2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
56 # include "http-ntlm.h"
69 extern char *version_string;
73 static char *create_authorization_line (const char *, const char *,
74 const char *, const char *,
75 const char *, bool *);
76 static char *basic_authentication_encode (const char *, const char *);
77 static bool known_authentication_scheme_p (const char *, const char *);
78 static void ensure_extension (struct http_stat *, const char *, int *);
79 static void load_cookies (void);
82 # define MIN(x, y) ((x) > (y) ? (y) : (x))
86 static bool cookies_loaded_p;
87 static struct cookie_jar *wget_cookie_jar;
89 #define TEXTHTML_S "text/html"
90 #define TEXTXHTML_S "application/xhtml+xml"
91 #define TEXTCSS_S "text/css"
93 /* Some status code validation macros: */
94 #define H_20X(x) (((x) >= 200) && ((x) < 300))
95 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
96 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
97 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
98 || (x) == HTTP_STATUS_SEE_OTHER \
99 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
101 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
102 /* Successful 2xx. */
103 #define HTTP_STATUS_OK 200
104 #define HTTP_STATUS_CREATED 201
105 #define HTTP_STATUS_ACCEPTED 202
106 #define HTTP_STATUS_NO_CONTENT 204
107 #define HTTP_STATUS_PARTIAL_CONTENTS 206
109 /* Redirection 3xx. */
110 #define HTTP_STATUS_MULTIPLE_CHOICES 300
111 #define HTTP_STATUS_MOVED_PERMANENTLY 301
112 #define HTTP_STATUS_MOVED_TEMPORARILY 302
113 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
114 #define HTTP_STATUS_NOT_MODIFIED 304
115 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
117 /* Client error 4xx. */
118 #define HTTP_STATUS_BAD_REQUEST 400
119 #define HTTP_STATUS_UNAUTHORIZED 401
120 #define HTTP_STATUS_FORBIDDEN 403
121 #define HTTP_STATUS_NOT_FOUND 404
122 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
124 /* Server errors 5xx. */
125 #define HTTP_STATUS_INTERNAL 500
126 #define HTTP_STATUS_NOT_IMPLEMENTED 501
127 #define HTTP_STATUS_BAD_GATEWAY 502
128 #define HTTP_STATUS_UNAVAILABLE 503
131 rel_none, rel_name, rel_value, rel_both
138 struct request_header {
140 enum rp release_policy;
142 int hcount, hcapacity;
147 /* Create a new, empty request. At least request_set_method must be
148 called before the request can be used. */
150 static struct request *
153 struct request *req = xnew0 (struct request);
155 req->headers = xnew_array (struct request_header, req->hcapacity);
159 /* Set the request's method and its arguments. METH should be a
160 literal string (or it should outlive the request) because it will
161 not be freed. ARG will be freed by request_free. */
164 request_set_method (struct request *req, const char *meth, char *arg)
170 /* Return the method string passed with the last call to
171 request_set_method. */
174 request_method (const struct request *req)
179 /* Free one header according to the release policy specified with
180 request_set_header. */
183 release_header (struct request_header *hdr)
185 switch (hdr->release_policy)
202 /* Set the request named NAME to VALUE. Specifically, this means that
203 a "NAME: VALUE\r\n" header line will be used in the request. If a
204 header with the same name previously existed in the request, its
205 value will be replaced by this one. A NULL value means do nothing.
207 RELEASE_POLICY determines whether NAME and VALUE should be released
208 (freed) with request_free. Allowed values are:
210 - rel_none - don't free NAME or VALUE
211 - rel_name - free NAME when done
212 - rel_value - free VALUE when done
213 - rel_both - free both NAME and VALUE when done
215 Setting release policy is useful when arguments come from different
216 sources. For example:
218 // Don't free literal strings!
219 request_set_header (req, "Pragma", "no-cache", rel_none);
221 // Don't free a global variable, we'll need it later.
222 request_set_header (req, "Referer", opt.referer, rel_none);
224 // Value freshly allocated, free it when done.
225 request_set_header (req, "Range",
226 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
231 request_set_header (struct request *req, char *name, char *value,
232 enum rp release_policy)
234 struct request_header *hdr;
239 /* A NULL value is a no-op; if freeing the name is requested,
240 free it now to avoid leaks. */
241 if (release_policy == rel_name || release_policy == rel_both)
246 for (i = 0; i < req->hcount; i++)
248 hdr = &req->headers[i];
249 if (0 == strcasecmp (name, hdr->name))
251 /* Replace existing header. */
252 release_header (hdr);
255 hdr->release_policy = release_policy;
260 /* Install new header. */
262 if (req->hcount >= req->hcapacity)
264 req->hcapacity <<= 1;
265 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
267 hdr = &req->headers[req->hcount++];
270 hdr->release_policy = release_policy;
273 /* Like request_set_header, but sets the whole header line, as
274 provided by the user using the `--header' option. For example,
275 request_set_user_header (req, "Foo: bar") works just like
276 request_set_header (req, "Foo", "bar"). */
279 request_set_user_header (struct request *req, const char *header)
282 const char *p = strchr (header, ':');
285 BOUNDED_TO_ALLOCA (header, p, name);
287 while (c_isspace (*p))
289 request_set_header (req, xstrdup (name), (char *) p, rel_name);
292 /* Remove the header with specified name from REQ. Returns true if
293 the header was actually removed, false otherwise. */
296 request_remove_header (struct request *req, char *name)
299 for (i = 0; i < req->hcount; i++)
301 struct request_header *hdr = &req->headers[i];
302 if (0 == strcasecmp (name, hdr->name))
304 release_header (hdr);
305 /* Move the remaining headers by one. */
306 if (i < req->hcount - 1)
307 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
315 #define APPEND(p, str) do { \
316 int A_len = strlen (str); \
317 memcpy (p, str, A_len); \
321 /* Construct the request and write it to FD using fd_write. */
324 request_send (const struct request *req, int fd)
326 char *request_string, *p;
327 int i, size, write_error;
329 /* Count the request size. */
332 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
333 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
335 for (i = 0; i < req->hcount; i++)
337 struct request_header *hdr = &req->headers[i];
338 /* NAME ": " VALUE "\r\n" */
339 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
345 p = request_string = alloca_array (char, size);
347 /* Generate the request. */
349 APPEND (p, req->method); *p++ = ' ';
350 APPEND (p, req->arg); *p++ = ' ';
351 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
353 for (i = 0; i < req->hcount; i++)
355 struct request_header *hdr = &req->headers[i];
356 APPEND (p, hdr->name);
357 *p++ = ':', *p++ = ' ';
358 APPEND (p, hdr->value);
359 *p++ = '\r', *p++ = '\n';
362 *p++ = '\r', *p++ = '\n', *p++ = '\0';
363 assert (p - request_string == size);
367 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
369 /* Send the request to the server. */
371 write_error = fd_write (fd, request_string, size - 1, -1);
373 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
378 /* Release the resources used by REQ. */
381 request_free (struct request *req)
384 xfree_null (req->arg);
385 for (i = 0; i < req->hcount; i++)
386 release_header (&req->headers[i]);
387 xfree_null (req->headers);
391 static struct hash_table *basic_authed_hosts;
393 /* Find out if this host has issued a Basic challenge yet; if so, give
394 * it the username, password. A temporary measure until we can get
395 * proper authentication in place. */
398 maybe_send_basic_creds (const char *hostname, const char *user,
399 const char *passwd, struct request *req)
401 bool do_challenge = false;
403 if (opt.auth_without_challenge)
405 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
408 else if (basic_authed_hosts
409 && hash_table_contains(basic_authed_hosts, hostname))
411 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
416 DEBUGP(("Host %s has not issued a general basic challenge.\n",
421 request_set_header (req, "Authorization",
422 basic_authentication_encode (user, passwd),
429 register_basic_auth_host (const char *hostname)
431 if (!basic_authed_hosts)
433 basic_authed_hosts = make_nocase_string_hash_table (1);
435 if (!hash_table_contains(basic_authed_hosts, hostname))
437 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
438 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
443 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
444 PROMISED_SIZE bytes are sent over the wire -- if the file is
445 longer, read only that much; if the file is shorter, report an error. */
448 post_file (int sock, const char *file_name, wgint promised_size)
450 static char chunk[8192];
455 DEBUGP (("[writing POST file %s ... ", file_name));
457 fp = fopen (file_name, "rb");
460 while (!feof (fp) && written < promised_size)
463 int length = fread (chunk, 1, sizeof (chunk), fp);
466 towrite = MIN (promised_size - written, length);
467 write_error = fd_write (sock, chunk, towrite, -1);
477 /* If we've written less than was promised, report a (probably
478 nonsensical) error rather than break the promise. */
479 if (written < promised_size)
485 assert (written == promised_size);
486 DEBUGP (("done]\n"));
490 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
491 If so, return the pointer to the position after the line, otherwise
492 return NULL. This is used as callback to fd_read_hunk. The data
493 between START and PEEKED has been read and cannot be "unread"; the
494 data after PEEKED has only been peeked. */
497 response_head_terminator (const char *start, const char *peeked, int peeklen)
501 /* If at first peek, verify whether HUNK starts with "HTTP". If
502 not, this is a HTTP/0.9 request and we must bail out without
504 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
507 /* Look for "\n[\r]\n", and return the following position if found.
