2 Copyright (C) 1996-2006 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
55 # include "http-ntlm.h"
68 extern char *version_string;
71 # define MIN(x, y) ((x) > (y) ? (y) : (x))
75 static bool cookies_loaded_p;
76 static struct cookie_jar *wget_cookie_jar;
78 #define TEXTHTML_S "text/html"
79 #define TEXTXHTML_S "application/xhtml+xml"
81 /* Some status code validation macros: */
82 #define H_20X(x) (((x) >= 200) && ((x) < 300))
83 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
84 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
85 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
86 || (x) == HTTP_STATUS_SEE_OTHER \
87 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
89 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
91 #define HTTP_STATUS_OK 200
92 #define HTTP_STATUS_CREATED 201
93 #define HTTP_STATUS_ACCEPTED 202
94 #define HTTP_STATUS_NO_CONTENT 204
95 #define HTTP_STATUS_PARTIAL_CONTENTS 206
97 /* Redirection 3xx. */
98 #define HTTP_STATUS_MULTIPLE_CHOICES 300
99 #define HTTP_STATUS_MOVED_PERMANENTLY 301
100 #define HTTP_STATUS_MOVED_TEMPORARILY 302
101 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
102 #define HTTP_STATUS_NOT_MODIFIED 304
103 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
105 /* Client error 4xx. */
106 #define HTTP_STATUS_BAD_REQUEST 400
107 #define HTTP_STATUS_UNAUTHORIZED 401
108 #define HTTP_STATUS_FORBIDDEN 403
109 #define HTTP_STATUS_NOT_FOUND 404
110 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
112 /* Server errors 5xx. */
113 #define HTTP_STATUS_INTERNAL 500
114 #define HTTP_STATUS_NOT_IMPLEMENTED 501
115 #define HTTP_STATUS_BAD_GATEWAY 502
116 #define HTTP_STATUS_UNAVAILABLE 503
119 rel_none, rel_name, rel_value, rel_both
126 struct request_header {
128 enum rp release_policy;
130 int hcount, hcapacity;
133 /* Create a new, empty request. At least request_set_method must be
134 called before the request can be used. */
136 static struct request *
139 struct request *req = xnew0 (struct request);
141 req->headers = xnew_array (struct request_header, req->hcapacity);
145 /* Set the request's method and its arguments. METH should be a
146 literal string (or it should outlive the request) because it will
147 not be freed. ARG will be freed by request_free. */
150 request_set_method (struct request *req, const char *meth, char *arg)
156 /* Return the method string passed with the last call to
157 request_set_method. */
160 request_method (const struct request *req)
165 /* Free one header according to the release policy specified with
166 request_set_header. */
169 release_header (struct request_header *hdr)
171 switch (hdr->release_policy)
188 /* Set the request named NAME to VALUE. Specifically, this means that
189 a "NAME: VALUE\r\n" header line will be used in the request. If a
190 header with the same name previously existed in the request, its
191 value will be replaced by this one. A NULL value means do nothing.
193 RELEASE_POLICY determines whether NAME and VALUE should be released
194 (freed) with request_free. Allowed values are:
196 - rel_none - don't free NAME or VALUE
197 - rel_name - free NAME when done
198 - rel_value - free VALUE when done
199 - rel_both - free both NAME and VALUE when done
201 Setting release policy is useful when arguments come from different
202 sources. For example:
204 // Don't free literal strings!
205 request_set_header (req, "Pragma", "no-cache", rel_none);
207 // Don't free a global variable, we'll need it later.
208 request_set_header (req, "Referer", opt.referer, rel_none);
210 // Value freshly allocated, free it when done.
211 request_set_header (req, "Range",
212 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
217 request_set_header (struct request *req, char *name, char *value,
218 enum rp release_policy)
220 struct request_header *hdr;
225 /* A NULL value is a no-op; if freeing the name is requested,
226 free it now to avoid leaks. */
227 if (release_policy == rel_name || release_policy == rel_both)
232 for (i = 0; i < req->hcount; i++)
234 hdr = &req->headers[i];
235 if (0 == strcasecmp (name, hdr->name))
237 /* Replace existing header. */
238 release_header (hdr);
241 hdr->release_policy = release_policy;
246 /* Install new header. */
248 if (req->hcount >= req->hcapacity)
250 req->hcapacity <<= 1;
251 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
253 hdr = &req->headers[req->hcount++];
256 hdr->release_policy = release_policy;
259 /* Like request_set_header, but sets the whole header line, as
260 provided by the user using the `--header' option. For example,
261 request_set_user_header (req, "Foo: bar") works just like
262 request_set_header (req, "Foo", "bar"). */
265 request_set_user_header (struct request *req, const char *header)
268 const char *p = strchr (header, ':');
271 BOUNDED_TO_ALLOCA (header, p, name);
275 request_set_header (req, xstrdup (name), (char *) p, rel_name);
278 /* Remove the header with specified name from REQ. Returns true if
279 the header was actually removed, false otherwise. */
282 request_remove_header (struct request *req, char *name)
285 for (i = 0; i < req->hcount; i++)
287 struct request_header *hdr = &req->headers[i];
288 if (0 == strcasecmp (name, hdr->name))
290 release_header (hdr);
291 /* Move the remaining headers by one. */
292 if (i < req->hcount - 1)
293 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
301 #define APPEND(p, str) do { \
302 int A_len = strlen (str); \
303 memcpy (p, str, A_len); \
307 /* Construct the request and write it to FD using fd_write. */
310 request_send (const struct request *req, int fd)
312 char *request_string, *p;
313 int i, size, write_error;
315 /* Count the request size. */
318 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
319 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
321 for (i = 0; i < req->hcount; i++)
323 struct request_header *hdr = &req->headers[i];
324 /* NAME ": " VALUE "\r\n" */
325 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
331 p = request_string = alloca_array (char, size);
333 /* Generate the request. */
335 APPEND (p, req->method); *p++ = ' ';
336 APPEND (p, req->arg); *p++ = ' ';
337 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
339 for (i = 0; i < req->hcount; i++)
341 struct request_header *hdr = &req->headers[i];
342 APPEND (p, hdr->name);
343 *p++ = ':', *p++ = ' ';
344 APPEND (p, hdr->value);
345 *p++ = '\r', *p++ = '\n';
348 *p++ = '\r', *p++ = '\n', *p++ = '\0';
349 assert (p - request_string == size);
353 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
355 /* Send the request to the server. */
357 write_error = fd_write (fd, request_string, size - 1, -1);
359 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
364 /* Release the resources used by REQ. */
367 request_free (struct request *req)
370 xfree_null (req->arg);
371 for (i = 0; i < req->hcount; i++)
372 release_header (&req->headers[i]);
373 xfree_null (req->headers);
377 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
378 PROMISED_SIZE bytes are sent over the wire -- if the file is
379 longer, read only that much; if the file is shorter, report an error. */
382 post_file (int sock, const char *file_name, wgint promised_size)
384 static char chunk[8192];
389 DEBUGP (("[writing POST file %s ... ", file_name));
391 fp = fopen (file_name, "rb");
394 while (!feof (fp) && written < promised_size)
397 int length = fread (chunk, 1, sizeof (chunk), fp);
400 towrite = MIN (promised_size - written, length);
401 write_error = fd_write (sock, chunk, towrite, -1);
411 /* If we've written less than was promised, report a (probably
412 nonsensical) error rather than break the promise. */
413 if (written < promised_size)
419 assert (written == promised_size);
420 DEBUGP (("done]\n"));
424 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
425 If so, return the pointer to the position after the line, otherwise
426 return NULL. This is used as callback to fd_read_hunk. The data
427 between START and PEEKED has been read and cannot be "unread"; the
428 data after PEEKED has only been peeked. */
431 response_head_terminator (const char *start, const char *peeked, int peeklen)
435 /* If at first peek, verify whether HUNK starts with "HTTP". If
436 not, this is a HTTP/0.9 request and we must bail out without
438 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
441 /* Look for "\n[\r]\n", and return the following position if found.
442 Start two chars before the current to cover the possibility that
443 part of the terminator (e.g. "\n\r") arrived in the previous
445 p = peeked - start < 2 ? start : peeked - 2;
446 end = peeked + peeklen;
448 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
449 for (; p < end - 2; p++)
452 if (p[1] == '\r' && p[2] == '\n')
454 else if (p[1] == '\n')
457 /* p==end-2: check for \n\n directly preceding END. */
458 if (p[0] == '\n' && p[1] == '\n')
464 /* The maximum size of a single HTTP response we care to read. Rather
465 than being a limit of the reader implementation, this limit
466 prevents Wget from slurping all available memory upon encountering
467 malicious or buggy server output, thus protecting the user. Define
468 it to 0 to remove the limit. */
470 #define HTTP_RESPONSE_MAX_SIZE 65536
472 /* Read the HTTP request head from FD and return it. The error
473 conditions are the same as with fd_read_hunk.
