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"
73 #endif /* def __VMS */
76 static char *create_authorization_line (const char *, const char *,
77 const char *, const char *,
78 const char *, bool *);
79 static char *basic_authentication_encode (const char *, const char *);
80 static bool known_authentication_scheme_p (const char *, const char *);
81 static void load_cookies (void);
84 # define MIN(x, y) ((x) > (y) ? (y) : (x))
88 static bool cookies_loaded_p;
89 static struct cookie_jar *wget_cookie_jar;
91 #define TEXTHTML_S "text/html"
92 #define TEXTXHTML_S "application/xhtml+xml"
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", 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", 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_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
820 /* Print the server response, line by line, omitting the trailing CRLF
821 from individual header lines, and prefixed with PREFIX. */
824 print_server_response (const struct response *resp, const char *prefix)
829 for (i = 0; resp->headers[i + 1]; i++)
831 const char *b = resp->headers[i];
832 const char *e = resp->headers[i + 1];
834 if (b < e && e[-1] == '\n')
836 if (b < e && e[-1] == '\r')
838 print_response_line(prefix, b, e);
842 /* Parse the `Content-Range' header and extract the information it
843 contains. Returns true if successful, false otherwise. */
845 parse_content_range (const char *hdr, wgint *first_byte_ptr,
846 wgint *last_byte_ptr, wgint *entity_length_ptr)
850 /* Ancient versions of Netscape proxy server, presumably predating
851 rfc2068, sent out `Content-Range' without the "bytes"
853 if (0 == strncasecmp (hdr, "bytes", 5))
856 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
860 while (c_isspace (*hdr))
865 if (!c_isdigit (*hdr))
867 for (num = 0; c_isdigit (*hdr); hdr++)
868 num = 10 * num + (*hdr - '0');
869 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
871 *first_byte_ptr = num;
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
877 *last_byte_ptr = num;
882 for (num = 0; c_isdigit (*hdr); hdr++)
883 num = 10 * num + (*hdr - '0');
884 *entity_length_ptr = num;
888 /* Read the body of the request, but don't store it anywhere and don't
889 display a progress gauge. This is useful for reading the bodies of
890 administrative responses to which we will soon issue another
891 request. The response is not useful to the user, but reading it
892 allows us to continue using the same connection to the server.
894 If reading fails, false is returned, true otherwise. In debug
895 mode, the body is displayed for debugging purposes. */
898 skip_short_body (int fd, wgint contlen)
901 SKIP_SIZE = 512, /* size of the download buffer */
902 SKIP_THRESHOLD = 4096 /* the largest size we read */
904 char dlbuf[SKIP_SIZE + 1];
905 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
907 /* We shouldn't get here with unknown contlen. (This will change
908 with HTTP/1.1, which supports "chunked" transfer.) */
909 assert (contlen != -1);
911 /* If the body is too large, it makes more sense to simply close the
912 connection than to try to read the body. */
913 if (contlen > SKIP_THRESHOLD)
916 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
920 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
923 /* Don't normally report the error since this is an
924 optimization that should be invisible to the user. */
925 DEBUGP (("] aborting (%s).\n",
926 ret < 0 ? fd_errstr (fd) : "EOF received"));
930 /* Safe even if %.*s bogusly expects terminating \0 because
931 we've zero-terminated dlbuf above. */
932 DEBUGP (("%.*s", ret, dlbuf));
935 DEBUGP (("] done.\n"));
939 /* Extract a parameter from the string (typically an HTTP header) at
940 **SOURCE and advance SOURCE to the next parameter. Return false
941 when there are no more parameters to extract. The name of the
942 parameter is returned in NAME, and the value in VALUE. If the
943 parameter has no value, the token's value is zeroed out.
945 For example, if *SOURCE points to the string "attachment;
946 filename=\"foo bar\"", the first call to this function will return
947 the token named "attachment" and no value, and the second call will
948 return the token named "filename" and value "foo bar". The third
949 call will return false, indicating no more valid tokens. */
952 extract_param (const char **source, param_token *name, param_token *value,
955 const char *p = *source;
957 while (c_isspace (*p)) ++p;
961 return false; /* no error; nothing more to extract */
966 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
968 if (name->b == name->e)
969 return false; /* empty name: error */
970 while (c_isspace (*p)) ++p;
971 if (*p == separator || !*p) /* no value */
974 if (*p == separator) ++p;
979 return false; /* error */
981 /* *p is '=', extract value */
983 while (c_isspace (*p)) ++p;
984 if (*p == '"') /* quoted */
987 while (*p && *p != '"') ++p;
991 /* Currently at closing quote; find the end of param. */
992 while (c_isspace (*p)) ++p;
993 while (*p && *p != separator) ++p;
997 /* garbage after closed quote, e.g. foo="bar"baz */
1003 while (*p && *p != separator) ++p;
1005 while (value->e != value->b && c_isspace (value->e[-1]))
1007 if (*p == separator) ++p;
1014 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1016 /* Parse the contents of the `Content-Disposition' header, extracting
1017 the information useful to Wget. Content-Disposition is a header
1018 borrowed from MIME; when used in HTTP, it typically serves for
1019 specifying the desired file name of the resource. For example:
1021 Content-Disposition: attachment; filename="flora.jpg"
1023 Wget will skip the tokens it doesn't care about, such as
1024 "attachment" in the previous example; it will also skip other
1025 unrecognized params. If the header is syntactically correct and
1026 contains a file name, a copy of the file name is stored in
1027 *filename and true is returned. Otherwise, the function returns
1030 The file name is stripped of directory components and must not be
1034 parse_content_disposition (const char *hdr, char **filename)
1036 param_token name, value;
1037 while (extract_param (&hdr, &name, &value, ';'))
1038 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1040 /* Make the file name begin at the last slash or backslash. */
1041 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1042 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1043 if (last_slash && last_bs)
1044 value.b = 1 + MAX (last_slash, last_bs);
1045 else if (last_slash || last_bs)
1046 value.b = 1 + (last_slash ? last_slash : last_bs);
1047 if (value.b == value.e)
1049 /* Start with the directory prefix, if specified. */
1052 int prefix_length = strlen (opt.dir_prefix);
1053 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1058 total_length = prefix_length + (value.e - value.b);
1059 *filename = xmalloc (total_length + 1);
1060 strcpy (*filename, opt.dir_prefix);
1062 (*filename)[prefix_length - 1] = '/';
1063 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1064 (*filename)[total_length] = '\0';
1067 *filename = strdupdelim (value.b, value.e);
1073 /* Persistent connections. Currently, we cache the most recently used
1074 connection as persistent, provided that the HTTP server agrees to
1075 make it such. The persistence data is stored in the variables
1076 below. Ideally, it should be possible to cache an arbitrary fixed
1077 number of these connections. */
1079 /* Whether a persistent connection is active. */
1080 static bool pconn_active;
1083 /* The socket of the connection. */
1086 /* Host and port of the currently active persistent connection. */
1090 /* Whether a ssl handshake has occoured on this connection. */
1093 /* Whether the connection was authorized. This is only done by
1094 NTLM, which authorizes *connections* rather than individual
1095 requests. (That practice is peculiar for HTTP, but it is a
1096 useful optimization.) */
1100 /* NTLM data of the current connection. */
1101 struct ntlmdata ntlm;
1105 /* Mark the persistent connection as invalid and free the resources it
1106 uses. This is used by the CLOSE_* macros after they forcefully
1107 close a registered persistent connection. */
1110 invalidate_persistent (void)
1112 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1113 pconn_active = false;
1114 fd_close (pconn.socket);
1119 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1120 persistent. This will enable someone to use the same connection
1121 later. In the context of HTTP, this must be called only AFTER the
1122 response has been received and the server has promised that the
1123 connection will remain alive.
