2 Copyright (C) 2005 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. */
53 # include "http-ntlm.h"
61 extern char *version_string;
62 extern LARGE_INT total_downloaded_bytes;
64 extern FILE *output_stream;
65 extern int output_stream_regular;
68 # define MIN(x, y) ((x) > (y) ? (y) : (x))
72 static int cookies_loaded_p;
73 static struct cookie_jar *wget_cookie_jar;
75 #define TEXTHTML_S "text/html"
76 #define TEXTXHTML_S "application/xhtml+xml"
78 /* Some status code validation macros: */
79 #define H_20X(x) (((x) >= 200) && ((x) < 300))
80 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
81 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
82 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
83 || (x) == HTTP_STATUS_SEE_OTHER \
84 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
86 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
88 #define HTTP_STATUS_OK 200
89 #define HTTP_STATUS_CREATED 201
90 #define HTTP_STATUS_ACCEPTED 202
91 #define HTTP_STATUS_NO_CONTENT 204
92 #define HTTP_STATUS_PARTIAL_CONTENTS 206
94 /* Redirection 3xx. */
95 #define HTTP_STATUS_MULTIPLE_CHOICES 300
96 #define HTTP_STATUS_MOVED_PERMANENTLY 301
97 #define HTTP_STATUS_MOVED_TEMPORARILY 302
98 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
99 #define HTTP_STATUS_NOT_MODIFIED 304
100 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
102 /* Client error 4xx. */
103 #define HTTP_STATUS_BAD_REQUEST 400
104 #define HTTP_STATUS_UNAUTHORIZED 401
105 #define HTTP_STATUS_FORBIDDEN 403
106 #define HTTP_STATUS_NOT_FOUND 404
107 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
109 /* Server errors 5xx. */
110 #define HTTP_STATUS_INTERNAL 500
111 #define HTTP_STATUS_NOT_IMPLEMENTED 501
112 #define HTTP_STATUS_BAD_GATEWAY 502
113 #define HTTP_STATUS_UNAVAILABLE 503
116 rel_none, rel_name, rel_value, rel_both
123 struct request_header {
125 enum rp release_policy;
127 int hcount, hcapacity;
130 /* Create a new, empty request. At least request_set_method must be
131 called before the request can be used. */
133 static struct request *
136 struct request *req = xnew0 (struct request);
138 req->headers = xnew_array (struct request_header, req->hcapacity);
142 /* Set the request's method and its arguments. METH should be a
143 literal string (or it should outlive the request) because it will
144 not be freed. ARG will be freed by request_free. */
147 request_set_method (struct request *req, const char *meth, char *arg)
153 /* Return the method string passed with the last call to
154 request_set_method. */
157 request_method (const struct request *req)
162 /* Free one header according to the release policy specified with
163 request_set_header. */
166 release_header (struct request_header *hdr)
168 switch (hdr->release_policy)
185 /* Set the request named NAME to VALUE. Specifically, this means that
186 a "NAME: VALUE\r\n" header line will be used in the request. If a
187 header with the same name previously existed in the request, its
188 value will be replaced by this one. A NULL value means do nothing.
190 RELEASE_POLICY determines whether NAME and VALUE should be released
191 (freed) with request_free. Allowed values are:
193 - rel_none - don't free NAME or VALUE
194 - rel_name - free NAME when done
195 - rel_value - free VALUE when done
196 - rel_both - free both NAME and VALUE when done
198 Setting release policy is useful when arguments come from different
199 sources. For example:
201 // Don't free literal strings!
202 request_set_header (req, "Pragma", "no-cache", rel_none);
204 // Don't free a global variable, we'll need it later.
205 request_set_header (req, "Referer", opt.referer, rel_none);
207 // Value freshly allocated, free it when done.
