2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
57 # include "http-ntlm.h"
70 extern char *version_string;
74 static char *create_authorization_line (const char *, const char *,
75 const char *, const char *,
76 const char *, bool *);
77 static char *basic_authentication_encode (const char *, const char *);
78 static bool known_authentication_scheme_p (const char *, const char *);
79 static void ensure_extension (struct http_stat *, const char *, int *);
80 static void load_cookies (void);
83 # define MIN(x, y) ((x) > (y) ? (y) : (x))
87 static bool cookies_loaded_p;
88 static struct cookie_jar *wget_cookie_jar;
90 #define TEXTHTML_S "text/html"
91 #define TEXTXHTML_S "application/xhtml+xml"
92 #define TEXTCSS_S "text/css"
94 /* Some status code validation macros: */
95 #define H_20X(x) (((x) >= 200) && ((x) < 300))
96 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
97 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
98 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
99 || (x) == HTTP_STATUS_SEE_OTHER \
100 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
102 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
103 /* Successful 2xx. */
104 #define HTTP_STATUS_OK 200
105 #define HTTP_STATUS_CREATED 201
106 #define HTTP_STATUS_ACCEPTED 202
107 #define HTTP_STATUS_NO_CONTENT 204
108 #define HTTP_STATUS_PARTIAL_CONTENTS 206
110 /* Redirection 3xx. */
111 #define HTTP_STATUS_MULTIPLE_CHOICES 300
112 #define HTTP_STATUS_MOVED_PERMANENTLY 301
113 #define HTTP_STATUS_MOVED_TEMPORARILY 302
114 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
115 #define HTTP_STATUS_NOT_MODIFIED 304
116 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
118 /* Client error 4xx. */
119 #define HTTP_STATUS_BAD_REQUEST 400
120 #define HTTP_STATUS_UNAUTHORIZED 401
121 #define HTTP_STATUS_FORBIDDEN 403
122 #define HTTP_STATUS_NOT_FOUND 404
123 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
125 /* Server errors 5xx. */
126 #define HTTP_STATUS_INTERNAL 500
127 #define HTTP_STATUS_NOT_IMPLEMENTED 501
128 #define HTTP_STATUS_BAD_GATEWAY 502
129 #define HTTP_STATUS_UNAVAILABLE 503
132 rel_none, rel_name, rel_value, rel_both
139 struct request_header {
141 enum rp release_policy;
143 int hcount, hcapacity;
146 /* Create a new, empty request. At least request_set_method must be
147 called before the request can be used. */
149 static struct request *
152 struct request *req = xnew0 (struct request);
154 req->headers = xnew_array (struct request_header, req->hcapacity);
158 /* Set the request's method and its arguments. METH should be a
159 literal string (or it should outlive the request) because it will
160 not be freed. ARG will be freed by request_free. */
163 request_set_method (struct request *req, const char *meth, char *arg)
169 /* Return the method string passed with the last call to
170 request_set_method. */
173 request_method (const struct request *req)
178 /* Free one header according to the release policy specified with
179 request_set_header. */
182 release_header (struct request_header *hdr)
184 switch (hdr->release_policy)
201 /* Set the request named NAME to VALUE. Specifically, this means that
202 a "NAME: VALUE\r\n" header line will be used in the request. If a
203 header with the same name previously existed in the request, its
204 value will be replaced by this one. A NULL value means do nothing.
206 RELEASE_POLICY determines whether NAME and VALUE should be released
207 (freed) with request_free. Allowed values are:
209 - rel_none - don't free NAME or VALUE
210 - rel_name - free NAME when done
211 - rel_value - free VALUE when done
212 - rel_both - free both NAME and VALUE when done
214 Setting release policy is useful when arguments come from different
215 sources. For example:
217 // Don't free literal strings!
218 request_set_header (req, "Pragma", "no-cache", rel_none);
220 // Don't free a global variable, we'll need it later.
221 request_set_header (req, "Referer", opt.referer, rel_none);
223 // Value freshly allocated, free it when done.
224 request_set_header (req, "Range",
225 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
230 request_set_header (struct request *req, char *name, char *value,
231 enum rp release_policy)
233 struct request_header *hdr;
238 /* A NULL value is a no-op; if freeing the name is requested,
239 free it now to avoid leaks. */
240 if (release_policy == rel_name || release_policy == rel_both)
245 for (i = 0; i < req->hcount; i++)
247 hdr = &req->headers[i];
248 if (0 == strcasecmp (name, hdr->name))
250 /* Replace existing header. */
251 release_header (hdr);
254 hdr->release_policy = release_policy;
259 /* Install new header. */
261 if (req->hcount >= req->hcapacity)
263 req->hcapacity <<= 1;
264 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
266 hdr = &req->headers[req->hcount++];
269 hdr->release_policy = release_policy;
272 /* Like request_set_header, but sets the whole header line, as
273 provided by the user using the `--header' option. For example,
274 request_set_user_header (req, "Foo: bar") works just like
275 request_set_header (req, "Foo", "bar"). */
278 request_set_user_header (struct request *req, const char *header)
281 const char *p = strchr (header, ':');
284 BOUNDED_TO_ALLOCA (header, p, name);
286 while (c_isspace (*p))
288 request_set_header (req, xstrdup (name), (char *) p, rel_name);
291 /* Remove the header with specified name from REQ. Returns true if
292 the header was actually removed, false otherwise. */
295 request_remove_header (struct request *req, char *name)
298 for (i = 0; i < req->hcount; i++)
300 struct request_header *hdr = &req->headers[i];
301 if (0 == strcasecmp (name, hdr->name))
303 release_header (hdr);
304 /* Move the remaining headers by one. */
305 if (i < req->hcount - 1)
306 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
314 #define APPEND(p, str) do { \
315 int A_len = strlen (str); \
316 memcpy (p, str, A_len); \
320 /* Construct the request and write it to FD using fd_write. */
323 request_send (const struct request *req, int fd)
325 char *request_string, *p;
326 int i, size, write_error;
328 /* Count the request size. */
331 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
332 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
334 for (i = 0; i < req->hcount; i++)
336 struct request_header *hdr = &req->headers[i];
337 /* NAME ": " VALUE "\r\n" */
338 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
344 p = request_string = alloca_array (char, size);
346 /* Generate the request. */
348 APPEND (p, req->method); *p++ = ' ';
349 APPEND (p, req->arg); *p++ = ' ';
350 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
352 for (i = 0; i < req->hcount; i++)
354 struct request_header *hdr = &req->headers[i];
355 APPEND (p, hdr->name);
356 *p++ = ':', *p++ = ' ';
357 APPEND (p, hdr->value);
358 *p++ = '\r', *p++ = '\n';
361 *p++ = '\r', *p++ = '\n', *p++ = '\0';
362 assert (p - request_string == size);
366 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
368 /* Send the request to the server. */
370 write_error = fd_write (fd, request_string, size - 1, -1);
372 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
377 /* Release the resources used by REQ. */
380 request_free (struct request *req)
383 xfree_null (req->arg);
384 for (i = 0; i < req->hcount; i++)
385 release_header (&req->headers[i]);
386 xfree_null (req->headers);
390 static struct hash_table *basic_authed_hosts;
392 /* Find out if this host has issued a Basic challenge yet; if so, give
393 * it the username, password. A temporary measure until we can get
394 * proper authentication in place. */
397 maybe_send_basic_creds (const char *hostname, const char *user,
398 const char *passwd, struct request *req)
400 bool do_challenge = false;
402 if (opt.auth_without_challenge)
404 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
407 else if (basic_authed_hosts
408 && hash_table_contains(basic_authed_hosts, hostname))
410 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
415 DEBUGP(("Host %s has not issued a general basic challenge.\n",
420 request_set_header (req, "Authorization",
421 basic_authentication_encode (user, passwd),
428 register_basic_auth_host (const char *hostname)
430 if (!basic_authed_hosts)
432 basic_authed_hosts = make_nocase_string_hash_table (1);
434 if (!hash_table_contains(basic_authed_hosts, hostname))
436 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
437 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
442 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
443 PROMISED_SIZE bytes are sent over the wire -- if the file is
444 longer, read only that much; if the file is shorter, report an error. */
447 post_file (int sock, const char *file_name, wgint promised_size)
449 static char chunk[8192];
454 DEBUGP (("[writing POST file %s ... ", file_name));
456 fp = fopen (file_name, "rb");
459 while (!feof (fp) && written < promised_size)
462 int length = fread (chunk, 1, sizeof (chunk), fp);
465 towrite = MIN (promised_size - written, length);
466 write_error = fd_write (sock, chunk, towrite, -1);
476 /* If we've written less than was promised, report a (probably
477 nonsensical) error rather than break the promise. */
478 if (written < promised_size)
484 assert (written == promised_size);
485 DEBUGP (("done]\n"));
489 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
490 If so, return the pointer to the position after the line, otherwise
491 return NULL. This is used as callback to fd_read_hunk. The data
492 between START and PEEKED has been read and cannot be "unread"; the
493 data after PEEKED has only been peeked. */
496 response_head_terminator (const char *start, const char *peeked, int peeklen)
500 /* If at first peek, verify whether HUNK starts with "HTTP". If
501 not, this is a HTTP/0.9 request and we must bail out without
503 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
506 /* Look for "\n[\r]\n", and return the following position if found.