1125 If a previous connection was persistent, it is closed. */
1128 register_persistent (const char *host, int port, int fd, bool ssl)
1132 if (pconn.socket == fd)
1134 /* The connection FD is already registered. */
1139 /* The old persistent connection is still active; close it
1140 first. This situation arises whenever a persistent
1141 connection exists, but we then connect to a different
1142 host, and try to register a persistent connection to that
1144 invalidate_persistent ();
1148 pconn_active = true;
1150 pconn.host = xstrdup (host);
1153 pconn.authorized = false;
1155 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1158 /* Return true if a persistent connection is available for connecting
1162 persistent_available_p (const char *host, int port, bool ssl,
1163 bool *host_lookup_failed)
1165 /* First, check whether a persistent connection is active at all. */
1169 /* If we want SSL and the last connection wasn't or vice versa,
1170 don't use it. Checking for host and port is not enough because
1171 HTTP and HTTPS can apparently coexist on the same port. */
1172 if (ssl != pconn.ssl)
1175 /* If we're not connecting to the same port, we're not interested. */
1176 if (port != pconn.port)
1179 /* If the host is the same, we're in business. If not, there is
1180 still hope -- read below. */
1181 if (0 != strcasecmp (host, pconn.host))
1183 /* Check if pconn.socket is talking to HOST under another name.
1184 This happens often when both sites are virtual hosts
1185 distinguished only by name and served by the same network
1186 interface, and hence the same web server (possibly set up by
1187 the ISP and serving many different web sites). This
1188 admittedly unconventional optimization does not contradict
1189 HTTP and works well with popular server software. */
1193 struct address_list *al;
1196 /* Don't try to talk to two different SSL sites over the same
1197 secure connection! (Besides, it's not clear that
1198 name-based virtual hosting is even possible with SSL.) */
1201 /* If pconn.socket's peer is one of the IP addresses HOST
1202 resolves to, pconn.socket is for all intents and purposes
1203 already talking to HOST. */
1205 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1207 /* Can't get the peer's address -- something must be very
1208 wrong with the connection. */
1209 invalidate_persistent ();
1212 al = lookup_host (host, 0);
1215 *host_lookup_failed = true;
1219 found = address_list_contains (al, &ip);
1220 address_list_release (al);
1225 /* The persistent connection's peer address was found among the
1226 addresses HOST resolved to; therefore, pconn.sock is in fact
1227 already talking to HOST -- no need to reconnect. */
1230 /* Finally, check whether the connection is still open. This is
1231 important because most servers implement liberal (short) timeout
1232 on persistent connections. Wget can of course always reconnect
1233 if the connection doesn't work out, but it's nicer to know in
1234 advance. This test is a logical followup of the first test, but
1235 is "expensive" and therefore placed at the end of the list.
1237 (Current implementation of test_socket_open has a nice side
1238 effect that it treats sockets with pending data as "closed".
1239 This is exactly what we want: if a broken server sends message
1240 body in response to HEAD, or if it sends more than conent-length
1241 data, we won't reuse the corrupted connection.) */
1243 if (!test_socket_open (pconn.socket))
1245 /* Oops, the socket is no longer open. Now that we know that,
1246 let's invalidate the persistent connection before returning
1248 invalidate_persistent ();
1255 /* The idea behind these two CLOSE macros is to distinguish between
1256 two cases: one when the job we've been doing is finished, and we
1257 want to close the connection and leave, and two when something is
1258 seriously wrong and we're closing the connection as part of
1261 In case of keep_alive, CLOSE_FINISH should leave the connection
1262 open, while CLOSE_INVALIDATE should still close it.
1264 Note that the semantics of the flag `keep_alive' is "this
1265 connection *will* be reused (the server has promised not to close
1266 the connection once we're done)", while the semantics of
1267 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1268 active, registered connection". */
1270 #define CLOSE_FINISH(fd) do { \
1273 if (pconn_active && (fd) == pconn.socket) \
1274 invalidate_persistent (); \
1283 #define CLOSE_INVALIDATE(fd) do { \
1284 if (pconn_active && (fd) == pconn.socket) \
1285 invalidate_persistent (); \
1293 wgint len; /* received length */
1294 wgint contlen; /* expected length */
1295 wgint restval; /* the restart value */
1296 int res; /* the result of last read */
1297 char *rderrmsg; /* error message from read error */
1298 char *newloc; /* new location (redirection) */
1299 char *remote_time; /* remote time-stamp string */
1300 char *error; /* textual HTTP error */
1301 int statcode; /* status code */
1302 wgint rd_size; /* amount of data read from socket */
1303 double dltime; /* time it took to download the data */
1304 const char *referer; /* value of the referer header. */
1305 char *local_file; /* local file name. */
1306 bool existence_checked; /* true if we already checked for a file's
1307 existence after having begun to download
1308 (needed in gethttp for when connection is
1309 interrupted/restarted. */
1310 bool timestamp_checked; /* true if pre-download time-stamping checks
1311 * have already been performed */
1312 char *orig_file_name; /* name of file to compare for time-stamping
1313 * (might be != local_file if -K is set) */
1314 wgint orig_file_size; /* size of file to compare for time-stamping */
1315 time_t orig_file_tstamp; /* time-stamp of file to compare for
1320 free_hstat (struct http_stat *hs)
1322 xfree_null (hs->newloc);
1323 xfree_null (hs->remote_time);
1324 xfree_null (hs->error);
1325 xfree_null (hs->rderrmsg);
1326 xfree_null (hs->local_file);
1327 xfree_null (hs->orig_file_name);
1329 /* Guard against being called twice. */
1331 hs->remote_time = NULL;
1335 #define BEGINS_WITH(line, string_constant) \
1336 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1337 && (c_isspace (line[sizeof (string_constant) - 1]) \
1338 || !line[sizeof (string_constant) - 1]))
1341 #define SET_USER_AGENT(req) do { \
1342 if (!opt.useragent) \
1343 request_set_header (req, "User-Agent", \
1344 aprintf ("Wget/%s (VMS %s %s)", \
1345 VERSION_STRING, vms_arch(), vms_vers()), \
1347 else if (*opt.useragent) \
1348 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1350 #else /* def __VMS */
1351 #define SET_USER_AGENT(req) do { \
1352 if (!opt.useragent) \
1353 request_set_header (req, "User-Agent", \
1354 aprintf ("Wget/%s (%s)", \
1355 VERSION_STRING, OS_TYPE), \
1357 else if (*opt.useragent) \
1358 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1360 #endif /* def __VMS [else] */
1362 /* The flags that allow clobbering the file (opening with "wb").