475 To support HTTP/0.9 responses, this function tries to make sure
476 that the data begins with "HTTP". If this is not the case, no data
477 is read and an empty request is returned, so that the remaining
478 data can be treated as body. */
481 read_http_response_head (int fd)
483 return fd_read_hunk (fd, response_head_terminator, 512,
484 HTTP_RESPONSE_MAX_SIZE);
488 /* The response data. */
491 /* The array of pointers that indicate where each header starts.
492 For example, given this HTTP response:
499 The headers are located like this:
501 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
503 headers[0] headers[1] headers[2] headers[3]
505 I.e. headers[0] points to the beginning of the request,
506 headers[1] points to the end of the first header and the
507 beginning of the second one, etc. */
509 const char **headers;
512 /* Create a new response object from the text of the HTTP response,
513 available in HEAD. That text is automatically split into
514 constituent header lines for fast retrieval using
517 static struct response *
518 resp_new (const char *head)
523 struct response *resp = xnew0 (struct response);
528 /* Empty head means that we're dealing with a headerless
529 (HTTP/0.9) response. In that case, don't set HEADERS at
534 /* Split HEAD into header lines, so that resp_header_* functions
535 don't need to do this over and over again. */
541 DO_REALLOC (resp->headers, size, count + 1, const char *);
542 resp->headers[count++] = hdr;
544 /* Break upon encountering an empty line. */
545 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
548 /* Find the end of HDR, including continuations. */
551 const char *end = strchr (hdr, '\n');
557 while (*hdr == ' ' || *hdr == '\t');
559 DO_REALLOC (resp->headers, size, count + 1, const char *);
560 resp->headers[count] = NULL;
565 /* Locate the header named NAME in the request data, starting with
566 position START. This allows the code to loop through the request
567 data, filtering for all requests of a given name. Returns the
568 found position, or -1 for failure. The code that uses this
569 function typically looks like this:
571 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
572 ... do something with header ...
574 If you only care about one header, use resp_header_get instead of
578 resp_header_locate (const struct response *resp, const char *name, int start,
579 const char **begptr, const char **endptr)
582 const char **headers = resp->headers;
585 if (!headers || !headers[1])
588 name_len = strlen (name);
594 for (; headers[i + 1]; i++)
596 const char *b = headers[i];
597 const char *e = headers[i + 1];
599 && b[name_len] == ':'
600 && 0 == strncasecmp (b, name, name_len))
603 while (b < e && ISSPACE (*b))
605 while (b < e && ISSPACE (e[-1]))
615 /* Find and retrieve the header named NAME in the request data. If
616 found, set *BEGPTR to its starting, and *ENDPTR to its ending
617 position, and return true. Otherwise return false.
619 This function is used as a building block for resp_header_copy
620 and resp_header_strdup. */
623 resp_header_get (const struct response *resp, const char *name,
624 const char **begptr, const char **endptr)
626 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
630 /* Copy the response header named NAME to buffer BUF, no longer than
631 BUFSIZE (BUFSIZE includes the terminating 0). If the header
632 exists, true is returned, false otherwise. If there should be no
633 limit on the size of the header, use resp_header_strdup instead.
635 If BUFSIZE is 0, no data is copied, but the boolean indication of
636 whether the header is present is still returned. */
639 resp_header_copy (const struct response *resp, const char *name,
640 char *buf, int bufsize)
643 if (!resp_header_get (resp, name, &b, &e))
647 int len = MIN (e - b, bufsize - 1);
648 memcpy (buf, b, len);
654 /* Return the value of header named NAME in RESP, allocated with
655 malloc. If such a header does not exist in RESP, return NULL. */
658 resp_header_strdup (const struct response *resp, const char *name)
661 if (!resp_header_get (resp, name, &b, &e))
663 return strdupdelim (b, e);
666 /* Parse the HTTP status line, which is of format:
668 HTTP-Version SP Status-Code SP Reason-Phrase
670 The function returns the status-code, or -1 if the status line
671 appears malformed. The pointer to "reason-phrase" message is
672 returned in *MESSAGE. */
675 resp_status (const struct response *resp, char **message)
682 /* For a HTTP/0.9 response, assume status 200. */
684 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
688 p = resp->headers[0];
689 end = resp->headers[1];
695 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
699 /* Match the HTTP version. This is optional because Gnutella
700 servers have been reported to not specify HTTP version. */
701 if (p < end && *p == '/')
704 while (p < end && ISDIGIT (*p))
706 if (p < end && *p == '.')
708 while (p < end && ISDIGIT (*p))
712 while (p < end && ISSPACE (*p))
714 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
717 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
722 while (p < end && ISSPACE (*p))
724 while (p < end && ISSPACE (end[-1]))
726 *message = strdupdelim (p, end);
732 /* Release the resources used by RESP. */
735 resp_free (struct response *resp)
737 xfree_null (resp->headers);
741 /* Print a single line of response, the characters [b, e). We tried
743 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
744 but that failed to escape the non-printable characters and, in fact,
745 caused crashes in UTF-8 locales. */
748 print_response_line(const char *prefix, const char *b, const char *e)
751 BOUNDED_TO_ALLOCA(b, e, copy);
752 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
755 /* Print the server response, line by line, omitting the trailing CRLF
756 from individual header lines, and prefixed with PREFIX. */
759 print_server_response (const struct response *resp, const char *prefix)
764 for (i = 0; resp->headers[i + 1]; i++)
766 const char *b = resp->headers[i];
767 const char *e = resp->headers[i + 1];
769 if (b < e && e[-1] == '\n')
771 if (b < e && e[-1] == '\r')
773 print_response_line(prefix, b, e);
777 /* Parse the `Content-Range' header and extract the information it
778 contains. Returns true if successful, false otherwise. */
780 parse_content_range (const char *hdr, wgint *first_byte_ptr,
781 wgint *last_byte_ptr, wgint *entity_length_ptr)
785 /* Ancient versions of Netscape proxy server, presumably predating
786 rfc2068, sent out `Content-Range' without the "bytes"
788 if (0 == strncasecmp (hdr, "bytes", 5))
791 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
795 while (ISSPACE (*hdr))
802 for (num = 0; ISDIGIT (*hdr); hdr++)
803 num = 10 * num + (*hdr - '0');
804 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
806 *first_byte_ptr = num;
808 for (num = 0; ISDIGIT (*hdr); hdr++)
809 num = 10 * num + (*hdr - '0');
810 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
812 *last_byte_ptr = num;
814 for (num = 0; ISDIGIT (*hdr); hdr++)
815 num = 10 * num + (*hdr - '0');
816 *entity_length_ptr = num;
820 /* Read the body of the request, but don't store it anywhere and don't
821 display a progress gauge. This is useful for reading the bodies of
822 administrative responses to which we will soon issue another
823 request. The response is not useful to the user, but reading it
824 allows us to continue using the same connection to the server.
826 If reading fails, false is returned, true otherwise. In debug
827 mode, the body is displayed for debugging purposes. */
830 skip_short_body (int fd, wgint contlen)
833 SKIP_SIZE = 512, /* size of the download buffer */
834 SKIP_THRESHOLD = 4096 /* the largest size we read */
836 char dlbuf[SKIP_SIZE + 1];
837 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
839 /* We shouldn't get here with unknown contlen. (This will change
840 with HTTP/1.1, which supports "chunked" transfer.) */
841 assert (contlen != -1);
843 /* If the body is too large, it makes more sense to simply close the
844 connection than to try to read the body. */
845 if (contlen > SKIP_THRESHOLD)
848 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
852 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
855 /* Don't normally report the error since this is an
856 optimization that should be invisible to the user. */
857 DEBUGP (("] aborting (%s).\n",
858 ret < 0 ? fd_errstr (fd) : "EOF received"));
862 /* Safe even if %.*s bogusly expects terminating \0 because
863 we've zero-terminated dlbuf above. */
864 DEBUGP (("%.*s", ret, dlbuf));
867 DEBUGP (("] done.\n"));
871 /* Extract a parameter from the string (typically an HTTP header) at
872 **SOURCE and advance SOURCE to the next parameter. Return false
873 when there are no more parameters to extract. The name of the
874 parameter is returned in NAME, and the value in VALUE. If the
875 parameter has no value, the token's value is zeroed out.