508 Start two chars before the current to cover the possibility that
509 part of the terminator (e.g. "\n\r") arrived in the previous
511 p = peeked - start < 2 ? start : peeked - 2;
512 end = peeked + peeklen;
514 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
515 for (; p < end - 2; p++)
518 if (p[1] == '\r' && p[2] == '\n')
520 else if (p[1] == '\n')
523 /* p==end-2: check for \n\n directly preceding END. */
524 if (p[0] == '\n' && p[1] == '\n')
530 /* The maximum size of a single HTTP response we care to read. Rather
531 than being a limit of the reader implementation, this limit
532 prevents Wget from slurping all available memory upon encountering
533 malicious or buggy server output, thus protecting the user. Define
534 it to 0 to remove the limit. */
536 #define HTTP_RESPONSE_MAX_SIZE 65536
538 /* Read the HTTP request head from FD and return it. The error
539 conditions are the same as with fd_read_hunk.
541 To support HTTP/0.9 responses, this function tries to make sure
542 that the data begins with "HTTP". If this is not the case, no data
543 is read and an empty request is returned, so that the remaining
544 data can be treated as body. */
547 read_http_response_head (int fd)
549 return fd_read_hunk (fd, response_head_terminator, 512,
550 HTTP_RESPONSE_MAX_SIZE);
554 /* The response data. */
557 /* The array of pointers that indicate where each header starts.
558 For example, given this HTTP response:
565 The headers are located like this:
567 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
569 headers[0] headers[1] headers[2] headers[3]
571 I.e. headers[0] points to the beginning of the request,
572 headers[1] points to the end of the first header and the
573 beginning of the second one, etc. */
575 const char **headers;
578 /* Create a new response object from the text of the HTTP response,
579 available in HEAD. That text is automatically split into
580 constituent header lines for fast retrieval using
583 static struct response *
584 resp_new (const char *head)
589 struct response *resp = xnew0 (struct response);
594 /* Empty head means that we're dealing with a headerless
595 (HTTP/0.9) response. In that case, don't set HEADERS at
600 /* Split HEAD into header lines, so that resp_header_* functions
601 don't need to do this over and over again. */
607 DO_REALLOC (resp->headers, size, count + 1, const char *);
608 resp->headers[count++] = hdr;
610 /* Break upon encountering an empty line. */
611 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
614 /* Find the end of HDR, including continuations. */
617 const char *end = strchr (hdr, '\n');
623 while (*hdr == ' ' || *hdr == '\t');
625 DO_REALLOC (resp->headers, size, count + 1, const char *);
626 resp->headers[count] = NULL;
631 /* Locate the header named NAME in the request data, starting with
632 position START. This allows the code to loop through the request
633 data, filtering for all requests of a given name. Returns the
634 found position, or -1 for failure. The code that uses this
635 function typically looks like this:
637 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
638 ... do something with header ...
640 If you only care about one header, use resp_header_get instead of
644 resp_header_locate (const struct response *resp, const char *name, int start,
645 const char **begptr, const char **endptr)
648 const char **headers = resp->headers;
651 if (!headers || !headers[1])
654 name_len = strlen (name);
660 for (; headers[i + 1]; i++)
662 const char *b = headers[i];
663 const char *e = headers[i + 1];
665 && b[name_len] == ':'
666 && 0 == strncasecmp (b, name, name_len))
669 while (b < e && c_isspace (*b))
671 while (b < e && c_isspace (e[-1]))
681 /* Find and retrieve the header named NAME in the request data. If
682 found, set *BEGPTR to its starting, and *ENDPTR to its ending
683 position, and return true. Otherwise return false.
685 This function is used as a building block for resp_header_copy
686 and resp_header_strdup. */
689 resp_header_get (const struct response *resp, const char *name,
690 const char **begptr, const char **endptr)
692 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
696 /* Copy the response header named NAME to buffer BUF, no longer than
697 BUFSIZE (BUFSIZE includes the terminating 0). If the header
698 exists, true is returned, false otherwise. If there should be no
699 limit on the size of the header, use resp_header_strdup instead.
701 If BUFSIZE is 0, no data is copied, but the boolean indication of
702 whether the header is present is still returned. */
705 resp_header_copy (const struct response *resp, const char *name,
706 char *buf, int bufsize)
709 if (!resp_header_get (resp, name, &b, &e))
713 int len = MIN (e - b, bufsize - 1);
714 memcpy (buf, b, len);
720 /* Return the value of header named NAME in RESP, allocated with
721 malloc. If such a header does not exist in RESP, return NULL. */
724 resp_header_strdup (const struct response *resp, const char *name)
727 if (!resp_header_get (resp, name, &b, &e))
729 return strdupdelim (b, e);
732 /* Parse the HTTP status line, which is of format:
734 HTTP-Version SP Status-Code SP Reason-Phrase
736 The function returns the status-code, or -1 if the status line
737 appears malformed. The pointer to "reason-phrase" message is
738 returned in *MESSAGE. */
741 resp_status (const struct response *resp, char **message)
748 /* For a HTTP/0.9 response, assume status 200. */
750 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
754 p = resp->headers[0];
755 end = resp->headers[1];
761 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
765 /* Match the HTTP version. This is optional because Gnutella
766 servers have been reported to not specify HTTP version. */
767 if (p < end && *p == '/')
770 while (p < end && c_isdigit (*p))
772 if (p < end && *p == '.')
774 while (p < end && c_isdigit (*p))
778 while (p < end && c_isspace (*p))
780 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
783 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
788 while (p < end && c_isspace (*p))
790 while (p < end && c_isspace (end[-1]))
792 *message = strdupdelim (p, end);
798 /* Release the resources used by RESP. */
801 resp_free (struct response *resp)
803 xfree_null (resp->headers);
807 /* Print a single line of response, the characters [b, e). We tried
809 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
810 but that failed to escape the non-printable characters and, in fact,
811 caused crashes in UTF-8 locales. */
814 print_response_line(const char *prefix, const char *b, const char *e)
817 BOUNDED_TO_ALLOCA(b, e, copy);
818 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
819 quotearg_style (escape_quoting_style, copy));
822 /* Print the server response, line by line, omitting the trailing CRLF
823 from individual header lines, and prefixed with PREFIX. */
826 print_server_response (const struct response *resp, const char *prefix)
831 for (i = 0; resp->headers[i + 1]; i++)
833 const char *b = resp->headers[i];
834 const char *e = resp->headers[i + 1];
836 if (b < e && e[-1] == '\n')
838 if (b < e && e[-1] == '\r')
840 print_response_line(prefix, b, e);
844 /* Parse the `Content-Range' header and extract the information it
845 contains. Returns true if successful, false otherwise. */
847 parse_content_range (const char *hdr, wgint *first_byte_ptr,
848 wgint *last_byte_ptr, wgint *entity_length_ptr)
852 /* Ancient versions of Netscape proxy server, presumably predating
853 rfc2068, sent out `Content-Range' without the "bytes"
855 if (0 == strncasecmp (hdr, "bytes", 5))
858 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
862 while (c_isspace (*hdr))
867 if (!c_isdigit (*hdr))
869 for (num = 0; c_isdigit (*hdr); hdr++)
870 num = 10 * num + (*hdr - '0');
871 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
873 *first_byte_ptr = num;
875 for (num = 0; c_isdigit (*hdr); hdr++)
876 num = 10 * num + (*hdr - '0');
877 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
879 *last_byte_ptr = num;
884 for (num = 0; c_isdigit (*hdr); hdr++)
885 num = 10 * num + (*hdr - '0');
886 *entity_length_ptr = num;
890 /* Read the body of the request, but don't store it anywhere and don't
891 display a progress gauge. This is useful for reading the bodies of
892 administrative responses to which we will soon issue another
893 request. The response is not useful to the user, but reading it
894 allows us to continue using the same connection to the server.
896 If reading fails, false is returned, true otherwise. In debug
897 mode, the body is displayed for debugging purposes. */
900 skip_short_body (int fd, wgint contlen)
903 SKIP_SIZE = 512, /* size of the download buffer */
904 SKIP_THRESHOLD = 4096 /* the largest size we read */
906 char dlbuf[SKIP_SIZE + 1];
907 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
909 /* We shouldn't get here with unknown contlen. (This will change
910 with HTTP/1.1, which supports "chunked" transfer.) */
911 assert (contlen != -1);
913 /* If the body is too large, it makes more sense to simply close the
914 connection than to try to read the body. */
915 if (contlen > SKIP_THRESHOLD)
918 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
922 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
925 /* Don't normally report the error since this is an
926 optimization that should be invisible to the user. */
927 DEBUGP (("] aborting (%s).\n",
928 ret < 0 ? fd_errstr (fd) : "EOF received"));
932 /* Safe even if %.*s bogusly expects terminating \0 because
933 we've zero-terminated dlbuf above. */
934 DEBUGP (("%.*s", ret, dlbuf));
937 DEBUGP (("] done.\n"));
941 /* Extract a parameter from the string (typically an HTTP header) at
942 **SOURCE and advance SOURCE to the next parameter. Return false
943 when there are no more parameters to extract. The name of the
944 parameter is returned in NAME, and the value in VALUE. If the
945 parameter has no value, the token's value is zeroed out.