1363 Defined here to avoid repetition later. #### This will require
1365 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1366 || opt.dirstruct || opt.output_document)
1368 /* Retrieve a document through HTTP protocol. It recognizes status
1369 code, and correctly handles redirections. It closes the network
1370 socket. If it receives an error from the functions below it, it
1371 will print it if there is enough information to do so (almost
1372 always), returning the error to the caller (i.e. http_loop).
1374 Various HTTP parameters are stored to hs.
1376 If PROXY is non-NULL, the connection will be made to the proxy
1377 server, and u->url will be requested. */
1379 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1381 struct request *req;
1384 char *user, *passwd;
1388 wgint contlen, contrange;
1395 /* Set to 1 when the authorization has already been sent and should
1396 not be tried again. */
1397 bool auth_finished = false;
1399 /* Set to 1 when just globally-set Basic authorization has been sent;
1400 * should prevent further Basic negotiations, but not other
1402 bool basic_auth_finished = false;
1404 /* Whether NTLM authentication is used for this request. */
1405 bool ntlm_seen = false;
1407 /* Whether our connection to the remote host is through SSL. */
1408 bool using_ssl = false;
1410 /* Whether a HEAD request will be issued (as opposed to GET or
1412 bool head_only = !!(*dt & HEAD_ONLY);
1415 struct response *resp;
1419 /* Whether this connection will be kept alive after the HTTP request
1423 /* Whether keep-alive should be inhibited.
1425 RFC 2068 requests that 1.0 clients not send keep-alive requests
1426 to proxies. This is because many 1.0 proxies do not interpret
1427 the Connection header and transfer it to the remote server,
1428 causing it to not close the connection and leave both the proxy
1429 and the client hanging. */
1430 bool inhibit_keep_alive =
1431 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1433 /* Headers sent when using POST. */
1434 wgint post_data_size = 0;
1436 bool host_lookup_failed = false;
1439 if (u->scheme == SCHEME_HTTPS)
1441 /* Initialize the SSL context. After this has once been done,
1442 it becomes a no-op. */
1445 scheme_disable (SCHEME_HTTPS);
1446 logprintf (LOG_NOTQUIET,
1447 _("Disabling SSL due to encountered errors.\n"));
1448 return SSLINITFAILED;
1451 #endif /* HAVE_SSL */
1453 /* Initialize certain elements of struct http_stat. */
1457 hs->rderrmsg = NULL;
1459 hs->remote_time = NULL;
1464 /* Prepare the request to send. */
1466 req = request_new ();
1469 const char *meth = "GET";
1472 else if (opt.post_file_name || opt.post_data)
1474 /* Use the full path, i.e. one that includes the leading slash and
1475 the query string. E.g. if u->path is "foo/bar" and u->query is
1476 "param=value", full_path will be "/foo/bar?param=value". */
1479 /* When using SSL over proxy, CONNECT establishes a direct
1480 connection to the HTTPS server. Therefore use the same
1481 argument as when talking to the server directly. */
1482 && u->scheme != SCHEME_HTTPS
1485 meth_arg = xstrdup (u->url);
1487 meth_arg = url_full_path (u);
1488 request_set_method (req, meth, meth_arg);
1491 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1492 if (*dt & SEND_NOCACHE)
1493 request_set_header (req, "Pragma", "no-cache", rel_none);
1495 request_set_header (req, "Range",
1496 aprintf ("bytes=%s-",
1497 number_to_static_string (hs->restval)),
1499 SET_USER_AGENT (req);
1500 request_set_header (req, "Accept", "*/*", rel_none);
1502 /* Find the username and password for authentication. */
1505 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1506 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1507 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1510 && !u->user) /* We only do "site-wide" authentication with "global"
1511 user/password values; URL user/password info overrides. */
1513 /* If this is a host for which we've already received a Basic
1514 * challenge, we'll go ahead and send Basic authentication creds. */
1515 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1521 char *proxy_user, *proxy_passwd;
1522 /* For normal username and password, URL components override
1523 command-line/wgetrc parameters. With proxy
1524 authentication, it's the reverse, because proxy URLs are
1525 normally the "permanent" ones, so command-line args
1526 should take precedence. */
1527 if (opt.proxy_user && opt.proxy_passwd)
1529 proxy_user = opt.proxy_user;
1530 proxy_passwd = opt.proxy_passwd;
1534 proxy_user = proxy->user;
1535 proxy_passwd = proxy->passwd;
1537 /* #### This does not appear right. Can't the proxy request,
1538 say, `Digest' authentication? */
1539 if (proxy_user && proxy_passwd)
1540 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1542 /* If we're using a proxy, we will be connecting to the proxy
1546 /* Proxy authorization over SSL is handled below. */
1548 if (u->scheme != SCHEME_HTTPS)
1550 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1553 /* Generate the Host header, HOST:PORT. Take into account that:
1555 - Broken server-side software often doesn't recognize the PORT
1556 argument, so we must generate "Host: www.server.com" instead of
1557 "Host: www.server.com:80" (and likewise for https port).
1559 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1560 becomes ambiguous and needs to be rewritten as "Host:
1561 [3ffe:8100:200:2::2]:1234". */
1563 /* Formats arranged for hfmt[add_port][add_squares]. */
1564 static const char *hfmt[][2] = {
1565 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1567 int add_port = u->port != scheme_default_port (u->scheme);
1568 int add_squares = strchr (u->host, ':') != NULL;
1569 request_set_header (req, "Host",
1570 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1574 if (!inhibit_keep_alive)
1575 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1578 request_set_header (req, "Cookie",
1579 cookie_header (wget_cookie_jar,
1580 u->host, u->port, u->path,
1582 u->scheme == SCHEME_HTTPS
1589 if (opt.post_data || opt.post_file_name)
1591 request_set_header (req, "Content-Type",
1592 "application/x-www-form-urlencoded", rel_none);
1594 post_data_size = strlen (opt.post_data);
1597 post_data_size = file_size (opt.post_file_name);
1598 if (post_data_size == -1)
1600 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1601 opt.post_file_name, strerror (errno));
1605 request_set_header (req, "Content-Length",
1606 xstrdup (number_to_static_string (post_data_size)),
1610 /* Add the user headers. */
1611 if (opt.user_headers)
1614 for (i = 0; opt.user_headers[i]; i++)
1615 request_set_user_header (req, opt.user_headers[i]);