877 For example, if *SOURCE points to the string "attachment;
878 filename=\"foo bar\"", the first call to this function will return
879 the token named "attachment" and no value, and the second call will
880 return the token named "filename" and value "foo bar". The third
881 call will return false, indicating no more valid tokens. */
884 extract_param (const char **source, param_token *name, param_token *value,
887 const char *p = *source;
889 while (ISSPACE (*p)) ++p;
893 return false; /* no error; nothing more to extract */
898 while (*p && !ISSPACE (*p) && *p != '=' && *p != separator) ++p;
900 if (name->b == name->e)
901 return false; /* empty name: error */
902 while (ISSPACE (*p)) ++p;
903 if (*p == separator || !*p) /* no value */
906 if (*p == separator) ++p;
911 return false; /* error */
913 /* *p is '=', extract value */
915 while (ISSPACE (*p)) ++p;
916 if (*p == '"') /* quoted */
919 while (*p && *p != '"') ++p;
923 /* Currently at closing quote; find the end of param. */
924 while (ISSPACE (*p)) ++p;
925 while (*p && *p != separator) ++p;
929 /* garbage after closed quote, e.g. foo="bar"baz */
935 while (*p && *p != separator) ++p;
937 while (value->e != value->b && ISSPACE (value->e[-1]))
939 if (*p == separator) ++p;
946 #define MAX(p, q) ((p) > (q) ? (p) : (q))
948 /* Parse the contents of the `Content-Disposition' header, extracting
949 the information useful to Wget. Content-Disposition is a header
950 borrowed from MIME; when used in HTTP, it typically serves for
951 specifying the desired file name of the resource. For example:
953 Content-Disposition: attachment; filename="flora.jpg"
955 Wget will skip the tokens it doesn't care about, such as
956 "attachment" in the previous example; it will also skip other
957 unrecognized params. If the header is syntactically correct and
958 contains a file name, a copy of the file name is stored in
959 *filename and true is returned. Otherwise, the function returns
962 The file name is stripped of directory components and must not be
966 parse_content_disposition (const char *hdr, char **filename)
968 param_token name, value;
969 while (extract_param (&hdr, &name, &value, ';'))
970 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
972 /* Make the file name begin at the last slash or backslash. */
973 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
974 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
975 if (last_slash && last_bs)
976 value.b = 1 + MAX (last_slash, last_bs);
977 else if (last_slash || last_bs)
978 value.b = 1 + (last_slash ? last_slash : last_bs);
979 if (value.b == value.e)
981 /* Start with the directory prefix, if specified. */
984 int prefix_length = strlen (opt.dir_prefix);
985 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
990 total_length = prefix_length + (value.e - value.b);
991 *filename = xmalloc (total_length + 1);
992 strcpy (*filename, opt.dir_prefix);
994 (*filename)[prefix_length - 1] = '/';
995 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
996 (*filename)[total_length] = '\0';
999 *filename = strdupdelim (value.b, value.e);
1005 /* Persistent connections. Currently, we cache the most recently used
1006 connection as persistent, provided that the HTTP server agrees to
1007 make it such. The persistence data is stored in the variables
1008 below. Ideally, it should be possible to cache an arbitrary fixed
1009 number of these connections. */
1011 /* Whether a persistent connection is active. */
1012 static bool pconn_active;
1015 /* The socket of the connection. */
1018 /* Host and port of the currently active persistent connection. */
1022 /* Whether a ssl handshake has occoured on this connection. */
1025 /* Whether the connection was authorized. This is only done by
1026 NTLM, which authorizes *connections* rather than individual
1027 requests. (That practice is peculiar for HTTP, but it is a
1028 useful optimization.) */
1032 /* NTLM data of the current connection. */
1033 struct ntlmdata ntlm;
1037 /* Mark the persistent connection as invalid and free the resources it
1038 uses. This is used by the CLOSE_* macros after they forcefully
1039 close a registered persistent connection. */
1042 invalidate_persistent (void)
1044 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1045 pconn_active = false;
1046 fd_close (pconn.socket);
1051 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1052 persistent. This will enable someone to use the same connection
1053 later. In the context of HTTP, this must be called only AFTER the
1054 response has been received and the server has promised that the
1055 connection will remain alive.
1057 If a previous connection was persistent, it is closed. */
1060 register_persistent (const char *host, int port, int fd, bool ssl)
1064 if (pconn.socket == fd)
1066 /* The connection FD is already registered. */
1071 /* The old persistent connection is still active; close it
1072 first. This situation arises whenever a persistent
1073 connection exists, but we then connect to a different
1074 host, and try to register a persistent connection to that
1076 invalidate_persistent ();
1080 pconn_active = true;
1082 pconn.host = xstrdup (host);
1085 pconn.authorized = false;
1087 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1090 /* Return true if a persistent connection is available for connecting
1094 persistent_available_p (const char *host, int port, bool ssl,
1095 bool *host_lookup_failed)
1097 /* First, check whether a persistent connection is active at all. */
1101 /* If we want SSL and the last connection wasn't or vice versa,
1102 don't use it. Checking for host and port is not enough because
1103 HTTP and HTTPS can apparently coexist on the same port. */
1104 if (ssl != pconn.ssl)
1107 /* If we're not connecting to the same port, we're not interested. */
1108 if (port != pconn.port)
1111 /* If the host is the same, we're in business. If not, there is
1112 still hope -- read below. */
1113 if (0 != strcasecmp (host, pconn.host))
1115 /* Check if pconn.socket is talking to HOST under another name.
1116 This happens often when both sites are virtual hosts
1117 distinguished only by name and served by the same network
1118 interface, and hence the same web server (possibly set up by
1119 the ISP and serving many different web sites). This
1120 admittedly unconventional optimization does not contradict
1121 HTTP and works well with popular server software. */
1125 struct address_list *al;
1128 /* Don't try to talk to two different SSL sites over the same
1129 secure connection! (Besides, it's not clear that
1130 name-based virtual hosting is even possible with SSL.) */
1133 /* If pconn.socket's peer is one of the IP addresses HOST
1134 resolves to, pconn.socket is for all intents and purposes
1135 already talking to HOST. */
1137 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1139 /* Can't get the peer's address -- something must be very
1140 wrong with the connection. */
1141 invalidate_persistent ();
1144 al = lookup_host (host, 0);
1147 *host_lookup_failed = true;
1151 found = address_list_contains (al, &ip);
1152 address_list_release (al);
1157 /* The persistent connection's peer address was found among the
1158 addresses HOST resolved to; therefore, pconn.sock is in fact
1159 already talking to HOST -- no need to reconnect. */
1162 /* Finally, check whether the connection is still open. This is
1163 important because most servers implement liberal (short) timeout
1164 on persistent connections. Wget can of course always reconnect
1165 if the connection doesn't work out, but it's nicer to know in
1166 advance. This test is a logical followup of the first test, but
1167 is "expensive" and therefore placed at the end of the list.
1169 (Current implementation of test_socket_open has a nice side
1170 effect that it treats sockets with pending data as "closed".
1171 This is exactly what we want: if a broken server sends message
1172 body in response to HEAD, or if it sends more than conent-length
1173 data, we won't reuse the corrupted connection.) */
1175 if (!test_socket_open (pconn.socket))
1177 /* Oops, the socket is no longer open. Now that we know that,
1178 let's invalidate the persistent connection before returning
1180 invalidate_persistent ();
1187 /* The idea behind these two CLOSE macros is to distinguish between
1188 two cases: one when the job we've been doing is finished, and we
1189 want to close the connection and leave, and two when something is
1190 seriously wrong and we're closing the connection as part of
1193 In case of keep_alive, CLOSE_FINISH should leave the connection
1194 open, while CLOSE_INVALIDATE should still close it.
1196 Note that the semantics of the flag `keep_alive' is "this
1197 connection *will* be reused (the server has promised not to close
1198 the connection once we're done)", while the semantics of
1199 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1200 active, registered connection". */
1202 #define CLOSE_FINISH(fd) do { \
1205 if (pconn_active && (fd) == pconn.socket) \
1206 invalidate_persistent (); \
1215 #define CLOSE_INVALIDATE(fd) do { \
1216 if (pconn_active && (fd) == pconn.socket) \
1217 invalidate_persistent (); \
1225 wgint len; /* received length */
1226 wgint contlen; /* expected length */
1227 wgint restval; /* the restart value */
1228 int res; /* the result of last read */
1229 char *rderrmsg; /* error message from read error */
1230 char *newloc; /* new location (redirection) */
1231 char *remote_time; /* remote time-stamp string */
1232 char *error; /* textual HTTP error */
1233 int statcode; /* status code */
1234 wgint rd_size; /* amount of data read from socket */
1235 double dltime; /* time it took to download the data */
1236 const char *referer; /* value of the referer header. */
1237 char *local_file; /* local file name. */
1238 bool timestamp_checked; /* true if pre-download time-stamping checks
1239 * have already been performed */
1240 char *orig_file_name; /* name of file to compare for time-stamping
1241 * (might be != local_file if -K is set) */
1242 wgint orig_file_size; /* size of file to compare for time-stamping */
1243 time_t orig_file_tstamp; /* time-stamp of file to compare for
1248 free_hstat (struct http_stat *hs)
1250 xfree_null (hs->newloc);
1251 xfree_null (hs->remote_time);
1252 xfree_null (hs->error);
1253 xfree_null (hs->rderrmsg);
1254 xfree_null (hs->local_file);
1255 xfree_null (hs->orig_file_name);
1257 /* Guard against being called twice. */
1259 hs->remote_time = NULL;
1263 static char *create_authorization_line (const char *, const char *,
1264 const char *, const char *,
1265 const char *, bool *);
1266 static char *basic_authentication_encode (const char *, const char *);
1267 static bool known_authentication_scheme_p (const char *, const char *);
1268 static void load_cookies (void);
1270 #define BEGINS_WITH(line, string_constant) \
1271 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1272 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1273 || !line[sizeof (string_constant) - 1]))
1275 #define SET_USER_AGENT(req) do { \
1276 if (!opt.useragent) \
1277 request_set_header (req, "User-Agent", \
1278 aprintf ("Wget/%s", version_string), rel_value); \
1279 else if (*opt.useragent) \
1280 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1283 /* The flags that allow clobbering the file (opening with "wb").
