authorization");
1670 else if (host_lookup_failed)
1673 logprintf(LOG_NOTQUIET,
1674 _("%s: unable to resolve host address %s\n"),
1675 exec_name, quote (relevant->host));
1682 sock = connect_to_host (conn->host, conn->port);
1691 return (retryable_socket_connect_error (errno)
1692 ? CONERROR : CONIMPOSSIBLE);
1696 if (proxy && u->scheme == SCHEME_HTTPS)
1698 /* When requesting SSL URLs through proxies, use the
1699 CONNECT method to request passthrough. */
1700 struct request *connreq = request_new ();
1701 request_set_method (connreq, "CONNECT",
1702 aprintf ("%s:%d", u->host, u->port));
1703 SET_USER_AGENT (connreq);
1706 request_set_header (connreq, "Proxy-Authorization",
1707 proxyauth, rel_value);
1708 /* Now that PROXYAUTH is part of the CONNECT request,
1709 zero it out so we don't send proxy authorization with
1710 the regular request below. */
1713 /* Examples in rfc2817 use the Host header in CONNECT
1714 requests. I don't see how that gains anything, given
1715 that the contents of Host would be exactly the same as
1716 the contents of CONNECT. */
1718 write_error = request_send (connreq, sock);
1719 request_free (connreq);
1720 if (write_error < 0)
1722 CLOSE_INVALIDATE (sock);
1726 head = read_http_response_head (sock);
1729 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1731 CLOSE_INVALIDATE (sock);
1740 DEBUGP (("proxy responded with: [%s]\n", head));
1742 resp = resp_new (head);
1743 statcode = resp_status (resp, &message);
1744 hs->message = xstrdup (message);
1747 if (statcode != 200)
1750 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1751 message ? quotearg_style (escape_quoting_style, message) : "?");
1752 xfree_null (message);
1755 xfree_null (message);
1757 /* SOCK is now *really* connected to u->host, so update CONN
1758 to reflect this. That way register_persistent will
1759 register SOCK as being connected to u->host:u->port. */
1763 if (conn->scheme == SCHEME_HTTPS)
1765 if (!ssl_connect_wget (sock))
1770 else if (!ssl_check_certificate (sock, u->host))
1773 return VERIFCERTERR;
1777 #endif /* HAVE_SSL */
1780 /* Send the request to server. */
1781 write_error = request_send (req, sock);
1783 if (write_error >= 0)
1787 DEBUGP (("[POST data: %s]\n", opt.post_data));
1788 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1790 else if (opt.post_file_name && post_data_size != 0)
1791 write_error = post_file (sock, opt.post_file_name, post_data_size);
1794 if (write_error < 0)
1796 CLOSE_INVALIDATE (sock);
1800 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1801 proxy ? "Proxy" : "HTTP");
1806 head = read_http_response_head (sock);
1811 logputs (LOG_NOTQUIET, _("No data received.\n"));
1812 CLOSE_INVALIDATE (sock);
1818 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1820 CLOSE_INVALIDATE (sock);
1825 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1827 resp = resp_new (head);
1829 /* Check for status line. */
1831 statcode = resp_status (resp, &message);
1832 hs->message = xstrdup (message);
1833 if (!opt.server_response)
1834 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1835 message ? quotearg_style (escape_quoting_style, message) : "");
1838 logprintf (LOG_VERBOSE, "\n");
1839 print_server_response (resp, " ");
1842 if (!opt.ignore_length
1843 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1847 parsed = str_to_wgint (hdrval, NULL, 10);
1848 if (parsed == WGINT_MAX && errno == ERANGE)
1851 #### If Content-Length is out of range, it most likely
1852 means that the file is larger than 2G and that we're
1853 compiled without LFS. In that case we should probably
1854 refuse to even attempt to download the file. */
1857 else if (parsed < 0)
1859 /* Negative Content-Length; nonsensical, so we can't
1860 assume any information about the content to receive. */
1867 /* Check for keep-alive related responses. */
1868 if (!inhibit_keep_alive && contlen != -1)
1870 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1872 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1874 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1879 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1883 const char *scbeg, *scend;
1884 /* The jar should have been created by now. */
1885 assert (wget_cookie_jar != NULL);
1887 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1888 &scbeg, &scend)) != -1;
1891 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1892 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1893 u->path, set_cookie);
1898 /* The server has promised that it will not close the connection
1899 when we're done. This means that we can register it. */
1900 register_persistent (conn->host, conn->port, sock, using_ssl);
1902 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1904 /* Authorization is required. */
1905 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1906 CLOSE_FINISH (sock);
1908 CLOSE_INVALIDATE (sock);
1909 pconn.authorized = false;
1910 if (!auth_finished && (user && passwd))
1912 /* IIS sends multiple copies of WWW-Authenticate, one with
1913 the value "negotiate", and other(s) with data. Loop over
1914 all the occurrences and pick the one we recognize. */
1916 const char *wabeg, *waend;
1917 char *www_authenticate = NULL;
1919 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1920 &wabeg, &waend)) != -1;
1922 if (known_authentication_scheme_p (wabeg, waend))
1924 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1928 if (!www_authenticate)
1930 /* If the authentication header is missing or
1931 unrecognized, there's no sense in retrying. */
1932 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1934 else if (!basic_auth_finished
1935 || !BEGINS_WITH (www_authenticate, "Basic"))
1938 pth = url_full_path (u);
1939 request_set_header (req, "Authorization",
1940 create_authorization_line (www_authenticate,
1942 request_method (req),
1946 if (BEGINS_WITH (www_authenticate, "NTLM"))
1948 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1950 /* Need to register this host as using basic auth,
1951 * so we automatically send creds next time. */
1952 register_basic_auth_host (u->host);
1955 xfree_null (message);
1958 goto retry_with_auth;
1962 /* We already did Basic auth, and it failed. Gotta
1966 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1968 xfree_null (message);
1973 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1975 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1977 pconn.authorized = true;
1980 /* Determine the local filename if needed. Notice that if -O is used
1981 * hstat.local_file is set by http_loop to the argument of -O. */
1982 if (!hs->local_file)
1984 /* Honor Content-Disposition whether possible. */
1985 if (!opt.content_disposition
1986 || !resp_header_copy (resp, "Content-Disposition",
1987 hdrval, sizeof (hdrval))
1988 || !parse_content_disposition (hdrval, &hs->local_file))
1990 /* The Content-Disposition header is missing or broken.
