2 Copyright (C) 2003 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
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, 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. */
34 #include <sys/types.h>
45 #if TIME_WITH_SYS_TIME
46 # include <sys/time.h>
50 # include <sys/time.h>
67 # include "gen_sslfunc.h"
75 extern char *version_string;
76 extern LARGE_INT total_downloaded_bytes;
78 extern FILE *output_stream;
79 extern int output_stream_regular;
82 # define MIN(x, y) ((x) > (y) ? (y) : (x))
86 static int cookies_loaded_p;
87 struct cookie_jar *wget_cookie_jar;
89 #define TEXTHTML_S "text/html"
90 #define TEXTXHTML_S "application/xhtml+xml"
92 /* Some status code validation macros: */
93 #define H_20X(x) (((x) >= 200) && ((x) < 300))
94 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
95 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
96 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
97 || (x) == HTTP_STATUS_SEE_OTHER \
98 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
100 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
101 /* Successful 2xx. */
102 #define HTTP_STATUS_OK 200
103 #define HTTP_STATUS_CREATED 201
104 #define HTTP_STATUS_ACCEPTED 202
105 #define HTTP_STATUS_NO_CONTENT 204
106 #define HTTP_STATUS_PARTIAL_CONTENTS 206
108 /* Redirection 3xx. */
109 #define HTTP_STATUS_MULTIPLE_CHOICES 300
110 #define HTTP_STATUS_MOVED_PERMANENTLY 301
111 #define HTTP_STATUS_MOVED_TEMPORARILY 302
112 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
113 #define HTTP_STATUS_NOT_MODIFIED 304
114 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
116 /* Client error 4xx. */
117 #define HTTP_STATUS_BAD_REQUEST 400
118 #define HTTP_STATUS_UNAUTHORIZED 401
119 #define HTTP_STATUS_FORBIDDEN 403
120 #define HTTP_STATUS_NOT_FOUND 404
121 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
123 /* Server errors 5xx. */
124 #define HTTP_STATUS_INTERNAL 500
125 #define HTTP_STATUS_NOT_IMPLEMENTED 501
126 #define HTTP_STATUS_BAD_GATEWAY 502
127 #define HTTP_STATUS_UNAVAILABLE 503
130 rel_none, rel_name, rel_value, rel_both
137 struct request_header {
139 enum rp release_policy;
141 int hcount, hcapacity;
144 /* Create a new, empty request. At least request_set_method must be
145 called before the request can be used. */
147 static struct request *
150 struct request *req = xnew0 (struct request);
152 req->headers = xnew_array (struct request_header, req->hcapacity);
156 /* Set the request's method and its arguments. METH should be a
157 literal string (or it should outlive the request) because it will
158 not be freed. ARG will be freed by request_free. */
161 request_set_method (struct request *req, const char *meth, char *arg)
167 /* Return the method string passed with the last call to
168 request_set_method. */
171 request_method (const struct request *req)
176 /* Free one header according to the release policy specified with
177 request_set_header. */
180 release_header (struct request_header *hdr)
182 switch (hdr->release_policy)
199 /* Set the request named NAME to VALUE. Specifically, this means that
200 a "NAME: VALUE\r\n" header line will be used in the request. If a
201 header with the same name previously existed in the request, its
202 value will be replaced by this one.
204 RELEASE_POLICY determines whether NAME and VALUE should be released
205 (freed) with request_free. Allowed values are:
207 - rel_none - don't free NAME or VALUE
208 - rel_name - free NAME when done
209 - rel_value - free VALUE when done
210 - rel_both - free both NAME and VALUE when done
212 Setting release policy is useful when arguments come from different
213 sources. For example:
215 // Don't free literal strings!
216 request_set_header (req, "Pragma", "no-cache", rel_none);
218 // Don't free a global variable, we'll need it later.
219 request_set_header (req, "Referer", opt.referer, rel_none);
221 // Value freshly allocated, free it when done.
222 request_set_header (req, "Range", aprintf ("bytes=%ld-", hs->restval),
227 request_set_header (struct request *req, char *name, char *value,
228 enum rp release_policy)
230 struct request_header *hdr;
234 for (i = 0; i < req->hcount; i++)
236 hdr = &req->headers[i];
237 if (0 == strcasecmp (name, hdr->name))
239 /* Replace existing header. */
240 release_header (hdr);
243 hdr->release_policy = release_policy;
248 /* Install new header. */
250 if (req->hcount >= req->hcount)
252 req->hcapacity <<= 1;
253 req->headers = xrealloc (req->headers,
254 req->hcapacity * sizeof (struct request_header));
256 hdr = &req->headers[req->hcount++];
259 hdr->release_policy = release_policy;
262 /* Like request_set_header, but sets the whole header line, as
263 provided by the user using the `--header' option. For example,
264 request_set_user_header (req, "Foo: bar") works just like
265 request_set_header (req, "Foo", "bar"). */
268 request_set_user_header (struct request *req, const char *header)
271 const char *p = strchr (header, ':');
274 BOUNDED_TO_ALLOCA (header, p, name);
278 request_set_header (req, xstrdup (name), (char *) p, rel_name);
281 #define APPEND(p, str) do { \
282 int A_len = strlen (str); \
283 memcpy (p, str, A_len); \
287 /* Construct the request and write it to FD using fd_write. */
290 request_send (const struct request *req, int fd)
292 char *request_string, *p;
293 int i, size, write_error;
295 /* Count the request size. */
298 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
299 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
301 for (i = 0; i < req->hcount; i++)
303 struct request_header *hdr = &req->headers[i];
304 /* NAME ": " VALUE "\r\n" */
305 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
311 p = request_string = alloca_array (char, size);
313 /* Generate the request. */
315 APPEND (p, req->method); *p++ = ' ';
316 APPEND (p, req->arg); *p++ = ' ';
317 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
319 for (i = 0; i < req->hcount; i++)
321 struct request_header *hdr = &req->headers[i];
322 APPEND (p, hdr->name);
323 *p++ = ':', *p++ = ' ';
324 APPEND (p, hdr->value);
325 *p++ = '\r', *p++ = '\n';
328 *p++ = '\r', *p++ = '\n', *p++ = '\0';
329 assert (p - request_string == size);
333 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
335 /* Send the request to the server. */
337 write_error = fd_write (fd, request_string, size - 1, -1);
339 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
344 /* Release the resources used by REQ. */
347 request_free (struct request *req)
350 xfree_null (req->arg);
351 for (i = 0; i < req->hcount; i++)
352 release_header (&req->headers[i]);
353 xfree_null (req->headers);
357 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
358 PROMISED_SIZE bytes are sent over the wire -- if the file is
359 longer, read only that much; if the file is shorter, report an error. */
362 post_file (int sock, const char *file_name, long promised_size)
364 static char chunk[8192];
369 DEBUGP (("[writing POST file %s ... ", file_name));
371 fp = fopen (file_name, "rb");
374 while (!feof (fp) && written < promised_size)
377 int length = fread (chunk, 1, sizeof (chunk), fp);
380 towrite = MIN (promised_size - written, length);
381 write_error = fd_write (sock, chunk, towrite, -1);
391 /* If we've written less than was promised, report a (probably
392 nonsensical) error rather than break the promise. */
393 if (written < promised_size)
399 assert (written == promised_size);
400 DEBUGP (("done]\n"));
405 head_terminator (const char *hunk, int oldlen, int peeklen)
407 const char *start, *end;
409 /* If at first peek, verify whether HUNK starts with "HTTP". If
410 not, this is a HTTP/0.9 request and we must bail out without
412 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
418 start = hunk + oldlen - 4;
419 end = hunk + oldlen + peeklen;
421 for (; start < end - 1; start++)
428 if (start[1] == '\n')
434 /* Read the HTTP request head from FD and return it. The error
435 conditions are the same as with fd_read_hunk.