1284 Defined here to avoid repetition later. #### This will require
1286 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1287 || opt.dirstruct || opt.output_document)
1289 /* Retrieve a document through HTTP protocol. It recognizes status
1290 code, and correctly handles redirections. It closes the network
1291 socket. If it receives an error from the functions below it, it
1292 will print it if there is enough information to do so (almost
1293 always), returning the error to the caller (i.e. http_loop).
1295 Various HTTP parameters are stored to hs.
1297 If PROXY is non-NULL, the connection will be made to the proxy
1298 server, and u->url will be requested. */
1300 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1302 struct request *req;
1305 char *user, *passwd;
1309 wgint contlen, contrange;
1316 /* Set to 1 when the authorization has failed permanently and should
1317 not be tried again. */
1318 bool auth_finished = false;
1320 /* Whether NTLM authentication is used for this request. */
1321 bool ntlm_seen = false;
1323 /* Whether our connection to the remote host is through SSL. */
1324 bool using_ssl = false;
1326 /* Whether a HEAD request will be issued (as opposed to GET or
1328 bool head_only = !!(*dt & HEAD_ONLY);
1331 struct response *resp;
1335 /* Whether this connection will be kept alive after the HTTP request
1339 /* Whether keep-alive should be inhibited.
1341 RFC 2068 requests that 1.0 clients not send keep-alive requests
1342 to proxies. This is because many 1.0 proxies do not interpret
1343 the Connection header and transfer it to the remote server,
1344 causing it to not close the connection and leave both the proxy
1345 and the client hanging. */
1346 bool inhibit_keep_alive =
1347 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1349 /* Headers sent when using POST. */
1350 wgint post_data_size = 0;
1352 bool host_lookup_failed = false;
1355 if (u->scheme == SCHEME_HTTPS)
1357 /* Initialize the SSL context. After this has once been done,
1358 it becomes a no-op. */
1361 scheme_disable (SCHEME_HTTPS);
1362 logprintf (LOG_NOTQUIET,
1363 _("Disabling SSL due to encountered errors.\n"));
1364 return SSLINITFAILED;
1367 #endif /* HAVE_SSL */
1369 /* Initialize certain elements of struct http_stat. */
1373 hs->rderrmsg = NULL;
1375 hs->remote_time = NULL;
1380 /* Prepare the request to send. */
1382 req = request_new ();
1385 const char *meth = "GET";
1388 else if (opt.post_file_name || opt.post_data)
1390 /* Use the full path, i.e. one that includes the leading slash and
1391 the query string. E.g. if u->path is "foo/bar" and u->query is
1392 "param=value", full_path will be "/foo/bar?param=value". */
1395 /* When using SSL over proxy, CONNECT establishes a direct
1396 connection to the HTTPS server. Therefore use the same
1397 argument as when talking to the server directly. */
1398 && u->scheme != SCHEME_HTTPS
1401 meth_arg = xstrdup (u->url);
1403 meth_arg = url_full_path (u);
1404 request_set_method (req, meth, meth_arg);
1407 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1408 if (*dt & SEND_NOCACHE)
1409 request_set_header (req, "Pragma", "no-cache", rel_none);
1411 request_set_header (req, "Range",
1412 aprintf ("bytes=%s-",
1413 number_to_static_string (hs->restval)),
1415 SET_USER_AGENT (req);
1416 request_set_header (req, "Accept", "*/*", rel_none);
1418 /* Find the username and password for authentication. */
1421 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1422 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1423 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1427 /* We have the username and the password, but haven't tried
1428 any authorization yet. Let's see if the "Basic" method
1429 works. If not, we'll come back here and construct a
1430 proper authorization method with the right challenges.
1432 If we didn't employ this kind of logic, every URL that
1433 requires authorization would have to be processed twice,
1434 which is very suboptimal and generates a bunch of false
1435 "unauthorized" errors in the server log.
1437 #### But this logic also has a serious problem when used
1438 with stronger authentications: we *first* transmit the
1439 username and the password in clear text, and *then* attempt a
1440 stronger authentication scheme. That cannot be right! We
1441 are only fortunate that almost everyone still uses the
1442 `Basic' scheme anyway.
1444 There should be an option to prevent this from happening, for
1445 those who use strong authentication schemes and value their
1447 request_set_header (req, "Authorization",
1448 basic_authentication_encode (user, passwd),
1455 char *proxy_user, *proxy_passwd;
1456 /* For normal username and password, URL components override
1457 command-line/wgetrc parameters. With proxy
1458 authentication, it's the reverse, because proxy URLs are
1459 normally the "permanent" ones, so command-line args
1460 should take precedence. */
1461 if (opt.proxy_user && opt.proxy_passwd)
1463 proxy_user = opt.proxy_user;
1464 proxy_passwd = opt.proxy_passwd;
1468 proxy_user = proxy->user;
1469 proxy_passwd = proxy->passwd;
1471 /* #### This does not appear right. Can't the proxy request,
1472 say, `Digest' authentication? */
1473 if (proxy_user && proxy_passwd)
1474 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1476 /* If we're using a proxy, we will be connecting to the proxy
1480 /* Proxy authorization over SSL is handled below. */
1482 if (u->scheme != SCHEME_HTTPS)
1484 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1487 /* Generate the Host header, HOST:PORT. Take into account that:
1489 - Broken server-side software often doesn't recognize the PORT
1490 argument, so we must generate "Host: www.server.com" instead of
1491 "Host: www.server.com:80" (and likewise for https port).