507 Start two chars before the current to cover the possibility that
508 part of the terminator (e.g. "\n\r") arrived in the previous
510 p = peeked - start < 2 ? start : peeked - 2;
511 end = peeked + peeklen;
513 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
514 for (; p < end - 2; p++)
517 if (p[1] == '\r' && p[2] == '\n')
519 else if (p[1] == '\n')
522 /* p==end-2: check for \n\n directly preceding END. */
523 if (p[0] == '\n' && p[1] == '\n')
529 /* The maximum size of a single HTTP response we care to read. Rather
530 than being a limit of the reader implementation, this limit
531 prevents Wget from slurping all available memory upon encountering
532 malicious or buggy server output, thus protecting the user. Define
533 it to 0 to remove the limit. */
535 #define HTTP_RESPONSE_MAX_SIZE 65536
537 /* Read the HTTP request head from FD and return it. The error
538 conditions are the same as with fd_read_hunk.
540 To support HTTP/0.9 responses, this function tries to make sure
541 that the data begins with "HTTP". If this is not the case, no data
542 is read and an empty request is returned, so that the remaining
543 data can be treated as body. */
546 read_http_response_head (int fd)
548 return fd_read_hunk (fd, response_head_terminator, 512,
549 HTTP_RESPONSE_MAX_SIZE);
553 /* The response data. */
556 /* The array of pointers that indicate where each header starts.
557 For example, given this HTTP response:
564 The headers are located like this:
566 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
568 headers[0] headers[1] headers[2] headers[3]
570 I.e. headers[0] points to the beginning of the request,
571 headers[1] points to the end of the first header and the
572 beginning of the second one, etc. */
574 const char **headers;
577 /* Create a new response object from the text of the HTTP response,
578 available in HEAD. That text is automatically split into
579 constituent header lines for fast retrieval using
582 static struct response *
583 resp_new (const char *head)
588 struct response *resp = xnew0 (struct response);
593 /* Empty head means that we're dealing with a headerless
594 (HTTP/0.9) response. In that case, don't set HEADERS at
599 /* Split HEAD into header lines, so that resp_header_* functions
600 don't need to do this over and over again. */
606 DO_REALLOC (resp->headers, size, count + 1, const char *);
607 resp->headers[count++] = hdr;
609 /* Break upon encountering an empty line. */
610 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
613 /* Find the end of HDR, including continuations. */
616 const char *end = strchr (hdr, '\n');
622 while (*hdr == ' ' || *hdr == '\t');
624 DO_REALLOC (resp->headers, size, count + 1, const char *);
625 resp->headers[count] = NULL;
630 /* Locate the header named NAME in the request data, starting with
631 position START. This allows the code to loop through the request
632 data, filtering for all requests of a given name. Returns the
633 found position, or -1 for failure. The code that uses this
634 function typically looks like this:
636 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
637 ... do something with header ...
639 If you only care about one header, use resp_header_get instead of
643 resp_header_locate (const struct response *resp, const char *name, int start,
644 const char **begptr, const char **endptr)
647 const char **headers = resp->headers;
650 if (!headers || !headers[1])
653 name_len = strlen (name);
659 for (; headers[i + 1]; i++)
661 const char *b = headers[i];
662 const char *e = headers[i + 1];
664 && b[name_len] == ':'
665 && 0 == strncasecmp (b, name, name_len))
668 while (b < e && c_isspace (*b))
670 while (b < e && c_isspace (e[-1]))
680 /* Find and retrieve the header named NAME in the request data. If
681 found, set *BEGPTR to its starting, and *ENDPTR to its ending
682 position, and return true. Otherwise return false.
684 This function is used as a building block for resp_header_copy
685 and resp_header_strdup. */
688 resp_header_get (const struct response *resp, const char *name,
689 const char **begptr, const char **endptr)
691 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
695 /* Copy the response header named NAME to buffer BUF, no longer than
696 BUFSIZE (BUFSIZE includes the terminating 0). If the header
697 exists, true is returned, false otherwise. If there should be no
698 limit on the size of the header, use resp_header_strdup instead.
700 If BUFSIZE is 0, no data is copied, but the boolean indication of
701 whether the header is present is still returned. */
704 resp_header_copy (const struct response *resp, const char *name,
705 char *buf, int bufsize)
708 if (!resp_header_get (resp, name, &b, &e))
712 int len = MIN (e - b, bufsize - 1);
713 memcpy (buf, b, len);
719 /* Return the value of header named NAME in RESP, allocated with
720 malloc. If such a header does not exist in RESP, return NULL. */
723 resp_header_strdup (const struct response *resp, const char *name)
726 if (!resp_header_get (resp, name, &b, &e))
728 return strdupdelim (b, e);
731 /* Parse the HTTP status line, which is of format:
733 HTTP-Version SP Status-Code SP Reason-Phrase
735 The function returns the status-code, or -1 if the status line
736 appears malformed. The pointer to "reason-phrase" message is
737 returned in *MESSAGE. */
740 resp_status (const struct response *resp, char **message)
747 /* For a HTTP/0.9 response, assume status 200. */
749 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
753 p = resp->headers[0];
754 end = resp->headers[1];
760 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
764 /* Match the HTTP version. This is optional because Gnutella
765 servers have been reported to not specify HTTP version. */
766 if (p < end && *p == '/')
769 while (p < end && c_isdigit (*p))
771 if (p < end && *p == '.')
773 while (p < end && c_isdigit (*p))
777 while (p < end && c_isspace (*p))
779 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
782 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
787 while (p < end && c_isspace (*p))
789 while (p < end && c_isspace (end[-1]))
791 *message = strdupdelim (p, end);
797 /* Release the resources used by RESP. */
800 resp_free (struct response *resp)
802 xfree_null (resp->headers);
806 /* Print a single line of response, the characters [b, e). We tried
808 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
809 but that failed to escape the non-printable characters and, in fact,
810 caused crashes in UTF-8 locales. */
813 print_response_line(const char *prefix, const char *b, const char *e)
816 BOUNDED_TO_ALLOCA(b, e, copy);
817 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
818 quotearg_style (escape_quoting_style, copy));
821 /* Print the server response, line by line, omitting the trailing CRLF
822 from individual header lines, and prefixed with PREFIX. */
825 print_server_response (const struct response *resp, const char *prefix)
830 for (i = 0; resp->headers[i + 1]; i++)
832 const char *b = resp->headers[i];
833 const char *e = resp->headers[i + 1];
835 if (b < e && e[-1] == '\n')
837 if (b < e && e[-1] == '\r')
839 print_response_line(prefix, b, e);
843 /* Parse the `Content-Range' header and extract the information it
844 contains. Returns true if successful, false otherwise. */
846 parse_content_range (const char *hdr, wgint *first_byte_ptr,
847 wgint *last_byte_ptr, wgint *entity_length_ptr)
851 /* Ancient versions of Netscape proxy server, presumably predating
852 rfc2068, sent out `Content-Range' without the "bytes"
854 if (0 == strncasecmp (hdr, "bytes", 5))
857 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
861 while (c_isspace (*hdr))
866 if (!c_isdigit (*hdr))
868 for (num = 0; c_isdigit (*hdr); hdr++)
869 num = 10 * num + (*hdr - '0');
870 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
872 *first_byte_ptr = num;
874 for (num = 0; c_isdigit (*hdr); hdr++)
875 num = 10 * num + (*hdr - '0');
876 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
878 *last_byte_ptr = num;
883 for (num = 0; c_isdigit (*hdr); hdr++)
884 num = 10 * num + (*hdr - '0');
885 *entity_length_ptr = num;
889 /* Read the body of the request, but don't store it anywhere and don't
890 display a progress gauge. This is useful for reading the bodies of
891 administrative responses to which we will soon issue another
892 request. The response is not useful to the user, but reading it
893 allows us to continue using the same connection to the server.
895 If reading fails, false is returned, true otherwise. In debug
896 mode, the body is displayed for debugging purposes. */
899 skip_short_body (int fd, wgint contlen)
902 SKIP_SIZE = 512, /* size of the download buffer */
903 SKIP_THRESHOLD = 4096 /* the largest size we read */
905 char dlbuf[SKIP_SIZE + 1];
906 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
908 /* We shouldn't get here with unknown contlen. (This will change
909 with HTTP/1.1, which supports "chunked" transfer.) */
910 assert (contlen != -1);
912 /* If the body is too large, it makes more sense to simply close the
913 connection than to try to read the body. */
914 if (contlen > SKIP_THRESHOLD)
917 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
921 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
924 /* Don't normally report the error since this is an
925 optimization that should be invisible to the user. */
926 DEBUGP (("] aborting (%s).\n",
927 ret < 0 ? fd_errstr (fd) : "EOF received"));
931 /* Safe even if %.*s bogusly expects terminating \0 because
932 we've zero-terminated dlbuf above. */
933 DEBUGP (("%.*s", ret, dlbuf));
936 DEBUGP (("] done.\n"));
940 /* Extract a parameter from the string (typically an HTTP header) at
941 **SOURCE and advance SOURCE to the next parameter. Return false
942 when there are no more parameters to extract. The name of the
943 parameter is returned in NAME, and the value in VALUE. If the
944 parameter has no value, the token's value is zeroed out.