1991 * Choose unique file name according to given URL. */
1992 hs->local_file = url_file_name (u);
1996 /* TODO: perform this check only once. */
1997 if (!hs->existence_checked && file_exists_p (hs->local_file))
1999 if (opt.noclobber && !opt.output_document)
2001 /* If opt.noclobber is turned on and file already exists, do not
2002 retrieve the file. But if the output_document was given, then this
2003 test was already done and the file didn't exist. Hence the !opt.output_document */
2004 logprintf (LOG_VERBOSE, _("\
2005 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2006 /* If the file is there, we suppose it's retrieved OK. */
2009 /* #### Bogusness alert. */
2010 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2011 if (has_html_suffix_p (hs->local_file))
2015 xfree_null (message);
2016 return RETRUNNEEDED;
2018 else if (!ALLOW_CLOBBER)
2020 char *unique = unique_name (hs->local_file, true);
2021 if (unique != hs->local_file)
2022 xfree (hs->local_file);
2023 hs->local_file = unique;
2026 hs->existence_checked = true;
2028 /* Support timestamping */
2029 /* TODO: move this code out of gethttp. */
2030 if (opt.timestamping && !hs->timestamp_checked)
2032 size_t filename_len = strlen (hs->local_file);
2033 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2034 bool local_dot_orig_file_exists = false;
2035 char *local_filename = NULL;
2038 if (opt.backup_converted)
2039 /* If -K is specified, we'll act on the assumption that it was specified
2040 last time these files were downloaded as well, and instead of just
2041 comparing local file X against server file X, we'll compare local
2042 file X.orig (if extant, else X) against server file X. If -K
2043 _wasn't_ specified last time, or the server contains files called
2044 *.orig, -N will be back to not operating correctly with -k. */
2046 /* Would a single s[n]printf() call be faster? --dan
2048 Definitely not. sprintf() is horribly slow. It's a
2049 different question whether the difference between the two
2050 affects a program. Usually I'd say "no", but at one
2051 point I profiled Wget, and found that a measurable and
2052 non-negligible amount of time was lost calling sprintf()
2053 in url.c. Replacing sprintf with inline calls to
2054 strcpy() and number_to_string() made a difference.