1493 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1494 becomes ambiguous and needs to be rewritten as "Host:
1495 [3ffe:8100:200:2::2]:1234". */
1497 /* Formats arranged for hfmt[add_port][add_squares]. */
1498 static const char *hfmt[][2] = {
1499 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1501 int add_port = u->port != scheme_default_port (u->scheme);
1502 int add_squares = strchr (u->host, ':') != NULL;
1503 request_set_header (req, "Host",
1504 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1508 if (!inhibit_keep_alive)
1509 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1512 request_set_header (req, "Cookie",
1513 cookie_header (wget_cookie_jar,
1514 u->host, u->port, u->path,
1516 u->scheme == SCHEME_HTTPS
1523 if (opt.post_data || opt.post_file_name)
1525 request_set_header (req, "Content-Type",
1526 "application/x-www-form-urlencoded", rel_none);
1528 post_data_size = strlen (opt.post_data);
1531 post_data_size = file_size (opt.post_file_name);
1532 if (post_data_size == -1)
1534 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1535 opt.post_file_name, strerror (errno));
1539 request_set_header (req, "Content-Length",
1540 xstrdup (number_to_static_string (post_data_size)),
1544 /* Add the user headers. */
1545 if (opt.user_headers)
1548 for (i = 0; opt.user_headers[i]; i++)
1549 request_set_user_header (req, opt.user_headers[i]);
1553 /* We need to come back here when the initial attempt to retrieve
1554 without authorization header fails. (Expected to happen at least
1555 for the Digest authorization scheme.) */
1559 /* Establish the connection. */
1561 if (!inhibit_keep_alive)
1563 /* Look for a persistent connection to target host, unless a
1564 proxy is used. The exception is when SSL is in use, in which
1565 case the proxy is nothing but a passthrough to the target
1566 host, registered as a connection to the latter. */
1567 struct url *relevant = conn;
1569 if (u->scheme == SCHEME_HTTPS)
1573 if (persistent_available_p (relevant->host, relevant->port,
1575 relevant->scheme == SCHEME_HTTPS,
1579 &host_lookup_failed))
1581 sock = pconn.socket;
1582 using_ssl = pconn.ssl;
1583 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1584 escnonprint (pconn.host), pconn.port);
1585 DEBUGP (("Reusing fd %d.\n", sock));
1586 if (pconn.authorized)
1587 /* If the connection is already authorized, the "Basic"
1588 authorization added by code above is unnecessary and
1590 request_remove_header (req, "Authorization");
1596 /* In its current implementation, persistent_available_p will
1597 look up conn->host in some cases. If that lookup failed, we
1598 don't need to bother with connect_to_host. */
1599 if (host_lookup_failed)
1605 sock = connect_to_host (conn->host, conn->port);
1614 return (retryable_socket_connect_error (errno)
1615 ? CONERROR : CONIMPOSSIBLE);
1619 if (proxy && u->scheme == SCHEME_HTTPS)
1621 /* When requesting SSL URLs through proxies, use the
1622 CONNECT method to request passthrough. */
1623 struct request *connreq = request_new ();
1624 request_set_method (connreq, "CONNECT",
1625 aprintf ("%s:%d", u->host, u->port));
1626 SET_USER_AGENT (connreq);
1629 request_set_header (connreq, "Proxy-Authorization",
1630 proxyauth, rel_value);
1631 /* Now that PROXYAUTH is part of the CONNECT request,
1632 zero it out so we don't send proxy authorization with
1633 the regular request below. */
1636 /* Examples in rfc2817 use the Host header in CONNECT
1637 requests. I don't see how that gains anything, given
1638 that the contents of Host would be exactly the same as
1639 the contents of CONNECT. */
1641 write_error = request_send (connreq, sock);
1642 request_free (connreq);
1643 if (write_error < 0)
1645 CLOSE_INVALIDATE (sock);
1649 head = read_http_response_head (sock);
1652 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1654 CLOSE_INVALIDATE (sock);
1663 DEBUGP (("proxy responded with: [%s]\n", head));
1665 resp = resp_new (head);
1666 statcode = resp_status (resp, &message);
1669 if (statcode != 200)
1672 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1673 message ? escnonprint (message) : "?");
1674 xfree_null (message);
1677 xfree_null (message);
1679 /* SOCK is now *really* connected to u->host, so update CONN
1680 to reflect this. That way register_persistent will
1681 register SOCK as being connected to u->host:u->port. */
1685 if (conn->scheme == SCHEME_HTTPS)
1687 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1694 #endif /* HAVE_SSL */
1697 /* Send the request to server. */
1698 write_error = request_send (req, sock);
1700 if (write_error >= 0)
1704 DEBUGP (("[POST data: %s]\n", opt.post_data));
1705 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1707 else if (opt.post_file_name && post_data_size != 0)
1708 write_error = post_file (sock, opt.post_file_name, post_data_size);
1711 if (write_error < 0)
1713 CLOSE_INVALIDATE (sock);
1717 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1718 proxy ? "Proxy" : "HTTP");
1723 head = read_http_response_head (sock);
1728 logputs (LOG_NOTQUIET, _("No data received.\n"));
1729 CLOSE_INVALIDATE (sock);
1735 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1737 CLOSE_INVALIDATE (sock);
1742 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1744 resp = resp_new (head);
1746 /* Check for status line. */
1748 statcode = resp_status (resp, &message);
1749 if (!opt.server_response)
1750 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1751 message ? escnonprint (message) : "");
1754 logprintf (LOG_VERBOSE, "\n");
1755 print_server_response (resp, " ");
1758 /* Determine the local filename if needed. Notice that if -O is used
1759 * hstat.local_file is set by http_loop to the argument of -O. */
1760 if (!hs->local_file)
1762 /* Honor Content-Disposition whether possible. */
1763 if (!opt.content_disposition
1764 || !resp_header_copy (resp, "Content-Disposition",
1765 hdrval, sizeof (hdrval))
1766 || !parse_content_disposition (hdrval, &hs->local_file))
1768 /* The Content-Disposition header is missing or broken.
1769 * Choose unique file name according to given URL. */
1770 hs->local_file = url_file_name (u);
1774 /* TODO: perform this check only once. */
1775 if (file_exists_p (hs->local_file))
1779 /* If opt.noclobber is turned on and file already exists, do not
1780 retrieve the file */
1781 logprintf (LOG_VERBOSE, _("\
1782 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1783 /* If the file is there, we suppose it's retrieved OK. */
1786 /* #### Bogusness alert. */
1787 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1788 if (has_html_suffix_p (hs->local_file))
1793 else if (!ALLOW_CLOBBER)
1795 char *unique = unique_name (hs->local_file, true);
1796 if (unique != hs->local_file)
1797 xfree (hs->local_file);
1798 hs->local_file = unique;
1802 /* Support timestamping */
1803 /* TODO: move this code out of gethttp. */
1804 if (opt.timestamping && !hs->timestamp_checked)
1806 size_t filename_len = strlen (hs->local_file);
1807 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1808 bool local_dot_orig_file_exists = false;
1809 char *local_filename = NULL;
1812 if (opt.backup_converted)
1813 /* If -K is specified, we'll act on the assumption that it was specified
1814 last time these files were downloaded as well, and instead of just
1815 comparing local file X against server file X, we'll compare local
1816 file X.orig (if extant, else X) against server file X. If -K
1817 _wasn't_ specified last time, or the server contains files called
1818 *.orig, -N will be back to not operating correctly with -k. */
1820 /* Would a single s[n]printf() call be faster? --dan
1822 Definitely not. sprintf() is horribly slow. It's a
1823 different question whether the difference between the two
1824 affects a program. Usually I'd say "no", but at one
1825 point I profiled Wget, and found that a measurable and
1826 non-negligible amount of time was lost calling sprintf()
1827 in url.c. Replacing sprintf with inline calls to
1828 strcpy() and number_to_string() made a difference.
1830 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1831 memcpy (filename_plus_orig_suffix + filename_len,
1832 ".orig", sizeof (".orig"));
1834 /* Try to stat() the .orig file. */
1835 if (stat (filename_plus_orig_suffix, &st) == 0)
1837 local_dot_orig_file_exists = true;
1838 local_filename = filename_plus_orig_suffix;
1842 if (!local_dot_orig_file_exists)
1843 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1844 if (stat (hs->local_file, &st) == 0)
1845 local_filename = hs->local_file;
1847 if (local_filename != NULL)
1848 /* There was a local file, so we'll check later to see if the version
1849 the server has is the same version we already have, allowing us to
1852 hs->orig_file_name = xstrdup (local_filename);
1853 hs->orig_file_size = st.st_size;
1854 hs->orig_file_tstamp = st.st_mtime;
1856 /* Modification time granularity is 2 seconds for Windows, so
1857 increase local time by 1 second for later comparison. */
1858 ++hs->orig_file_tstamp;
1863 if (!opt.ignore_length
1864 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1868 parsed = str_to_wgint (hdrval, NULL, 10);
1869 if (parsed == WGINT_MAX && errno == ERANGE)
1871 #### If Content-Length is out of range, it most likely
1872 means that the file is larger than 2G and that we're
1873 compiled without LFS. In that case we should probably
1874 refuse to even attempt to download the file. */
1880 /* Check for keep-alive related responses. */
1881 if (!inhibit_keep_alive && contlen != -1)
1883 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1885 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1887 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1892 /* The server has promised that it will not close the connection
1893 when we're done. This means that we can register it. */
1894 register_persistent (conn->host, conn->port, sock, using_ssl);
1896 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1898 /* Authorization is required. */
1899 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1900 CLOSE_FINISH (sock);
1902 CLOSE_INVALIDATE (sock);
1903 pconn.authorized = false;
1904 if (!auth_finished && (user && passwd))
1906 /* IIS sends multiple copies of WWW-Authenticate, one with
1907 the value "negotiate", and other(s) with data. Loop over
1908 all the occurrences and pick the one we recognize. */
1910 const char *wabeg, *waend;
1911 char *www_authenticate = NULL;
1913 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1914 &wabeg, &waend)) != -1;
1916 if (known_authentication_scheme_p (wabeg, waend))
1918 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1922 if (!www_authenticate)
1923 /* If the authentication header is missing or
1924 unrecognized, there's no sense in retrying. */