437 To support HTTP/0.9 responses, this function tries to make sure
438 that the data begins with "HTTP". If this is not the case, no data
439 is read and an empty request is returned, so that the remaining
440 data can be treated as body. */
443 fd_read_http_head (int fd)
445 return fd_read_hunk (fd, head_terminator, 512);
449 /* The response data. */
452 /* The array of pointers that indicate where each header starts.
453 For example, given this HTTP response:
460 The headers are located like this:
462 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
464 headers[0] headers[1] headers[2] headers[3]
466 I.e. headers[0] points to the beginning of the request,
467 headers[1] points to the end of the first header and the
468 beginning of the second one, etc. */
470 const char **headers;
473 /* Create a new response object from the text of the HTTP response,
474 available in HEAD. That text is automatically split into
475 constituent header lines for fast retrieval using
476 response_header_*. */
478 static struct response *
479 response_new (const char *head)
484 struct response *resp = xnew0 (struct response);
489 /* Empty head means that we're dealing with a headerless
490 (HTTP/0.9) response. In that case, don't set HEADERS at
495 /* Split HEAD into header lines, so that response_header_* functions
496 don't need to do this over and over again. */
502 DO_REALLOC (resp->headers, size, count + 1, const char *);
503 resp->headers[count++] = hdr;
505 /* Break upon encountering an empty line. */
506 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
509 /* Find the end of HDR, including continuations. */
512 const char *end = strchr (hdr, '\n');
518 while (*hdr == ' ' || *hdr == '\t');
520 DO_REALLOC (resp->headers, size, count + 1, const char *);
521 resp->headers[count++] = NULL;
526 /* Locate the header named NAME in the request data. If found, set
527 *BEGPTR to its starting, and *ENDPTR to its ending position, and
528 return 1. Otherwise return 0.
530 This function is used as a building block for response_header_copy
531 and response_header_strdup. */
534 response_header_bounds (const struct response *resp, const char *name,
535 const char **begptr, const char **endptr)
538 const char **headers = resp->headers;
541 if (!headers || !headers[1])
544 name_len = strlen (name);
546 for (i = 1; headers[i + 1]; i++)
548 const char *b = headers[i];
549 const char *e = headers[i + 1];
551 && b[name_len] == ':'
552 && 0 == strncasecmp (b, name, name_len))
555 while (b < e && ISSPACE (*b))
557 while (b < e && ISSPACE (e[-1]))
567 /* Copy the response header named NAME to buffer BUF, no longer than
568 BUFSIZE (BUFSIZE includes the terminating 0). If the header
569 exists, 1 is returned, otherwise 0. If there should be no limit on
570 the size of the header, use response_header_strdup instead.
572 If BUFSIZE is 0, no data is copied, but the boolean indication of
573 whether the header is present is still returned. */
576 response_header_copy (const struct response *resp, const char *name,
577 char *buf, int bufsize)
580 if (!response_header_bounds (resp, name, &b, &e))
584 int len = MIN (e - b, bufsize - 1);
585 memcpy (buf, b, len);
591 /* Return the value of header named NAME in RESP, allocated with
592 malloc. If such a header does not exist in RESP, return NULL. */
595 response_header_strdup (const struct response *resp, const char *name)
598 if (!response_header_bounds (resp, name, &b, &e))
600 return strdupdelim (b, e);
603 /* Parse the HTTP status line, which is of format:
605 HTTP-Version SP Status-Code SP Reason-Phrase
607 The function returns the status-code, or -1 if the status line
608 appears malformed. The pointer to "reason-phrase" message is
609 returned in *MESSAGE. */
612 response_status (const struct response *resp, char **message)
619 /* For a HTTP/0.9 response, assume status 200. */
621 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
625 p = resp->headers[0];
626 end = resp->headers[1];
632 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
636 /* Match the HTTP version. This is optional because Gnutella
637 servers have been reported to not specify HTTP version. */
638 if (p < end && *p == '/')
641 while (p < end && ISDIGIT (*p))
643 if (p < end && *p == '.')
645 while (p < end && ISDIGIT (*p))
649 while (p < end && ISSPACE (*p))
651 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
654 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
659 while (p < end && ISSPACE (*p))
661 while (p < end && ISSPACE (end[-1]))
663 *message = strdupdelim (p, end);
669 /* Release the resources used by RESP. */
672 response_free (struct response *resp)
674 xfree_null (resp->headers);
678 /* Print [b, e) to the log, omitting the trailing CRLF. */
681 print_server_response_1 (const char *prefix, const char *b, const char *e)
684 if (b < e && e[-1] == '\n')
686 if (b < e && e[-1] == '\r')
688 BOUNDED_TO_ALLOCA (b, e, ln);
689 logprintf (LOG_VERBOSE, "%s%s\n", prefix, ln);
692 /* Print the server response, line by line, omitting the trailing CR
693 characters, prefixed with PREFIX. */
696 print_server_response (const struct response *resp, const char *prefix)
701 for (i = 0; resp->headers[i + 1]; i++)
702 print_server_response_1 (prefix, resp->headers[i], resp->headers[i + 1]);
705 /* Parse the `Content-Range' header and extract the information it
706 contains. Returns 1 if successful, -1 otherwise. */
708 parse_content_range (const char *hdr, long *first_byte_ptr,
709 long *last_byte_ptr, long *entity_length_ptr)
713 /* Ancient versions of Netscape proxy server, presumably predating
714 rfc2068, sent out `Content-Range' without the "bytes"
716 if (!strncasecmp (hdr, "bytes", 5))
719 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
723 while (ISSPACE (*hdr))
730 for (num = 0; ISDIGIT (*hdr); hdr++)
731 num = 10 * num + (*hdr - '0');
732 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
734 *first_byte_ptr = num;
736 for (num = 0; ISDIGIT (*hdr); hdr++)
737 num = 10 * num + (*hdr - '0');
738 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
740 *last_byte_ptr = num;
742 for (num = 0; ISDIGIT (*hdr); hdr++)
743 num = 10 * num + (*hdr - '0');
744 *entity_length_ptr = num;
748 /* Read the body of the request, but don't store it anywhere and don't
749 display a progress gauge. This is useful for reading the error
750 responses whose bodies don't need to be displayed or logged, but
751 which need to be read anyway. */
754 skip_short_body (int fd, long contlen)
756 /* Skipping the body doesn't make sense if the content length is
757 unknown because, in that case, persistent connections cannot be
758 used. (#### This is not the case with HTTP/1.1 where they can
759 still be used with the magic of the "chunked" transfer!) */
762 DEBUGP (("Skipping %ld bytes of body data... ", contlen));
767 int ret = fd_read (fd, dlbuf, MIN (contlen, sizeof (dlbuf)), -1);
772 DEBUGP (("done.\n"));
775 /* Persistent connections. Currently, we cache the most recently used
776 connection as persistent, provided that the HTTP server agrees to
777 make it such. The persistence data is stored in the variables
778 below. Ideally, it should be possible to cache an arbitrary fixed
779 number of these connections. */
781 /* Whether a persistent connection is active. */
782 static int pconn_active;
785 /* The socket of the connection. */
788 /* Host and port of the currently active persistent connection. */
792 /* Whether a ssl handshake has occoured on this connection. */
796 /* Mark the persistent connection as invalid and free the resources it
797 uses. This is used by the CLOSE_* macros after they forcefully
798 close a registered persistent connection. */
801 invalidate_persistent (void)
803 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
805 fd_close (pconn.socket);
810 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
811 persistent. This will enable someone to use the same connection
812 later. In the context of HTTP, this must be called only AFTER the
813 response has been received and the server has promised that the
814 connection will remain alive.