946 For example, if *SOURCE points to the string "attachment;
947 filename=\"foo bar\"", the first call to this function will return
948 the token named "attachment" and no value, and the second call will
949 return the token named "filename" and value "foo bar". The third
950 call will return false, indicating no more valid tokens. */
953 extract_param (const char **source, param_token *name, param_token *value,
956 const char *p = *source;
958 while (c_isspace (*p)) ++p;
962 return false; /* no error; nothing more to extract */
967 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
969 if (name->b == name->e)
970 return false; /* empty name: error */
971 while (c_isspace (*p)) ++p;
972 if (*p == separator || !*p) /* no value */
975 if (*p == separator) ++p;
980 return false; /* error */
982 /* *p is '=', extract value */
984 while (c_isspace (*p)) ++p;
985 if (*p == '"') /* quoted */
988 while (*p && *p != '"') ++p;
992 /* Currently at closing quote; find the end of param. */
993 while (c_isspace (*p)) ++p;
994 while (*p && *p != separator) ++p;
998 /* garbage after closed quote, e.g. foo="bar"baz */
1004 while (*p && *p != separator) ++p;
1006 while (value->e != value->b && c_isspace (value->e[-1]))
1008 if (*p == separator) ++p;
1015 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1017 /* Parse the contents of the `Content-Disposition' header, extracting
1018 the information useful to Wget. Content-Disposition is a header
1019 borrowed from MIME; when used in HTTP, it typically serves for
1020 specifying the desired file name of the resource. For example:
1022 Content-Disposition: attachment; filename="flora.jpg"
1024 Wget will skip the tokens it doesn't care about, such as
1025 "attachment" in the previous example; it will also skip other
1026 unrecognized params. If the header is syntactically correct and
1027 contains a file name, a copy of the file name is stored in
1028 *filename and true is returned. Otherwise, the function returns
1031 The file name is stripped of directory components and must not be
1035 parse_content_disposition (const char *hdr, char **filename)
1037 param_token name, value;
1038 while (extract_param (&hdr, &name, &value, ';'))
1039 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1041 /* Make the file name begin at the last slash or backslash. */
1042 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1043 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1044 if (last_slash && last_bs)
1045 value.b = 1 + MAX (last_slash, last_bs);
1046 else if (last_slash || last_bs)
1047 value.b = 1 + (last_slash ? last_slash : last_bs);
1048 if (value.b == value.e)
1050 /* Start with the directory prefix, if specified. */
1053 int prefix_length = strlen (opt.dir_prefix);
1054 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1059 total_length = prefix_length + (value.e - value.b);
1060 *filename = xmalloc (total_length + 1);
1061 strcpy (*filename, opt.dir_prefix);
1063 (*filename)[prefix_length - 1] = '/';
1064 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1065 (*filename)[total_length] = '\0';
1068 *filename = strdupdelim (value.b, value.e);
1074 /* Persistent connections. Currently, we cache the most recently used
1075 connection as persistent, provided that the HTTP server agrees to
1076 make it such. The persistence data is stored in the variables
1077 below. Ideally, it should be possible to cache an arbitrary fixed
1078 number of these connections. */
1080 /* Whether a persistent connection is active. */
1081 static bool pconn_active;
1084 /* The socket of the connection. */
1087 /* Host and port of the currently active persistent connection. */
1091 /* Whether a ssl handshake has occoured on this connection. */
1094 /* Whether the connection was authorized. This is only done by
1095 NTLM, which authorizes *connections* rather than individual
1096 requests. (That practice is peculiar for HTTP, but it is a
1097 useful optimization.) */
1101 /* NTLM data of the current connection. */
1102 struct ntlmdata ntlm;
1106 /* Mark the persistent connection as invalid and free the resources it
1107 uses. This is used by the CLOSE_* macros after they forcefully
1108 close a registered persistent connection. */
1111 invalidate_persistent (void)
1113 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1114 pconn_active = false;
1115 fd_close (pconn.socket);
1120 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1121 persistent. This will enable someone to use the same connection
1122 later. In the context of HTTP, this must be called only AFTER the
1123 response has been received and the server has promised that the
1124 connection will remain alive.
1126 If a previous connection was persistent, it is closed. */
1129 register_persistent (const char *host, int port, int fd, bool ssl)
1133 if (pconn.socket == fd)
1135 /* The connection FD is already registered. */
1140 /* The old persistent connection is still active; close it
1141 first. This situation arises whenever a persistent
1142 connection exists, but we then connect to a different
1143 host, and try to register a persistent connection to that
1145 invalidate_persistent ();
1149 pconn_active = true;
1151 pconn.host = xstrdup (host);
1154 pconn.authorized = false;
1156 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1159 /* Return true if a persistent connection is available for connecting
1163 persistent_available_p (const char *host, int port, bool ssl,
1164 bool *host_lookup_failed)
1166 /* First, check whether a persistent connection is active at all. */
1170 /* If we want SSL and the last connection wasn't or vice versa,
1171 don't use it. Checking for host and port is not enough because
1172 HTTP and HTTPS can apparently coexist on the same port. */
1173 if (ssl != pconn.ssl)
1176 /* If we're not connecting to the same port, we're not interested. */
1177 if (port != pconn.port)
1180 /* If the host is the same, we're in business. If not, there is
1181 still hope -- read below. */
1182 if (0 != strcasecmp (host, pconn.host))
1184 /* Check if pconn.socket is talking to HOST under another name.
1185 This happens often when both sites are virtual hosts
1186 distinguished only by name and served by the same network
1187 interface, and hence the same web server (possibly set up by
1188 the ISP and serving many different web sites). This
1189 admittedly unconventional optimization does not contradict
1190 HTTP and works well with popular server software. */
1194 struct address_list *al;
1197 /* Don't try to talk to two different SSL sites over the same
1198 secure connection! (Besides, it's not clear that
1199 name-based virtual hosting is even possible with SSL.) */
1202 /* If pconn.socket's peer is one of the IP addresses HOST
1203 resolves to, pconn.socket is for all intents and purposes
1204 already talking to HOST. */
1206 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1208 /* Can't get the peer's address -- something must be very
1209 wrong with the connection. */
1210 invalidate_persistent ();
1213 al = lookup_host (host, 0);
1216 *host_lookup_failed = true;
1220 found = address_list_contains (al, &ip);
1221 address_list_release (al);
1226 /* The persistent connection's peer address was found among the
1227 addresses HOST resolved to; therefore, pconn.sock is in fact
1228 already talking to HOST -- no need to reconnect. */
1231 /* Finally, check whether the connection is still open. This is
1232 important because most servers implement liberal (short) timeout
1233 on persistent connections. Wget can of course always reconnect
1234 if the connection doesn't work out, but it's nicer to know in
1235 advance. This test is a logical followup of the first test, but
1236 is "expensive" and therefore placed at the end of the list.
1238 (Current implementation of test_socket_open has a nice side
1239 effect that it treats sockets with pending data as "closed".
1240 This is exactly what we want: if a broken server sends message
1241 body in response to HEAD, or if it sends more than conent-length
1242 data, we won't reuse the corrupted connection.) */
1244 if (!test_socket_open (pconn.socket))
1246 /* Oops, the socket is no longer open. Now that we know that,
1247 let's invalidate the persistent connection before returning
1249 invalidate_persistent ();
1256 /* The idea behind these two CLOSE macros is to distinguish between
1257 two cases: one when the job we've been doing is finished, and we
1258 want to close the connection and leave, and two when something is
1259 seriously wrong and we're closing the connection as part of
1262 In case of keep_alive, CLOSE_FINISH should leave the connection
1263 open, while CLOSE_INVALIDATE should still close it.
1265 Note that the semantics of the flag `keep_alive' is "this
1266 connection *will* be reused (the server has promised not to close
1267 the connection once we're done)", while the semantics of
1268 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1269 active, registered connection". */
1271 #define CLOSE_FINISH(fd) do { \
1274 if (pconn_active && (fd) == pconn.socket) \
1275 invalidate_persistent (); \
1284 #define CLOSE_INVALIDATE(fd) do { \
1285 if (pconn_active && (fd) == pconn.socket) \
1286 invalidate_persistent (); \
1294 wgint len; /* received length */
1295 wgint contlen; /* expected length */
1296 wgint restval; /* the restart value */
1297 int res; /* the result of last read */
1298 char *rderrmsg; /* error message from read error */
1299 char *newloc; /* new location (redirection) */
1300 char *remote_time; /* remote time-stamp string */
1301 char *error; /* textual HTTP error */
1302 int statcode; /* status code */
1303 char *message; /* status message */
1304 wgint rd_size; /* amount of data read from socket */
1305 double dltime; /* time it took to download the data */
1306 const char *referer; /* value of the referer header. */
1307 char *local_file; /* local file name. */
1308 bool existence_checked; /* true if we already checked for a file's
1309 existence after having begun to download
1310 (needed in gethttp for when connection is
1311 interrupted/restarted. */
1312 bool timestamp_checked; /* true if pre-download time-stamping checks
1313 * have already been performed */
1314 char *orig_file_name; /* name of file to compare for time-stamping
1315 * (might be != local_file if -K is set) */
1316 wgint orig_file_size; /* size of file to compare for time-stamping */
1317 time_t orig_file_tstamp; /* time-stamp of file to compare for
1322 free_hstat (struct http_stat *hs)
1324 xfree_null (hs->newloc);
1325 xfree_null (hs->remote_time);
1326 xfree_null (hs->error);
1327 xfree_null (hs->rderrmsg);
1328 xfree_null (hs->local_file);
1329 xfree_null (hs->orig_file_name);
1330 xfree_null (hs->message);
1332 /* Guard against being called twice. */
1334 hs->remote_time = NULL;
1338 #define BEGINS_WITH(line, string_constant) \
1339 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1340 && (c_isspace (line[sizeof (string_constant) - 1]) \
1341 || !line[sizeof (string_constant) - 1]))
1343 #define SET_USER_AGENT(req) do { \
1344 if (!opt.useragent) \
1345 request_set_header (req, "User-Agent", \
1346 aprintf ("Wget/%s", version_string), rel_value); \
1347 else if (*opt.useragent) \
1348 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1351 /* The flags that allow clobbering the file (opening with "wb").