1619 /* We need to come back here when the initial attempt to retrieve
1620 without authorization header fails. (Expected to happen at least
1621 for the Digest authorization scheme.) */
1625 /* Establish the connection. */
1627 if (!inhibit_keep_alive)
1629 /* Look for a persistent connection to target host, unless a
1630 proxy is used. The exception is when SSL is in use, in which
1631 case the proxy is nothing but a passthrough to the target
1632 host, registered as a connection to the latter. */
1633 struct url *relevant = conn;
1635 if (u->scheme == SCHEME_HTTPS)
1639 if (persistent_available_p (relevant->host, relevant->port,
1641 relevant->scheme == SCHEME_HTTPS,
1645 &host_lookup_failed))
1647 sock = pconn.socket;
1648 using_ssl = pconn.ssl;
1649 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1650 escnonprint (pconn.host), pconn.port);
1651 DEBUGP (("Reusing fd %d.\n", sock));
1652 if (pconn.authorized)
1653 /* If the connection is already authorized, the "Basic"
1654 authorization added by code above is unnecessary and
1656 request_remove_header (req, "Authorization");
1658 else if (host_lookup_failed)
1661 logprintf(LOG_NOTQUIET,
1662 _("%s: unable to resolve host address `%s'\n"),
1663 exec_name, relevant->host);
1670 sock = connect_to_host (conn->host, conn->port);
1679 return (retryable_socket_connect_error (errno)
1680 ? CONERROR : CONIMPOSSIBLE);
1684 if (proxy && u->scheme == SCHEME_HTTPS)
1686 /* When requesting SSL URLs through proxies, use the
1687 CONNECT method to request passthrough. */
1688 struct request *connreq = request_new ();
1689 request_set_method (connreq, "CONNECT",
1690 aprintf ("%s:%d", u->host, u->port));
1691 SET_USER_AGENT (connreq);
1694 request_set_header (connreq, "Proxy-Authorization",
1695 proxyauth, rel_value);
1696 /* Now that PROXYAUTH is part of the CONNECT request,
1697 zero it out so we don't send proxy authorization with
1698 the regular request below. */
1701 /* Examples in rfc2817 use the Host header in CONNECT
1702 requests. I don't see how that gains anything, given
1703 that the contents of Host would be exactly the same as
1704 the contents of CONNECT. */
1706 write_error = request_send (connreq, sock);
1707 request_free (connreq);
1708 if (write_error < 0)
1710 CLOSE_INVALIDATE (sock);
1714 head = read_http_response_head (sock);
1717 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1719 CLOSE_INVALIDATE (sock);
1728 DEBUGP (("proxy responded with: [%s]\n", head));
1730 resp = resp_new (head);
1731 statcode = resp_status (resp, &message);
1734 if (statcode != 200)
1737 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1738 message ? escnonprint (message) : "?");
1739 xfree_null (message);
1742 xfree_null (message);
1744 /* SOCK is now *really* connected to u->host, so update CONN
1745 to reflect this. That way register_persistent will
1746 register SOCK as being connected to u->host:u->port. */
1750 if (conn->scheme == SCHEME_HTTPS)
1752 if (!ssl_connect_wget (sock) || !ssl_check_certificate (sock, u->host))
1759 #endif /* HAVE_SSL */
1762 /* Send the request to server. */
1763 write_error = request_send (req, sock);
1765 if (write_error >= 0)
1769 DEBUGP (("[POST data: %s]\n", opt.post_data));
1770 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1772 else if (opt.post_file_name && post_data_size != 0)
1773 write_error = post_file (sock, opt.post_file_name, post_data_size);
1776 if (write_error < 0)
1778 CLOSE_INVALIDATE (sock);
1782 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1783 proxy ? "Proxy" : "HTTP");
1788 head = read_http_response_head (sock);
1793 logputs (LOG_NOTQUIET, _("No data received.\n"));
1794 CLOSE_INVALIDATE (sock);
1800 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1802 CLOSE_INVALIDATE (sock);
1807 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1809 resp = resp_new (head);
1811 /* Check for status line. */
1813 statcode = resp_status (resp, &message);
1814 if (!opt.server_response)
1815 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1816 message ? escnonprint (message) : "");
1819 logprintf (LOG_VERBOSE, "\n");
1820 print_server_response (resp, " ");
1823 /* Determine the local filename if needed. Notice that if -O is used
1824 * hstat.local_file is set by http_loop to the argument of -O. */
1825 if (!hs->local_file)
1827 /* Honor Content-Disposition whether possible. */
1828 if (!opt.content_disposition
1829 || !resp_header_copy (resp, "Content-Disposition",
1830 hdrval, sizeof (hdrval))
1831 || !parse_content_disposition (hdrval, &hs->local_file))
1833 /* The Content-Disposition header is missing or broken.
1834 * Choose unique file name according to given URL. */
1835 hs->local_file = url_file_name (u);
1839 /* TODO: perform this check only once. */
1840 if (!hs->existence_checked && file_exists_p (hs->local_file))
1844 /* If opt.noclobber is turned on and file already exists, do not
1845 retrieve the file */
1846 logprintf (LOG_VERBOSE, _("\
1847 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1848 /* If the file is there, we suppose it's retrieved OK. */
1851 /* #### Bogusness alert. */
1852 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1853 if (has_html_suffix_p (hs->local_file))
1856 return RETRUNNEEDED;
1858 else if (!ALLOW_CLOBBER)
1860 char *unique = unique_name (hs->local_file, true);
1861 if (unique != hs->local_file)
1862 xfree (hs->local_file);
1863 hs->local_file = unique;
1866 hs->existence_checked = true;
1868 /* Support timestamping */
1869 /* TODO: move this code out of gethttp. */
1870 if (opt.timestamping && !hs->timestamp_checked)
1872 size_t filename_len = strlen (hs->local_file);
1873 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
1874 bool local_dot_orig_file_exists = false;
1875 char *local_filename = NULL;
1878 if (opt.backup_converted)
1879 /* If -K is specified, we'll act on the assumption that it was specified
1880 last time these files were downloaded as well, and instead of just
1881 comparing local file X against server file X, we'll compare local
1882 file X.orig (if extant, else X) against server file X. If -K
1883 _wasn't_ specified last time, or the server contains files called
1884 *.orig, -N will be back to not operating correctly with -k. */
1886 /* Would a single s[n]printf() call be faster? --dan
1888 Definitely not. sprintf() is horribly slow. It's a
1889 different question whether the difference between the two
1890 affects a program. Usually I'd say "no", but at one
1891 point I profiled Wget, and found that a measurable and
1892 non-negligible amount of time was lost calling sprintf()
1893 in url.c. Replacing sprintf with inline calls to
1894 strcpy() and number_to_string() made a difference.