816 If a previous connection was persistent, it is closed. */
819 register_persistent (const char *host, int port, int fd, int ssl)
823 if (pconn.socket == fd)
825 /* The connection FD is already registered. */
830 /* The old persistent connection is still active; close it
831 first. This situation arises whenever a persistent
832 connection exists, but we then connect to a different
833 host, and try to register a persistent connection to that
835 invalidate_persistent ();
841 pconn.host = xstrdup (host);
845 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
848 /* Return non-zero if a persistent connection is available for
849 connecting to HOST:PORT. */
852 persistent_available_p (const char *host, int port, int ssl,
853 int *host_lookup_failed)
855 /* First, check whether a persistent connection is active at all. */
859 /* If we want SSL and the last connection wasn't or vice versa,
860 don't use it. Checking for host and port is not enough because
861 HTTP and HTTPS can apparently coexist on the same port. */
862 if (ssl != pconn.ssl)
865 /* If we're not connecting to the same port, we're not interested. */
866 if (port != pconn.port)
869 /* If the host is the same, we're in business. If not, there is
870 still hope -- read below. */
871 if (0 != strcasecmp (host, pconn.host))
873 /* If pconn.socket is already talking to HOST, we needn't
874 reconnect. This happens often when both sites are virtual
875 hosts distinguished only by name and served by the same
876 network interface, and hence the same web server (possibly
877 set up by the ISP and serving many different web sites).
878 This admittedly non-standard optimization does not contradict
879 HTTP and works well with popular server software. */
883 struct address_list *al;
886 /* Don't try to talk to two different SSL sites over the same
887 secure connection! (Besides, it's not clear if name-based
888 virtual hosting is even possible with SSL.) */
891 /* If pconn.socket's peer is one of the IP addresses HOST
892 resolves to, pconn.socket is for all intents and purposes
893 already talking to HOST. */
895 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
897 /* Can't get the peer's address -- something must be very
898 wrong with the connection. */
899 invalidate_persistent ();
902 al = lookup_host (host, 0);
905 *host_lookup_failed = 1;
909 found = address_list_contains (al, &ip);
910 address_list_release (al);
915 /* The persistent connection's peer address was found among the
916 addresses HOST resolved to; therefore, pconn.sock is in fact
917 already talking to HOST -- no need to reconnect. */
920 /* Finally, check whether the connection is still open. This is
921 important because most server implement a liberal (short) timeout
922 on persistent connections. Wget can of course always reconnect
923 if the connection doesn't work out, but it's nicer to know in
924 advance. This test is a logical followup of the first test, but
925 is "expensive" and therefore placed at the end of the list. */
927 if (!test_socket_open (pconn.socket))
929 /* Oops, the socket is no longer open. Now that we know that,
930 let's invalidate the persistent connection before returning
932 invalidate_persistent ();
939 /* The idea behind these two CLOSE macros is to distinguish between
940 two cases: one when the job we've been doing is finished, and we
941 want to close the connection and leave, and two when something is
942 seriously wrong and we're closing the connection as part of
945 In case of keep_alive, CLOSE_FINISH should leave the connection
946 open, while CLOSE_INVALIDATE should still close it.
948 Note that the semantics of the flag `keep_alive' is "this
949 connection *will* be reused (the server has promised not to close
950 the connection once we're done)", while the semantics of
951 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
952 active, registered connection". */
954 #define CLOSE_FINISH(fd) do { \
957 if (pconn_active && (fd) == pconn.socket) \
958 invalidate_persistent (); \
967 #define CLOSE_INVALIDATE(fd) do { \
968 if (pconn_active && (fd) == pconn.socket) \
969 invalidate_persistent (); \
977 long len; /* received length */
978 long contlen; /* expected length */
979 long restval; /* the restart value */
980 int res; /* the result of last read */
981 char *newloc; /* new location (redirection) */
982 char *remote_time; /* remote time-stamp string */
983 char *error; /* textual HTTP error */
984 int statcode; /* status code */
985 long rd_size; /* amount of data read from socket */
986 double dltime; /* time it took to download the data */
987 const char *referer; /* value of the referer header. */
988 char **local_file; /* local file. */
992 free_hstat (struct http_stat *hs)
994 xfree_null (hs->newloc);
995 xfree_null (hs->remote_time);
996 xfree_null (hs->error);
998 /* Guard against being called twice. */
1000 hs->remote_time = NULL;
1004 static char *create_authorization_line PARAMS ((const char *, const char *,
1005 const char *, const char *,
1007 static char *basic_authentication_encode PARAMS ((const char *, const char *));
1008 static int known_authentication_scheme_p PARAMS ((const char *));
1010 time_t http_atotm PARAMS ((const char *));
1012 #define BEGINS_WITH(line, string_constant) \
1013 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1014 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1015 || !line[sizeof (string_constant) - 1]))
1017 /* Retrieve a document through HTTP protocol. It recognizes status
1018 code, and correctly handles redirections. It closes the network
1019 socket. If it receives an error from the functions below it, it
1020 will print it if there is enough information to do so (almost
1021 always), returning the error to the caller (i.e. http_loop).
1023 Various HTTP parameters are stored to hs.
1025 If PROXY is non-NULL, the connection will be made to the proxy
1026 server, and u->url will be requested. */
1028 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1030 struct request *req;
1033 char *user, *passwd;
1037 long contlen, contrange;
1044 /* Whether authorization has been already tried. */
1045 int auth_tried_already = 0;
1047 /* Whether our connection to the remote host is through SSL. */
1051 struct response *resp;
1055 /* Whether this connection will be kept alive after the HTTP request
1059 /* Whether keep-alive should be inhibited. */
1060 int inhibit_keep_alive = !opt.http_keep_alive || opt.ignore_length;
1062 /* Headers sent when using POST. */
1063 long post_data_size = 0;
1065 int host_lookup_failed = 0;
1068 if (u->scheme == SCHEME_HTTPS)
1070 /* Initialize the SSL context. After this has once been done,
1071 it becomes a no-op. */
1072 switch (ssl_init ())
1074 case SSLERRCTXCREATE:
1076 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
1077 return SSLERRCTXCREATE;
1078 case SSLERRCERTFILE:
1079 /* try without certfile */
1080 logprintf (LOG_NOTQUIET,
1081 _("Failed to load certificates from %s\n"),
1083 logprintf (LOG_NOTQUIET,
1084 _("Trying without the specified certificate\n"));
1087 logprintf (LOG_NOTQUIET,
1088 _("Failed to get certificate key from %s\n"),
1090 logprintf (LOG_NOTQUIET,
1091 _("Trying without the specified certificate\n"));
1097 #endif /* HAVE_SSL */
1099 if (!(*dt & HEAD_ONLY))
1100 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1101 know the local filename so we can save to it. */
1102 assert (*hs->local_file != NULL);
1104 auth_tried_already = 0;
1106 /* Initialize certain elements of struct http_stat. */
1111 hs->remote_time = NULL;
1116 /* Prepare the request to send. */
1118 req = request_new ();
1120 const char *meth = "GET";
1121 if (*dt & HEAD_ONLY)
1123 else if (opt.post_file_name || opt.post_data)
1125 /* Use the full path, i.e. one that includes the leading slash and
1126 the query string. E.g. if u->path is "foo/bar" and u->query is
1127 "param=value", full_path will be "/foo/bar?param=value". */
1128 request_set_method (req, meth,
1129 proxy ? xstrdup (u->url) : url_full_path (u));
1132 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1133 if (*dt & SEND_NOCACHE)
1134 request_set_header (req, "Pragma", "no-cache", rel_none);
1136 request_set_header (req, "Range",
1137 aprintf ("bytes=%ld-", hs->restval), rel_value);
1139 request_set_header (req, "User-Agent", opt.useragent, rel_none);
1141 request_set_header (req, "User-Agent",
1142 aprintf ("Wget/%s", version_string), rel_value);
1143 request_set_header (req, "Accept", "*/*", rel_none);
1145 /* Find the username and password for authentication. */
1148 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1149 user = user ? user : opt.http_user;
1150 passwd = passwd ? passwd : opt.http_passwd;
1154 /* We have the username and the password, but haven't tried
1155 any authorization yet. Let's see if the "Basic" method
1156 works. If not, we'll come back here and construct a
1157 proper authorization method with the right challenges.