947 For example, if *SOURCE points to the string "attachment;
948 filename=\"foo bar\"", the first call to this function will return
949 the token named "attachment" and no value, and the second call will
950 return the token named "filename" and value "foo bar". The third
951 call will return false, indicating no more valid tokens. */
954 extract_param (const char **source, param_token *name, param_token *value,
957 const char *p = *source;
959 while (c_isspace (*p)) ++p;
963 return false; /* no error; nothing more to extract */
968 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
970 if (name->b == name->e)
971 return false; /* empty name: error */
972 while (c_isspace (*p)) ++p;
973 if (*p == separator || !*p) /* no value */
976 if (*p == separator) ++p;
981 return false; /* error */
983 /* *p is '=', extract value */
985 while (c_isspace (*p)) ++p;
986 if (*p == '"') /* quoted */
989 while (*p && *p != '"') ++p;
993 /* Currently at closing quote; find the end of param. */
994 while (c_isspace (*p)) ++p;
995 while (*p && *p != separator) ++p;
999 /* garbage after closed quote, e.g. foo="bar"baz */
1005 while (*p && *p != separator) ++p;
1007 while (value->e != value->b && c_isspace (value->e[-1]))
1009 if (*p == separator) ++p;
1016 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1018 /* Parse the contents of the `Content-Disposition' header, extracting
1019 the information useful to Wget. Content-Disposition is a header
1020 borrowed from MIME; when used in HTTP, it typically serves for
1021 specifying the desired file name of the resource. For example:
1023 Content-Disposition: attachment; filename="flora.jpg"
1025 Wget will skip the tokens it doesn't care about, such as
1026 "attachment" in the previous example; it will also skip other
1027 unrecognized params. If the header is syntactically correct and
1028 contains a file name, a copy of the file name is stored in
1029 *filename and true is returned. Otherwise, the function returns
1032 The file name is stripped of directory components and must not be
1036 parse_content_disposition (const char *hdr, char **filename)
1038 param_token name, value;
1039 while (extract_param (&hdr, &name, &value, ';'))
1040 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1042 /* Make the file name begin at the last slash or backslash. */
1043 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1044 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1045 if (last_slash && last_bs)
1046 value.b = 1 + MAX (last_slash, last_bs);
1047 else if (last_slash || last_bs)
1048 value.b = 1 + (last_slash ? last_slash : last_bs);
1049 if (value.b == value.e)
1051 /* Start with the directory prefix, if specified. */
1054 int prefix_length = strlen (opt.dir_prefix);
1055 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1060 total_length = prefix_length + (value.e - value.b);
1061 *filename = xmalloc (total_length + 1);
1062 strcpy (*filename, opt.dir_prefix);
1064 (*filename)[prefix_length - 1] = '/';
1065 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1066 (*filename)[total_length] = '\0';
1069 *filename = strdupdelim (value.b, value.e);
1075 /* Persistent connections. Currently, we cache the most recently used
1076 connection as persistent, provided that the HTTP server agrees to
1077 make it such. The persistence data is stored in the variables
1078 below. Ideally, it should be possible to cache an arbitrary fixed
1079 number of these connections. */
1081 /* Whether a persistent connection is active. */
1082 static bool pconn_active;
1085 /* The socket of the connection. */
1088 /* Host and port of the currently active persistent connection. */
1092 /* Whether a ssl handshake has occoured on this connection. */
1095 /* Whether the connection was authorized. This is only done by
1096 NTLM, which authorizes *connections* rather than individual
1097 requests. (That practice is peculiar for HTTP, but it is a
1098 useful optimization.) */
1102 /* NTLM data of the current connection. */
1103 struct ntlmdata ntlm;
1107 /* Mark the persistent connection as invalid and free the resources it
1108 uses. This is used by the CLOSE_* macros after they forcefully
1109 close a registered persistent connection. */
1112 invalidate_persistent (void)
1114 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1115 pconn_active = false;
1116 fd_close (pconn.socket);
1121 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1122 persistent. This will enable someone to use the same connection
1123 later. In the context of HTTP, this must be called only AFTER the
1124 response has been received and the server has promised that the
1125 connection will remain alive.
1127 If a previous connection was persistent, it is closed. */
1130 register_persistent (const char *host, int port, int fd, bool ssl)
1134 if (pconn.socket == fd)
1136 /* The connection FD is already registered. */
1141 /* The old persistent connection is still active; close it
1142 first. This situation arises whenever a persistent
1143 connection exists, but we then connect to a different
1144 host, and try to register a persistent connection to that
1146 invalidate_persistent ();
1150 pconn_active = true;
1152 pconn.host = xstrdup (host);
1155 pconn.authorized = false;
1157 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1160 /* Return true if a persistent connection is available for connecting
1164 persistent_available_p (const char *host, int port, bool ssl,
1165 bool *host_lookup_failed)
1167 /* First, check whether a persistent connection is active at all. */
1171 /* If we want SSL and the last connection wasn't or vice versa,
1172 don't use it. Checking for host and port is not enough because
1173 HTTP and HTTPS can apparently coexist on the same port. */
1174 if (ssl != pconn.ssl)
1177 /* If we're not connecting to the same port, we're not interested. */
1178 if (port != pconn.port)
1181 /* If the host is the same, we're in business. If not, there is
1182 still hope -- read below. */
1183 if (0 != strcasecmp (host, pconn.host))
1185 /* Check if pconn.socket is talking to HOST under another name.
1186 This happens often when both sites are virtual hosts
1187 distinguished only by name and served by the same network
1188 interface, and hence the same web server (possibly set up by
1189 the ISP and serving many different web sites). This
1190 admittedly unconventional optimization does not contradict
1191 HTTP and works well with popular server software. */
1195 struct address_list *al;
1198 /* Don't try to talk to two different SSL sites over the same
1199 secure connection! (Besides, it's not clear that
1200 name-based virtual hosting is even possible with SSL.) */
1203 /* If pconn.socket's peer is one of the IP addresses HOST
1204 resolves to, pconn.socket is for all intents and purposes
1205 already talking to HOST. */
1207 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1209 /* Can't get the peer's address -- something must be very
1210 wrong with the connection. */
1211 invalidate_persistent ();
1214 al = lookup_host (host, 0);
1217 *host_lookup_failed = true;
1221 found = address_list_contains (al, &ip);
1222 address_list_release (al);
1227 /* The persistent connection's peer address was found among the
1228 addresses HOST resolved to; therefore, pconn.sock is in fact
1229 already talking to HOST -- no need to reconnect. */
1232 /* Finally, check whether the connection is still open. This is
1233 important because most servers implement liberal (short) timeout
1234 on persistent connections. Wget can of course always reconnect
1235 if the connection doesn't work out, but it's nicer to know in
1236 advance. This test is a logical followup of the first test, but
1237 is "expensive" and therefore placed at the end of the list.
1239 (Current implementation of test_socket_open has a nice side
1240 effect that it treats sockets with pending data as "closed".
1241 This is exactly what we want: if a broken server sends message
1242 body in response to HEAD, or if it sends more than conent-length
1243 data, we won't reuse the corrupted connection.) */
1245 if (!test_socket_open (pconn.socket))
1247 /* Oops, the socket is no longer open. Now that we know that,
1248 let's invalidate the persistent connection before returning
1250 invalidate_persistent ();
1257 /* The idea behind these two CLOSE macros is to distinguish between
1258 two cases: one when the job we've been doing is finished, and we
1259 want to close the connection and leave, and two when something is
1260 seriously wrong and we're closing the connection as part of
1263 In case of keep_alive, CLOSE_FINISH should leave the connection
1264 open, while CLOSE_INVALIDATE should still close it.
1266 Note that the semantics of the flag `keep_alive' is "this
1267 connection *will* be reused (the server has promised not to close
1268 the connection once we're done)", while the semantics of
1269 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1270 active, registered connection". */
1272 #define CLOSE_FINISH(fd) do { \
1275 if (pconn_active && (fd) == pconn.socket) \
1276 invalidate_persistent (); \
1285 #define CLOSE_INVALIDATE(fd) do { \
1286 if (pconn_active && (fd) == pconn.socket) \
1287 invalidate_persistent (); \
1295 wgint len; /* received length */
1296 wgint contlen; /* expected length */
1297 wgint restval; /* the restart value */
1298 int res; /* the result of last read */
1299 char *rderrmsg; /* error message from read error */
1300 char *newloc; /* new location (redirection) */
1301 char *remote_time; /* remote time-stamp string */
1302 char *error; /* textual HTTP error */
1303 int statcode; /* status code */
1304 char *message; /* status message */
1305 wgint rd_size; /* amount of data read from socket */
1306 double dltime; /* time it took to download the data */
1307 const char *referer; /* value of the referer header. */
1308 char *local_file; /* local file name. */
1309 bool existence_checked; /* true if we already checked for a file's
1310 existence after having begun to download
1311 (needed in gethttp for when connection is
1312 interrupted/restarted. */
1313 bool timestamp_checked; /* true if pre-download time-stamping checks
1314 * have already been performed */
1315 char *orig_file_name; /* name of file to compare for time-stamping
1316 * (might be != local_file if -K is set) */
1317 wgint orig_file_size; /* size of file to compare for time-stamping */
1318 time_t orig_file_tstamp; /* time-stamp of file to compare for
1323 free_hstat (struct http_stat *hs)
1325 xfree_null (hs->newloc);
1326 xfree_null (hs->remote_time);
1327 xfree_null (hs->error);
1328 xfree_null (hs->rderrmsg);
1329 xfree_null (hs->local_file);
1330 xfree_null (hs->orig_file_name);
1331 xfree_null (hs->message);
1333 /* Guard against being called twice. */
1335 hs->remote_time = NULL;
1339 #define BEGINS_WITH(line, string_constant) \
1340 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1341 && (c_isspace (line[sizeof (string_constant) - 1]) \
1342 || !line[sizeof (string_constant) - 1]))
1344 #define SET_USER_AGENT(req) do { \
1345 if (!opt.useragent) \
1346 request_set_header (req, "User-Agent", \
1347 aprintf ("Wget/%s", version_string), rel_value); \
1348 else if (*opt.useragent) \
1349 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1352 /* The flags that allow clobbering the file (opening with "wb").