2056 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2057 memcpy (filename_plus_orig_suffix + filename_len,
2058 ORIG_SFX, sizeof (ORIG_SFX));
2060 /* Try to stat() the .orig file. */
2061 if (stat (filename_plus_orig_suffix, &st) == 0)
2063 local_dot_orig_file_exists = true;
2064 local_filename = filename_plus_orig_suffix;
2068 if (!local_dot_orig_file_exists)
2069 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2070 if (stat (hs->local_file, &st) == 0)
2071 local_filename = hs->local_file;
2073 if (local_filename != NULL)
2074 /* There was a local file, so we'll check later to see if the version
2075 the server has is the same version we already have, allowing us to
2078 hs->orig_file_name = xstrdup (local_filename);
2079 hs->orig_file_size = st.st_size;
2080 hs->orig_file_tstamp = st.st_mtime;
2082 /* Modification time granularity is 2 seconds for Windows, so
2083 increase local time by 1 second for later comparison. */
2084 ++hs->orig_file_tstamp;
2091 hs->statcode = statcode;
2093 hs->error = xstrdup (_("Malformed status line"));
2095 hs->error = xstrdup (_("(no description)"));
2097 hs->error = xstrdup (message);
2098 xfree_null (message);
2100 type = resp_header_strdup (resp, "Content-Type");
2103 char *tmp = strchr (type, ';');
2106 /* sXXXav: only needed if IRI support is enabled */
2107 char *tmp2 = tmp + 1;
2109 while (tmp > type && c_isspace (tmp[-1]))
2113 /* Try to get remote encoding if needed */
2114 if (opt.enable_iri && !opt.encoding_remote)
2116 tmp = parse_charset (tmp2);
2118 set_content_encoding (iri, tmp);
2122 hs->newloc = resp_header_strdup (resp, "Location");
2123 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2125 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2127 wgint first_byte_pos, last_byte_pos, entity_length;
2128 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2131 contrange = first_byte_pos;
2132 contlen = last_byte_pos - first_byte_pos + 1;
2137 /* 20x responses are counted among successful by default. */
2138 if (H_20X (statcode))
2141 /* Return if redirected. */
2142 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2144 /* RFC2068 says that in case of the 300 (multiple choices)
2145 response, the server can output a preferred URL through
2146 `Location' header; otherwise, the request should be treated
2147 like GET. So, if the location is set, it will be a
2148 redirection; otherwise, just proceed normally. */
2149 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2153 logprintf (LOG_VERBOSE,
2154 _("Location: %s%s\n"),
2155 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2156 hs->newloc ? _(" [following]") : "");
2157 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2158 CLOSE_FINISH (sock);
2160 CLOSE_INVALIDATE (sock);
2167 /* If content-type is not given, assume text/html. This is because
2168 of the multitude of broken CGI's that "forget" to generate the
2171 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2172 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2178 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2183 if (opt.adjust_extension)
2186 /* -E / --adjust-extension / adjust_extension = on was specified,
2187 and this is a text/html file. If some case-insensitive
2188 variation on ".htm[l]" isn't already the file's suffix,
2191 ensure_extension (hs, ".html", dt);
2193 else if (*dt & TEXTCSS)
2195 ensure_extension (hs, ".css", dt);
2199 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2200 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2201 && contrange == 0 && hs->restval >= contlen)
2204 /* If `-c' is in use and the file has been fully downloaded (or
2205 the remote file has shrunk), Wget effectively requests bytes
2206 after the end of file and the server response with 416
2207 (or 200 with a <= Content-Length. */
2208 logputs (LOG_VERBOSE, _("\
2209 \n The file is already fully retrieved; nothing to do.\n\n"));
2210 /* In case the caller inspects. */
2213 /* Mark as successfully retrieved. */
2216 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2217 might be more bytes in the body. */
2219 return RETRUNNEEDED;
2221 if ((contrange != 0 && contrange != hs->restval)
2222 || (H_PARTIAL (statcode) && !contrange))
2224 /* The Range request was somehow misunderstood by the server.
2227 CLOSE_INVALIDATE (sock);
2234 hs->contlen = contlen + contrange;
2240 /* No need to print this output if the body won't be
2241 downloaded at all, or if the original server response is
2243 logputs (LOG_VERBOSE, _("Length: "));
2246 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2247 if (contlen + contrange >= 1024)
2248 logprintf (LOG_VERBOSE, " (%s)",
2249 human_readable (contlen + contrange));
2252 if (contlen >= 1024)
2253 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2254 number_to_static_string (contlen),
2255 human_readable (contlen));
2257 logprintf (LOG_VERBOSE, _(", %s remaining"),
2258 number_to_static_string (contlen));
2262 logputs (LOG_VERBOSE,
2263 opt.ignore_length ? _("ignored") : _("unspecified"));
2265 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2267 logputs (LOG_VERBOSE, "\n");
2271 type = NULL; /* We don't need it any more. */
2273 /* Return if we have no intention of further downloading. */
2274 if (!(*dt & RETROKF) || head_only)
2276 /* In case the caller cares to look... */
2281 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2282 servers not to send body in response to a HEAD request, and
2283 those that do will likely be caught by test_socket_open.
2284 If not, they can be worked around using
2285 `--no-http-keep-alive'. */
2286 CLOSE_FINISH (sock);
2287 else if (keep_alive && skip_short_body (sock, contlen))
2288 /* Successfully skipped the body; also keep using the socket. */
2289 CLOSE_FINISH (sock);
2291 CLOSE_INVALIDATE (sock);
2293 return RETRFINISHED;
2297 For VMS, define common fopen() optional arguments.
2300 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2301 # define FOPEN_BIN_FLAG 3
2302 #else /* def __VMS */
2303 # define FOPEN_BIN_FLAG true
2304 #endif /* def __VMS [else] */
2306 /* Open the local file. */
2309 mkalldirs (hs->local_file);
2311 rotate_backups (hs->local_file);
2318 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2319 #else /* def __VMS */
2320 fp = fopen (hs->local_file, "ab");
2321 #endif /* def __VMS [else] */
2323 else if (ALLOW_CLOBBER)
2329 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2330 #else /* def __VMS */
2331 fp = fopen (hs->local_file, "wb");
2332 #endif /* def __VMS [else] */
2336 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2337 if (!fp && errno == EEXIST)
2339 /* We cannot just invent a new name and use it (which is
2340 what functions like unique_create typically do)
2341 because we told the user we'd use this name.