1925 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1926 else if (BEGINS_WITH (www_authenticate, "Basic"))
1927 /* If the authentication scheme is "Basic", which we send
1928 by default, there's no sense in retrying either. (This
1929 should be changed when we stop sending "Basic" data by
1935 pth = url_full_path (u);
1936 request_set_header (req, "Authorization",
1937 create_authorization_line (www_authenticate,
1939 request_method (req),
1943 if (BEGINS_WITH (www_authenticate, "NTLM"))
1946 goto retry_with_auth;
1949 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1953 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1955 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1957 pconn.authorized = true;
1961 hs->statcode = statcode;
1963 hs->error = xstrdup (_("Malformed status line"));
1965 hs->error = xstrdup (_("(no description)"));
1967 hs->error = xstrdup (message);
1968 xfree_null (message);
1970 type = resp_header_strdup (resp, "Content-Type");
1973 char *tmp = strchr (type, ';');
1976 while (tmp > type && ISSPACE (tmp[-1]))
1981 hs->newloc = resp_header_strdup (resp, "Location");
1982 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1984 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1988 const char *scbeg, *scend;
1989 /* The jar should have been created by now. */
1990 assert (wget_cookie_jar != NULL);
1992 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1993 &scbeg, &scend)) != -1;
1996 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1997 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1998 u->path, set_cookie);
2002 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2004 wgint first_byte_pos, last_byte_pos, entity_length;
2005 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2007 contrange = first_byte_pos;
2011 /* 20x responses are counted among successful by default. */
2012 if (H_20X (statcode))
2015 /* Return if redirected. */
2016 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2018 /* RFC2068 says that in case of the 300 (multiple choices)
2019 response, the server can output a preferred URL through
2020 `Location' header; otherwise, the request should be treated
2021 like GET. So, if the location is set, it will be a
2022 redirection; otherwise, just proceed normally. */
2023 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2027 logprintf (LOG_VERBOSE,
2028 _("Location: %s%s\n"),
2029 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2030 hs->newloc ? _(" [following]") : "");
2031 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2032 CLOSE_FINISH (sock);
2034 CLOSE_INVALIDATE (sock);
2040 /* If content-type is not given, assume text/html. This is because
2041 of the multitude of broken CGI's that "forget" to generate the
2044 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2045 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2050 if (opt.html_extension && (*dt & TEXTHTML))
2051 /* -E / --html-extension / html_extension = on was specified, and this is a
2052 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2053 already the file's suffix, tack on ".html". */
2055 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2057 if (last_period_in_local_filename == NULL
2058 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2059 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2061 int local_filename_len = strlen (hs->local_file);
2062 /* Resize the local file, allowing for ".html" preceded by
2063 optional ".NUMBER". */
2064 hs->local_file = xrealloc (hs->local_file,
2065 local_filename_len + 24 + sizeof (".html"));
2066 strcpy(hs->local_file + local_filename_len, ".html");
2067 /* If clobbering is not allowed and the file, as named,
2068 exists, tack on ".NUMBER.html" instead. */
2069 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2073 sprintf (hs->local_file + local_filename_len,
2074 ".%d.html", ext_num++);
2075 while (file_exists_p (hs->local_file));
2077 *dt |= ADDED_HTML_EXTENSION;
2081 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2083 /* If `-c' is in use and the file has been fully downloaded (or
2084 the remote file has shrunk), Wget effectively requests bytes
2085 after the end of file and the server response with 416. */
2086 logputs (LOG_VERBOSE, _("\
2087 \n The file is already fully retrieved; nothing to do.\n\n"));
2088 /* In case the caller inspects. */
2091 /* Mark as successfully retrieved. */
2094 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2095 might be more bytes in the body. */
2096 return RETRUNNEEDED;
2098 if ((contrange != 0 && contrange != hs->restval)
2099 || (H_PARTIAL (statcode) && !contrange))
2101 /* The Range request was somehow misunderstood by the server.
2104 CLOSE_INVALIDATE (sock);
2107 hs->contlen = contlen + contrange;
2113 /* No need to print this output if the body won't be
2114 downloaded at all, or if the original server response is
2116 logputs (LOG_VERBOSE, _("Length: "));
2119 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2120 if (contlen + contrange >= 1024)
2121 logprintf (LOG_VERBOSE, " (%s)",
2122 human_readable (contlen + contrange));
2125 if (contlen >= 1024)
2126 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2127 number_to_static_string (contlen),
2128 human_readable (contlen));
2130 logprintf (LOG_VERBOSE, _(", %s remaining"),
2131 number_to_static_string (contlen));
2135 logputs (LOG_VERBOSE,
2136 opt.ignore_length ? _("ignored") : _("unspecified"));
2138 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2140 logputs (LOG_VERBOSE, "\n");
2144 type = NULL; /* We don't need it any more. */
2146 /* Return if we have no intention of further downloading. */
2147 if (!(*dt & RETROKF) || head_only)
2149 /* In case the caller cares to look... */
2154 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2155 servers not to send body in response to a HEAD request, and
2156 those that do will likely be caught by test_socket_open.
2157 If not, they can be worked around using
2158 `--no-http-keep-alive'. */
2159 CLOSE_FINISH (sock);
2160 else if (keep_alive && skip_short_body (sock, contlen))
2161 /* Successfully skipped the body; also keep using the socket. */
2162 CLOSE_FINISH (sock);
2164 CLOSE_INVALIDATE (sock);
2165 return RETRFINISHED;
2168 /* Open the local file. */
2171 mkalldirs (hs->local_file);
2173 rotate_backups (hs->local_file);
2175 fp = fopen (hs->local_file, "ab");
2176 else if (ALLOW_CLOBBER)
2177 fp = fopen (hs->local_file, "wb");
2180 fp = fopen_excl (hs->local_file, true);
2181 if (!fp && errno == EEXIST)
2183 /* We cannot just invent a new name and use it (which is
2184 what functions like unique_create typically do)
2185 because we told the user we'd use this name.
2186 Instead, return and retry the download. */
2187 logprintf (LOG_NOTQUIET,
2188 _("%s has sprung into existence.\n"),
2190 CLOSE_INVALIDATE (sock);
2191 return FOPEN_EXCL_ERR;
2196 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2197 CLOSE_INVALIDATE (sock);
2204 /* Print fetch message, if opt.verbose. */
2207 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2208 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2211 /* This confuses the timestamping code that checks for file size.
2212 #### The timestamping code should be smarter about file size. */
2213 if (opt.save_headers && hs->restval == 0)
2214 fwrite (head, 1, strlen (head), fp);
2216 /* Now we no longer need to store the response header. */
2219 /* Download the request body. */
2222 /* If content-length is present, read that much; otherwise, read
2223 until EOF. The HTTP spec doesn't require the server to
2224 actually close the connection when it's done sending data. */
2225 flags |= rb_read_exactly;
2226 if (hs->restval > 0 && contrange == 0)
2227 /* If the server ignored our range request, instruct fd_read_body
2228 to skip the first RESTVAL bytes of body. */
2229 flags |= rb_skip_startpos;
2230 hs->len = hs->restval;
2232 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2233 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2237 CLOSE_FINISH (sock);
2241 hs->rderrmsg = xstrdup (fd_errstr (sock));
2242 CLOSE_INVALIDATE (sock);
2249 return RETRFINISHED;
2252 /* The genuine HTTP loop! This is the part where the retrieval is
2253 retried, and retried, and retried, and... */
2255 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2256 int *dt, struct url *proxy)
2259 bool got_head = false; /* used for time-stamping and filename detection */
2260 bool got_name = false;
2263 uerr_t err, ret = TRYLIMEXC;
2264 time_t tmr = -1; /* remote time-stamp */
2265 wgint local_size = 0; /* the size of the local file */
2266 struct http_stat hstat; /* HTTP status */
2269 /* Assert that no value for *LOCAL_FILE was passed. */
2270 assert (local_file == NULL || *local_file == NULL);
2272 /* Set LOCAL_FILE parameter. */
2273 if (local_file && opt.output_document)
2274 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2276 /* Reset NEWLOC parameter. */
2279 /* This used to be done in main(), but it's a better idea to do it
2280 here so that we don't go through the hoops if we're just using
2285 /* Warn on (likely bogus) wildcard usage in HTTP. */
2286 if (opt.ftp_glob && has_wildcards_p (u->path))
2287 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2289 /* Setup hstat struct. */
2291 hstat.referer = referer;
2293 if (opt.output_document)
2295 hstat.local_file = xstrdup (opt.output_document);
2299 /* Reset the counter. */
2302 /* Reset the document type. */
2308 /* Increment the pass counter. */
2310 sleep_between_retrievals (count);
2312 /* Get the current time string. */
2313 tms = time_str (time (NULL));
2315 if (opt.spider && !got_head)
2316 logprintf (LOG_VERBOSE, _("\
2317 Spider mode enabled. Check if remote file exists.\n"));
2319 /* Print fetch message, if opt.verbose. */
2322 char *hurl = url_string (u, true);
2327 sprintf (tmp, _("(try:%2d)"), count);
2328 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2333 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2338 ws_changetitle (hurl);
2343 /* Default document type is empty. However, if spider mode is
2344 on or time-stamping is employed, HEAD_ONLY commands is
2345 encoded within *dt. */
2346 if (((opt.spider || opt.timestamping) && !got_head) || !got_name)
2351 /* Decide whether or not to restart. */
2354 && stat (hstat.local_file, &st) == 0
2355 && S_ISREG (st.st_mode))
2356 /* When -c is used, continue from on-disk size. (Can't use
2357 hstat.len even if count>1 because we don't want a failed
2358 first attempt to clobber existing data.) */
2359 hstat.restval = st.st_size;
2361 /* otherwise, continue where the previous try left off */
2362 hstat.restval = hstat.len;
2366 /* Decide whether to send the no-cache directive. We send it in
2368 a) we're using a proxy, and we're past our first retrieval.