1352 Defined here to avoid repetition later. #### This will require
1354 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1355 || opt.dirstruct || opt.output_document)
1357 /* Retrieve a document through HTTP protocol. It recognizes status
1358 code, and correctly handles redirections. It closes the network
1359 socket. If it receives an error from the functions below it, it
1360 will print it if there is enough information to do so (almost
1361 always), returning the error to the caller (i.e. http_loop).
1363 Various HTTP parameters are stored to hs.
1365 If PROXY is non-NULL, the connection will be made to the proxy
1366 server, and u->url will be requested. */
1368 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1370 struct request *req;
1373 char *user, *passwd;
1377 wgint contlen, contrange;
1384 /* Set to 1 when the authorization has already been sent and should
1385 not be tried again. */
1386 bool auth_finished = false;
1388 /* Set to 1 when just globally-set Basic authorization has been sent;
1389 * should prevent further Basic negotiations, but not other
1391 bool basic_auth_finished = false;
1393 /* Whether NTLM authentication is used for this request. */
1394 bool ntlm_seen = false;
1396 /* Whether our connection to the remote host is through SSL. */
1397 bool using_ssl = false;
1399 /* Whether a HEAD request will be issued (as opposed to GET or
1401 bool head_only = !!(*dt & HEAD_ONLY);
1404 struct response *resp;
1408 /* Whether this connection will be kept alive after the HTTP request
1412 /* Whether keep-alive should be inhibited.
1414 RFC 2068 requests that 1.0 clients not send keep-alive requests
1415 to proxies. This is because many 1.0 proxies do not interpret
1416 the Connection header and transfer it to the remote server,
1417 causing it to not close the connection and leave both the proxy
1418 and the client hanging. */
1419 bool inhibit_keep_alive =
1420 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1422 /* Headers sent when using POST. */
1423 wgint post_data_size = 0;
1425 bool host_lookup_failed = false;
1428 if (u->scheme == SCHEME_HTTPS)
1430 /* Initialize the SSL context. After this has once been done,
1431 it becomes a no-op. */
1434 scheme_disable (SCHEME_HTTPS);
1435 logprintf (LOG_NOTQUIET,
1436 _("Disabling SSL due to encountered errors.\n"));
1437 return SSLINITFAILED;
1440 #endif /* HAVE_SSL */
1442 /* Initialize certain elements of struct http_stat. */
1446 hs->rderrmsg = NULL;
1448 hs->remote_time = NULL;
1454 /* Prepare the request to send. */
1456 req = request_new ();
1459 const char *meth = "GET";
1462 else if (opt.post_file_name || opt.post_data)
1464 /* Use the full path, i.e. one that includes the leading slash and
1465 the query string. E.g. if u->path is "foo/bar" and u->query is
1466 "param=value", full_path will be "/foo/bar?param=value". */
1469 /* When using SSL over proxy, CONNECT establishes a direct
1470 connection to the HTTPS server. Therefore use the same
1471 argument as when talking to the server directly. */
1472 && u->scheme != SCHEME_HTTPS
1475 meth_arg = xstrdup (u->url);
1477 meth_arg = url_full_path (u);
1478 request_set_method (req, meth, meth_arg);
1481 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1482 if (*dt & SEND_NOCACHE)
1483 request_set_header (req, "Pragma", "no-cache", rel_none);
1485 request_set_header (req, "Range",
1486 aprintf ("bytes=%s-",
1487 number_to_static_string (hs->restval)),
1489 SET_USER_AGENT (req);
1490 request_set_header (req, "Accept", "*/*", rel_none);
1492 /* Find the username and password for authentication. */
1495 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1496 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1497 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1500 && !u->user) /* We only do "site-wide" authentication with "global"
1501 user/password values; URL user/password info overrides. */
1503 /* If this is a host for which we've already received a Basic
1504 * challenge, we'll go ahead and send Basic authentication creds. */
1505 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1508 /* Generate the Host header, HOST:PORT. Take into account that:
1510 - Broken server-side software often doesn't recognize the PORT
1511 argument, so we must generate "Host: www.server.com" instead of
1512 "Host: www.server.com:80" (and likewise for https port).
1514 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1515 becomes ambiguous and needs to be rewritten as "Host:
1516 [3ffe:8100:200:2::2]:1234". */
1518 /* Formats arranged for hfmt[add_port][add_squares]. */
1519 static const char *hfmt[][2] = {
1520 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1522 int add_port = u->port != scheme_default_port (u->scheme);
1523 int add_squares = strchr (u->host, ':') != NULL;
1524 request_set_header (req, "Host",
1525 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1529 if (!inhibit_keep_alive)
1530 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1533 request_set_header (req, "Cookie",
1534 cookie_header (wget_cookie_jar,
1535 u->host, u->port, u->path,
1537 u->scheme == SCHEME_HTTPS
1544 if (opt.post_data || opt.post_file_name)
1546 request_set_header (req, "Content-Type",
1547 "application/x-www-form-urlencoded", rel_none);
1549 post_data_size = strlen (opt.post_data);
1552 post_data_size = file_size (opt.post_file_name);
1553 if (post_data_size == -1)
1555 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1556 quote (opt.post_file_name), strerror (errno));
1560 request_set_header (req, "Content-Length",
1561 xstrdup (number_to_static_string (post_data_size)),
1565 /* Add the user headers. */
1566 if (opt.user_headers)
1569 for (i = 0; opt.user_headers[i]; i++)
1570 request_set_user_header (req, opt.user_headers[i]);
1574 /* We need to come back here when the initial attempt to retrieve
1575 without authorization header fails. (Expected to happen at least
1576 for the Digest authorization scheme.) */
1581 char *proxy_user, *proxy_passwd;
1582 /* For normal username and password, URL components override
1583 command-line/wgetrc parameters. With proxy
1584 authentication, it's the reverse, because proxy URLs are
1585 normally the "permanent" ones, so command-line args
1586 should take precedence. */
1587 if (opt.proxy_user && opt.proxy_passwd)
1589 proxy_user = opt.proxy_user;
1590 proxy_passwd = opt.proxy_passwd;
1594 proxy_user = proxy->user;
1595 proxy_passwd = proxy->passwd;
1597 /* #### This does not appear right. Can't the proxy request,
1598 say, `Digest' authentication? */
1599 if (proxy_user && proxy_passwd)
1600 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1602 /* If we're using a proxy, we will be connecting to the proxy
1606 /* Proxy authorization over SSL is handled below. */
1608 if (u->scheme != SCHEME_HTTPS)
1610 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1615 /* Establish the connection. */
1617 if (!inhibit_keep_alive)
1619 /* Look for a persistent connection to target host, unless a
1620 proxy is used. The exception is when SSL is in use, in which
1621 case the proxy is nothing but a passthrough to the target
1622 host, registered as a connection to the latter. */
1623 struct url *relevant = conn;
1625 if (u->scheme == SCHEME_HTTPS)
1629 if (persistent_available_p (relevant->host, relevant->port,
1631 relevant->scheme == SCHEME_HTTPS,
1635 &host_lookup_failed))
1637 sock = pconn.socket;
1638 using_ssl = pconn.ssl;
1639 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1640 quotearg_style (escape_quoting_style, pconn.host),
1642 DEBUGP (("Reusing fd %d.\n", sock));
1643 if (pconn.authorized)
1644 /* If the connection is already authorized, the "Basic"
1645 authorization added by code above is unnecessary and
1647 request_remove_header (req, "Authorization");
1649 else if (host_lookup_failed)
1652 logprintf(LOG_NOTQUIET,
1653 _("%s: unable to resolve host address %s\n"),
1654 exec_name, quote (relevant->host));
1661 sock = connect_to_host (conn->host, conn->port);
1670 return (retryable_socket_connect_error (errno)
1671 ? CONERROR : CONIMPOSSIBLE);
1675 if (proxy && u->scheme == SCHEME_HTTPS)
1677 /* When requesting SSL URLs through proxies, use the
1678 CONNECT method to request passthrough. */
1679 struct request *connreq = request_new ();
1680 request_set_method (connreq, "CONNECT",
1681 aprintf ("%s:%d", u->host, u->port));
1682 SET_USER_AGENT (connreq);
1685 request_set_header (connreq, "Proxy-Authorization",
1686 proxyauth, rel_value);
1687 /* Now that PROXYAUTH is part of the CONNECT request,
1688 zero it out so we don't send proxy authorization with
1689 the regular request below. */
1692 /* Examples in rfc2817 use the Host header in CONNECT
1693 requests. I don't see how that gains anything, given
1694 that the contents of Host would be exactly the same as
1695 the contents of CONNECT. */
1697 write_error = request_send (connreq, sock);
1698 request_free (connreq);
1699 if (write_error < 0)
1701 CLOSE_INVALIDATE (sock);
1705 head = read_http_response_head (sock);
1708 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1710 CLOSE_INVALIDATE (sock);
1719 DEBUGP (("proxy responded with: [%s]\n", head));
1721 resp = resp_new (head);
1722 statcode = resp_status (resp, &message);
1723 hs->message = xstrdup (message);
1726 if (statcode != 200)
1729 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1730 message ? quotearg_style (escape_quoting_style, message) : "?");
1731 xfree_null (message);
1734 xfree_null (message);
1736 /* SOCK is now *really* connected to u->host, so update CONN
1737 to reflect this. That way register_persistent will
1738 register SOCK as being connected to u->host:u->port. */
1742 if (conn->scheme == SCHEME_HTTPS)
1744 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1751 #endif /* HAVE_SSL */
1754 /* Send the request to server. */
1755 write_error = request_send (req, sock);
1757 if (write_error >= 0)
1761 DEBUGP (("[POST data: %s]\n", opt.post_data));
1762 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1764 else if (opt.post_file_name && post_data_size != 0)
1765 write_error = post_file (sock, opt.post_file_name, post_data_size);
1768 if (write_error < 0)
1770 CLOSE_INVALIDATE (sock);
1774 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1775 proxy ? "Proxy" : "HTTP");
1780 head = read_http_response_head (sock);
1785 logputs (LOG_NOTQUIET, _("No data received.\n"));
1786 CLOSE_INVALIDATE (sock);
1792 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1794 CLOSE_INVALIDATE (sock);
1799 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1801 resp = resp_new (head);
1803 /* Check for status line. */
1805 statcode = resp_status (resp, &message);
1806 hs->message = xstrdup (message);
1807 if (!opt.server_response)
1808 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1809 message ? quotearg_style (escape_quoting_style, message) : "");
1812 logprintf (LOG_VERBOSE, "\n");
1813 print_server_response (resp, " ");
1816 /* Determine the local filename if needed. Notice that if -O is used
1817 * hstat.local_file is set by http_loop to the argument of -O. */
1818 if (!hs->local_file)
1820 /* Honor Content-Disposition whether possible. */
1821 if (!opt.content_disposition
1822 || !resp_header_copy (resp, "Content-Disposition",
1823 hdrval, sizeof (hdrval))
1824 || !parse_content_disposition (hdrval, &hs->local_file))
1826 /* The Content-Disposition header is missing or broken.