2342 Instead, return and retry the download. */
2343 logprintf (LOG_NOTQUIET,
2344 _("%s has sprung into existence.\n"),
2346 CLOSE_INVALIDATE (sock);
2348 return FOPEN_EXCL_ERR;
2353 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2354 CLOSE_INVALIDATE (sock);
2362 /* Print fetch message, if opt.verbose. */
2365 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2366 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2369 /* This confuses the timestamping code that checks for file size.
2370 #### The timestamping code should be smarter about file size. */
2371 if (opt.save_headers && hs->restval == 0)
2372 fwrite (head, 1, strlen (head), fp);
2374 /* Now we no longer need to store the response header. */
2377 /* Download the request body. */
2380 /* If content-length is present, read that much; otherwise, read
2381 until EOF. The HTTP spec doesn't require the server to
2382 actually close the connection when it's done sending data. */
2383 flags |= rb_read_exactly;
2384 if (hs->restval > 0 && contrange == 0)
2385 /* If the server ignored our range request, instruct fd_read_body
2386 to skip the first RESTVAL bytes of body. */
2387 flags |= rb_skip_startpos;
2388 hs->len = hs->restval;
2390 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2391 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2395 CLOSE_FINISH (sock);
2399 hs->rderrmsg = xstrdup (fd_errstr (sock));
2400 CLOSE_INVALIDATE (sock);
2407 return RETRFINISHED;
2410 /* The genuine HTTP loop! This is the part where the retrieval is
2411 retried, and retried, and retried, and... */
2413 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2414 int *dt, struct url *proxy, struct iri *iri)
2417 bool got_head = false; /* used for time-stamping and filename detection */
2418 bool time_came_from_head = false;
2419 bool got_name = false;
2422 uerr_t err, ret = TRYLIMEXC;
2423 time_t tmr = -1; /* remote time-stamp */
2424 struct http_stat hstat; /* HTTP status */
2426 bool send_head_first = true;
2429 /* Assert that no value for *LOCAL_FILE was passed. */
2430 assert (local_file == NULL || *local_file == NULL);
2432 /* Set LOCAL_FILE parameter. */
2433 if (local_file && opt.output_document)
2434 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2436 /* Reset NEWLOC parameter. */
2439 /* This used to be done in main(), but it's a better idea to do it
2440 here so that we don't go through the hoops if we're just using
2445 /* Warn on (likely bogus) wildcard usage in HTTP. */
2446 if (opt.ftp_glob && has_wildcards_p (u->path))
2447 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2449 /* Setup hstat struct. */
2451 hstat.referer = referer;
2453 if (opt.output_document)
2455 hstat.local_file = xstrdup (opt.output_document);
2458 else if (!opt.content_disposition)
2460 hstat.local_file = url_file_name (u);
2464 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2465 * screaming for some refactoring. */
2466 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2468 /* If opt.noclobber is turned on and file already exists, do not
2469 retrieve the file. But if the output_document was given, then this
2470 test was already done and the file didn't exist. Hence the !opt.output_document */
2471 logprintf (LOG_VERBOSE, _("\
2472 File %s already there; not retrieving.\n\n"),
2473 quote (hstat.local_file));
2474 /* If the file is there, we suppose it's retrieved OK. */
2477 /* #### Bogusness alert. */
2478 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2479 if (has_html_suffix_p (hstat.local_file))
2486 /* Reset the counter. */
2489 /* Reset the document type. */
2492 /* Skip preliminary HEAD request if we're not in spider mode AND
2493 * if -O was given or HTTP Content-Disposition support is disabled. */
2495 && (got_name || !opt.content_disposition))
2496 send_head_first = false;
2498 /* Send preliminary HEAD request if -N is given and we have an existing
2499 * destination file. */
2500 file_name = url_file_name (u);
2501 if (opt.timestamping
2502 && !opt.content_disposition
2503 && file_exists_p (file_name))
2504 send_head_first = true;
2510 /* Increment the pass counter. */
2512 sleep_between_retrievals (count);
2514 /* Get the current time string. */
2515 tms = datetime_str (time (NULL));
2517 if (opt.spider && !got_head)
2518 logprintf (LOG_VERBOSE, _("\
2519 Spider mode enabled. Check if remote file exists.\n"));
2521 /* Print fetch message, if opt.verbose. */
2524 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2529 sprintf (tmp, _("(try:%2d)"), count);
2530 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2535 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2540 ws_changetitle (hurl);
2545 /* Default document type is empty. However, if spider mode is
2546 on or time-stamping is employed, HEAD_ONLY commands is
2547 encoded within *dt. */
2548 if (send_head_first && !got_head)
2553 /* Decide whether or not to restart. */
2556 && stat (hstat.local_file, &st) == 0
2557 && S_ISREG (st.st_mode))
2558 /* When -c is used, continue from on-disk size. (Can't use
2559 hstat.len even if count>1 because we don't want a failed
2560 first attempt to clobber existing data.) */
2561 hstat.restval = st.st_size;
2563 /* otherwise, continue where the previous try left off */
2564 hstat.restval = hstat.len;
2568 /* Decide whether to send the no-cache directive. We send it in
2570 a) we're using a proxy, and we're past our first retrieval.