1353 Defined here to avoid repetition later. #### This will require
1355 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1356 || opt.dirstruct || opt.output_document)
1358 /* Retrieve a document through HTTP protocol. It recognizes status
1359 code, and correctly handles redirections. It closes the network
1360 socket. If it receives an error from the functions below it, it
1361 will print it if there is enough information to do so (almost
1362 always), returning the error to the caller (i.e. http_loop).
1364 Various HTTP parameters are stored to hs.
1366 If PROXY is non-NULL, the connection will be made to the proxy
1367 server, and u->url will be requested. */
1369 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1372 struct request *req;
1375 char *user, *passwd;
1379 wgint contlen, contrange;
1386 /* Set to 1 when the authorization has already been sent and should
1387 not be tried again. */
1388 bool auth_finished = false;
1390 /* Set to 1 when just globally-set Basic authorization has been sent;
1391 * should prevent further Basic negotiations, but not other
1393 bool basic_auth_finished = false;
1395 /* Whether NTLM authentication is used for this request. */
1396 bool ntlm_seen = false;
1398 /* Whether our connection to the remote host is through SSL. */
1399 bool using_ssl = false;
1401 /* Whether a HEAD request will be issued (as opposed to GET or
1403 bool head_only = !!(*dt & HEAD_ONLY);
1406 struct response *resp;
1410 /* Whether this connection will be kept alive after the HTTP request
1414 /* Whether keep-alive should be inhibited.
1416 RFC 2068 requests that 1.0 clients not send keep-alive requests
1417 to proxies. This is because many 1.0 proxies do not interpret
1418 the Connection header and transfer it to the remote server,
1419 causing it to not close the connection and leave both the proxy
1420 and the client hanging. */
1421 bool inhibit_keep_alive =
1422 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1424 /* Headers sent when using POST. */
1425 wgint post_data_size = 0;
1427 bool host_lookup_failed = false;
1430 if (u->scheme == SCHEME_HTTPS)
1432 /* Initialize the SSL context. After this has once been done,
1433 it becomes a no-op. */
1436 scheme_disable (SCHEME_HTTPS);
1437 logprintf (LOG_NOTQUIET,
1438 _("Disabling SSL due to encountered errors.\n"));
1439 return SSLINITFAILED;
1442 #endif /* HAVE_SSL */
1444 /* Initialize certain elements of struct http_stat. */
1448 hs->rderrmsg = NULL;
1450 hs->remote_time = NULL;
1456 /* Prepare the request to send. */
1458 req = request_new ();
1461 const char *meth = "GET";
1464 else if (opt.post_file_name || opt.post_data)
1466 /* Use the full path, i.e. one that includes the leading slash and
1467 the query string. E.g. if u->path is "foo/bar" and u->query is
1468 "param=value", full_path will be "/foo/bar?param=value". */
1471 /* When using SSL over proxy, CONNECT establishes a direct
1472 connection to the HTTPS server. Therefore use the same
1473 argument as when talking to the server directly. */
1474 && u->scheme != SCHEME_HTTPS
1477 meth_arg = xstrdup (u->url);
1479 meth_arg = url_full_path (u);
1480 request_set_method (req, meth, meth_arg);
1483 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1484 if (*dt & SEND_NOCACHE)
1485 request_set_header (req, "Pragma", "no-cache", rel_none);
1487 request_set_header (req, "Range",
1488 aprintf ("bytes=%s-",
1489 number_to_static_string (hs->restval)),
1491 SET_USER_AGENT (req);
1492 request_set_header (req, "Accept", "*/*", rel_none);
1494 /* Find the username and password for authentication. */
1497 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1498 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1499 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1501 /* We only do "site-wide" authentication with "global" user/password
1502 * values unless --auth-no-challange has been requested; URL user/password
1503 * info overrides. */
1504 if (user && passwd && (!u->user || opt.auth_without_challenge))
1506 /* If this is a host for which we've already received a Basic
1507 * challenge, we'll go ahead and send Basic authentication creds. */
1508 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1511 /* Generate the Host header, HOST:PORT. Take into account that:
1513 - Broken server-side software often doesn't recognize the PORT
1514 argument, so we must generate "Host: www.server.com" instead of
1515 "Host: www.server.com:80" (and likewise for https port).
1517 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1518 becomes ambiguous and needs to be rewritten as "Host:
1519 [3ffe:8100:200:2::2]:1234". */
1521 /* Formats arranged for hfmt[add_port][add_squares]. */
1522 static const char *hfmt[][2] = {
1523 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1525 int add_port = u->port != scheme_default_port (u->scheme);
1526 int add_squares = strchr (u->host, ':') != NULL;
1527 request_set_header (req, "Host",
1528 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1532 if (!inhibit_keep_alive)
1533 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1536 request_set_header (req, "Cookie",
1537 cookie_header (wget_cookie_jar,
1538 u->host, u->port, u->path,
1540 u->scheme == SCHEME_HTTPS
1547 if (opt.post_data || opt.post_file_name)
1549 request_set_header (req, "Content-Type",
1550 "application/x-www-form-urlencoded", rel_none);
1552 post_data_size = strlen (opt.post_data);
1555 post_data_size = file_size (opt.post_file_name);
1556 if (post_data_size == -1)
1558 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1559 quote (opt.post_file_name), strerror (errno));
1563 request_set_header (req, "Content-Length",
1564 xstrdup (number_to_static_string (post_data_size)),
1568 /* Add the user headers. */
1569 if (opt.user_headers)
1572 for (i = 0; opt.user_headers[i]; i++)
1573 request_set_user_header (req, opt.user_headers[i]);
1577 /* We need to come back here when the initial attempt to retrieve
1578 without authorization header fails. (Expected to happen at least
1579 for the Digest authorization scheme.) */
1584 char *proxy_user, *proxy_passwd;
1585 /* For normal username and password, URL components override
1586 command-line/wgetrc parameters. With proxy
1587 authentication, it's the reverse, because proxy URLs are
1588 normally the "permanent" ones, so command-line args
1589 should take precedence. */
1590 if (opt.proxy_user && opt.proxy_passwd)
1592 proxy_user = opt.proxy_user;
1593 proxy_passwd = opt.proxy_passwd;
1597 proxy_user = proxy->user;
1598 proxy_passwd = proxy->passwd;
1600 /* #### This does not appear right. Can't the proxy request,
1601 say, `Digest' authentication? */
1602 if (proxy_user && proxy_passwd)
1603 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1605 /* If we're using a proxy, we will be connecting to the proxy
1609 /* Proxy authorization over SSL is handled below. */
1611 if (u->scheme != SCHEME_HTTPS)
1613 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1618 /* Establish the connection. */
1620 if (!inhibit_keep_alive)
1622 /* Look for a persistent connection to target host, unless a
1623 proxy is used. The exception is when SSL is in use, in which
1624 case the proxy is nothing but a passthrough to the target
1625 host, registered as a connection to the latter. */
1626 struct url *relevant = conn;
1628 if (u->scheme == SCHEME_HTTPS)
1632 if (persistent_available_p (relevant->host, relevant->port,
1634 relevant->scheme == SCHEME_HTTPS,
1638 &host_lookup_failed))
1640 sock = pconn.socket;
1641 using_ssl = pconn.ssl;
1642 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1643 quotearg_style (escape_quoting_style, pconn.host),
1645 DEBUGP (("Reusing fd %d.\n", sock));
1646 if (pconn.authorized)
1647 /* If the connection is already authorized, the "Basic"
1648 authorization added by code above is unnecessary and
1650 request_remove_header (req, "Authorization");
1652 else if (host_lookup_failed)
1655 logprintf(LOG_NOTQUIET,
1656 _("%s: unable to resolve host address %s\n"),
1657 exec_name, quote (relevant->host));
1664 sock = connect_to_host (conn->host, conn->port);
1673 return (retryable_socket_connect_error (errno)
1674 ? CONERROR : CONIMPOSSIBLE);
1678 if (proxy && u->scheme == SCHEME_HTTPS)
1680 /* When requesting SSL URLs through proxies, use the
1681 CONNECT method to request passthrough. */
1682 struct request *connreq = request_new ();
1683 request_set_method (connreq, "CONNECT",
1684 aprintf ("%s:%d", u->host, u->port));
1685 SET_USER_AGENT (connreq);
1688 request_set_header (connreq, "Proxy-Authorization",
1689 proxyauth, rel_value);
1690 /* Now that PROXYAUTH is part of the CONNECT request,
1691 zero it out so we don't send proxy authorization with
1692 the regular request below. */
1695 /* Examples in rfc2817 use the Host header in CONNECT
1696 requests. I don't see how that gains anything, given
1697 that the contents of Host would be exactly the same as
1698 the contents of CONNECT. */
1700 write_error = request_send (connreq, sock);
1701 request_free (connreq);
1702 if (write_error < 0)
1704 CLOSE_INVALIDATE (sock);
1708 head = read_http_response_head (sock);
1711 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1713 CLOSE_INVALIDATE (sock);
1722 DEBUGP (("proxy responded with: [%s]\n", head));
1724 resp = resp_new (head);
1725 statcode = resp_status (resp, &message);
1726 hs->message = xstrdup (message);
1729 if (statcode != 200)
1732 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1733 message ? quotearg_style (escape_quoting_style, message) : "?");
1734 xfree_null (message);
1737 xfree_null (message);
1739 /* SOCK is now *really* connected to u->host, so update CONN
1740 to reflect this. That way register_persistent will
1741 register SOCK as being connected to u->host:u->port. */
1745 if (conn->scheme == SCHEME_HTTPS)
1747 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1754 #endif /* HAVE_SSL */
1757 /* Send the request to server. */
1758 write_error = request_send (req, sock);
1760 if (write_error >= 0)
1764 DEBUGP (("[POST data: %s]\n", opt.post_data));
1765 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1767 else if (opt.post_file_name && post_data_size != 0)
1768 write_error = post_file (sock, opt.post_file_name, post_data_size);
1771 if (write_error < 0)
1773 CLOSE_INVALIDATE (sock);
1777 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1778 proxy ? "Proxy" : "HTTP");
1783 head = read_http_response_head (sock);
1788 logputs (LOG_NOTQUIET, _("No data received.\n"));
1789 CLOSE_INVALIDATE (sock);
1795 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1797 CLOSE_INVALIDATE (sock);
1802 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1804 resp = resp_new (head);
1806 /* Check for status line. */
1808 statcode = resp_status (resp, &message);
1809 hs->message = xstrdup (message);
1810 if (!opt.server_response)
1811 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1812 message ? quotearg_style (escape_quoting_style, message) : "");
1815 logprintf (LOG_VERBOSE, "\n");
1816 print_server_response (resp, " ");
1819 /* Determine the local filename if needed. Notice that if -O is used
1820 * hstat.local_file is set by http_loop to the argument of -O. */
1821 if (!hs->local_file)
1823 /* Honor Content-Disposition whether possible. */
1824 if (!opt.content_disposition
1825 || !resp_header_copy (resp, "Content-Disposition",
1826 hdrval, sizeof (hdrval))
1827 || !parse_content_disposition (hdrval, &hs->local_file))
1829 /* The Content-Disposition header is missing or broken.