1896 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1897 memcpy (filename_plus_orig_suffix + filename_len,
1898 ORIG_SFX, sizeof (ORIG_SFX));
1900 /* Try to stat() the .orig file. */
1901 if (stat (filename_plus_orig_suffix, &st) == 0)
1903 local_dot_orig_file_exists = true;
1904 local_filename = filename_plus_orig_suffix;
1908 if (!local_dot_orig_file_exists)
1909 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1910 if (stat (hs->local_file, &st) == 0)
1911 local_filename = hs->local_file;
1913 if (local_filename != NULL)
1914 /* There was a local file, so we'll check later to see if the version
1915 the server has is the same version we already have, allowing us to
1918 hs->orig_file_name = xstrdup (local_filename);
1919 hs->orig_file_size = st.st_size;
1920 hs->orig_file_tstamp = st.st_mtime;
1922 /* Modification time granularity is 2 seconds for Windows, so
1923 increase local time by 1 second for later comparison. */
1924 ++hs->orig_file_tstamp;
1929 if (!opt.ignore_length
1930 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1934 parsed = str_to_wgint (hdrval, NULL, 10);
1935 if (parsed == WGINT_MAX && errno == ERANGE)
1938 #### If Content-Length is out of range, it most likely
1939 means that the file is larger than 2G and that we're
1940 compiled without LFS. In that case we should probably
1941 refuse to even attempt to download the file. */
1944 else if (parsed < 0)
1946 /* Negative Content-Length; nonsensical, so we can't
1947 assume any information about the content to receive. */
1954 /* Check for keep-alive related responses. */
1955 if (!inhibit_keep_alive && contlen != -1)
1957 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1959 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1961 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1966 /* The server has promised that it will not close the connection
1967 when we're done. This means that we can register it. */
1968 register_persistent (conn->host, conn->port, sock, using_ssl);
1970 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1972 /* Authorization is required. */
1973 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1974 CLOSE_FINISH (sock);
1976 CLOSE_INVALIDATE (sock);
1977 pconn.authorized = false;
1978 if (!auth_finished && (user && passwd))
1980 /* IIS sends multiple copies of WWW-Authenticate, one with
1981 the value "negotiate", and other(s) with data. Loop over
1982 all the occurrences and pick the one we recognize. */
1984 const char *wabeg, *waend;
1985 char *www_authenticate = NULL;
1987 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1988 &wabeg, &waend)) != -1;
1990 if (known_authentication_scheme_p (wabeg, waend))
1992 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1996 if (!www_authenticate)
1998 /* If the authentication header is missing or
1999 unrecognized, there's no sense in retrying. */
2000 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2002 else if (!basic_auth_finished
2003 || !BEGINS_WITH (www_authenticate, "Basic"))
2006 pth = url_full_path (u);
2007 request_set_header (req, "Authorization",
2008 create_authorization_line (www_authenticate,
2010 request_method (req),
2014 if (BEGINS_WITH (www_authenticate, "NTLM"))
2016 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2018 /* Need to register this host as using basic auth,
2019 * so we automatically send creds next time. */
2020 register_basic_auth_host (u->host);
2023 goto retry_with_auth;
2027 /* We already did Basic auth, and it failed. Gotta
2031 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2035 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2037 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2039 pconn.authorized = true;
2043 hs->statcode = statcode;
2045 hs->error = xstrdup (_("Malformed status line"));
2047 hs->error = xstrdup (_("(no description)"));
2049 hs->error = xstrdup (message);
2050 xfree_null (message);
2052 type = resp_header_strdup (resp, "Content-Type");
2055 char *tmp = strchr (type, ';');
2058 while (tmp > type && c_isspace (tmp[-1]))
2063 hs->newloc = resp_header_strdup (resp, "Location");
2064 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2066 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2070 const char *scbeg, *scend;
2071 /* The jar should have been created by now. */
2072 assert (wget_cookie_jar != NULL);
2074 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2075 &scbeg, &scend)) != -1;
2078 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2079 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2080 u->path, set_cookie);
2084 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2086 wgint first_byte_pos, last_byte_pos, entity_length;
2087 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2090 contrange = first_byte_pos;
2091 contlen = last_byte_pos - first_byte_pos + 1;
2096 /* 20x responses are counted among successful by default. */
2097 if (H_20X (statcode))
2100 /* Return if redirected. */
2101 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2103 /* RFC2068 says that in case of the 300 (multiple choices)
2104 response, the server can output a preferred URL through
2105 `Location' header; otherwise, the request should be treated
2106 like GET. So, if the location is set, it will be a
2107 redirection; otherwise, just proceed normally. */
2108 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2112 logprintf (LOG_VERBOSE,
2113 _("Location: %s%s\n"),
2114 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2115 hs->newloc ? _(" [following]") : "");
2116 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2117 CLOSE_FINISH (sock);
2119 CLOSE_INVALIDATE (sock);
2125 /* If content-type is not given, assume text/html. This is because
2126 of the multitude of broken CGI's that "forget" to generate the
2129 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2130 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2135 if (opt.html_extension && (*dt & TEXTHTML))
2136 /* -E / --html-extension / html_extension = on was specified, and this is a
2137 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2138 already the file's suffix, tack on ".html". */
2140 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2142 if (last_period_in_local_filename == NULL
2143 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2144 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2146 int local_filename_len = strlen (hs->local_file);
2147 /* Resize the local file, allowing for ".html" preceded by
2148 optional ".NUMBER". */
2149 hs->local_file = xrealloc (hs->local_file,
2150 local_filename_len + 24 + sizeof (".html"));
2151 strcpy(hs->local_file + local_filename_len, ".html");
2152 /* If clobbering is not allowed and the file, as named,
2153 exists, tack on ".NUMBER.html" instead. */
2154 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2158 sprintf (hs->local_file + local_filename_len,
2159 ".%d.html", ext_num++);
2160 while (file_exists_p (hs->local_file));
2162 *dt |= ADDED_HTML_EXTENSION;
2166 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2168 /* If `-c' is in use and the file has been fully downloaded (or
2169 the remote file has shrunk), Wget effectively requests bytes
2170 after the end of file and the server response with 416. */
2171 logputs (LOG_VERBOSE, _("\
2172 \n The file is already fully retrieved; nothing to do.\n\n"));
2173 /* In case the caller inspects. */
2176 /* Mark as successfully retrieved. */
2179 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2180 might be more bytes in the body. */
2181 return RETRUNNEEDED;
2183 if ((contrange != 0 && contrange != hs->restval)
2184 || (H_PARTIAL (statcode) && !contrange))
2186 /* The Range request was somehow misunderstood by the server.
2189 CLOSE_INVALIDATE (sock);
2195 hs->contlen = contlen + contrange;
2201 /* No need to print this output if the body won't be
2202 downloaded at all, or if the original server response is
2204 logputs (LOG_VERBOSE, _("Length: "));
2207 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2208 if (contlen + contrange >= 1024)
2209 logprintf (LOG_VERBOSE, " (%s)",
2210 human_readable (contlen + contrange));
2213 if (contlen >= 1024)
2214 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2215 number_to_static_string (contlen),
2216 human_readable (contlen));
2218 logprintf (LOG_VERBOSE, _(", %s remaining"),
2219 number_to_static_string (contlen));
2223 logputs (LOG_VERBOSE,
2224 opt.ignore_length ? _("ignored") : _("unspecified"));
2226 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2228 logputs (LOG_VERBOSE, "\n");
2232 type = NULL; /* We don't need it any more. */
2234 /* Return if we have no intention of further downloading. */
2235 if (!(*dt & RETROKF) || head_only)
2237 /* In case the caller cares to look... */
2242 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2243 servers not to send body in response to a HEAD request, and
2244 those that do will likely be caught by test_socket_open.
2245 If not, they can be worked around using
2246 `--no-http-keep-alive'. */
2247 CLOSE_FINISH (sock);
2248 else if (keep_alive && skip_short_body (sock, contlen))
2249 /* Successfully skipped the body; also keep using the socket. */
2250 CLOSE_FINISH (sock);
2252 CLOSE_INVALIDATE (sock);
2253 return RETRFINISHED;
2257 For VMS, define common fopen() optional arguments.
2260 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2261 # define FOPEN_BIN_FLAG 3
2262 #else /* def __VMS */
2263 # define FOPEN_BIN_FLAG 1
2264 #endif /* def __VMS [else] */
2266 /* Open the local file. */
2269 mkalldirs (hs->local_file);
2271 rotate_backups (hs->local_file);
2278 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2279 #else /* def __VMS */
2280 fp = fopen (hs->local_file, "ab");
2281 #endif /* def __VMS [else] */
2283 else if (ALLOW_CLOBBER)
2289 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2290 #else /* def __VMS */
2291 fp = fopen (hs->local_file, "wb");
2292 #endif /* def __VMS [else] */
2296 fp = fopen_excl (hs->local_file, true);
2297 if (!fp && errno == EEXIST)
2299 /* We cannot just invent a new name and use it (which is
2300 what functions like unique_create typically do)
2301 because we told the user we'd use this name.