1827 * Choose unique file name according to given URL. */
1828 hs->local_file = url_file_name (u);
1832 /* TODO: perform this check only once. */
1833 if (!hs->existence_checked && file_exists_p (hs->local_file))
1835 if (opt.noclobber && !opt.output_document)
1837 /* If opt.noclobber is turned on and file already exists, do not
1838 retrieve the file. But if the output_document was given, then this
1839 test was already done and the file didn't exist. Hence the !opt.output_document */
1840 logprintf (LOG_VERBOSE, _("\
1841 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1842 /* If the file is there, we suppose it's retrieved OK. */
1845 /* #### Bogusness alert. */
1846 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1847 if (has_html_suffix_p (hs->local_file))
1850 return RETRUNNEEDED;
1852 else if (!ALLOW_CLOBBER)
1854 char *unique = unique_name (hs->local_file, true);
1855 if (unique != hs->local_file)
1856 xfree (hs->local_file);
1857 hs->local_file = unique;
1860 hs->existence_checked = true;
1862 /* Support timestamping */
1863 /* TODO: move this code out of gethttp. */
1864 if (opt.timestamping && !hs->timestamp_checked)
1866 size_t filename_len = strlen (hs->local_file);
1867 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1868 bool local_dot_orig_file_exists = false;
1869 char *local_filename = NULL;
1872 if (opt.backup_converted)
1873 /* If -K is specified, we'll act on the assumption that it was specified
1874 last time these files were downloaded as well, and instead of just
1875 comparing local file X against server file X, we'll compare local
1876 file X.orig (if extant, else X) against server file X. If -K
1877 _wasn't_ specified last time, or the server contains files called
1878 *.orig, -N will be back to not operating correctly with -k. */
1880 /* Would a single s[n]printf() call be faster? --dan
1882 Definitely not. sprintf() is horribly slow. It's a
1883 different question whether the difference between the two
1884 affects a program. Usually I'd say "no", but at one
1885 point I profiled Wget, and found that a measurable and
1886 non-negligible amount of time was lost calling sprintf()
1887 in url.c. Replacing sprintf with inline calls to
1888 strcpy() and number_to_string() made a difference.
1890 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1891 memcpy (filename_plus_orig_suffix + filename_len,
1892 ".orig", sizeof (".orig"));
1894 /* Try to stat() the .orig file. */
1895 if (stat (filename_plus_orig_suffix, &st) == 0)
1897 local_dot_orig_file_exists = true;
1898 local_filename = filename_plus_orig_suffix;
1902 if (!local_dot_orig_file_exists)
1903 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1904 if (stat (hs->local_file, &st) == 0)
1905 local_filename = hs->local_file;
1907 if (local_filename != NULL)
1908 /* There was a local file, so we'll check later to see if the version
1909 the server has is the same version we already have, allowing us to
1912 hs->orig_file_name = xstrdup (local_filename);
1913 hs->orig_file_size = st.st_size;
1914 hs->orig_file_tstamp = st.st_mtime;
1916 /* Modification time granularity is 2 seconds for Windows, so
1917 increase local time by 1 second for later comparison. */
1918 ++hs->orig_file_tstamp;
1923 if (!opt.ignore_length
1924 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1928 parsed = str_to_wgint (hdrval, NULL, 10);
1929 if (parsed == WGINT_MAX && errno == ERANGE)
1932 #### If Content-Length is out of range, it most likely
1933 means that the file is larger than 2G and that we're
1934 compiled without LFS. In that case we should probably
1935 refuse to even attempt to download the file. */
1938 else if (parsed < 0)
1940 /* Negative Content-Length; nonsensical, so we can't
1941 assume any information about the content to receive. */
1948 /* Check for keep-alive related responses. */
1949 if (!inhibit_keep_alive && contlen != -1)
1951 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1953 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1955 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1960 /* The server has promised that it will not close the connection
1961 when we're done. This means that we can register it. */
1962 register_persistent (conn->host, conn->port, sock, using_ssl);
1964 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1966 /* Authorization is required. */
1967 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1968 CLOSE_FINISH (sock);
1970 CLOSE_INVALIDATE (sock);
1971 pconn.authorized = false;
1972 if (!auth_finished && (user && passwd))
1974 /* IIS sends multiple copies of WWW-Authenticate, one with
1975 the value "negotiate", and other(s) with data. Loop over
1976 all the occurrences and pick the one we recognize. */
1978 const char *wabeg, *waend;
1979 char *www_authenticate = NULL;
1981 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1982 &wabeg, &waend)) != -1;
1984 if (known_authentication_scheme_p (wabeg, waend))
1986 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1990 if (!www_authenticate)
1992 /* If the authentication header is missing or
1993 unrecognized, there's no sense in retrying. */
1994 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1996 else if (!basic_auth_finished
1997 || !BEGINS_WITH (www_authenticate, "Basic"))
2000 pth = url_full_path (u);
2001 request_set_header (req, "Authorization",
2002 create_authorization_line (www_authenticate,
2004 request_method (req),
2008 if (BEGINS_WITH (www_authenticate, "NTLM"))
2010 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2012 /* Need to register this host as using basic auth,
2013 * so we automatically send creds next time. */
2014 register_basic_auth_host (u->host);
2017 goto retry_with_auth;
2021 /* We already did Basic auth, and it failed. Gotta
2025 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2029 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2031 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2033 pconn.authorized = true;
2037 hs->statcode = statcode;
2039 hs->error = xstrdup (_("Malformed status line"));
2041 hs->error = xstrdup (_("(no description)"));
2043 hs->error = xstrdup (message);
2044 xfree_null (message);
2046 type = resp_header_strdup (resp, "Content-Type");
2049 char *tmp = strchr (type, ';');
2052 /* sXXXav: only needed if IRI support is enabled */
2053 char *tmp2 = tmp + 1;
2055 while (tmp > type && c_isspace (tmp[-1]))
2059 /* Try to get remote encoding if needed */
2060 if (opt.enable_iri && !opt.encoding_remote)
2061 set_current_charset (parse_charset (tmp2));
2064 hs->newloc = resp_header_strdup (resp, "Location");
2065 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2067 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2071 const char *scbeg, *scend;
2072 /* The jar should have been created by now. */
2073 assert (wget_cookie_jar != NULL);
2075 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2076 &scbeg, &scend)) != -1;
2079 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2080 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2081 u->path, set_cookie);
2085 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2087 wgint first_byte_pos, last_byte_pos, entity_length;
2088 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2091 contrange = first_byte_pos;
2092 contlen = last_byte_pos - first_byte_pos + 1;
2097 /* 20x responses are counted among successful by default. */
2098 if (H_20X (statcode))
2101 /* Return if redirected. */
2102 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2104 /* RFC2068 says that in case of the 300 (multiple choices)
2105 response, the server can output a preferred URL through
2106 `Location' header; otherwise, the request should be treated
2107 like GET. So, if the location is set, it will be a
2108 redirection; otherwise, just proceed normally. */
2109 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2113 logprintf (LOG_VERBOSE,
2114 _("Location: %s%s\n"),
2115 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2116 hs->newloc ? _(" [following]") : "");
2117 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2118 CLOSE_FINISH (sock);
2120 CLOSE_INVALIDATE (sock);
2126 /* If content-type is not given, assume text/html. This is because
2127 of the multitude of broken CGI's that "forget" to generate the
2130 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2131 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2137 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2142 if (opt.html_extension)
2145 /* -E / --html-extension / html_extension = on was specified,
2146 and this is a text/html file. If some case-insensitive
2147 variation on ".htm[l]" isn't already the file's suffix,
2150 ensure_extension (hs, ".html", dt);
2152 else if (*dt & TEXTCSS)
2154 ensure_extension (hs, ".css", dt);
2158 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2160 /* If `-c' is in use and the file has been fully downloaded (or
2161 the remote file has shrunk), Wget effectively requests bytes
2162 after the end of file and the server response with 416. */
2163 logputs (LOG_VERBOSE, _("\
2164 \n The file is already fully retrieved; nothing to do.\n\n"));
2165 /* In case the caller inspects. */
2168 /* Mark as successfully retrieved. */
2171 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2172 might be more bytes in the body. */
2173 return RETRUNNEEDED;
2175 if ((contrange != 0 && contrange != hs->restval)
2176 || (H_PARTIAL (statcode) && !contrange))
2178 /* The Range request was somehow misunderstood by the server.