1830 * Choose unique file name according to given URL. */
1831 hs->local_file = url_file_name (u);
1835 /* TODO: perform this check only once. */
1836 if (!hs->existence_checked && file_exists_p (hs->local_file))
1838 if (opt.noclobber && !opt.output_document)
1840 /* If opt.noclobber is turned on and file already exists, do not
1841 retrieve the file. But if the output_document was given, then this
1842 test was already done and the file didn't exist. Hence the !opt.output_document */
1843 logprintf (LOG_VERBOSE, _("\
1844 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1845 /* If the file is there, we suppose it's retrieved OK. */
1848 /* #### Bogusness alert. */
1849 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1850 if (has_html_suffix_p (hs->local_file))
1853 return RETRUNNEEDED;
1855 else if (!ALLOW_CLOBBER)
1857 char *unique = unique_name (hs->local_file, true);
1858 if (unique != hs->local_file)
1859 xfree (hs->local_file);
1860 hs->local_file = unique;
1863 hs->existence_checked = true;
1865 /* Support timestamping */
1866 /* TODO: move this code out of gethttp. */
1867 if (opt.timestamping && !hs->timestamp_checked)
1869 size_t filename_len = strlen (hs->local_file);
1870 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1871 bool local_dot_orig_file_exists = false;
1872 char *local_filename = NULL;
1875 if (opt.backup_converted)
1876 /* If -K is specified, we'll act on the assumption that it was specified
1877 last time these files were downloaded as well, and instead of just
1878 comparing local file X against server file X, we'll compare local
1879 file X.orig (if extant, else X) against server file X. If -K
1880 _wasn't_ specified last time, or the server contains files called
1881 *.orig, -N will be back to not operating correctly with -k. */
1883 /* Would a single s[n]printf() call be faster? --dan
1885 Definitely not. sprintf() is horribly slow. It's a
1886 different question whether the difference between the two
1887 affects a program. Usually I'd say "no", but at one
1888 point I profiled Wget, and found that a measurable and
1889 non-negligible amount of time was lost calling sprintf()
1890 in url.c. Replacing sprintf with inline calls to
1891 strcpy() and number_to_string() made a difference.
1893 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1894 memcpy (filename_plus_orig_suffix + filename_len,
1895 ".orig", sizeof (".orig"));
1897 /* Try to stat() the .orig file. */
1898 if (stat (filename_plus_orig_suffix, &st) == 0)
1900 local_dot_orig_file_exists = true;
1901 local_filename = filename_plus_orig_suffix;
1905 if (!local_dot_orig_file_exists)
1906 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1907 if (stat (hs->local_file, &st) == 0)
1908 local_filename = hs->local_file;
1910 if (local_filename != NULL)
1911 /* There was a local file, so we'll check later to see if the version
1912 the server has is the same version we already have, allowing us to
1915 hs->orig_file_name = xstrdup (local_filename);
1916 hs->orig_file_size = st.st_size;
1917 hs->orig_file_tstamp = st.st_mtime;
1919 /* Modification time granularity is 2 seconds for Windows, so
1920 increase local time by 1 second for later comparison. */
1921 ++hs->orig_file_tstamp;
1926 if (!opt.ignore_length
1927 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1931 parsed = str_to_wgint (hdrval, NULL, 10);
1932 if (parsed == WGINT_MAX && errno == ERANGE)
1935 #### If Content-Length is out of range, it most likely
1936 means that the file is larger than 2G and that we're
1937 compiled without LFS. In that case we should probably
1938 refuse to even attempt to download the file. */
1941 else if (parsed < 0)
1943 /* Negative Content-Length; nonsensical, so we can't
1944 assume any information about the content to receive. */
1951 /* Check for keep-alive related responses. */
1952 if (!inhibit_keep_alive && contlen != -1)
1954 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1956 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1958 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1963 /* The server has promised that it will not close the connection
1964 when we're done. This means that we can register it. */
1965 register_persistent (conn->host, conn->port, sock, using_ssl);
1967 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1969 /* Authorization is required. */
1970 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1971 CLOSE_FINISH (sock);
1973 CLOSE_INVALIDATE (sock);
1974 pconn.authorized = false;
1975 if (!auth_finished && (user && passwd))
1977 /* IIS sends multiple copies of WWW-Authenticate, one with
1978 the value "negotiate", and other(s) with data. Loop over
1979 all the occurrences and pick the one we recognize. */
1981 const char *wabeg, *waend;
1982 char *www_authenticate = NULL;
1984 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1985 &wabeg, &waend)) != -1;
1987 if (known_authentication_scheme_p (wabeg, waend))
1989 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1993 if (!www_authenticate)
1995 /* If the authentication header is missing or
1996 unrecognized, there's no sense in retrying. */
1997 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1999 else if (!basic_auth_finished
2000 || !BEGINS_WITH (www_authenticate, "Basic"))
2003 pth = url_full_path (u);
2004 request_set_header (req, "Authorization",
2005 create_authorization_line (www_authenticate,
2007 request_method (req),
2011 if (BEGINS_WITH (www_authenticate, "NTLM"))
2013 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2015 /* Need to register this host as using basic auth,
2016 * so we automatically send creds next time. */
2017 register_basic_auth_host (u->host);
2020 goto retry_with_auth;
2024 /* We already did Basic auth, and it failed. Gotta
2028 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2032 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2034 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2036 pconn.authorized = true;
2040 hs->statcode = statcode;
2042 hs->error = xstrdup (_("Malformed status line"));
2044 hs->error = xstrdup (_("(no description)"));
2046 hs->error = xstrdup (message);
2047 xfree_null (message);
2049 type = resp_header_strdup (resp, "Content-Type");
2052 char *tmp = strchr (type, ';');
2055 /* sXXXav: only needed if IRI support is enabled */
2056 char *tmp2 = tmp + 1;
2058 while (tmp > type && c_isspace (tmp[-1]))
2062 /* Try to get remote encoding if needed */
2063 if (opt.enable_iri && !opt.encoding_remote)
2065 tmp = parse_charset (tmp2);
2067 set_content_encoding (iri, tmp);
2071 hs->newloc = resp_header_strdup (resp, "Location");
2072 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2074 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2078 const char *scbeg, *scend;
2079 /* The jar should have been created by now. */
2080 assert (wget_cookie_jar != NULL);
2082 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2083 &scbeg, &scend)) != -1;
2086 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2087 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2088 u->path, set_cookie);
2092 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2094 wgint first_byte_pos, last_byte_pos, entity_length;
2095 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2098 contrange = first_byte_pos;
2099 contlen = last_byte_pos - first_byte_pos + 1;
2104 /* 20x responses are counted among successful by default. */
2105 if (H_20X (statcode))
2108 /* Return if redirected. */
2109 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2111 /* RFC2068 says that in case of the 300 (multiple choices)
2112 response, the server can output a preferred URL through
2113 `Location' header; otherwise, the request should be treated
2114 like GET. So, if the location is set, it will be a
2115 redirection; otherwise, just proceed normally. */
2116 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2120 logprintf (LOG_VERBOSE,
2121 _("Location: %s%s\n"),
2122 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2123 hs->newloc ? _(" [following]") : "");
2124 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2125 CLOSE_FINISH (sock);
2127 CLOSE_INVALIDATE (sock);
2133 /* If content-type is not given, assume text/html. This is because
2134 of the multitude of broken CGI's that "forget" to generate the
2137 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2138 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2144 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2149 if (opt.html_extension)
2152 /* -E / --html-extension / html_extension = on was specified,
2153 and this is a text/html file. If some case-insensitive
2154 variation on ".htm[l]" isn't already the file's suffix,
2157 ensure_extension (hs, ".html", dt);
2159 else if (*dt & TEXTCSS)
2161 ensure_extension (hs, ".css", dt);
2165 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2166 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2167 && contrange == 0 && hs->restval >= contlen)
2170 /* If `-c' is in use and the file has been fully downloaded (or
2171 the remote file has shrunk), Wget effectively requests bytes
2172 after the end of file and the server response with 416
2173 (or 200 with a <= Content-Length. */
2174 logputs (LOG_VERBOSE, _("\
2175 \n The file is already fully retrieved; nothing to do.\n\n"));
2176 /* In case the caller inspects. */
2179 /* Mark as successfully retrieved. */
2182 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2183 might be more bytes in the body. */
2184 return RETRUNNEEDED;
2186 if ((contrange != 0 && contrange != hs->restval)
2187 || (H_PARTIAL (statcode) && !contrange))
2189 /* The Range request was somehow misunderstood by the server.