1159 If we didn't employ this kind of logic, every URL that
1160 requires authorization would have to be processed twice,
1161 which is very suboptimal and generates a bunch of false
1162 "unauthorized" errors in the server log.
1164 #### But this logic also has a serious problem when used
1165 with stronger authentications: we *first* transmit the
1166 username and the password in clear text, and *then* attempt a
1167 stronger authentication scheme. That cannot be right! We
1168 are only fortunate that almost everyone still uses the
1169 `Basic' scheme anyway.
1171 There should be an option to prevent this from happening, for
1172 those who use strong authentication schemes and value their
1174 request_set_header (req, "Authorization",
1175 basic_authentication_encode (user, passwd),
1182 char *proxy_user, *proxy_passwd;
1183 /* For normal username and password, URL components override
1184 command-line/wgetrc parameters. With proxy
1185 authentication, it's the reverse, because proxy URLs are
1186 normally the "permanent" ones, so command-line args
1187 should take precedence. */
1188 if (opt.proxy_user && opt.proxy_passwd)
1190 proxy_user = opt.proxy_user;
1191 proxy_passwd = opt.proxy_passwd;
1195 proxy_user = proxy->user;
1196 proxy_passwd = proxy->passwd;
1198 /* #### This does not appear right. Can't the proxy request,
1199 say, `Digest' authentication? */
1200 if (proxy_user && proxy_passwd)
1201 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1203 /* If we're using a proxy, we will be connecting to the proxy
1207 /* Proxy authorization over SSL is handled below. */
1209 if (u->scheme != SCHEME_HTTPS)
1211 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1215 /* Whether we need to print the host header with braces around
1216 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1217 usual "Host: symbolic-name:1234". */
1218 int squares = strchr (u->host, ':') != NULL;
1219 if (u->port == scheme_default_port (u->scheme))
1220 request_set_header (req, "Host",
1221 aprintf (squares ? "[%s]" : "%s", u->host),
1224 request_set_header (req, "Host",
1225 aprintf (squares ? "[%s]:%d" : "%s:%d",
1230 if (!inhibit_keep_alive)
1231 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1234 request_set_header (req, "Cookie",
1235 cookie_header (wget_cookie_jar,
1236 u->host, u->port, u->path,
1238 u->scheme == SCHEME_HTTPS
1245 if (opt.post_data || opt.post_file_name)
1247 request_set_header (req, "Content-Type",
1248 "application/x-www-form-urlencoded", rel_none);
1250 post_data_size = strlen (opt.post_data);
1253 post_data_size = file_size (opt.post_file_name);
1254 if (post_data_size == -1)
1256 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1257 opt.post_file_name);
1261 request_set_header (req, "Content-Length",
1262 aprintf ("%ld", post_data_size), rel_value);
1265 /* Add the user headers. */
1266 if (opt.user_headers)
1269 for (i = 0; opt.user_headers[i]; i++)
1270 request_set_user_header (req, opt.user_headers[i]);
1274 /* We need to come back here when the initial attempt to retrieve
1275 without authorization header fails. (Expected to happen at least
1276 for the Digest authorization scheme.) */
1280 /* Establish the connection. */
1282 if (!inhibit_keep_alive)
1284 /* Look for a persistent connection to target host, unless a
1285 proxy is used. The exception is when SSL is in use, in which
1286 case the proxy is nothing but a passthrough to the target
1287 host, registered as a connection to the latter. */
1288 struct url *relevant = conn;
1290 if (u->scheme == SCHEME_HTTPS)
1294 if (persistent_available_p (relevant->host, relevant->port,
1296 relevant->scheme == SCHEME_HTTPS,
1300 &host_lookup_failed))
1302 sock = pconn.socket;
1303 using_ssl = pconn.ssl;
1304 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1305 pconn.host, pconn.port);
1306 DEBUGP (("Reusing fd %d.\n", sock));
1312 /* In its current implementation, persistent_available_p will
1313 look up conn->host in some cases. If that lookup failed, we
1314 don't need to bother with connect_to_host. */
1315 if (host_lookup_failed)
1318 sock = connect_to_host (conn->host, conn->port);
1322 return (retryable_socket_connect_error (errno)
1323 ? CONERROR : CONIMPOSSIBLE);
1326 if (proxy && u->scheme == SCHEME_HTTPS)
1328 /* When requesting SSL URLs through proxies, use the
1329 CONNECT method to request passthrough. */
1330 struct request *connreq = request_new ();
1331 request_set_method (connreq, "CONNECT",
1332 aprintf ("%s:%d", u->host, u->port));
1335 request_set_header (connreq, "Proxy-Authorization",
1336 proxyauth, rel_value);
1337 /* Now that PROXYAUTH is part of the CONNECT request,
1338 zero it out so we don't send proxy authorization with
1339 the regular request below. */
1343 write_error = request_send (connreq, sock);
1344 request_free (connreq);
1345 if (write_error < 0)
1347 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1349 CLOSE_INVALIDATE (sock);
1353 head = fd_read_http_head (sock);
1356 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1358 CLOSE_INVALIDATE (sock);
1367 DEBUGP (("proxy responded with: [%s]\n", head));
1369 resp = response_new (head);
1370 statcode = response_status (resp, &message);
1371 response_free (resp);
1372 if (statcode != 200)
1375 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1376 message ? message : "?");
1377 xfree_null (message);
1382 /* SOCK is now *really* connected to u->host, so update CONN
1383 to reflect this. That way register_persistent will
1384 register SOCK as being connected to u->host:u->port. */
1388 if (conn->scheme == SCHEME_HTTPS)
1390 if (!ssl_connect (sock))
1397 #endif /* HAVE_SSL */
1400 /* Send the request to server. */
1401 write_error = request_send (req, sock);
1403 if (write_error >= 0)
1407 DEBUGP (("[POST data: %s]\n", opt.post_data));
1408 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1410 else if (opt.post_file_name && post_data_size != 0)
1411 write_error = post_file (sock, opt.post_file_name, post_data_size);
1414 if (write_error < 0)
1416 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1418 CLOSE_INVALIDATE (sock);
1422 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1423 proxy ? "Proxy" : "HTTP");
1430 head = fd_read_http_head (sock);
1435 logputs (LOG_NOTQUIET, _("No data received.\n"));
1436 CLOSE_INVALIDATE (sock);
1442 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1444 CLOSE_INVALIDATE (sock);
1449 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1451 resp = response_new (head);
1453 /* Check for status line. */
1455 statcode = response_status (resp, &message);
1456 if (!opt.server_response)
1457 logprintf (LOG_VERBOSE, "%2d %s\n", statcode, message ? message : "");
1460 logprintf (LOG_VERBOSE, "\n");
1461 print_server_response (resp, " ");
1464 if (!opt.ignore_length
1465 && response_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1466 contlen = strtol (hdrval, NULL, 10);
1468 /* Check for keep-alive related responses. */
1469 if (!inhibit_keep_alive && contlen != -1)
1471 if (response_header_copy (resp, "Keep-Alive", NULL, 0))
1473 else if (response_header_copy (resp, "Connection", hdrval,
1476 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1481 /* The server has promised that it will not close the connection
1482 when we're done. This means that we can register it. */
1483 register_persistent (conn->host, conn->port, sock, using_ssl);
1485 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1487 /* Authorization is required. */
1488 skip_short_body (sock, contlen);
1489 CLOSE_FINISH (sock);
1490 if (auth_tried_already || !(user && passwd))
1492 /* If we have tried it already, then there is not point
1494 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1498 char *www_authenticate = response_header_strdup (resp,
1499 "WWW-Authenticate");
1500 /* If the authentication scheme is unknown or if it's the
1501 "Basic" authentication (which we try by default), there's
1502 no sense in retrying. */
1503 if (!www_authenticate
1504 || !known_authentication_scheme_p (www_authenticate)
1505 || BEGINS_WITH (www_authenticate, "Basic"))
1507 xfree_null (www_authenticate);
1508 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1513 auth_tried_already = 1;
1514 pth = url_full_path (u);
1515 request_set_header (req, "Authorization",
1516 create_authorization_line (www_authenticate,
1518 request_method (req),
1522 xfree (www_authenticate);
1523 goto retry_with_auth;
1531 hs->statcode = statcode;
1533 hs->error = xstrdup (_("Malformed status line"));
1535 hs->error = xstrdup (_("(no description)"));
1537 hs->error = xstrdup (message);
1539 type = response_header_strdup (resp, "Content-Type");