2181 CLOSE_INVALIDATE (sock);
2187 hs->contlen = contlen + contrange;
2193 /* No need to print this output if the body won't be
2194 downloaded at all, or if the original server response is
2196 logputs (LOG_VERBOSE, _("Length: "));
2199 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2200 if (contlen + contrange >= 1024)
2201 logprintf (LOG_VERBOSE, " (%s)",
2202 human_readable (contlen + contrange));
2205 if (contlen >= 1024)
2206 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2207 number_to_static_string (contlen),
2208 human_readable (contlen));
2210 logprintf (LOG_VERBOSE, _(", %s remaining"),
2211 number_to_static_string (contlen));
2215 logputs (LOG_VERBOSE,
2216 opt.ignore_length ? _("ignored") : _("unspecified"));
2218 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2220 logputs (LOG_VERBOSE, "\n");
2224 type = NULL; /* We don't need it any more. */
2226 /* Return if we have no intention of further downloading. */
2227 if (!(*dt & RETROKF) || head_only)
2229 /* In case the caller cares to look... */
2234 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2235 servers not to send body in response to a HEAD request, and
2236 those that do will likely be caught by test_socket_open.
2237 If not, they can be worked around using
2238 `--no-http-keep-alive'. */
2239 CLOSE_FINISH (sock);
2240 else if (keep_alive && skip_short_body (sock, contlen))
2241 /* Successfully skipped the body; also keep using the socket. */
2242 CLOSE_FINISH (sock);
2244 CLOSE_INVALIDATE (sock);
2245 return RETRFINISHED;
2248 /* Open the local file. */
2251 mkalldirs (hs->local_file);
2253 rotate_backups (hs->local_file);
2255 fp = fopen (hs->local_file, "ab");
2256 else if (ALLOW_CLOBBER)
2257 fp = fopen (hs->local_file, "wb");
2260 fp = fopen_excl (hs->local_file, true);
2261 if (!fp && errno == EEXIST)
2263 /* We cannot just invent a new name and use it (which is
2264 what functions like unique_create typically do)
2265 because we told the user we'd use this name.
2266 Instead, return and retry the download. */
2267 logprintf (LOG_NOTQUIET,
2268 _("%s has sprung into existence.\n"),
2270 CLOSE_INVALIDATE (sock);
2271 return FOPEN_EXCL_ERR;
2276 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2277 CLOSE_INVALIDATE (sock);
2284 /* Print fetch message, if opt.verbose. */
2287 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2288 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2291 /* This confuses the timestamping code that checks for file size.
2292 #### The timestamping code should be smarter about file size. */
2293 if (opt.save_headers && hs->restval == 0)
2294 fwrite (head, 1, strlen (head), fp);
2296 /* Now we no longer need to store the response header. */
2299 /* Download the request body. */
2302 /* If content-length is present, read that much; otherwise, read
2303 until EOF. The HTTP spec doesn't require the server to
2304 actually close the connection when it's done sending data. */
2305 flags |= rb_read_exactly;
2306 if (hs->restval > 0 && contrange == 0)
2307 /* If the server ignored our range request, instruct fd_read_body
2308 to skip the first RESTVAL bytes of body. */
2309 flags |= rb_skip_startpos;
2310 hs->len = hs->restval;
2312 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2313 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2317 CLOSE_FINISH (sock);
2321 hs->rderrmsg = xstrdup (fd_errstr (sock));
2322 CLOSE_INVALIDATE (sock);
2329 return RETRFINISHED;
2332 /* The genuine HTTP loop! This is the part where the retrieval is
2333 retried, and retried, and retried, and... */
2335 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2336 int *dt, struct url *proxy)
2339 bool got_head = false; /* used for time-stamping and filename detection */
2340 bool time_came_from_head = false;
2341 bool got_name = false;
2344 uerr_t err, ret = TRYLIMEXC;
2345 time_t tmr = -1; /* remote time-stamp */
2346 struct http_stat hstat; /* HTTP status */
2348 bool send_head_first = true;
2350 /* Assert that no value for *LOCAL_FILE was passed. */
2351 assert (local_file == NULL || *local_file == NULL);
2353 /* Set LOCAL_FILE parameter. */
2354 if (local_file && opt.output_document)
2355 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2357 /* Reset NEWLOC parameter. */
2360 /* This used to be done in main(), but it's a better idea to do it
2361 here so that we don't go through the hoops if we're just using
2366 /* Warn on (likely bogus) wildcard usage in HTTP. */
2367 if (opt.ftp_glob && has_wildcards_p (u->path))
2368 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2370 /* Setup hstat struct. */
2372 hstat.referer = referer;
2374 if (opt.output_document)
2376 hstat.local_file = xstrdup (opt.output_document);
2379 else if (!opt.content_disposition)
2381 hstat.local_file = url_file_name (u);
2385 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2386 * screaming for some refactoring. */
2387 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2389 /* If opt.noclobber is turned on and file already exists, do not
2390 retrieve the file. But if the output_document was given, then this
2391 test was already done and the file didn't exist. Hence the !opt.output_document */
2392 logprintf (LOG_VERBOSE, _("\
2393 File %s already there; not retrieving.\n\n"),
2394 quote (hstat.local_file));
2395 /* If the file is there, we suppose it's retrieved OK. */
2398 /* #### Bogusness alert. */
2399 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2400 if (has_html_suffix_p (hstat.local_file))
2407 /* Reset the counter. */
2410 /* Reset the document type. */
2413 /* Skip preliminary HEAD request if we're not in spider mode AND
2414 * if -O was given or HTTP Content-Disposition support is disabled. */
2416 && (got_name || !opt.content_disposition))
2417 send_head_first = false;
2419 /* Send preliminary HEAD request if -N is given and we have an existing
2420 * destination file. */
2421 if (opt.timestamping
2422 && !opt.content_disposition
2423 && file_exists_p (url_file_name (u)))
2424 send_head_first = true;
2429 /* Increment the pass counter. */
2431 sleep_between_retrievals (count);
2433 /* Get the current time string. */
2434 tms = datetime_str (time (NULL));
2436 if (opt.spider && !got_head)
2437 logprintf (LOG_VERBOSE, _("\
2438 Spider mode enabled. Check if remote file exists.\n"));
2440 /* Print fetch message, if opt.verbose. */
2443 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2448 sprintf (tmp, _("(try:%2d)"), count);
2449 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2454 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2459 ws_changetitle (hurl);
2464 /* Default document type is empty. However, if spider mode is
2465 on or time-stamping is employed, HEAD_ONLY commands is
2466 encoded within *dt. */
2467 if (send_head_first && !got_head)
2472 /* Decide whether or not to restart. */
2475 && stat (hstat.local_file, &st) == 0
2476 && S_ISREG (st.st_mode))
2477 /* When -c is used, continue from on-disk size. (Can't use
2478 hstat.len even if count>1 because we don't want a failed
2479 first attempt to clobber existing data.) */
2480 hstat.restval = st.st_size;
2482 /* otherwise, continue where the previous try left off */
2483 hstat.restval = hstat.len;
2487 /* Decide whether to send the no-cache directive. We send it in
2489 a) we're using a proxy, and we're past our first retrieval.
2490 Some proxies are notorious for caching incomplete data, so
2491 we require a fresh get.