2192 CLOSE_INVALIDATE (sock);
2198 hs->contlen = contlen + contrange;
2204 /* No need to print this output if the body won't be
2205 downloaded at all, or if the original server response is
2207 logputs (LOG_VERBOSE, _("Length: "));
2210 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2211 if (contlen + contrange >= 1024)
2212 logprintf (LOG_VERBOSE, " (%s)",
2213 human_readable (contlen + contrange));
2216 if (contlen >= 1024)
2217 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2218 number_to_static_string (contlen),
2219 human_readable (contlen));
2221 logprintf (LOG_VERBOSE, _(", %s remaining"),
2222 number_to_static_string (contlen));
2226 logputs (LOG_VERBOSE,
2227 opt.ignore_length ? _("ignored") : _("unspecified"));
2229 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2231 logputs (LOG_VERBOSE, "\n");
2235 type = NULL; /* We don't need it any more. */
2237 /* Return if we have no intention of further downloading. */
2238 if (!(*dt & RETROKF) || head_only)
2240 /* In case the caller cares to look... */
2245 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2246 servers not to send body in response to a HEAD request, and
2247 those that do will likely be caught by test_socket_open.
2248 If not, they can be worked around using
2249 `--no-http-keep-alive'. */
2250 CLOSE_FINISH (sock);
2251 else if (keep_alive && skip_short_body (sock, contlen))
2252 /* Successfully skipped the body; also keep using the socket. */
2253 CLOSE_FINISH (sock);
2255 CLOSE_INVALIDATE (sock);
2256 return RETRFINISHED;
2259 /* Open the local file. */
2262 mkalldirs (hs->local_file);
2264 rotate_backups (hs->local_file);
2266 fp = fopen (hs->local_file, "ab");
2267 else if (ALLOW_CLOBBER)
2268 fp = fopen (hs->local_file, "wb");
2271 fp = fopen_excl (hs->local_file, true);
2272 if (!fp && errno == EEXIST)
2274 /* We cannot just invent a new name and use it (which is
2275 what functions like unique_create typically do)
2276 because we told the user we'd use this name.
2277 Instead, return and retry the download. */
2278 logprintf (LOG_NOTQUIET,
2279 _("%s has sprung into existence.\n"),
2281 CLOSE_INVALIDATE (sock);
2282 return FOPEN_EXCL_ERR;
2287 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2288 CLOSE_INVALIDATE (sock);
2295 /* Print fetch message, if opt.verbose. */
2298 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2299 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2302 /* This confuses the timestamping code that checks for file size.
2303 #### The timestamping code should be smarter about file size. */
2304 if (opt.save_headers && hs->restval == 0)
2305 fwrite (head, 1, strlen (head), fp);
2307 /* Now we no longer need to store the response header. */
2310 /* Download the request body. */
2313 /* If content-length is present, read that much; otherwise, read
2314 until EOF. The HTTP spec doesn't require the server to
2315 actually close the connection when it's done sending data. */
2316 flags |= rb_read_exactly;
2317 if (hs->restval > 0 && contrange == 0)
2318 /* If the server ignored our range request, instruct fd_read_body
2319 to skip the first RESTVAL bytes of body. */
2320 flags |= rb_skip_startpos;
2321 hs->len = hs->restval;
2323 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2324 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2328 CLOSE_FINISH (sock);
2332 hs->rderrmsg = xstrdup (fd_errstr (sock));
2333 CLOSE_INVALIDATE (sock);
2340 return RETRFINISHED;
2343 /* The genuine HTTP loop! This is the part where the retrieval is
2344 retried, and retried, and retried, and... */
2346 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2347 int *dt, struct url *proxy, struct iri *iri)
2350 bool got_head = false; /* used for time-stamping and filename detection */
2351 bool time_came_from_head = false;
2352 bool got_name = false;
2355 uerr_t err, ret = TRYLIMEXC;
2356 time_t tmr = -1; /* remote time-stamp */
2357 struct http_stat hstat; /* HTTP status */
2359 bool send_head_first = true;
2361 /* Assert that no value for *LOCAL_FILE was passed. */
2362 assert (local_file == NULL || *local_file == NULL);
2364 /* Set LOCAL_FILE parameter. */
2365 if (local_file && opt.output_document)
2366 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2368 /* Reset NEWLOC parameter. */
2371 /* This used to be done in main(), but it's a better idea to do it
2372 here so that we don't go through the hoops if we're just using
2377 /* Warn on (likely bogus) wildcard usage in HTTP. */
2378 if (opt.ftp_glob && has_wildcards_p (u->path))
2379 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2381 /* Setup hstat struct. */
2383 hstat.referer = referer;
2385 if (opt.output_document)
2387 hstat.local_file = xstrdup (opt.output_document);
2390 else if (!opt.content_disposition)
2392 hstat.local_file = url_file_name (u);
2396 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2397 * screaming for some refactoring. */
2398 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2400 /* If opt.noclobber is turned on and file already exists, do not
2401 retrieve the file. But if the output_document was given, then this
2402 test was already done and the file didn't exist. Hence the !opt.output_document */
2403 logprintf (LOG_VERBOSE, _("\
2404 File %s already there; not retrieving.\n\n"),
2405 quote (hstat.local_file));
2406 /* If the file is there, we suppose it's retrieved OK. */
2409 /* #### Bogusness alert. */
2410 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2411 if (has_html_suffix_p (hstat.local_file))
2418 /* Reset the counter. */
2421 /* Reset the document type. */
2424 /* Skip preliminary HEAD request if we're not in spider mode AND
2425 * if -O was given or HTTP Content-Disposition support is disabled. */
2427 && (got_name || !opt.content_disposition))
2428 send_head_first = false;
2430 /* Send preliminary HEAD request if -N is given and we have an existing
2431 * destination file. */
2432 if (opt.timestamping
2433 && !opt.content_disposition
2434 && file_exists_p (url_file_name (u)))
2435 send_head_first = true;
2440 /* Increment the pass counter. */
2442 sleep_between_retrievals (count);
2444 /* Get the current time string. */
2445 tms = datetime_str (time (NULL));
2447 if (opt.spider && !got_head)
2448 logprintf (LOG_VERBOSE, _("\
2449 Spider mode enabled. Check if remote file exists.\n"));
2451 /* Print fetch message, if opt.verbose. */
2454 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2459 sprintf (tmp, _("(try:%2d)"), count);
2460 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2465 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2470 ws_changetitle (hurl);
2475 /* Default document type is empty. However, if spider mode is
2476 on or time-stamping is employed, HEAD_ONLY commands is
2477 encoded within *dt. */
2478 if (send_head_first && !got_head)
2483 /* Decide whether or not to restart. */
2486 && stat (hstat.local_file, &st) == 0
2487 && S_ISREG (st.st_mode))
2488 /* When -c is used, continue from on-disk size. (Can't use
2489 hstat.len even if count>1 because we don't want a failed
2490 first attempt to clobber existing data.) */
2491 hstat.restval = st.st_size;
2493 /* otherwise, continue where the previous try left off */
2494 hstat.restval = hstat.len;
2498 /* Decide whether to send the no-cache directive. We send it in
2500 a) we're using a proxy, and we're past our first retrieval.
2501 Some proxies are notorious for caching incomplete data, so
2502 we require a fresh get.