1542 char *tmp = strchr (type, ';');
1545 while (tmp > type && ISSPACE (tmp[-1]))
1550 hs->newloc = response_header_strdup (resp, "Location");
1551 hs->remote_time = response_header_strdup (resp, "Last-Modified");
1553 char *set_cookie = response_header_strdup (resp, "Set-Cookie");
1556 /* The jar should have been created by now. */
1557 assert (wget_cookie_jar != NULL);
1558 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1563 if (response_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1565 long first_byte_pos, last_byte_pos, entity_length;
1566 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1568 contrange = first_byte_pos;
1570 response_free (resp);
1572 /* 20x responses are counted among successful by default. */
1573 if (H_20X (statcode))
1576 /* Return if redirected. */
1577 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1579 /* RFC2068 says that in case of the 300 (multiple choices)
1580 response, the server can output a preferred URL through
1581 `Location' header; otherwise, the request should be treated
1582 like GET. So, if the location is set, it will be a
1583 redirection; otherwise, just proceed normally. */
1584 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1588 logprintf (LOG_VERBOSE,
1589 _("Location: %s%s\n"),
1590 hs->newloc ? hs->newloc : _("unspecified"),
1591 hs->newloc ? _(" [following]") : "");
1593 skip_short_body (sock, contlen);
1594 CLOSE_FINISH (sock);
1600 /* If content-type is not given, assume text/html. This is because
1601 of the multitude of broken CGI's that "forget" to generate the
1604 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1605 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1610 if (opt.html_extension && (*dt & TEXTHTML))
1611 /* -E / --html-extension / html_extension = on was specified, and this is a
1612 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1613 already the file's suffix, tack on ".html". */
1615 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1617 if (last_period_in_local_filename == NULL
1618 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1619 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1621 size_t local_filename_len = strlen(*hs->local_file);
1623 *hs->local_file = xrealloc(*hs->local_file,
1624 local_filename_len + sizeof(".html"));
1625 strcpy(*hs->local_file + local_filename_len, ".html");
1627 *dt |= ADDED_HTML_EXTENSION;
1631 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1633 /* If `-c' is in use and the file has been fully downloaded (or
1634 the remote file has shrunk), Wget effectively requests bytes
1635 after the end of file and the server response with 416. */
1636 logputs (LOG_VERBOSE, _("\
1637 \n The file is already fully retrieved; nothing to do.\n\n"));
1638 /* In case the caller inspects. */
1641 /* Mark as successfully retrieved. */
1644 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1645 might be more bytes in the body. */
1646 return RETRUNNEEDED;
1648 if ((contrange != 0 && contrange != hs->restval)
1649 || (H_PARTIAL (statcode) && !contrange))
1651 /* The Range request was somehow misunderstood by the server.
1654 CLOSE_INVALIDATE (sock);
1657 hs->contlen = contlen + contrange;
1663 /* No need to print this output if the body won't be
1664 downloaded at all, or if the original server response is
1666 logputs (LOG_VERBOSE, _("Length: "));
1669 logputs (LOG_VERBOSE, legible (contlen + contrange));
1671 logprintf (LOG_VERBOSE, _(" (%s to go)"), legible (contlen));
1674 logputs (LOG_VERBOSE,
1675 opt.ignore_length ? _("ignored") : _("unspecified"));
1677 logprintf (LOG_VERBOSE, " [%s]\n", type);
1679 logputs (LOG_VERBOSE, "\n");
1683 type = NULL; /* We don't need it any more. */
1685 /* Return if we have no intention of further downloading. */
1686 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1688 /* In case the caller cares to look... */
1692 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1693 servers not to send body in response to a HEAD request. If
1694 you encounter such a server (more likely a broken CGI), use
1695 `--no-http-keep-alive'. */
1696 CLOSE_FINISH (sock);
1697 return RETRFINISHED;
1700 /* Open the local file. */
1703 mkalldirs (*hs->local_file);
1705 rotate_backups (*hs->local_file);
1706 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1709 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1710 CLOSE_INVALIDATE (sock);
1717 /* #### This confuses the timestamping code that checks for file
1718 size. Maybe we should save some additional information? */
1719 if (opt.save_headers)
1720 fwrite (head, 1, strlen (head), fp);
1722 /* Download the request body. */
1725 flags |= rb_read_exactly;
1726 if (hs->restval > 0 && contrange == 0)
1727 /* If the server ignored our range request, instruct fd_read_body
1728 to skip the first RESTVAL bytes of body. */
1729 flags |= rb_skip_startpos;
1730 hs->len = hs->restval;
1732 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1733 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1737 CLOSE_FINISH (sock);
1739 CLOSE_INVALIDATE (sock);
1742 /* Close or flush the file. We have to be careful to check for
1743 error here. Checking the result of fwrite() is not enough --
1744 errors could go unnoticed! */
1747 flush_res = fclose (fp);
1749 flush_res = fflush (fp);
1750 if (flush_res == EOF)
1755 return RETRFINISHED;
1758 /* The genuine HTTP loop! This is the part where the retrieval is
1759 retried, and retried, and retried, and... */
1761 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1762 int *dt, struct url *proxy)
1765 int use_ts, got_head = 0; /* time-stamping info */
1766 char *filename_plus_orig_suffix;
1767 char *local_filename = NULL;
1768 char *tms, *locf, *tmrate;
1770 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1771 long local_size = 0; /* the size of the local file */
1772 size_t filename_len;
1773 struct http_stat hstat; /* HTTP status */
1777 /* This used to be done in main(), but it's a better idea to do it
1778 here so that we don't go through the hoops if we're just using
1782 if (!wget_cookie_jar)
1783 wget_cookie_jar = cookie_jar_new ();
1784 if (opt.cookies_input && !cookies_loaded_p)
1786 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1787 cookies_loaded_p = 1;
1793 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1794 has_wildcards_p because it would also warn on `?', and we know that
1795 shows up in CGI paths a *lot*. */
1796 if (strchr (u->url, '*'))
1797 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1801 /* Determine the local filename. */
1802 if (local_file && *local_file)
1803 hstat.local_file = local_file;
1804 else if (local_file && !opt.output_document)
1806 *local_file = url_file_name (u);
1807 hstat.local_file = local_file;
1811 dummy = url_file_name (u);
1812 hstat.local_file = &dummy;
1813 /* be honest about where we will save the file */
1814 if (local_file && opt.output_document)
1815 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
1818 if (!opt.output_document)
1819 locf = *hstat.local_file;
1821 locf = opt.output_document;
1823 hstat.referer = referer;
1825 filename_len = strlen (*hstat.local_file);
1826 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1828 if (opt.noclobber && file_exists_p (*hstat.local_file))
1830 /* If opt.noclobber is turned on and file already exists, do not
1831 retrieve the file */
1832 logprintf (LOG_VERBOSE, _("\
1833 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1834 /* If the file is there, we suppose it's retrieved OK. */
1837 /* #### Bogusness alert. */
1838 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1839 if (has_html_suffix_p (*hstat.local_file))
1847 if (opt.timestamping)
1849 int local_dot_orig_file_exists = 0;
1851 if (opt.backup_converted)
1852 /* If -K is specified, we'll act on the assumption that it was specified
1853 last time these files were downloaded as well, and instead of just
1854 comparing local file X against server file X, we'll compare local
1855 file X.orig (if extant, else X) against server file X. If -K
1856 _wasn't_ specified last time, or the server contains files called
1857 *.orig, -N will be back to not operating correctly with -k. */
1859 /* Would a single s[n]printf() call be faster? --dan
1861 Definitely not. sprintf() is horribly slow. It's a
1862 different question whether the difference between the two
1863 affects a program. Usually I'd say "no", but at one
1864 point I profiled Wget, and found that a measurable and
1865 non-negligible amount of time was lost calling sprintf()
1866 in url.c. Replacing sprintf with inline calls to
1867 strcpy() and long_to_string() made a difference.