2302 Instead, return and retry the download. */
2303 logprintf (LOG_NOTQUIET,
2304 _("%s has sprung into existence.\n"),
2306 CLOSE_INVALIDATE (sock);
2307 return FOPEN_EXCL_ERR;
2312 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2313 CLOSE_INVALIDATE (sock);
2320 /* Print fetch message, if opt.verbose. */
2323 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2324 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2327 /* This confuses the timestamping code that checks for file size.
2328 #### The timestamping code should be smarter about file size. */
2329 if (opt.save_headers && hs->restval == 0)
2330 fwrite (head, 1, strlen (head), fp);
2332 /* Now we no longer need to store the response header. */
2335 /* Download the request body. */
2338 /* If content-length is present, read that much; otherwise, read
2339 until EOF. The HTTP spec doesn't require the server to
2340 actually close the connection when it's done sending data. */
2341 flags |= rb_read_exactly;
2342 if (hs->restval > 0 && contrange == 0)
2343 /* If the server ignored our range request, instruct fd_read_body
2344 to skip the first RESTVAL bytes of body. */
2345 flags |= rb_skip_startpos;
2346 hs->len = hs->restval;
2348 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2349 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2353 CLOSE_FINISH (sock);
2357 hs->rderrmsg = xstrdup (fd_errstr (sock));
2358 CLOSE_INVALIDATE (sock);
2365 return RETRFINISHED;
2368 /* The genuine HTTP loop! This is the part where the retrieval is
2369 retried, and retried, and retried, and... */
2371 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2372 int *dt, struct url *proxy)
2375 bool got_head = false; /* used for time-stamping and filename detection */
2376 bool time_came_from_head = false;
2377 bool got_name = false;
2380 uerr_t err, ret = TRYLIMEXC;
2381 time_t tmr = -1; /* remote time-stamp */
2382 struct http_stat hstat; /* HTTP status */
2384 bool send_head_first = true;
2386 /* Assert that no value for *LOCAL_FILE was passed. */
2387 assert (local_file == NULL || *local_file == NULL);
2389 /* Set LOCAL_FILE parameter. */
2390 if (local_file && opt.output_document)
2391 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2393 /* Reset NEWLOC parameter. */
2396 /* This used to be done in main(), but it's a better idea to do it
2397 here so that we don't go through the hoops if we're just using
2402 /* Warn on (likely bogus) wildcard usage in HTTP. */
2403 if (opt.ftp_glob && has_wildcards_p (u->path))
2404 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2406 /* Setup hstat struct. */
2408 hstat.referer = referer;
2410 if (opt.output_document)
2412 hstat.local_file = xstrdup (opt.output_document);
2415 else if (!opt.content_disposition)
2417 hstat.local_file = url_file_name (u);
2421 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2422 * screaming for some refactoring. */
2423 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber)
2425 /* If opt.noclobber is turned on and file already exists, do not
2426 retrieve the file */
2427 logprintf (LOG_VERBOSE, _("\
2428 File `%s' already there; not retrieving.\n\n"),
2430 /* If the file is there, we suppose it's retrieved OK. */
2433 /* #### Bogusness alert. */
2434 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2435 if (has_html_suffix_p (hstat.local_file))
2438 return RETRUNNEEDED;
2441 /* Reset the counter. */
2444 /* Reset the document type. */
2447 /* Skip preliminary HEAD request if we're not in spider mode AND
2448 * if -O was given or HTTP Content-Disposition support is disabled. */
2450 && (got_name || !opt.content_disposition))
2451 send_head_first = false;
2453 /* Send preliminary HEAD request if -N is given and we have an existing
2454 * destination file. */
2455 if (opt.timestamping
2456 && !opt.content_disposition
2457 && file_exists_p (url_file_name (u)))
2458 send_head_first = true;
2463 /* Increment the pass counter. */
2465 sleep_between_retrievals (count);
2467 /* Get the current time string. */
2468 tms = datetime_str (time (NULL));
2470 if (opt.spider && !got_head)
2471 logprintf (LOG_VERBOSE, _("\
2472 Spider mode enabled. Check if remote file exists.\n"));
2474 /* Print fetch message, if opt.verbose. */
2477 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2482 sprintf (tmp, _("(try:%2d)"), count);
2483 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2488 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2493 ws_changetitle (hurl);
2498 /* Default document type is empty. However, if spider mode is
2499 on or time-stamping is employed, HEAD_ONLY commands is
2500 encoded within *dt. */
2501 if (send_head_first && !got_head)
2506 /* Decide whether or not to restart. */
2509 && stat (hstat.local_file, &st) == 0
2510 && S_ISREG (st.st_mode))
2511 /* When -c is used, continue from on-disk size. (Can't use
2512 hstat.len even if count>1 because we don't want a failed
2513 first attempt to clobber existing data.) */
2514 hstat.restval = st.st_size;
2516 /* otherwise, continue where the previous try left off */
2517 hstat.restval = hstat.len;
2521 /* Decide whether to send the no-cache directive. We send it in
2523 a) we're using a proxy, and we're past our first retrieval.
2524 Some proxies are notorious for caching incomplete data, so
2525 we require a fresh get.