2503 b) caching is explicitly inhibited. */
2504 if ((proxy && count > 1) /* a */
2505 || !opt.allow_cache) /* b */
2506 *dt |= SEND_NOCACHE;
2508 *dt &= ~SEND_NOCACHE;
2510 /* Try fetching the document, or at least its head. */
2511 err = gethttp (u, &hstat, dt, proxy, iri);
2514 tms = datetime_str (time (NULL));
2516 /* Get the new location (with or without the redirection). */
2518 *newloc = xstrdup (hstat.newloc);
2522 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2523 case CONERROR: case READERR: case WRITEFAILED:
2524 case RANGEERR: case FOPEN_EXCL_ERR:
2525 /* Non-fatal errors continue executing the loop, which will
2526 bring them to "while" statement at the end, to judge
2527 whether the number of tries was exceeded. */
2528 printwhat (count, opt.ntry);
2530 case FWRITEERR: case FOPENERR:
2531 /* Another fatal error. */
2532 logputs (LOG_VERBOSE, "\n");
2533 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2534 quote (hstat.local_file), strerror (errno));
2535 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2536 case SSLINITFAILED: case CONTNOTSUPPORTED:
2537 /* Fatal errors just return from the function. */
2541 /* Another fatal error. */
2542 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2546 /* Return the new location to the caller. */
2549 logprintf (LOG_NOTQUIET,
2550 _("ERROR: Redirection (%d) without location.\n"),
2560 /* The file was already fully retrieved. */
2564 /* Deal with you later. */
2567 /* All possibilities should have been exhausted. */
2571 if (!(*dt & RETROKF))
2576 /* #### Ugly ugly ugly! */
2577 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2578 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2581 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2583 && (hstat.statcode == 500 || hstat.statcode == 501))
2588 /* Maybe we should always keep track of broken links, not just in
2590 * Don't log error if it was UTF-8 encoded because we will try
2591 * once unencoded. */
2592 else if (opt.spider && !iri->utf8_encode)
2594 /* #### Again: ugly ugly ugly! */
2596 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2597 nonexisting_url (hurl);
2598 logprintf (LOG_NOTQUIET, _("\
2599 Remote file does not exist -- broken link!!!\n"));
2603 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2604 tms, hstat.statcode,
2605 quotearg_style (escape_quoting_style, hstat.error));
2607 logputs (LOG_VERBOSE, "\n");
2613 /* Did we get the time-stamp? */
2616 got_head = true; /* no more time-stamping */
2618 if (opt.timestamping && !hstat.remote_time)
2620 logputs (LOG_NOTQUIET, _("\
2621 Last-modified header missing -- time-stamps turned off.\n"));
2623 else if (hstat.remote_time)
2625 /* Convert the date-string into struct tm. */
2626 tmr = http_atotm (hstat.remote_time);
2627 if (tmr == (time_t) (-1))
2628 logputs (LOG_VERBOSE, _("\
2629 Last-modified header invalid -- time-stamp ignored.\n"));
2630 if (*dt & HEAD_ONLY)
2631 time_came_from_head = true;
2634 if (send_head_first)
2636 /* The time-stamping section. */
2637 if (opt.timestamping)
2639 if (hstat.orig_file_name) /* Perform the following
2640 checks only if the file
2642 download already exists. */
2644 if (hstat.remote_time &&
2645 tmr != (time_t) (-1))
2647 /* Now time-stamping can be used validly.
2648 Time-stamping means that if the sizes of
2649 the local and remote file match, and local
2650 file is newer than the remote file, it will
2651 not be retrieved. Otherwise, the normal
2652 download procedure is resumed. */
2653 if (hstat.orig_file_tstamp >= tmr)
2655 if (hstat.contlen == -1
2656 || hstat.orig_file_size == hstat.contlen)
2658 logprintf (LOG_VERBOSE, _("\
2659 Server file no newer than local file %s -- not retrieving.\n\n"),
2660 quote (hstat.orig_file_name));
2666 logprintf (LOG_VERBOSE, _("\
2667 The sizes do not match (local %s) -- retrieving.\n"),
2668 number_to_static_string (hstat.orig_file_size));
2672 logputs (LOG_VERBOSE,
2673 _("Remote file is newer, retrieving.\n"));
2675 logputs (LOG_VERBOSE, "\n");
2679 /* free_hstat (&hstat); */
2680 hstat.timestamp_checked = true;
2685 bool finished = true;
2690 logputs (LOG_VERBOSE, _("\
2691 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2696 logprintf (LOG_VERBOSE, _("\
2697 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2698 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2705 logprintf (LOG_VERBOSE, _("\
2706 Remote file exists and could contain further links,\n\
2707 but recursion is disabled -- not retrieving.\n\n"));
2711 logprintf (LOG_VERBOSE, _("\
2712 Remote file exists.\n\n"));
2714 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2719 logprintf (LOG_NONVERBOSE,
2720 _("%s URL:%s %2d %s\n"),
2721 tms, u->url, hstat.statcode,
2722 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2729 count = 0; /* the retrieve count for HEAD is reset */
2731 } /* send_head_first */
2734 if ((tmr != (time_t) (-1))
2735 && ((hstat.len == hstat.contlen) ||
2736 ((hstat.res == 0) && (hstat.contlen == -1))))
2738 /* #### This code repeats in http.c and ftp.c. Move it to a
2740 const char *fl = NULL;
2741 if (opt.output_document)
2743 if (output_stream_regular)
2744 fl = opt.output_document;
2747 fl = hstat.local_file;
2751 /* Reparse time header, in case it's changed. */
2752 if (time_came_from_head
2753 && hstat.remote_time && hstat.remote_time[0])
2755 newtmr = http_atotm (hstat.remote_time);
2762 /* End of time-stamping section. */
2764 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2765 total_download_time += hstat.dltime;
2767 if (hstat.len == hstat.contlen)
2771 logprintf (LOG_VERBOSE,
2772 _("%s (%s) - %s saved [%s/%s]\n\n"),
2773 tms, tmrate, quote (hstat.local_file),
2774 number_to_static_string (hstat.len),
2775 number_to_static_string (hstat.contlen));
2776 logprintf (LOG_NONVERBOSE,
2777 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2779 number_to_static_string (hstat.len),
2780 number_to_static_string (hstat.contlen),
2781 hstat.local_file, count);
2784 total_downloaded_bytes += hstat.len;
2786 /* Remember that we downloaded the file for later ".orig" code. */
2787 if (*dt & ADDED_HTML_EXTENSION)
2788 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2790 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2795 else if (hstat.res == 0) /* No read error */
2797 if (hstat.contlen == -1) /* We don't know how much we were supposed
2798 to get, so assume we succeeded. */
2802 logprintf (LOG_VERBOSE,
2803 _("%s (%s) - %s saved [%s]\n\n"),
2804 tms, tmrate, quote (hstat.local_file),
2805 number_to_static_string (hstat.len));
2806 logprintf (LOG_NONVERBOSE,
2807 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2808 tms, u->url, number_to_static_string (hstat.len),
2809 hstat.local_file, count);
2812 total_downloaded_bytes += hstat.len;
2814 /* Remember that we downloaded the file for later ".orig" code. */
2815 if (*dt & ADDED_HTML_EXTENSION)
2816 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2818 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2823 else if (hstat.len < hstat.contlen) /* meaning we lost the
2824 connection too soon */
2826 logprintf (LOG_VERBOSE,
2827 _("%s (%s) - Connection closed at byte %s. "),
2828 tms, tmrate, number_to_static_string (hstat.len));
2829 printwhat (count, opt.ntry);
2832 else if (hstat.len != hstat.restval)
2833 /* Getting here would mean reading more data than
2834 requested with content-length, which we never do. */
2838 /* Getting here probably means that the content-length was
2839 * _less_ than the original, local size. We should probably
2840 * truncate or re-read, or something. FIXME */
2845 else /* from now on hstat.res can only be -1 */
2847 if (hstat.contlen == -1)
2849 logprintf (LOG_VERBOSE,
2850 _("%s (%s) - Read error at byte %s (%s)."),
2851 tms, tmrate, number_to_static_string (hstat.len),
2853 printwhat (count, opt.ntry);
2856 else /* hstat.res == -1 and contlen is given */
2858 logprintf (LOG_VERBOSE,
2859 _("%s (%s) - Read error at byte %s/%s (%s). "),
2861 number_to_static_string (hstat.len),
2862 number_to_static_string (hstat.contlen),
2864 printwhat (count, opt.ntry);
2870 while (!opt.ntry || (count < opt.ntry));
2874 *local_file = xstrdup (hstat.local_file);
2875 free_hstat (&hstat);
2880 /* Check whether the result of strptime() indicates success.
2881 strptime() returns the pointer to how far it got to in the string.
2882 The processing has been successful if the string is at `GMT' or
2883 `+X', or at the end of the string.
2885 In extended regexp parlance, the function returns 1 if P matches
2886 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2887 can return) is considered a failure and 0 is returned. */
2889 check_end (const char *p)
2893 while (c_isspace (*p))
2896 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2897 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2903 /* Convert the textual specification of time in TIME_STRING to the
2904 number of seconds since the Epoch.
2906 TIME_STRING can be in any of the three formats RFC2616 allows the
2907 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2908 as well as the time format used in the Set-Cookie header.