2571 Some proxies are notorious for caching incomplete data, so
2572 we require a fresh get.
2573 b) caching is explicitly inhibited. */
2574 if ((proxy && count > 1) /* a */
2575 || !opt.allow_cache) /* b */
2576 *dt |= SEND_NOCACHE;
2578 *dt &= ~SEND_NOCACHE;
2580 /* Try fetching the document, or at least its head. */
2581 err = gethttp (u, &hstat, dt, proxy, iri);
2584 tms = datetime_str (time (NULL));
2586 /* Get the new location (with or without the redirection). */
2588 *newloc = xstrdup (hstat.newloc);
2592 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2593 case CONERROR: case READERR: case WRITEFAILED:
2594 case RANGEERR: case FOPEN_EXCL_ERR:
2595 /* Non-fatal errors continue executing the loop, which will
2596 bring them to "while" statement at the end, to judge
2597 whether the number of tries was exceeded. */
2598 printwhat (count, opt.ntry);
2600 case FWRITEERR: case FOPENERR:
2601 /* Another fatal error. */
2602 logputs (LOG_VERBOSE, "\n");
2603 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2604 quote (hstat.local_file), strerror (errno));
2605 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2606 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2607 /* Fatal errors just return from the function. */
2611 /* Another fatal error. */
2612 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2616 /* Return the new location to the caller. */
2619 logprintf (LOG_NOTQUIET,
2620 _("ERROR: Redirection (%d) without location.\n"),
2630 /* The file was already fully retrieved. */
2634 /* Deal with you later. */
2637 /* All possibilities should have been exhausted. */
2641 if (!(*dt & RETROKF))
2646 /* #### Ugly ugly ugly! */
2647 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2648 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2651 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2653 && (hstat.statcode == 500 || hstat.statcode == 501))
2658 /* Maybe we should always keep track of broken links, not just in
2660 * Don't log error if it was UTF-8 encoded because we will try
2661 * once unencoded. */
2662 else if (opt.spider && !iri->utf8_encode)
2664 /* #### Again: ugly ugly ugly! */
2666 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2667 nonexisting_url (hurl);
2668 logprintf (LOG_NOTQUIET, _("\
2669 Remote file does not exist -- broken link!!!\n"));
2673 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2674 tms, hstat.statcode,
2675 quotearg_style (escape_quoting_style, hstat.error));
2677 logputs (LOG_VERBOSE, "\n");
2683 /* Did we get the time-stamp? */
2686 got_head = true; /* no more time-stamping */
2688 if (opt.timestamping && !hstat.remote_time)
2690 logputs (LOG_NOTQUIET, _("\
2691 Last-modified header missing -- time-stamps turned off.\n"));
2693 else if (hstat.remote_time)
2695 /* Convert the date-string into struct tm. */
2696 tmr = http_atotm (hstat.remote_time);
2697 if (tmr == (time_t) (-1))
2698 logputs (LOG_VERBOSE, _("\
2699 Last-modified header invalid -- time-stamp ignored.\n"));
2700 if (*dt & HEAD_ONLY)
2701 time_came_from_head = true;
2704 if (send_head_first)
2706 /* The time-stamping section. */
2707 if (opt.timestamping)
2709 if (hstat.orig_file_name) /* Perform the following
2710 checks only if the file
2712 download already exists. */
2714 if (hstat.remote_time &&
2715 tmr != (time_t) (-1))
2717 /* Now time-stamping can be used validly.