2369 Some proxies are notorious for caching incomplete data, so
2370 we require a fresh get.
2371 b) caching is explicitly inhibited. */
2372 if ((proxy && count > 1) /* a */
2373 || !opt.allow_cache) /* b */
2374 *dt |= SEND_NOCACHE;
2376 *dt &= ~SEND_NOCACHE;
2378 /* Try fetching the document, or at least its head. */
2379 err = gethttp (u, &hstat, dt, proxy);
2382 tms = time_str (time (NULL));
2384 /* Get the new location (with or without the redirection). */
2386 *newloc = xstrdup (hstat.newloc);
2390 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2391 case CONERROR: case READERR: case WRITEFAILED:
2392 case RANGEERR: case FOPEN_EXCL_ERR:
2393 /* Non-fatal errors continue executing the loop, which will
2394 bring them to "while" statement at the end, to judge
2395 whether the number of tries was exceeded. */
2396 printwhat (count, opt.ntry);
2398 case FWRITEERR: case FOPENERR:
2399 /* Another fatal error. */
2400 logputs (LOG_VERBOSE, "\n");
2401 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2402 hstat.local_file, strerror (errno));
2403 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2404 case SSLINITFAILED: case CONTNOTSUPPORTED:
2405 /* Fatal errors just return from the function. */
2409 /* Another fatal error. */
2410 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2414 /* Return the new location to the caller. */
2417 logprintf (LOG_NOTQUIET,
2418 _("ERROR: Redirection (%d) without location.\n"),
2428 /* The file was already fully retrieved. */
2432 /* Deal with you later. */
2435 /* All possibilities should have been exhausted. */
2439 if (!(*dt & RETROKF))
2444 /* #### Ugly ugly ugly! */
2445 hurl = url_string (u, true);
2446 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2448 /* Maybe we should always keep track of broken links, not just in
2452 /* #### Again: ugly ugly ugly! */
2454 hurl = url_string (u, true);
2455 nonexisting_url (hurl);
2456 logprintf (LOG_NOTQUIET, _("\
2457 Remote file does not exist -- broken link!!!\n"));
2461 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2462 tms, hstat.statcode, escnonprint (hstat.error));
2464 logputs (LOG_VERBOSE, "\n");
2470 /* Did we get the time-stamp? */
2473 bool restart_loop = false;
2475 if (opt.timestamping && !hstat.remote_time)
2477 logputs (LOG_NOTQUIET, _("\
2478 Last-modified header missing -- time-stamps turned off.\n"));
2480 else if (hstat.remote_time)
2482 /* Convert the date-string into struct tm. */
2483 tmr = http_atotm (hstat.remote_time);
2484 if (tmr == (time_t) (-1))
2485 logputs (LOG_VERBOSE, _("\
2486 Last-modified header invalid -- time-stamp ignored.\n"));
2489 /* The time-stamping section. */
2490 if (opt.timestamping)
2492 if (hstat.orig_file_name) /* Perform the following checks only
2493 if the file we're supposed to
2494 download already exists. */
2496 if (hstat.remote_time &&
2497 tmr != (time_t) (-1))
2499 /* Now time-stamping can be used validly. Time-stamping
2500 means that if the sizes of the local and remote file
2501 match, and local file is newer than the remote file,
2502 it will not be retrieved. Otherwise, the normal
2503 download procedure is resumed. */
2504 if (hstat.orig_file_tstamp >= tmr)
2506 if (hstat.contlen == -1
2507 || hstat.orig_file_size == hstat.contlen)
2509 logprintf (LOG_VERBOSE, _("\
2510 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2511 hstat.orig_file_name);
2517 logprintf (LOG_VERBOSE, _("\
2518 The sizes do not match (local %s) -- retrieving.\n"),
2519 number_to_static_string (local_size));
2523 logputs (LOG_VERBOSE,
2524 _("Remote file is newer, retrieving.\n"));
2526 logputs (LOG_VERBOSE, "\n");
2530 /* free_hstat (&hstat); */
2531 hstat.timestamp_checked = true;
2532 restart_loop = true;
2535 if (opt.always_rest)
2538 restart_loop = true;
2547 logputs (LOG_VERBOSE, _("\
2548 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2549 restart_loop = true;
2553 logprintf (LOG_VERBOSE, _("\
2554 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2561 logprintf (LOG_VERBOSE, _("\
2562 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2568 got_head = true; /* no more time-stamping */
2570 count = 0; /* the retrieve count for HEAD is reset */
2576 if ((tmr != (time_t) (-1))
2577 && ((hstat.len == hstat.contlen) ||
2578 ((hstat.res == 0) && (hstat.contlen == -1))))
2580 /* #### This code repeats in http.c and ftp.c. Move it to a
2582 const char *fl = NULL;
2583 if (opt.output_document)
2585 if (output_stream_regular)
2586 fl = opt.output_document;
2589 fl = hstat.local_file;
2593 /* End of time-stamping section. */
2595 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2596 total_download_time += hstat.dltime;
2598 if (hstat.len == hstat.contlen)
2602 logprintf (LOG_VERBOSE,
2603 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2604 tms, tmrate, hstat.local_file,
2605 number_to_static_string (hstat.len),
2606 number_to_static_string (hstat.contlen));
2607 logprintf (LOG_NONVERBOSE,
2608 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2610 number_to_static_string (hstat.len),
2611 number_to_static_string (hstat.contlen),
2612 hstat.local_file, count);
2615 total_downloaded_bytes += hstat.len;
2617 /* Remember that we downloaded the file for later ".orig" code. */
2618 if (*dt & ADDED_HTML_EXTENSION)
2619 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2621 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2626 else if (hstat.res == 0) /* No read error */
2628 if (hstat.contlen == -1) /* We don't know how much we were supposed
2629 to get, so assume we succeeded. */
2633 logprintf (LOG_VERBOSE,
2634 _("%s (%s) - `%s' saved [%s]\n\n"),
2635 tms, tmrate, hstat.local_file,
2636 number_to_static_string (hstat.len));
2637 logprintf (LOG_NONVERBOSE,
2638 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2639 tms, u->url, number_to_static_string (hstat.len),
2640 hstat.local_file, count);
2643 total_downloaded_bytes += hstat.len;
2645 /* Remember that we downloaded the file for later ".orig" code. */
2646 if (*dt & ADDED_HTML_EXTENSION)
2647 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2649 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2654 else if (hstat.len < hstat.contlen) /* meaning we lost the
2655 connection too soon */
2657 logprintf (LOG_VERBOSE,
2658 _("%s (%s) - Connection closed at byte %s. "),
2659 tms, tmrate, number_to_static_string (hstat.len));
2660 printwhat (count, opt.ntry);
2664 /* Getting here would mean reading more data than
2665 requested with content-length, which we never do. */
2668 else /* from now on hstat.res can only be -1 */
2670 if (hstat.contlen == -1)
2672 logprintf (LOG_VERBOSE,
2673 _("%s (%s) - Read error at byte %s (%s)."),
2674 tms, tmrate, number_to_static_string (hstat.len),
2676 printwhat (count, opt.ntry);
2679 else /* hstat.res == -1 and contlen is given */
2681 logprintf (LOG_VERBOSE,
2682 _("%s (%s) - Read error at byte %s/%s (%s). "),
2684 number_to_static_string (hstat.len),
2685 number_to_static_string (hstat.contlen),
2687 printwhat (count, opt.ntry);
2693 while (!opt.ntry || (count < opt.ntry));
2697 *local_file = xstrdup (hstat.local_file);
2698 free_hstat (&hstat);
2703 /* Check whether the result of strptime() indicates success.
2704 strptime() returns the pointer to how far it got to in the string.
2705 The processing has been successful if the string is at `GMT' or
2706 `+X', or at the end of the string.
2708 In extended regexp parlance, the function returns 1 if P matches
2709 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2710 can return) is considered a failure and 0 is returned. */
2712 check_end (const char *p)
2716 while (ISSPACE (*p))
2719 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2720 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2726 /* Convert the textual specification of time in TIME_STRING to the
2727 number of seconds since the Epoch.
2729 TIME_STRING can be in any of the three formats RFC2616 allows the
2730 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2731 as well as the time format used in the Set-Cookie header.