2492 b) caching is explicitly inhibited. */
2493 if ((proxy && count > 1) /* a */
2494 || !opt.allow_cache) /* b */
2495 *dt |= SEND_NOCACHE;
2497 *dt &= ~SEND_NOCACHE;
2499 /* Try fetching the document, or at least its head. */
2500 err = gethttp (u, &hstat, dt, proxy);
2503 tms = datetime_str (time (NULL));
2505 /* Get the new location (with or without the redirection). */
2507 *newloc = xstrdup (hstat.newloc);
2511 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2512 case CONERROR: case READERR: case WRITEFAILED:
2513 case RANGEERR: case FOPEN_EXCL_ERR:
2514 /* Non-fatal errors continue executing the loop, which will
2515 bring them to "while" statement at the end, to judge
2516 whether the number of tries was exceeded. */
2517 printwhat (count, opt.ntry);
2519 case FWRITEERR: case FOPENERR:
2520 /* Another fatal error. */
2521 logputs (LOG_VERBOSE, "\n");
2522 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2523 quote (hstat.local_file), strerror (errno));
2524 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2525 case SSLINITFAILED: case CONTNOTSUPPORTED:
2526 /* Fatal errors just return from the function. */
2530 /* Another fatal error. */
2531 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2535 /* Return the new location to the caller. */
2538 logprintf (LOG_NOTQUIET,
2539 _("ERROR: Redirection (%d) without location.\n"),
2549 /* The file was already fully retrieved. */
2553 /* Deal with you later. */
2556 /* All possibilities should have been exhausted. */
2560 if (!(*dt & RETROKF))
2565 /* #### Ugly ugly ugly! */
2566 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2567 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2570 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2572 && (hstat.statcode == 500 || hstat.statcode == 501))
2577 /* Maybe we should always keep track of broken links, not just in
2579 * Don't log error if it was utf8 encoded because we will try
2581 else if (opt.spider && !get_utf8_encode ())
2583 /* #### Again: ugly ugly ugly! */
2585 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2586 nonexisting_url (hurl);
2587 logprintf (LOG_NOTQUIET, _("\
2588 Remote file does not exist -- broken link!!!\n"));
2592 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2593 tms, hstat.statcode,
2594 quotearg_style (escape_quoting_style, hstat.error));
2596 logputs (LOG_VERBOSE, "\n");
2602 /* Did we get the time-stamp? */
2605 got_head = true; /* no more time-stamping */
2607 if (opt.timestamping && !hstat.remote_time)
2609 logputs (LOG_NOTQUIET, _("\
2610 Last-modified header missing -- time-stamps turned off.\n"));
2612 else if (hstat.remote_time)
2614 /* Convert the date-string into struct tm. */
2615 tmr = http_atotm (hstat.remote_time);
2616 if (tmr == (time_t) (-1))
2617 logputs (LOG_VERBOSE, _("\
2618 Last-modified header invalid -- time-stamp ignored.\n"));
2619 if (*dt & HEAD_ONLY)
2620 time_came_from_head = true;
2623 if (send_head_first)
2625 /* The time-stamping section. */
2626 if (opt.timestamping)
2628 if (hstat.orig_file_name) /* Perform the following
2629 checks only if the file
2631 download already exists. */
2633 if (hstat.remote_time &&
2634 tmr != (time_t) (-1))
2636 /* Now time-stamping can be used validly.
2637 Time-stamping means that if the sizes of
2638 the local and remote file match, and local
2639 file is newer than the remote file, it will
2640 not be retrieved. Otherwise, the normal
2641 download procedure is resumed. */
2642 if (hstat.orig_file_tstamp >= tmr)
2644 if (hstat.contlen == -1
2645 || hstat.orig_file_size == hstat.contlen)
2647 logprintf (LOG_VERBOSE, _("\
2648 Server file no newer than local file %s -- not retrieving.\n\n"),
2649 quote (hstat.orig_file_name));
2655 logprintf (LOG_VERBOSE, _("\
2656 The sizes do not match (local %s) -- retrieving.\n"),
2657 number_to_static_string (hstat.orig_file_size));
2661 logputs (LOG_VERBOSE,
2662 _("Remote file is newer, retrieving.\n"));
2664 logputs (LOG_VERBOSE, "\n");
2668 /* free_hstat (&hstat); */
2669 hstat.timestamp_checked = true;
2674 bool finished = true;
2679 logputs (LOG_VERBOSE, _("\
2680 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2685 logprintf (LOG_VERBOSE, _("\
2686 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2687 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2694 logprintf (LOG_VERBOSE, _("\
2695 Remote file exists and could contain further links,\n\
2696 but recursion is disabled -- not retrieving.\n\n"));
2700 logprintf (LOG_VERBOSE, _("\
2701 Remote file exists.\n\n"));
2703 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2708 logprintf (LOG_NONVERBOSE,
2709 _("%s URL:%s %2d %s\n"),
2710 tms, u->url, hstat.statcode,
2711 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2718 count = 0; /* the retrieve count for HEAD is reset */
2720 } /* send_head_first */
2723 if ((tmr != (time_t) (-1))
2724 && ((hstat.len == hstat.contlen) ||
2725 ((hstat.res == 0) && (hstat.contlen == -1))))
2727 /* #### This code repeats in http.c and ftp.c. Move it to a
2729 const char *fl = NULL;
2730 if (opt.output_document)
2732 if (output_stream_regular)
2733 fl = opt.output_document;
2736 fl = hstat.local_file;
2740 /* Reparse time header, in case it's changed. */
2741 if (time_came_from_head
2742 && hstat.remote_time && hstat.remote_time[0])
2744 newtmr = http_atotm (hstat.remote_time);
2751 /* End of time-stamping section. */
2753 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2754 total_download_time += hstat.dltime;
2756 if (hstat.len == hstat.contlen)
2760 logprintf (LOG_VERBOSE,
2761 _("%s (%s) - %s saved [%s/%s]\n\n"),
2762 tms, tmrate, quote (hstat.local_file),
2763 number_to_static_string (hstat.len),
2764 number_to_static_string (hstat.contlen));
2765 logprintf (LOG_NONVERBOSE,
2766 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2768 number_to_static_string (hstat.len),
2769 number_to_static_string (hstat.contlen),
2770 hstat.local_file, count);
2773 total_downloaded_bytes += hstat.len;
2775 /* Remember that we downloaded the file for later ".orig" code. */
2776 if (*dt & ADDED_HTML_EXTENSION)
2777 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2779 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2784 else if (hstat.res == 0) /* No read error */
2786 if (hstat.contlen == -1) /* We don't know how much we were supposed
2787 to get, so assume we succeeded. */
2791 logprintf (LOG_VERBOSE,
2792 _("%s (%s) - %s saved [%s]\n\n"),
2793 tms, tmrate, quote (hstat.local_file),
2794 number_to_static_string (hstat.len));
2795 logprintf (LOG_NONVERBOSE,
2796 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2797 tms, u->url, number_to_static_string (hstat.len),
2798 hstat.local_file, count);
2801 total_downloaded_bytes += hstat.len;
2803 /* Remember that we downloaded the file for later ".orig" code. */
2804 if (*dt & ADDED_HTML_EXTENSION)
2805 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2807 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2812 else if (hstat.len < hstat.contlen) /* meaning we lost the
2813 connection too soon */
2815 logprintf (LOG_VERBOSE,
2816 _("%s (%s) - Connection closed at byte %s. "),
2817 tms, tmrate, number_to_static_string (hstat.len));
2818 printwhat (count, opt.ntry);
2821 else if (hstat.len != hstat.restval)
2822 /* Getting here would mean reading more data than
2823 requested with content-length, which we never do. */
2827 /* Getting here probably means that the content-length was
2828 * _less_ than the original, local size. We should probably
2829 * truncate or re-read, or something. FIXME */
2834 else /* from now on hstat.res can only be -1 */
2836 if (hstat.contlen == -1)
2838 logprintf (LOG_VERBOSE,
2839 _("%s (%s) - Read error at byte %s (%s)."),
2840 tms, tmrate, number_to_static_string (hstat.len),
2842 printwhat (count, opt.ntry);
2845 else /* hstat.res == -1 and contlen is given */
2847 logprintf (LOG_VERBOSE,
2848 _("%s (%s) - Read error at byte %s/%s (%s). "),
2850 number_to_static_string (hstat.len),
2851 number_to_static_string (hstat.contlen),
2853 printwhat (count, opt.ntry);
2859 while (!opt.ntry || (count < opt.ntry));
2863 *local_file = xstrdup (hstat.local_file);
2864 free_hstat (&hstat);
2869 /* Check whether the result of strptime() indicates success.
2870 strptime() returns the pointer to how far it got to in the string.
2871 The processing has been successful if the string is at `GMT' or
2872 `+X', or at the end of the string.
2874 In extended regexp parlance, the function returns 1 if P matches
2875 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2876 can return) is considered a failure and 0 is returned. */
2878 check_end (const char *p)
2882 while (c_isspace (*p))
2885 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2886 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2892 /* Convert the textual specification of time in TIME_STRING to the
2893 number of seconds since the Epoch.
2895 TIME_STRING can be in any of the three formats RFC2616 allows the
2896 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2897 as well as the time format used in the Set-Cookie header.