2909 Timezones are ignored, and should be GMT.
2911 Return the computed time_t representation, or -1 if the conversion
2914 This function uses strptime with various string formats for parsing
2915 TIME_STRING. This results in a parser that is not as lenient in
2916 interpreting TIME_STRING as I would like it to be. Being based on
2917 strptime, it always allows shortened months, one-digit days, etc.,
2918 but due to the multitude of formats in which time can be
2919 represented, an ideal HTTP time parser would be even more
2920 forgiving. It should completely ignore things like week days and
2921 concentrate only on the various forms of representing years,
2922 months, days, hours, minutes, and seconds. For example, it would
2923 be nice if it accepted ISO 8601 out of the box.
2925 I've investigated free and PD code for this purpose, but none was
2926 usable. getdate was big and unwieldy, and had potential copyright
2927 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2928 distributed with phttpd, is excellent, but we cannot use it because
2929 it is not assigned to the FSF. So I stuck it with strptime. */
2932 http_atotm (const char *time_string)
2934 /* NOTE: Solaris strptime man page claims that %n and %t match white
2935 space, but that's not universally available. Instead, we simply
2936 use ` ' to mean "skip all WS", which works under all strptime
2937 implementations I've tested. */
2939 static const char *time_formats[] = {
2940 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2941 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2942 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2943 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2944 (used in Set-Cookie, defined in the
2945 Netscape cookie specification.) */
2947 const char *oldlocale;
2949 time_t ret = (time_t) -1;
2951 /* Solaris strptime fails to recognize English month names in
2952 non-English locales, which we work around by temporarily setting
2953 locale to C before invoking strptime. */
2954 oldlocale = setlocale (LC_TIME, NULL);
2955 setlocale (LC_TIME, "C");
2957 for (i = 0; i < countof (time_formats); i++)
2961 /* Some versions of strptime use the existing contents of struct
2962 tm to recalculate the date according to format. Zero it out
2963 to prevent stack garbage from influencing strptime. */
2966 if (check_end (strptime (time_string, time_formats[i], &t)))
2973 /* Restore the previous locale. */
2974 setlocale (LC_TIME, oldlocale);
2979 /* Authorization support: We support three authorization schemes:
2981 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2983 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2984 consisting of answering to the server's challenge with the proper
2987 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2988 Stenberg for libcurl. Like digest, NTLM is based on a
2989 challenge-response mechanism, but unlike digest, it is non-standard
2990 (authenticates TCP connections rather than requests), undocumented
2991 and Microsoft-specific. */
2993 /* Create the authentication header contents for the `Basic' scheme.
2994 This is done by encoding the string "USER:PASS" to base64 and
2995 prepending the string "Basic " in front of it. */
2998 basic_authentication_encode (const char *user, const char *passwd)
3001 int len1 = strlen (user) + 1 + strlen (passwd);
3003 t1 = (char *)alloca (len1 + 1);
3004 sprintf (t1, "%s:%s", user, passwd);
3006 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3007 base64_encode (t1, len1, t2);
3009 return concat_strings ("Basic ", t2, (char *) 0);
3012 #define SKIP_WS(x) do { \
3013 while (c_isspace (*(x))) \
3017 #ifdef ENABLE_DIGEST
3018 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3019 an array of 16 bytes containing the hash keys, and BUF should be a
3020 buffer of 33 writable characters (32 for hex digits plus one for
3021 zero termination). */
3023 dump_hash (char *buf, const unsigned char *hash)
3027 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3029 *buf++ = XNUM_TO_digit (*hash >> 4);
3030 *buf++ = XNUM_TO_digit (*hash & 0xf);
3035 /* Take the line apart to find the challenge, and compose a digest
3036 authorization header. See RFC2069 section 2.1.2. */
3038 digest_authentication_encode (const char *au, const char *user,
3039 const char *passwd, const char *method,
3042 static char *realm, *opaque, *nonce;
3047 { "realm", &realm },
3048 { "opaque", &opaque },
3052 param_token name, value;
3054 realm = opaque = nonce = NULL;
3056 au += 6; /* skip over `Digest' */
3057 while (extract_param (&au, &name, &value, ','))
3060 size_t namelen = name.e - name.b;
3061 for (i = 0; i < countof (options); i++)
3062 if (namelen == strlen (options[i].name)
3063 && 0 == strncmp (name.b, options[i].name,
3066 *options[i].variable = strdupdelim (value.b, value.e);
3070 if (!realm || !nonce || !user || !passwd || !path || !method)
3073 xfree_null (opaque);
3078 /* Calculate the digest value. */
3080 ALLOCA_MD5_CONTEXT (ctx);
3081 unsigned char hash[MD5_HASHLEN];
3082 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3083 char response_digest[MD5_HASHLEN * 2 + 1];
3085 /* A1BUF = H(user ":" realm ":" password) */
3087 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3088 gen_md5_update ((unsigned char *)":", 1, ctx);
3089 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3090 gen_md5_update ((unsigned char *)":", 1, ctx);
3091 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3092 gen_md5_finish (ctx, hash);
3093 dump_hash (a1buf, hash);
3095 /* A2BUF = H(method ":" path) */
3097 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3098 gen_md5_update ((unsigned char *)":", 1, ctx);
3099 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3100 gen_md5_finish (ctx, hash);
3101 dump_hash (a2buf, hash);
3103 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3105 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3106 gen_md5_update ((unsigned char *)":", 1, ctx);
3107 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3108 gen_md5_update ((unsigned char *)":", 1, ctx);
3109 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3110 gen_md5_finish (ctx, hash);
3111 dump_hash (response_digest, hash);
3113 res = xmalloc (strlen (user)
3118 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3119 + (opaque ? strlen (opaque) : 0)
3121 sprintf (res, "Digest \
3122 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3123 user, realm, nonce, path, response_digest);
3126 char *p = res + strlen (res);
3127 strcat (p, ", opaque=\"");
3134 #endif /* ENABLE_DIGEST */
3136 /* Computing the size of a string literal must take into account that
3137 value returned by sizeof includes the terminating \0. */
3138 #define STRSIZE(literal) (sizeof (literal) - 1)
3140 /* Whether chars in [b, e) begin with the literal string provided as
3141 first argument and are followed by whitespace or terminating \0.
3142 The comparison is case-insensitive. */
3143 #define STARTS(literal, b, e) \
3145 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3146 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3147 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3148 || c_isspace (b[STRSIZE (literal)])))
3151 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3153 return STARTS ("Basic", hdrbeg, hdrend)
3154 #ifdef ENABLE_DIGEST
3155 || STARTS ("Digest", hdrbeg, hdrend)
3158 || STARTS ("NTLM", hdrbeg, hdrend)
3165 /* Create the HTTP authorization request header. When the
3166 `WWW-Authenticate' response header is seen, according to the
3167 authorization scheme specified in that header (`Basic' and `Digest'
3168 are supported by the current implementation), produce an
3169 appropriate HTTP authorization request header. */
3171 create_authorization_line (const char *au, const char *user,
3172 const char *passwd, const char *method,
3173 const char *path, bool *finished)
3175 /* We are called only with known schemes, so we can dispatch on the
3177 switch (c_toupper (*au))
3179 case 'B': /* Basic */
3181 return basic_authentication_encode (user, passwd);
3182 #ifdef ENABLE_DIGEST
3183 case 'D': /* Digest */
3185 return digest_authentication_encode (au, user, passwd, method, path);
3188 case 'N': /* NTLM */
3189 if (!ntlm_input (&pconn.ntlm, au))
3194 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3197 /* We shouldn't get here -- this function should be only called
3198 with values approved by known_authentication_scheme_p. */
3206 if (!wget_cookie_jar)
3207 wget_cookie_jar = cookie_jar_new ();
3208 if (opt.cookies_input && !cookies_loaded_p)
3210 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3211 cookies_loaded_p = true;
3218 if (wget_cookie_jar)
3219 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3225 xfree_null (pconn.host);
3226 if (wget_cookie_jar)
3227 cookie_jar_delete (wget_cookie_jar);
3231 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3233 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3235 int len = strlen (ext);
3238 strncpy (shortext, ext, len - 1);
3239 shortext[len - 2] = '\0';
3242 if (last_period_in_local_filename == NULL
3243 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3244 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3246 int local_filename_len = strlen (hs->local_file);
3247 /* Resize the local file, allowing for ".html" preceded by
3248 optional ".NUMBER". */
3249 hs->local_file = xrealloc (hs->local_file,
3250 local_filename_len + 24 + len);
3251 strcpy (hs->local_file + local_filename_len, ext);
3252 /* If clobbering is not allowed and the file, as named,
3253 exists, tack on ".NUMBER.html" instead. */
3254 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3258 sprintf (hs->local_file + local_filename_len,
3259 ".%d%s", ext_num++, ext);
3260 while (file_exists_p (hs->local_file));
3262 *dt |= ADDED_HTML_EXTENSION;
3270 test_parse_content_disposition()
3275 char *opt_dir_prefix;
3279 { "filename=\"file.ext\"", NULL, "file.ext", true },
3280 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3281 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3282 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3283 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3284 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3285 { "attachment", NULL, NULL, false },
3286 { "attachment", "somedir", NULL, false },
3289 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3294 opt.dir_prefix = test_array[i].opt_dir_prefix;
3295 res = parse_content_disposition (test_array[i].hdrval, &filename);
3297 mu_assert ("test_parse_content_disposition: wrong result",
3298 res == test_array[i].result
3300 || 0 == strcmp (test_array[i].filename, filename)));
3306 #endif /* TESTING */
3309 * vim: et sts=2 sw=2 cino+={s