2718 Time-stamping means that if the sizes of
2719 the local and remote file match, and local
2720 file is newer than the remote file, it will
2721 not be retrieved. Otherwise, the normal
2722 download procedure is resumed. */
2723 if (hstat.orig_file_tstamp >= tmr)
2725 if (hstat.contlen == -1
2726 || hstat.orig_file_size == hstat.contlen)
2728 logprintf (LOG_VERBOSE, _("\
2729 Server file no newer than local file %s -- not retrieving.\n\n"),
2730 quote (hstat.orig_file_name));
2736 logprintf (LOG_VERBOSE, _("\
2737 The sizes do not match (local %s) -- retrieving.\n"),
2738 number_to_static_string (hstat.orig_file_size));
2742 logputs (LOG_VERBOSE,
2743 _("Remote file is newer, retrieving.\n"));
2745 logputs (LOG_VERBOSE, "\n");
2749 /* free_hstat (&hstat); */
2750 hstat.timestamp_checked = true;
2755 bool finished = true;
2760 logputs (LOG_VERBOSE, _("\
2761 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2766 logprintf (LOG_VERBOSE, _("\
2767 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2768 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2775 logprintf (LOG_VERBOSE, _("\
2776 Remote file exists and could contain further links,\n\
2777 but recursion is disabled -- not retrieving.\n\n"));
2781 logprintf (LOG_VERBOSE, _("\
2782 Remote file exists.\n\n"));
2784 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2789 logprintf (LOG_NONVERBOSE,
2790 _("%s URL: %s %2d %s\n"),
2791 tms, u->url, hstat.statcode,
2792 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2799 count = 0; /* the retrieve count for HEAD is reset */
2801 } /* send_head_first */
2804 if ((tmr != (time_t) (-1))
2805 && ((hstat.len == hstat.contlen) ||
2806 ((hstat.res == 0) && (hstat.contlen == -1))))
2808 const char *fl = NULL;
2809 set_local_file (&fl, hstat.local_file);
2813 /* Reparse time header, in case it's changed. */
2814 if (time_came_from_head
2815 && hstat.remote_time && hstat.remote_time[0])
2817 newtmr = http_atotm (hstat.remote_time);
2818 if (newtmr != (time_t)-1)
2824 /* End of time-stamping section. */
2826 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2827 total_download_time += hstat.dltime;
2829 if (hstat.len == hstat.contlen)
2833 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
2835 logprintf (LOG_VERBOSE,
2837 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
2838 : _("%s (%s) - %s saved [%s/%s]\n\n"),
2840 write_to_stdout ? "" : quote (hstat.local_file),
2841 number_to_static_string (hstat.len),
2842 number_to_static_string (hstat.contlen));
2843 logprintf (LOG_NONVERBOSE,
2844 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2846 number_to_static_string (hstat.len),
2847 number_to_static_string (hstat.contlen),
2848 hstat.local_file, count);
2851 total_downloaded_bytes += hstat.len;
2853 /* Remember that we downloaded the file for later ".orig" code. */
2854 if (*dt & ADDED_HTML_EXTENSION)
2855 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2857 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2862 else if (hstat.res == 0) /* No read error */
2864 if (hstat.contlen == -1) /* We don't know how much we were supposed
2865 to get, so assume we succeeded. */
2869 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
2871 logprintf (LOG_VERBOSE,
2873 ? _("%s (%s) - written to stdout %s[%s]\n\n")
2874 : _("%s (%s) - %s saved [%s]\n\n"),
2876 write_to_stdout ? "" : quote (hstat.local_file),
2877 number_to_static_string (hstat.len));
2878 logprintf (LOG_NONVERBOSE,
2879 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2880 tms, u->url, number_to_static_string (hstat.len),
2881 hstat.local_file, count);
2884 total_downloaded_bytes += hstat.len;
2886 /* Remember that we downloaded the file for later ".orig" code. */
2887 if (*dt & ADDED_HTML_EXTENSION)
2888 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2890 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2895 else if (hstat.len < hstat.contlen) /* meaning we lost the
2896 connection too soon */
2898 logprintf (LOG_VERBOSE,
2899 _("%s (%s) - Connection closed at byte %s. "),
2900 tms, tmrate, number_to_static_string (hstat.len));
2901 printwhat (count, opt.ntry);
2904 else if (hstat.len != hstat.restval)
2905 /* Getting here would mean reading more data than
2906 requested with content-length, which we never do. */
2910 /* Getting here probably means that the content-length was
2911 * _less_ than the original, local size. We should probably
2912 * truncate or re-read, or something. FIXME */
2917 else /* from now on hstat.res can only be -1 */
2919 if (hstat.contlen == -1)
2921 logprintf (LOG_VERBOSE,
2922 _("%s (%s) - Read error at byte %s (%s)."),
2923 tms, tmrate, number_to_static_string (hstat.len),
2925 printwhat (count, opt.ntry);
2928 else /* hstat.res == -1 and contlen is given */
2930 logprintf (LOG_VERBOSE,
2931 _("%s (%s) - Read error at byte %s/%s (%s). "),
2933 number_to_static_string (hstat.len),
2934 number_to_static_string (hstat.contlen),
2936 printwhat (count, opt.ntry);
2942 while (!opt.ntry || (count < opt.ntry));
2946 *local_file = xstrdup (hstat.local_file);
2947 free_hstat (&hstat);
2952 /* Check whether the result of strptime() indicates success.
2953 strptime() returns the pointer to how far it got to in the string.
2954 The processing has been successful if the string is at `GMT' or
2955 `+X', or at the end of the string.
2957 In extended regexp parlance, the function returns 1 if P matches
2958 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2959 can return) is considered a failure and 0 is returned. */
2961 check_end (const char *p)
2965 while (c_isspace (*p))
2968 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2969 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2975 /* Convert the textual specification of time in TIME_STRING to the
2976 number of seconds since the Epoch.
2978 TIME_STRING can be in any of the three formats RFC2616 allows the
2979 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2980 as well as the time format used in the Set-Cookie header.