2898 Timezones are ignored, and should be GMT.
2900 Return the computed time_t representation, or -1 if the conversion
2903 This function uses strptime with various string formats for parsing
2904 TIME_STRING. This results in a parser that is not as lenient in
2905 interpreting TIME_STRING as I would like it to be. Being based on
2906 strptime, it always allows shortened months, one-digit days, etc.,
2907 but due to the multitude of formats in which time can be
2908 represented, an ideal HTTP time parser would be even more
2909 forgiving. It should completely ignore things like week days and
2910 concentrate only on the various forms of representing years,
2911 months, days, hours, minutes, and seconds. For example, it would
2912 be nice if it accepted ISO 8601 out of the box.
2914 I've investigated free and PD code for this purpose, but none was
2915 usable. getdate was big and unwieldy, and had potential copyright
2916 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2917 distributed with phttpd, is excellent, but we cannot use it because
2918 it is not assigned to the FSF. So I stuck it with strptime. */
2921 http_atotm (const char *time_string)
2923 /* NOTE: Solaris strptime man page claims that %n and %t match white
2924 space, but that's not universally available. Instead, we simply
2925 use ` ' to mean "skip all WS", which works under all strptime
2926 implementations I've tested. */
2928 static const char *time_formats[] = {
2929 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2930 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2931 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2932 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2933 (used in Set-Cookie, defined in the
2934 Netscape cookie specification.) */
2936 const char *oldlocale;
2938 time_t ret = (time_t) -1;
2940 /* Solaris strptime fails to recognize English month names in
2941 non-English locales, which we work around by temporarily setting
2942 locale to C before invoking strptime. */
2943 oldlocale = setlocale (LC_TIME, NULL);
2944 setlocale (LC_TIME, "C");
2946 for (i = 0; i < countof (time_formats); i++)
2950 /* Some versions of strptime use the existing contents of struct
2951 tm to recalculate the date according to format. Zero it out
2952 to prevent stack garbage from influencing strptime. */
2955 if (check_end (strptime (time_string, time_formats[i], &t)))
2962 /* Restore the previous locale. */
2963 setlocale (LC_TIME, oldlocale);
2968 /* Authorization support: We support three authorization schemes:
2970 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2972 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2973 consisting of answering to the server's challenge with the proper
2976 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2977 Stenberg for libcurl. Like digest, NTLM is based on a
2978 challenge-response mechanism, but unlike digest, it is non-standard
2979 (authenticates TCP connections rather than requests), undocumented
2980 and Microsoft-specific. */
2982 /* Create the authentication header contents for the `Basic' scheme.
2983 This is done by encoding the string "USER:PASS" to base64 and
2984 prepending the string "Basic " in front of it. */
2987 basic_authentication_encode (const char *user, const char *passwd)
2990 int len1 = strlen (user) + 1 + strlen (passwd);
2992 t1 = (char *)alloca (len1 + 1);
2993 sprintf (t1, "%s:%s", user, passwd);
2995 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2996 base64_encode (t1, len1, t2);
2998 return concat_strings ("Basic ", t2, (char *) 0);
3001 #define SKIP_WS(x) do { \
3002 while (c_isspace (*(x))) \
3006 #ifdef ENABLE_DIGEST
3007 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3008 an array of 16 bytes containing the hash keys, and BUF should be a
3009 buffer of 33 writable characters (32 for hex digits plus one for
3010 zero termination). */
3012 dump_hash (char *buf, const unsigned char *hash)
3016 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3018 *buf++ = XNUM_TO_digit (*hash >> 4);
3019 *buf++ = XNUM_TO_digit (*hash & 0xf);
3024 /* Take the line apart to find the challenge, and compose a digest
3025 authorization header. See RFC2069 section 2.1.2. */
3027 digest_authentication_encode (const char *au, const char *user,
3028 const char *passwd, const char *method,
3031 static char *realm, *opaque, *nonce;
3036 { "realm", &realm },
3037 { "opaque", &opaque },
3041 param_token name, value;
3043 realm = opaque = nonce = NULL;
3045 au += 6; /* skip over `Digest' */
3046 while (extract_param (&au, &name, &value, ','))
3049 size_t namelen = name.e - name.b;
3050 for (i = 0; i < countof (options); i++)
3051 if (namelen == strlen (options[i].name)
3052 && 0 == strncmp (name.b, options[i].name,
3055 *options[i].variable = strdupdelim (value.b, value.e);
3059 if (!realm || !nonce || !user || !passwd || !path || !method)
3062 xfree_null (opaque);
3067 /* Calculate the digest value. */
3069 ALLOCA_MD5_CONTEXT (ctx);
3070 unsigned char hash[MD5_HASHLEN];
3071 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3072 char response_digest[MD5_HASHLEN * 2 + 1];
3074 /* A1BUF = H(user ":" realm ":" password) */
3076 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3077 gen_md5_update ((unsigned char *)":", 1, ctx);
3078 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3079 gen_md5_update ((unsigned char *)":", 1, ctx);
3080 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3081 gen_md5_finish (ctx, hash);
3082 dump_hash (a1buf, hash);
3084 /* A2BUF = H(method ":" path) */
3086 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3087 gen_md5_update ((unsigned char *)":", 1, ctx);
3088 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3089 gen_md5_finish (ctx, hash);
3090 dump_hash (a2buf, hash);
3092 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3094 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3095 gen_md5_update ((unsigned char *)":", 1, ctx);
3096 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3097 gen_md5_update ((unsigned char *)":", 1, ctx);
3098 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3099 gen_md5_finish (ctx, hash);
3100 dump_hash (response_digest, hash);
3102 res = xmalloc (strlen (user)
3107 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3108 + (opaque ? strlen (opaque) : 0)
3110 sprintf (res, "Digest \
3111 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3112 user, realm, nonce, path, response_digest);
3115 char *p = res + strlen (res);
3116 strcat (p, ", opaque=\"");
3123 #endif /* ENABLE_DIGEST */
3125 /* Computing the size of a string literal must take into account that
3126 value returned by sizeof includes the terminating \0. */
3127 #define STRSIZE(literal) (sizeof (literal) - 1)
3129 /* Whether chars in [b, e) begin with the literal string provided as
3130 first argument and are followed by whitespace or terminating \0.
3131 The comparison is case-insensitive. */
3132 #define STARTS(literal, b, e) \
3134 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3135 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3136 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3137 || c_isspace (b[STRSIZE (literal)])))
3140 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3142 return STARTS ("Basic", hdrbeg, hdrend)
3143 #ifdef ENABLE_DIGEST
3144 || STARTS ("Digest", hdrbeg, hdrend)
3147 || STARTS ("NTLM", hdrbeg, hdrend)
3154 /* Create the HTTP authorization request header. When the
3155 `WWW-Authenticate' response header is seen, according to the
3156 authorization scheme specified in that header (`Basic' and `Digest'
3157 are supported by the current implementation), produce an
3158 appropriate HTTP authorization request header. */
3160 create_authorization_line (const char *au, const char *user,
3161 const char *passwd, const char *method,
3162 const char *path, bool *finished)
3164 /* We are called only with known schemes, so we can dispatch on the
3166 switch (c_toupper (*au))
3168 case 'B': /* Basic */
3170 return basic_authentication_encode (user, passwd);
3171 #ifdef ENABLE_DIGEST
3172 case 'D': /* Digest */
3174 return digest_authentication_encode (au, user, passwd, method, path);
3177 case 'N': /* NTLM */
3178 if (!ntlm_input (&pconn.ntlm, au))
3183 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3186 /* We shouldn't get here -- this function should be only called
3187 with values approved by known_authentication_scheme_p. */
3195 if (!wget_cookie_jar)
3196 wget_cookie_jar = cookie_jar_new ();
3197 if (opt.cookies_input && !cookies_loaded_p)
3199 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3200 cookies_loaded_p = true;
3207 if (wget_cookie_jar)
3208 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3214 xfree_null (pconn.host);
3215 if (wget_cookie_jar)
3216 cookie_jar_delete (wget_cookie_jar);
3220 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3222 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3224 int len = strlen (ext);
3227 strncpy (shortext, ext, len - 1);
3228 shortext[len - 2] = '\0';
3231 if (last_period_in_local_filename == NULL
3232 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3233 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3235 int local_filename_len = strlen (hs->local_file);
3236 /* Resize the local file, allowing for ".html" preceded by
3237 optional ".NUMBER". */
3238 hs->local_file = xrealloc (hs->local_file,
3239 local_filename_len + 24 + len);
3240 strcpy (hs->local_file + local_filename_len, ext);
3241 /* If clobbering is not allowed and the file, as named,
3242 exists, tack on ".NUMBER.html" instead. */
3243 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3247 sprintf (hs->local_file + local_filename_len,
3248 ".%d%s", ext_num++, ext);
3249 while (file_exists_p (hs->local_file));
3251 *dt |= ADDED_HTML_EXTENSION;
3259 test_parse_content_disposition()
3264 char *opt_dir_prefix;
3268 { "filename=\"file.ext\"", NULL, "file.ext", true },
3269 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3270 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3271 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3272 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3273 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3274 { "attachment", NULL, NULL, false },
3275 { "attachment", "somedir", NULL, false },
3278 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3283 opt.dir_prefix = test_array[i].opt_dir_prefix;
3284 res = parse_content_disposition (test_array[i].hdrval, &filename);
3286 mu_assert ("test_parse_content_disposition: wrong result",
3287 res == test_array[i].result
3289 || 0 == strcmp (test_array[i].filename, filename)));
3295 #endif /* TESTING */
3298 * vim: et sts=2 sw=2 cino+={s