1869 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1870 memcpy (filename_plus_orig_suffix + filename_len,
1871 ".orig", sizeof (".orig"));
1873 /* Try to stat() the .orig file. */
1874 if (stat (filename_plus_orig_suffix, &st) == 0)
1876 local_dot_orig_file_exists = 1;
1877 local_filename = filename_plus_orig_suffix;
1881 if (!local_dot_orig_file_exists)
1882 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1883 if (stat (*hstat.local_file, &st) == 0)
1884 local_filename = *hstat.local_file;
1886 if (local_filename != NULL)
1887 /* There was a local file, so we'll check later to see if the version
1888 the server has is the same version we already have, allowing us to
1894 /* Modification time granularity is 2 seconds for Windows, so
1895 increase local time by 1 second for later comparison. */
1898 local_size = st.st_size;
1902 /* Reset the counter. */
1908 /* Increment the pass counter. */
1910 sleep_between_retrievals (count);
1911 /* Get the current time string. */
1912 tms = time_str (NULL);
1913 /* Print fetch message, if opt.verbose. */
1916 char *hurl = url_string (u, 1);
1920 sprintf (tmp, _("(try:%2d)"), count);
1921 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1922 tms, hurl, tmp, locf);
1924 ws_changetitle (hurl);
1929 /* Default document type is empty. However, if spider mode is
1930 on or time-stamping is employed, HEAD_ONLY commands is
1931 encoded within *dt. */
1932 if (opt.spider || (use_ts && !got_head))
1937 /* Decide whether or not to restart. */
1940 hstat.restval = hstat.len; /* continue where we left off */
1941 else if (opt.always_rest
1942 && stat (locf, &st) == 0
1943 && S_ISREG (st.st_mode))
1944 hstat.restval = st.st_size;
1946 /* Decide whether to send the no-cache directive. We send it in
1948 a) we're using a proxy, and we're past our first retrieval.
1949 Some proxies are notorious for caching incomplete data, so
1950 we require a fresh get.
1951 b) caching is explicitly inhibited. */
1952 if ((proxy && count > 1) /* a */
1953 || !opt.allow_cache /* b */
1955 *dt |= SEND_NOCACHE;
1957 *dt &= ~SEND_NOCACHE;
1959 /* Try fetching the document, or at least its head. */
1960 err = gethttp (u, &hstat, dt, proxy);
1962 /* It's unfortunate that wget determines the local filename before finding
1963 out the Content-Type of the file. Barring a major restructuring of the
1964 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1965 *hstat.local_file to tack on ".html". */
1966 if (!opt.output_document)
1967 locf = *hstat.local_file;
1970 tms = time_str (NULL);
1971 /* Get the new location (with or without the redirection). */
1973 *newloc = xstrdup (hstat.newloc);
1976 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1977 case CONERROR: case READERR: case WRITEFAILED:
1979 /* Non-fatal errors continue executing the loop, which will
1980 bring them to "while" statement at the end, to judge
1981 whether the number of tries was exceeded. */
1982 free_hstat (&hstat);
1983 printwhat (count, opt.ntry);
1986 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
1987 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
1988 /* Fatal errors just return from the function. */
1989 free_hstat (&hstat);
1993 case FWRITEERR: case FOPENERR:
1994 /* Another fatal error. */
1995 logputs (LOG_VERBOSE, "\n");
1996 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
1997 *hstat.local_file, strerror (errno));
1998 free_hstat (&hstat);
2003 /* Another fatal error. */
2004 logputs (LOG_VERBOSE, "\n");
2005 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2006 free_hstat (&hstat);
2011 /* Return the new location to the caller. */
2014 logprintf (LOG_NOTQUIET,
2015 _("ERROR: Redirection (%d) without location.\n"),
2017 free_hstat (&hstat);
2021 free_hstat (&hstat);
2026 /* The file was already fully retrieved. */
2027 free_hstat (&hstat);
2032 /* Deal with you later. */
2035 /* All possibilities should have been exhausted. */
2038 if (!(*dt & RETROKF))
2042 /* #### Ugly ugly ugly! */
2043 char *hurl = url_string (u, 1);
2044 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2047 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2048 tms, hstat.statcode, hstat.error);
2049 logputs (LOG_VERBOSE, "\n");
2050 free_hstat (&hstat);
2055 /* Did we get the time-stamp? */
2058 if (opt.timestamping && !hstat.remote_time)
2060 logputs (LOG_NOTQUIET, _("\
2061 Last-modified header missing -- time-stamps turned off.\n"));
2063 else if (hstat.remote_time)
2065 /* Convert the date-string into struct tm. */
2066 tmr = http_atotm (hstat.remote_time);
2067 if (tmr == (time_t) (-1))
2068 logputs (LOG_VERBOSE, _("\
2069 Last-modified header invalid -- time-stamp ignored.\n"));
2073 /* The time-stamping section. */
2078 use_ts = 0; /* no more time-stamping */
2079 count = 0; /* the retrieve count for HEAD is
2081 if (hstat.remote_time && tmr != (time_t) (-1))
2083 /* Now time-stamping can be used validly. Time-stamping
2084 means that if the sizes of the local and remote file
2085 match, and local file is newer than the remote file,
2086 it will not be retrieved. Otherwise, the normal
2087 download procedure is resumed. */
2089 (hstat.contlen == -1 || local_size == hstat.contlen))
2091 logprintf (LOG_VERBOSE, _("\
2092 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2094 free_hstat (&hstat);
2098 else if (tml >= tmr)
2099 logprintf (LOG_VERBOSE, _("\
2100 The sizes do not match (local %ld) -- retrieving.\n"), local_size);
2102 logputs (LOG_VERBOSE,
2103 _("Remote file is newer, retrieving.\n"));
2105 free_hstat (&hstat);
2108 if ((tmr != (time_t) (-1))
2110 && ((hstat.len == hstat.contlen) ||
2111 ((hstat.res == 0) &&
2112 ((hstat.contlen == -1) ||
2113 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2115 /* #### This code repeats in http.c and ftp.c. Move it to a
2117 const char *fl = NULL;
2118 if (opt.output_document)
2120 if (output_stream_regular)
2121 fl = opt.output_document;
2124 fl = *hstat.local_file;
2128 /* End of time-stamping section. */
2132 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
2137 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2139 if (hstat.len == hstat.contlen)
2143 logprintf (LOG_VERBOSE,
2144 _("%s (%s) - `%s' saved [%ld/%ld]\n\n"),
2145 tms, tmrate, locf, hstat.len, hstat.contlen);