2732 Timezones are ignored, and should be GMT.
2734 Return the computed time_t representation, or -1 if the conversion
2737 This function uses strptime with various string formats for parsing
2738 TIME_STRING. This results in a parser that is not as lenient in
2739 interpreting TIME_STRING as I would like it to be. Being based on
2740 strptime, it always allows shortened months, one-digit days, etc.,
2741 but due to the multitude of formats in which time can be
2742 represented, an ideal HTTP time parser would be even more
2743 forgiving. It should completely ignore things like week days and
2744 concentrate only on the various forms of representing years,
2745 months, days, hours, minutes, and seconds. For example, it would
2746 be nice if it accepted ISO 8601 out of the box.
2748 I've investigated free and PD code for this purpose, but none was
2749 usable. getdate was big and unwieldy, and had potential copyright
2750 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2751 distributed with phttpd, is excellent, but we cannot use it because
2752 it is not assigned to the FSF. So I stuck it with strptime. */
2755 http_atotm (const char *time_string)
2757 /* NOTE: Solaris strptime man page claims that %n and %t match white
2758 space, but that's not universally available. Instead, we simply
2759 use ` ' to mean "skip all WS", which works under all strptime
2760 implementations I've tested. */
2762 static const char *time_formats[] = {
2763 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2764 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2765 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2766 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2767 (used in Set-Cookie, defined in the
2768 Netscape cookie specification.) */
2770 const char *oldlocale;
2772 time_t ret = (time_t) -1;
2774 /* Solaris strptime fails to recognize English month names in
2775 non-English locales, which we work around by temporarily setting
2776 locale to C before invoking strptime. */
2777 oldlocale = setlocale (LC_TIME, NULL);
2778 setlocale (LC_TIME, "C");
2780 for (i = 0; i < countof (time_formats); i++)
2784 /* Some versions of strptime use the existing contents of struct
2785 tm to recalculate the date according to format. Zero it out
2786 to prevent stack garbage from influencing strptime. */
2789 if (check_end (strptime (time_string, time_formats[i], &t)))
2796 /* Restore the previous locale. */
2797 setlocale (LC_TIME, oldlocale);
2802 /* Authorization support: We support three authorization schemes:
2804 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2806 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2807 consisting of answering to the server's challenge with the proper
2810 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2811 Stenberg for libcurl. Like digest, NTLM is based on a
2812 challenge-response mechanism, but unlike digest, it is non-standard
2813 (authenticates TCP connections rather than requests), undocumented
2814 and Microsoft-specific. */
2816 /* Create the authentication header contents for the `Basic' scheme.
2817 This is done by encoding the string "USER:PASS" to base64 and
2818 prepending the string "Basic " in front of it. */
2821 basic_authentication_encode (const char *user, const char *passwd)
2824 int len1 = strlen (user) + 1 + strlen (passwd);
2826 t1 = (char *)alloca (len1 + 1);
2827 sprintf (t1, "%s:%s", user, passwd);
2829 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2830 base64_encode (t1, len1, t2);
2832 return concat_strings ("Basic ", t2, (char *) 0);
2835 #define SKIP_WS(x) do { \
2836 while (ISSPACE (*(x))) \
2840 #ifdef ENABLE_DIGEST
2841 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2842 an array of 16 bytes containing the hash keys, and BUF should be a
2843 buffer of 33 writable characters (32 for hex digits plus one for
2844 zero termination). */
2846 dump_hash (char *buf, const unsigned char *hash)
2850 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2852 *buf++ = XNUM_TO_digit (*hash >> 4);
2853 *buf++ = XNUM_TO_digit (*hash & 0xf);
2858 /* Take the line apart to find the challenge, and compose a digest
2859 authorization header. See RFC2069 section 2.1.2. */
2861 digest_authentication_encode (const char *au, const char *user,
2862 const char *passwd, const char *method,
2865 static char *realm, *opaque, *nonce;
2870 { "realm", &realm },
2871 { "opaque", &opaque },
2875 param_token name, value;
2877 realm = opaque = nonce = NULL;
2879 au += 6; /* skip over `Digest' */
2880 while (extract_param (&au, &name, &value, ','))
2883 for (i = 0; i < countof (options); i++)
2884 if (name.e - name.b == strlen (options[i].name)
2885 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2887 *options[i].variable = strdupdelim (value.b, value.e);
2891 if (!realm || !nonce || !user || !passwd || !path || !method)
2894 xfree_null (opaque);
2899 /* Calculate the digest value. */
2901 ALLOCA_MD5_CONTEXT (ctx);
2902 unsigned char hash[MD5_HASHLEN];
2903 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2904 char response_digest[MD5_HASHLEN * 2 + 1];
2906 /* A1BUF = H(user ":" realm ":" password) */
2908 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2909 gen_md5_update ((unsigned char *)":", 1, ctx);
2910 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2911 gen_md5_update ((unsigned char *)":", 1, ctx);
2912 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2913 gen_md5_finish (ctx, hash);
2914 dump_hash (a1buf, hash);
2916 /* A2BUF = H(method ":" path) */
2918 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2919 gen_md5_update ((unsigned char *)":", 1, ctx);
2920 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2921 gen_md5_finish (ctx, hash);
2922 dump_hash (a2buf, hash);
2924 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2926 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2927 gen_md5_update ((unsigned char *)":", 1, ctx);
2928 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2929 gen_md5_update ((unsigned char *)":", 1, ctx);
2930 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2931 gen_md5_finish (ctx, hash);
2932 dump_hash (response_digest, hash);
2934 res = xmalloc (strlen (user)
2939 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2940 + (opaque ? strlen (opaque) : 0)
2942 sprintf (res, "Digest \
2943 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2944 user, realm, nonce, path, response_digest);
2947 char *p = res + strlen (res);
2948 strcat (p, ", opaque=\"");
2955 #endif /* ENABLE_DIGEST */
2957 /* Computing the size of a string literal must take into account that
2958 value returned by sizeof includes the terminating \0. */
2959 #define STRSIZE(literal) (sizeof (literal) - 1)
2961 /* Whether chars in [b, e) begin with the literal string provided as
2962 first argument and are followed by whitespace or terminating \0.
2963 The comparison is case-insensitive. */
2964 #define STARTS(literal, b, e) \
2965 ((e) - (b) >= STRSIZE (literal) \
2966 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2967 && ((e) - (b) == STRSIZE (literal) \
2968 || ISSPACE (b[STRSIZE (literal)])))
2971 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2973 return STARTS ("Basic", hdrbeg, hdrend)
2974 #ifdef ENABLE_DIGEST
2975 || STARTS ("Digest", hdrbeg, hdrend)
2978 || STARTS ("NTLM", hdrbeg, hdrend)
2985 /* Create the HTTP authorization request header. When the
2986 `WWW-Authenticate' response header is seen, according to the
2987 authorization scheme specified in that header (`Basic' and `Digest'
2988 are supported by the current implementation), produce an
2989 appropriate HTTP authorization request header. */
2991 create_authorization_line (const char *au, const char *user,
2992 const char *passwd, const char *method,
2993 const char *path, bool *finished)
2995 /* We are called only with known schemes, so we can dispatch on the
2997 switch (TOUPPER (*au))
2999 case 'B': /* Basic */
3001 return basic_authentication_encode (user, passwd);
3002 #ifdef ENABLE_DIGEST
3003 case 'D': /* Digest */
3005 return digest_authentication_encode (au, user, passwd, method, path);
3008 case 'N': /* NTLM */
3009 if (!ntlm_input (&pconn.ntlm, au))
3014 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3017 /* We shouldn't get here -- this function should be only called
3018 with values approved by known_authentication_scheme_p. */
3026 if (!wget_cookie_jar)
3027 wget_cookie_jar = cookie_jar_new ();
3028 if (opt.cookies_input && !cookies_loaded_p)
3030 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3031 cookies_loaded_p = true;
3038 if (wget_cookie_jar)
3039 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3045 xfree_null (pconn.host);
3046 if (wget_cookie_jar)
3047 cookie_jar_delete (wget_cookie_jar);
3054 test_parse_content_disposition()
3059 char *opt_dir_prefix;
3063 { "filename=\"file.ext\"", NULL, "file.ext", true },
3064 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3065 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3066 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3067 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3068 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3069 { "attachment", NULL, NULL, false },
3070 { "attachment", "somedir", NULL, false },
3073 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3078 opt.dir_prefix = test_array[i].opt_dir_prefix;
3079 res = parse_content_disposition (test_array[i].hdrval, &filename);
3081 mu_assert ("test_parse_content_disposition: wrong result",
3082 res == test_array[i].result
3084 || 0 == strcmp (test_array[i].filename, filename)));
3090 #endif /* TESTING */