2981 Timezones are ignored, and should be GMT.
2983 Return the computed time_t representation, or -1 if the conversion
2986 This function uses strptime with various string formats for parsing
2987 TIME_STRING. This results in a parser that is not as lenient in
2988 interpreting TIME_STRING as I would like it to be. Being based on
2989 strptime, it always allows shortened months, one-digit days, etc.,
2990 but due to the multitude of formats in which time can be
2991 represented, an ideal HTTP time parser would be even more
2992 forgiving. It should completely ignore things like week days and
2993 concentrate only on the various forms of representing years,
2994 months, days, hours, minutes, and seconds. For example, it would
2995 be nice if it accepted ISO 8601 out of the box.
2997 I've investigated free and PD code for this purpose, but none was
2998 usable. getdate was big and unwieldy, and had potential copyright
2999 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3000 distributed with phttpd, is excellent, but we cannot use it because
3001 it is not assigned to the FSF. So I stuck it with strptime. */
3004 http_atotm (const char *time_string)
3006 /* NOTE: Solaris strptime man page claims that %n and %t match white
3007 space, but that's not universally available. Instead, we simply
3008 use ` ' to mean "skip all WS", which works under all strptime
3009 implementations I've tested. */
3011 static const char *time_formats[] = {
3012 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3013 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3014 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3015 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3016 (used in Set-Cookie, defined in the
3017 Netscape cookie specification.) */
3019 const char *oldlocale;
3020 char savedlocale[256];
3022 time_t ret = (time_t) -1;
3024 /* Solaris strptime fails to recognize English month names in
3025 non-English locales, which we work around by temporarily setting
3026 locale to C before invoking strptime. */
3027 oldlocale = setlocale (LC_TIME, NULL);
3030 size_t l = strlen (oldlocale);
3031 if (l >= sizeof savedlocale)
3032 savedlocale[0] = '\0';
3034 memcpy (savedlocale, oldlocale, l);
3036 else savedlocale[0] = '\0';
3038 setlocale (LC_TIME, "C");
3040 for (i = 0; i < countof (time_formats); i++)
3044 /* Some versions of strptime use the existing contents of struct
3045 tm to recalculate the date according to format. Zero it out
3046 to prevent stack garbage from influencing strptime. */
3049 if (check_end (strptime (time_string, time_formats[i], &t)))
3056 /* Restore the previous locale. */
3058 setlocale (LC_TIME, savedlocale);
3063 /* Authorization support: We support three authorization schemes:
3065 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3067 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3068 consisting of answering to the server's challenge with the proper
3071 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3072 Stenberg for libcurl. Like digest, NTLM is based on a
3073 challenge-response mechanism, but unlike digest, it is non-standard
3074 (authenticates TCP connections rather than requests), undocumented
3075 and Microsoft-specific. */
3077 /* Create the authentication header contents for the `Basic' scheme.
3078 This is done by encoding the string "USER:PASS" to base64 and
3079 prepending the string "Basic " in front of it. */
3082 basic_authentication_encode (const char *user, const char *passwd)
3085 int len1 = strlen (user) + 1 + strlen (passwd);
3087 t1 = (char *)alloca (len1 + 1);
3088 sprintf (t1, "%s:%s", user, passwd);
3090 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3091 base64_encode (t1, len1, t2);
3093 return concat_strings ("Basic ", t2, (char *) 0);
3096 #define SKIP_WS(x) do { \
3097 while (c_isspace (*(x))) \
3101 #ifdef ENABLE_DIGEST
3102 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3103 an array of 16 bytes containing the hash keys, and BUF should be a
3104 buffer of 33 writable characters (32 for hex digits plus one for
3105 zero termination). */
3107 dump_hash (char *buf, const unsigned char *hash)
3111 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3113 *buf++ = XNUM_TO_digit (*hash >> 4);
3114 *buf++ = XNUM_TO_digit (*hash & 0xf);
3119 /* Take the line apart to find the challenge, and compose a digest
3120 authorization header. See RFC2069 section 2.1.2. */
3122 digest_authentication_encode (const char *au, const char *user,
3123 const char *passwd, const char *method,
3126 static char *realm, *opaque, *nonce;
3131 { "realm", &realm },
3132 { "opaque", &opaque },
3136 param_token name, value;
3138 realm = opaque = nonce = NULL;
3140 au += 6; /* skip over `Digest' */
3141 while (extract_param (&au, &name, &value, ','))
3144 size_t namelen = name.e - name.b;
3145 for (i = 0; i < countof (options); i++)
3146 if (namelen == strlen (options[i].name)
3147 && 0 == strncmp (name.b, options[i].name,
3150 *options[i].variable = strdupdelim (value.b, value.e);
3154 if (!realm || !nonce || !user || !passwd || !path || !method)
3157 xfree_null (opaque);
3162 /* Calculate the digest value. */
3164 ALLOCA_MD5_CONTEXT (ctx);
3165 unsigned char hash[MD5_HASHLEN];
3166 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3167 char response_digest[MD5_HASHLEN * 2 + 1];
3169 /* A1BUF = H(user ":" realm ":" password) */
3171 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3172 gen_md5_update ((unsigned char *)":", 1, ctx);