2146 logprintf (LOG_NONVERBOSE,
2147 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2148 tms, u->url, hstat.len, hstat.contlen, locf, count);
2151 total_downloaded_bytes += hstat.len;
2153 /* Remember that we downloaded the file for later ".orig" code. */
2154 if (*dt & ADDED_HTML_EXTENSION)
2155 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2157 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2159 free_hstat (&hstat);
2163 else if (hstat.res == 0) /* No read error */
2165 if (hstat.contlen == -1) /* We don't know how much we were supposed
2166 to get, so assume we succeeded. */
2170 logprintf (LOG_VERBOSE,
2171 _("%s (%s) - `%s' saved [%ld]\n\n"),
2172 tms, tmrate, locf, hstat.len);
2173 logprintf (LOG_NONVERBOSE,
2174 "%s URL:%s [%ld] -> \"%s\" [%d]\n",
2175 tms, u->url, hstat.len, locf, count);
2178 total_downloaded_bytes += hstat.len;
2180 /* Remember that we downloaded the file for later ".orig" code. */
2181 if (*dt & ADDED_HTML_EXTENSION)
2182 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2184 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2186 free_hstat (&hstat);
2190 else if (hstat.len < hstat.contlen) /* meaning we lost the
2191 connection too soon */
2193 logprintf (LOG_VERBOSE,
2194 _("%s (%s) - Connection closed at byte %ld. "),
2195 tms, tmrate, hstat.len);
2196 printwhat (count, opt.ntry);
2197 free_hstat (&hstat);
2200 else if (!opt.kill_longer) /* meaning we got more than expected */
2202 logprintf (LOG_VERBOSE,
2203 _("%s (%s) - `%s' saved [%ld/%ld])\n\n"),
2204 tms, tmrate, locf, hstat.len, hstat.contlen);
2205 logprintf (LOG_NONVERBOSE,
2206 "%s URL:%s [%ld/%ld] -> \"%s\" [%d]\n",
2207 tms, u->url, hstat.len, hstat.contlen, locf, count);
2209 total_downloaded_bytes += hstat.len;
2211 /* Remember that we downloaded the file for later ".orig" code. */
2212 if (*dt & ADDED_HTML_EXTENSION)
2213 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2215 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2217 free_hstat (&hstat);
2221 else /* the same, but not accepted */
2223 logprintf (LOG_VERBOSE,
2224 _("%s (%s) - Connection closed at byte %ld/%ld. "),
2225 tms, tmrate, hstat.len, hstat.contlen);
2226 printwhat (count, opt.ntry);
2227 free_hstat (&hstat);
2231 else /* now hstat.res can only be -1 */
2233 if (hstat.contlen == -1)
2235 logprintf (LOG_VERBOSE,
2236 _("%s (%s) - Read error at byte %ld (%s)."),
2237 tms, tmrate, hstat.len, strerror (errno));
2238 printwhat (count, opt.ntry);
2239 free_hstat (&hstat);
2242 else /* hstat.res == -1 and contlen is given */
2244 logprintf (LOG_VERBOSE,
2245 _("%s (%s) - Read error at byte %ld/%ld (%s). "),
2246 tms, tmrate, hstat.len, hstat.contlen,
2248 printwhat (count, opt.ntry);
2249 free_hstat (&hstat);
2256 while (!opt.ntry || (count < opt.ntry));
2260 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2261 than local timezone.
2263 mktime is similar but assumes struct tm, also known as the
2264 "broken-down" form of time, is in local time zone. mktime_from_utc
2265 uses mktime to make the conversion understanding that an offset
2266 will be introduced by the local time assumption.
2268 mktime_from_utc then measures the introduced offset by applying
2269 gmtime to the initial result and applying mktime to the resulting
2270 "broken-down" form. The difference between the two mktime results
2271 is the measured offset which is then subtracted from the initial
2272 mktime result to yield a calendar time which is the value returned.
2274 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2275 consistent offset (the non DST offset) since tm and tm+o might be
2276 on opposite sides of a DST change.
2278 Some implementations of mktime return -1 for the nonexistent
2279 localtime hour at the beginning of DST. In this event, use
2280 mktime(tm - 1hr) + 3600.
2284 gmtime(t+o) --> tm+o
2285 mktime(tm+o) --> t+2o
2286 t+o - (t+2o - t+o) = t
2288 Note that glibc contains a function of the same purpose named
2289 `timegm' (reverse of gmtime). But obviously, it is not universally
2290 available, and unfortunately it is not straightforwardly
2291 extractable for use here. Perhaps configure should detect timegm
2292 and use it where available.
2294 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2295 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2296 Further improved by Roger with assistance from Edward J. Sabol
2297 based on input by Jamie Zawinski. */
2300 mktime_from_utc (struct tm *t)
2311 return -1; /* can't deal with output from strptime */
2322 return -1; /* can't deal with output from gmtime */
2325 return (tl - (tb - tl));
2328 /* Check whether the result of strptime() indicates success.
2329 strptime() returns the pointer to how far it got to in the string.
2330 The processing has been successful if the string is at `GMT' or
2331 `+X', or at the end of the string.
2333 In extended regexp parlance, the function returns 1 if P matches
2334 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2335 can return) is considered a failure and 0 is returned. */
2337 check_end (const char *p)
2341 while (ISSPACE (*p))
2344 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2345 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2351 /* Convert the textual specification of time in TIME_STRING to the
2352 number of seconds since the Epoch.
2354 TIME_STRING can be in any of the three formats RFC2068 allows the
2355 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2356 Timezones are ignored, and should be GMT.
2358 Return the computed time_t representation, or -1 if the conversion
2361 This function uses strptime with various string formats for parsing
2362 TIME_STRING. This results in a parser that is not as lenient in
2363 interpreting TIME_STRING as I would like it to be. Being based on
2364 strptime, it always allows shortened months, one-digit days, etc.,
2365 but due to the multitude of formats in which time can be
2366 represented, an ideal HTTP time parser would be even more
2367 forgiving. It should completely ignore things like week days and
2368 concentrate only on the various forms of representing years,
2369 months, days, hours, minutes, and seconds. For example, it would
2370 be nice if it accepted ISO 8601 out of the box.