2526 b) caching is explicitly inhibited. */
2527 if ((proxy && count > 1) /* a */
2528 || !opt.allow_cache) /* b */
2529 *dt |= SEND_NOCACHE;
2531 *dt &= ~SEND_NOCACHE;
2533 /* Try fetching the document, or at least its head. */
2534 err = gethttp (u, &hstat, dt, proxy);
2537 tms = datetime_str (time (NULL));
2539 /* Get the new location (with or without the redirection). */
2541 *newloc = xstrdup (hstat.newloc);
2545 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2546 case CONERROR: case READERR: case WRITEFAILED:
2547 case RANGEERR: case FOPEN_EXCL_ERR:
2548 /* Non-fatal errors continue executing the loop, which will
2549 bring them to "while" statement at the end, to judge
2550 whether the number of tries was exceeded. */
2551 printwhat (count, opt.ntry);
2553 case FWRITEERR: case FOPENERR:
2554 /* Another fatal error. */
2555 logputs (LOG_VERBOSE, "\n");
2556 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2557 hstat.local_file, strerror (errno));
2558 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2559 case SSLINITFAILED: case CONTNOTSUPPORTED:
2560 /* Fatal errors just return from the function. */
2564 /* Another fatal error. */
2565 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2569 /* Return the new location to the caller. */
2572 logprintf (LOG_NOTQUIET,
2573 _("ERROR: Redirection (%d) without location.\n"),
2583 /* The file was already fully retrieved. */
2587 /* Deal with you later. */
2590 /* All possibilities should have been exhausted. */
2594 if (!(*dt & RETROKF))
2599 /* #### Ugly ugly ugly! */
2600 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2601 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2604 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2606 && (hstat.statcode == 500 || hstat.statcode == 501))
2611 /* Maybe we should always keep track of broken links, not just in
2613 else if (opt.spider)
2615 /* #### Again: ugly ugly ugly! */
2617 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2618 nonexisting_url (hurl);
2619 logprintf (LOG_NOTQUIET, _("\
2620 Remote file does not exist -- broken link!!!\n"));
2624 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2625 tms, hstat.statcode, escnonprint (hstat.error));
2627 logputs (LOG_VERBOSE, "\n");
2633 /* Did we get the time-stamp? */
2636 got_head = true; /* no more time-stamping */
2638 if (opt.timestamping && !hstat.remote_time)
2640 logputs (LOG_NOTQUIET, _("\
2641 Last-modified header missing -- time-stamps turned off.\n"));
2643 else if (hstat.remote_time)
2645 /* Convert the date-string into struct tm. */
2646 tmr = http_atotm (hstat.remote_time);
2647 if (tmr == (time_t) (-1))
2648 logputs (LOG_VERBOSE, _("\
2649 Last-modified header invalid -- time-stamp ignored.\n"));
2650 if (*dt & HEAD_ONLY)
2651 time_came_from_head = true;
2654 if (send_head_first)
2656 /* The time-stamping section. */
2657 if (opt.timestamping)
2659 if (hstat.orig_file_name) /* Perform the following
2660 checks only if the file
2662 download already exists. */
2664 if (hstat.remote_time &&
2665 tmr != (time_t) (-1))
2667 /* Now time-stamping can be used validly.
2668 Time-stamping means that if the sizes of
2669 the local and remote file match, and local
2670 file is newer than the remote file, it will
2671 not be retrieved. Otherwise, the normal
2672 download procedure is resumed. */
2673 if (hstat.orig_file_tstamp >= tmr)
2675 if (hstat.contlen == -1
2676 || hstat.orig_file_size == hstat.contlen)
2678 logprintf (LOG_VERBOSE, _("\
2679 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2680 hstat.orig_file_name);
2686 logprintf (LOG_VERBOSE, _("\
2687 The sizes do not match (local %s) -- retrieving.\n"),
2688 number_to_static_string (hstat.orig_file_size));
2692 logputs (LOG_VERBOSE,
2693 _("Remote file is newer, retrieving.\n"));
2695 logputs (LOG_VERBOSE, "\n");
2699 /* free_hstat (&hstat); */
2700 hstat.timestamp_checked = true;
2709 logputs (LOG_VERBOSE, _("\
2710 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2714 logprintf (LOG_VERBOSE, _("\
2715 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2716 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2724 logprintf (LOG_VERBOSE, _("\
2725 Remote file exists and could contain further links,\n\
2726 but recursion is disabled -- not retrieving.\n\n"));
2730 logprintf (LOG_VERBOSE, _("\
2731 Remote file exists.\n\n"));
2733 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2740 count = 0; /* the retrieve count for HEAD is reset */
2742 } /* send_head_first */
2745 if ((tmr != (time_t) (-1))
2746 && ((hstat.len == hstat.contlen) ||
2747 ((hstat.res == 0) && (hstat.contlen == -1))))
2749 /* #### This code repeats in http.c and ftp.c. Move it to a
2751 const char *fl = NULL;
2752 if (opt.output_document)
2754 if (output_stream_regular)
2755 fl = opt.output_document;
2758 fl = hstat.local_file;
2762 /* Reparse time header, in case it's changed. */
2763 if (time_came_from_head
2764 && hstat.remote_time && hstat.remote_time[0])
2766 newtmr = http_atotm (hstat.remote_time);
2767 if (newtmr != (time_t)-1)
2773 /* End of time-stamping section. */
2775 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2776 total_download_time += hstat.dltime;
2778 if (hstat.len == hstat.contlen)
2782 logprintf (LOG_VERBOSE,
2783 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2784 tms, tmrate, hstat.local_file,
2785 number_to_static_string (hstat.len),
2786 number_to_static_string (hstat.contlen));
2787 logprintf (LOG_NONVERBOSE,
2788 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2790 number_to_static_string (hstat.len),
2791 number_to_static_string (hstat.contlen),
2792 hstat.local_file, count);
2795 total_downloaded_bytes += hstat.len;
2797 /* Remember that we downloaded the file for later ".orig" code. */
2798 if (*dt & ADDED_HTML_EXTENSION)
2799 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2801 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2806 else if (hstat.res == 0) /* No read error */
2808 if (hstat.contlen == -1) /* We don't know how much we were supposed
2809 to get, so assume we succeeded. */
2813 logprintf (LOG_VERBOSE,
2814 _("%s (%s) - `%s' saved [%s]\n\n"),
2815 tms, tmrate, hstat.local_file,
2816 number_to_static_string (hstat.len));
2817 logprintf (LOG_NONVERBOSE,
2818 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2819 tms, u->url, number_to_static_string (hstat.len),
2820 hstat.local_file, count);
2823 total_downloaded_bytes += hstat.len;
2825 /* Remember that we downloaded the file for later ".orig" code. */
2826 if (*dt & ADDED_HTML_EXTENSION)
2827 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2829 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2834 else if (hstat.len < hstat.contlen) /* meaning we lost the
2835 connection too soon */
2837 logprintf (LOG_VERBOSE,
2838 _("%s (%s) - Connection closed at byte %s. "),
2839 tms, tmrate, number_to_static_string (hstat.len));
2840 printwhat (count, opt.ntry);
2844 /* Getting here would mean reading more data than
2845 requested with content-length, which we never do. */
2848 else /* from now on hstat.res can only be -1 */
2850 if (hstat.contlen == -1)
2852 logprintf (LOG_VERBOSE,
2853 _("%s (%s) - Read error at byte %s (%s)."),
2854 tms, tmrate, number_to_static_string (hstat.len),
2856 printwhat (count, opt.ntry);
2859 else /* hstat.res == -1 and contlen is given */
2861 logprintf (LOG_VERBOSE,
2862 _("%s (%s) - Read error at byte %s/%s (%s). "),
2864 number_to_static_string (hstat.len),
2865 number_to_static_string (hstat.contlen),
2867 printwhat (count, opt.ntry);
2873 while (!opt.ntry || (count < opt.ntry));
2877 *local_file = xstrdup (hstat.local_file);
2878 free_hstat (&hstat);
2883 /* Check whether the result of strptime() indicates success.
2884 strptime() returns the pointer to how far it got to in the string.
2885 The processing has been successful if the string is at `GMT' or
2886 `+X', or at the end of the string.
2888 In extended regexp parlance, the function returns 1 if P matches
2889 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2890 can return) is considered a failure and 0 is returned. */
2892 check_end (const char *p)
2896 while (c_isspace (*p))
2899 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2900 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2906 /* Convert the textual specification of time in TIME_STRING to the
2907 number of seconds since the Epoch.