3173 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3174 gen_md5_update ((unsigned char *)":", 1, ctx);
3175 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3176 gen_md5_finish (ctx, hash);
3177 dump_hash (a1buf, hash);
3179 /* A2BUF = H(method ":" path) */
3181 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3182 gen_md5_update ((unsigned char *)":", 1, ctx);
3183 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3184 gen_md5_finish (ctx, hash);
3185 dump_hash (a2buf, hash);
3187 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3189 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3190 gen_md5_update ((unsigned char *)":", 1, ctx);
3191 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3192 gen_md5_update ((unsigned char *)":", 1, ctx);
3193 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3194 gen_md5_finish (ctx, hash);
3195 dump_hash (response_digest, hash);
3197 res = xmalloc (strlen (user)
3202 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3203 + (opaque ? strlen (opaque) : 0)
3205 sprintf (res, "Digest \
3206 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3207 user, realm, nonce, path, response_digest);
3210 char *p = res + strlen (res);
3211 strcat (p, ", opaque=\"");
3218 #endif /* ENABLE_DIGEST */
3220 /* Computing the size of a string literal must take into account that
3221 value returned by sizeof includes the terminating \0. */
3222 #define STRSIZE(literal) (sizeof (literal) - 1)
3224 /* Whether chars in [b, e) begin with the literal string provided as
3225 first argument and are followed by whitespace or terminating \0.
3226 The comparison is case-insensitive. */
3227 #define STARTS(literal, b, e) \
3229 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3230 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3231 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3232 || c_isspace (b[STRSIZE (literal)])))
3235 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3237 return STARTS ("Basic", hdrbeg, hdrend)
3238 #ifdef ENABLE_DIGEST
3239 || STARTS ("Digest", hdrbeg, hdrend)
3242 || STARTS ("NTLM", hdrbeg, hdrend)
3249 /* Create the HTTP authorization request header. When the
3250 `WWW-Authenticate' response header is seen, according to the
3251 authorization scheme specified in that header (`Basic' and `Digest'
3252 are supported by the current implementation), produce an
3253 appropriate HTTP authorization request header. */
3255 create_authorization_line (const char *au, const char *user,
3256 const char *passwd, const char *method,
3257 const char *path, bool *finished)
3259 /* We are called only with known schemes, so we can dispatch on the
3261 switch (c_toupper (*au))
3263 case 'B': /* Basic */
3265 return basic_authentication_encode (user, passwd);
3266 #ifdef ENABLE_DIGEST
3267 case 'D': /* Digest */
3269 return digest_authentication_encode (au, user, passwd, method, path);
3272 case 'N': /* NTLM */
3273 if (!ntlm_input (&pconn.ntlm, au))
3278 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3281 /* We shouldn't get here -- this function should be only called
3282 with values approved by known_authentication_scheme_p. */
3290 if (!wget_cookie_jar)
3291 wget_cookie_jar = cookie_jar_new ();
3292 if (opt.cookies_input && !cookies_loaded_p)
3294 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3295 cookies_loaded_p = true;
3302 if (wget_cookie_jar)
3303 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3309 xfree_null (pconn.host);
3310 if (wget_cookie_jar)
3311 cookie_jar_delete (wget_cookie_jar);
3315 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3317 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3319 int len = strlen (ext);
3322 strncpy (shortext, ext, len - 1);
3323 shortext[len - 2] = '\0';
3326 if (last_period_in_local_filename == NULL
3327 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3328 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3330 int local_filename_len = strlen (hs->local_file);
3331 /* Resize the local file, allowing for ".html" preceded by
3332 optional ".NUMBER". */
3333 hs->local_file = xrealloc (hs->local_file,
3334 local_filename_len + 24 + len);
3335 strcpy (hs->local_file + local_filename_len, ext);
3336 /* If clobbering is not allowed and the file, as named,
3337 exists, tack on ".NUMBER.html" instead. */
3338 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3342 sprintf (hs->local_file + local_filename_len,
3343 ".%d%s", ext_num++, ext);
3344 while (file_exists_p (hs->local_file));
3346 *dt |= ADDED_HTML_EXTENSION;
3354 test_parse_content_disposition()
3359 char *opt_dir_prefix;
3363 { "filename=\"file.ext\"", NULL, "file.ext", true },
3364 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3365 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3366 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3367 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3368 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3369 { "attachment", NULL, NULL, false },
3370 { "attachment", "somedir", NULL, false },
3373 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3378 opt.dir_prefix = test_array[i].opt_dir_prefix;
3379 res = parse_content_disposition (test_array[i].hdrval, &filename);
3381 mu_assert ("test_parse_content_disposition: wrong result",
3382 res == test_array[i].result
3384 || 0 == strcmp (test_array[i].filename, filename)));
3390 #endif /* TESTING */
3393 * vim: et sts=2 sw=2 cino+={s