2372 I've investigated free and PD code for this purpose, but none was
2373 usable. getdate was big and unwieldy, and had potential copyright
2374 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2375 distributed with phttpd, is excellent, but we cannot use it because
2376 it is not assigned to the FSF. So I stuck it with strptime. */
2379 http_atotm (const char *time_string)
2381 /* NOTE: Solaris strptime man page claims that %n and %t match white
2382 space, but that's not universally available. Instead, we simply
2383 use ` ' to mean "skip all WS", which works under all strptime
2384 implementations I've tested. */
2386 static const char *time_formats[] = {
2387 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2388 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2389 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2390 (google.com uses this for their cookies.) */
2391 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2397 /* According to Roger Beeman, we need to initialize tm_isdst, since
2398 strptime won't do it. */
2401 /* Note that under foreign locales Solaris strptime() fails to
2402 recognize English dates, which renders this function useless. We
2403 solve this by being careful not to affect LC_TIME when
2404 initializing locale.
2406 Another solution would be to temporarily set locale to C, invoke
2407 strptime(), and restore it back. This is slow and dirty,
2408 however, and locale support other than LC_MESSAGES can mess other
2409 things, so I rather chose to stick with just setting LC_MESSAGES.
2411 GNU strptime does not have this problem because it recognizes
2412 both international and local dates. */
2414 for (i = 0; i < countof (time_formats); i++)
2415 if (check_end (strptime (time_string, time_formats[i], &t)))
2416 return mktime_from_utc (&t);
2418 /* All formats have failed. */
2422 /* Authorization support: We support two authorization schemes:
2424 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2426 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2427 consisting of answering to the server's challenge with the proper
2430 /* How many bytes it will take to store LEN bytes in base64. */
2431 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2433 /* Encode the string S of length LENGTH to base64 format and place it
2434 to STORE. STORE will be 0-terminated, and must point to a writable
2435 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2437 base64_encode (const char *s, char *store, int length)
2439 /* Conversion table. */
2440 static char tbl[64] = {
2441 'A','B','C','D','E','F','G','H',
2442 'I','J','K','L','M','N','O','P',
2443 'Q','R','S','T','U','V','W','X',
2444 'Y','Z','a','b','c','d','e','f',
2445 'g','h','i','j','k','l','m','n',
2446 'o','p','q','r','s','t','u','v',
2447 'w','x','y','z','0','1','2','3',
2448 '4','5','6','7','8','9','+','/'
2451 unsigned char *p = (unsigned char *)store;
2453 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2454 for (i = 0; i < length; i += 3)
2456 *p++ = tbl[s[0] >> 2];
2457 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2458 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2459 *p++ = tbl[s[2] & 0x3f];
2462 /* Pad the result if necessary... */
2463 if (i == length + 1)
2465 else if (i == length + 2)
2466 *(p - 1) = *(p - 2) = '=';
2467 /* ...and zero-terminate it. */
2471 /* Create the authentication header contents for the `Basic' scheme.
2472 This is done by encoding the string `USER:PASS' in base64 and
2473 prepending `HEADER: Basic ' to it. */
2475 basic_authentication_encode (const char *user, const char *passwd)
2477 char *t1, *t2, *res;
2478 int len1 = strlen (user) + 1 + strlen (passwd);
2479 int len2 = BASE64_LENGTH (len1);
2481 t1 = (char *)alloca (len1 + 1);
2482 sprintf (t1, "%s:%s", user, passwd);
2484 t2 = (char *)alloca (len2 + 1);
2485 base64_encode (t1, t2, len1);
2487 res = (char *)xmalloc (6 + len2 + 1);
2488 sprintf (res, "Basic %s", t2);
2493 #define SKIP_WS(x) do { \
2494 while (ISSPACE (*(x))) \
2499 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2500 of a field in such a header. If the field is the one specified by
2501 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2502 digest authorization code), extract its value in the (char*)
2503 variable pointed by RET. Returns negative on a malformed header,
2504 or number of bytes that have been parsed by this call. */
2506 extract_header_attr (const char *au, const char *attr_name, char **ret)
2508 const char *cp, *ep;
2512 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2514 cp += strlen (attr_name);
2527 for (ep = cp; *ep && *ep != '\"'; ep++)
2532 *ret = strdupdelim (cp, ep);
2539 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2540 an array of 16 bytes containing the hash keys, and BUF should be a
2541 buffer of 33 writable characters (32 for hex digits plus one for
2542 zero termination). */
2544 dump_hash (unsigned char *buf, const unsigned char *hash)
2548 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2550 *buf++ = XNUM_TO_digit (*hash >> 4);
2551 *buf++ = XNUM_TO_digit (*hash & 0xf);
2556 /* Take the line apart to find the challenge, and compose a digest
2557 authorization header. See RFC2069 section 2.1.2. */
2559 digest_authentication_encode (const char *au, const char *user,
2560 const char *passwd, const char *method,
2563 static char *realm, *opaque, *nonce;
2568 { "realm", &realm },
2569 { "opaque", &opaque },
2574 realm = opaque = nonce = NULL;
2576 au += 6; /* skip over `Digest' */
2582 for (i = 0; i < countof (options); i++)
2584 int skip = extract_header_attr (au, options[i].name,
2585 options[i].variable);
2589 xfree_null (opaque);
2599 if (i == countof (options))
2601 while (*au && *au != '=')
2609 while (*au && *au != '\"')
2616 while (*au && *au != ',')
2621 if (!realm || !nonce || !user || !passwd || !path || !method)
2624 xfree_null (opaque);
2629 /* Calculate the digest value. */
2631 ALLOCA_MD5_CONTEXT (ctx);
2632 unsigned char hash[MD5_HASHLEN];
2633 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2634 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2636 /* A1BUF = H(user ":" realm ":" password) */
2638 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2639 gen_md5_update ((unsigned char *)":", 1, ctx);
2640 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2641 gen_md5_update ((unsigned char *)":", 1, ctx);
2642 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2643 gen_md5_finish (ctx, hash);
2644 dump_hash (a1buf, hash);
2646 /* A2BUF = H(method ":" path) */
2648 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2649 gen_md5_update ((unsigned char *)":", 1, ctx);
2650 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2651 gen_md5_finish (ctx, hash);
2652 dump_hash (a2buf, hash);
2654 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2656 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2657 gen_md5_update ((unsigned char *)":", 1, ctx);
2658 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2659 gen_md5_update ((unsigned char *)":", 1, ctx);
2660 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2661 gen_md5_finish (ctx, hash);
2662 dump_hash (response_digest, hash);
2664 res = (char*) xmalloc (strlen (user)
2669 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2670 + (opaque ? strlen (opaque) : 0)
2672 sprintf (res, "Digest \
2673 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2674 user, realm, nonce, path, response_digest);
2677 char *p = res + strlen (res);
2678 strcat (p, ", opaque=\"");
2685 #endif /* USE_DIGEST */
2688 #define BEGINS_WITH(line, string_constant) \
2689 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2690 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2691 || !line[sizeof (string_constant) - 1]))
2694 known_authentication_scheme_p (const char *au)
2696 return BEGINS_WITH (au, "Basic")
2697 || BEGINS_WITH (au, "Digest")
2698 || BEGINS_WITH (au, "NTLM");
2703 /* Create the HTTP authorization request header. When the
2704 `WWW-Authenticate' response header is seen, according to the
2705 authorization scheme specified in that header (`Basic' and `Digest'
2706 are supported by the current implementation), produce an
2707 appropriate HTTP authorization request header. */
2709 create_authorization_line (const char *au, const char *user,
2710 const char *passwd, const char *method,
2713 if (0 == strncasecmp (au, "Basic", 5))
2714 return basic_authentication_encode (user, passwd);
2716 if (0 == strncasecmp (au, "Digest", 6))
2717 return digest_authentication_encode (au, user, passwd, method, path);
2718 #endif /* USE_DIGEST */