2909 TIME_STRING can be in any of the three formats RFC2616 allows the
2910 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2911 as well as the time format used in the Set-Cookie header.
2912 Timezones are ignored, and should be GMT.
2914 Return the computed time_t representation, or -1 if the conversion
2917 This function uses strptime with various string formats for parsing
2918 TIME_STRING. This results in a parser that is not as lenient in
2919 interpreting TIME_STRING as I would like it to be. Being based on
2920 strptime, it always allows shortened months, one-digit days, etc.,
2921 but due to the multitude of formats in which time can be
2922 represented, an ideal HTTP time parser would be even more
2923 forgiving. It should completely ignore things like week days and
2924 concentrate only on the various forms of representing years,
2925 months, days, hours, minutes, and seconds. For example, it would
2926 be nice if it accepted ISO 8601 out of the box.
2928 I've investigated free and PD code for this purpose, but none was
2929 usable. getdate was big and unwieldy, and had potential copyright
2930 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2931 distributed with phttpd, is excellent, but we cannot use it because
2932 it is not assigned to the FSF. So I stuck it with strptime. */
2935 http_atotm (const char *time_string)
2937 /* NOTE: Solaris strptime man page claims that %n and %t match white
2938 space, but that's not universally available. Instead, we simply
2939 use ` ' to mean "skip all WS", which works under all strptime
2940 implementations I've tested. */
2942 static const char *time_formats[] = {
2943 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2944 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2945 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2946 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2947 (used in Set-Cookie, defined in the
2948 Netscape cookie specification.) */
2950 const char *oldlocale;
2952 time_t ret = (time_t) -1;
2954 /* Solaris strptime fails to recognize English month names in
2955 non-English locales, which we work around by temporarily setting
2956 locale to C before invoking strptime. */
2957 oldlocale = setlocale (LC_TIME, NULL);
2958 setlocale (LC_TIME, "C");
2960 for (i = 0; i < countof (time_formats); i++)
2964 /* Some versions of strptime use the existing contents of struct
2965 tm to recalculate the date according to format. Zero it out
2966 to prevent stack garbage from influencing strptime. */
2969 if (check_end (strptime (time_string, time_formats[i], &t)))
2976 /* Restore the previous locale. */
2977 setlocale (LC_TIME, oldlocale);
2982 /* Authorization support: We support three authorization schemes:
2984 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2986 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2987 consisting of answering to the server's challenge with the proper
2990 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2991 Stenberg for libcurl. Like digest, NTLM is based on a
2992 challenge-response mechanism, but unlike digest, it is non-standard
2993 (authenticates TCP connections rather than requests), undocumented
2994 and Microsoft-specific. */
2996 /* Create the authentication header contents for the `Basic' scheme.
2997 This is done by encoding the string "USER:PASS" to base64 and
2998 prepending the string "Basic " in front of it. */
3001 basic_authentication_encode (const char *user, const char *passwd)
3004 int len1 = strlen (user) + 1 + strlen (passwd);
3006 t1 = (char *)alloca (len1 + 1);
3007 sprintf (t1, "%s:%s", user, passwd);
3009 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3010 base64_encode (t1, len1, t2);
3012 return concat_strings ("Basic ", t2, (char *) 0);
3015 #define SKIP_WS(x) do { \
3016 while (c_isspace (*(x))) \
3020 #ifdef ENABLE_DIGEST
3021 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3022 an array of 16 bytes containing the hash keys, and BUF should be a
3023 buffer of 33 writable characters (32 for hex digits plus one for
3024 zero termination). */
3026 dump_hash (char *buf, const unsigned char *hash)
3030 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3032 *buf++ = XNUM_TO_digit (*hash >> 4);
3033 *buf++ = XNUM_TO_digit (*hash & 0xf);
3038 /* Take the line apart to find the challenge, and compose a digest
3039 authorization header. See RFC2069 section 2.1.2. */
3041 digest_authentication_encode (const char *au, const char *user,
3042 const char *passwd, const char *method,
3045 static char *realm, *opaque, *nonce;
3050 { "realm", &realm },
3051 { "opaque", &opaque },
3055 param_token name, value;
3057 realm = opaque = nonce = NULL;
3059 au += 6; /* skip over `Digest' */
3060 while (extract_param (&au, &name, &value, ','))
3063 for (i = 0; i < countof (options); i++)
3064 if (name.e - name.b == strlen (options[i].name)
3065 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
3067 *options[i].variable = strdupdelim (value.b, value.e);
3071 if (!realm || !nonce || !user || !passwd || !path || !method)
3074 xfree_null (opaque);
3079 /* Calculate the digest value. */
3081 ALLOCA_MD5_CONTEXT (ctx);
3082 unsigned char hash[MD5_HASHLEN];
3083 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3084 char response_digest[MD5_HASHLEN * 2 + 1];
3086 /* A1BUF = H(user ":" realm ":" password) */
3088 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3089 gen_md5_update ((unsigned char *)":", 1, ctx);
3090 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3091 gen_md5_update ((unsigned char *)":", 1, ctx);
3092 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3093 gen_md5_finish (ctx, hash);
3094 dump_hash (a1buf, hash);
3096 /* A2BUF = H(method ":" path) */
3098 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3099 gen_md5_update ((unsigned char *)":", 1, ctx);
3100 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3101 gen_md5_finish (ctx, hash);
3102 dump_hash (a2buf, hash);
3104 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3106 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3107 gen_md5_update ((unsigned char *)":", 1, ctx);
3108 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3109 gen_md5_update ((unsigned char *)":", 1, ctx);
3110 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3111 gen_md5_finish (ctx, hash);
3112 dump_hash (response_digest, hash);
3114 res = xmalloc (strlen (user)
3119 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3120 + (opaque ? strlen (opaque) : 0)
3122 sprintf (res, "Digest \
3123 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3124 user, realm, nonce, path, response_digest);
3127 char *p = res + strlen (res);
3128 strcat (p, ", opaque=\"");
3135 #endif /* ENABLE_DIGEST */
3137 /* Computing the size of a string literal must take into account that
3138 value returned by sizeof includes the terminating \0. */
3139 #define STRSIZE(literal) (sizeof (literal) - 1)
3141 /* Whether chars in [b, e) begin with the literal string provided as
3142 first argument and are followed by whitespace or terminating \0.
3143 The comparison is case-insensitive. */
3144 #define STARTS(literal, b, e) \
3145 ((e) - (b) >= STRSIZE (literal) \
3146 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3147 && ((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);
3234 test_parse_content_disposition()
3239 char *opt_dir_prefix;
3243 { "filename=\"file.ext\"", NULL, "file.ext", true },
3244 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3245 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3246 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3247 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3248 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3249 { "attachment", NULL, NULL, false },
3250 { "attachment", "somedir", NULL, false },
3253 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3258 opt.dir_prefix = test_array[i].opt_dir_prefix;
3259 res = parse_content_disposition (test_array[i].hdrval, &filename);
3261 mu_assert ("test_parse_content_disposition: wrong result",
3262 res == test_array[i].result
3264 || 0 == strcmp (test_array[i].filename, filename)));
3270 #endif /* TESTING */
3273 * vim: et sts=2 sw=2 cino+={s