208 request_set_header (req, "Range",
209 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
214 request_set_header (struct request *req, char *name, char *value,
215 enum rp release_policy)
217 struct request_header *hdr;
222 /* A NULL value is a no-op; if freeing the name is requested,
223 free it now to avoid leaks. */
224 if (release_policy == rel_name || release_policy == rel_both)
229 for (i = 0; i < req->hcount; i++)
231 hdr = &req->headers[i];
232 if (0 == strcasecmp (name, hdr->name))
234 /* Replace existing header. */
235 release_header (hdr);
238 hdr->release_policy = release_policy;
243 /* Install new header. */
245 if (req->hcount >= req->hcapacity)
247 req->hcapacity <<= 1;
248 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
250 hdr = &req->headers[req->hcount++];
253 hdr->release_policy = release_policy;
256 /* Like request_set_header, but sets the whole header line, as
257 provided by the user using the `--header' option. For example,
258 request_set_user_header (req, "Foo: bar") works just like
259 request_set_header (req, "Foo", "bar"). */
262 request_set_user_header (struct request *req, const char *header)
265 const char *p = strchr (header, ':');
268 BOUNDED_TO_ALLOCA (header, p, name);
272 request_set_header (req, xstrdup (name), (char *) p, rel_name);
275 /* Remove the header with specified name from REQ. Returns 1 if the
276 header was actually removed, 0 otherwise. */
279 request_remove_header (struct request *req, char *name)
282 for (i = 0; i < req->hcount; i++)
284 struct request_header *hdr = &req->headers[i];
285 if (0 == strcasecmp (name, hdr->name))
287 release_header (hdr);
288 /* Move the remaining headers by one. */
289 if (i < req->hcount - 1)
290 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
298 #define APPEND(p, str) do { \
299 int A_len = strlen (str); \
300 memcpy (p, str, A_len); \
304 /* Construct the request and write it to FD using fd_write. */
307 request_send (const struct request *req, int fd)
309 char *request_string, *p;
310 int i, size, write_error;
312 /* Count the request size. */
315 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
316 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
318 for (i = 0; i < req->hcount; i++)
320 struct request_header *hdr = &req->headers[i];
321 /* NAME ": " VALUE "\r\n" */
322 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
328 p = request_string = alloca_array (char, size);
330 /* Generate the request. */
332 APPEND (p, req->method); *p++ = ' ';
333 APPEND (p, req->arg); *p++ = ' ';
334 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
336 for (i = 0; i < req->hcount; i++)
338 struct request_header *hdr = &req->headers[i];
339 APPEND (p, hdr->name);
340 *p++ = ':', *p++ = ' ';
341 APPEND (p, hdr->value);
342 *p++ = '\r', *p++ = '\n';
345 *p++ = '\r', *p++ = '\n', *p++ = '\0';
346 assert (p - request_string == size);
350 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
352 /* Send the request to the server. */
354 write_error = fd_write (fd, request_string, size - 1, -1);
356 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
361 /* Release the resources used by REQ. */
364 request_free (struct request *req)
367 xfree_null (req->arg);
368 for (i = 0; i < req->hcount; i++)
369 release_header (&req->headers[i]);
370 xfree_null (req->headers);
374 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
375 PROMISED_SIZE bytes are sent over the wire -- if the file is
376 longer, read only that much; if the file is shorter, report an error. */
379 post_file (int sock, const char *file_name, wgint promised_size)
381 static char chunk[8192];
386 DEBUGP (("[writing POST file %s ... ", file_name));
388 fp = fopen (file_name, "rb");
391 while (!feof (fp) && written < promised_size)
394 int length = fread (chunk, 1, sizeof (chunk), fp);
397 towrite = MIN (promised_size - written, length);
398 write_error = fd_write (sock, chunk, towrite, -1);
408 /* If we've written less than was promised, report a (probably
409 nonsensical) error rather than break the promise. */
410 if (written < promised_size)
416 assert (written == promised_size);
417 DEBUGP (("done]\n"));
422 response_head_terminator (const char *hunk, int oldlen, int peeklen)
424 const char *start, *end;
426 /* If at first peek, verify whether HUNK starts with "HTTP". If
427 not, this is a HTTP/0.9 request and we must bail out without
429 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
435 start = hunk + oldlen - 4;
436 end = hunk + oldlen + peeklen;
438 for (; start < end - 1; start++)
445 if (start[1] == '\n')
451 /* The maximum size of a single HTTP response we care to read. This
452 is not meant to impose an arbitrary limit, but to protect the user
453 from Wget slurping up available memory upon encountering malicious
454 or buggy server output. Define it to 0 to remove the limit. */
456 #define HTTP_RESPONSE_MAX_SIZE 65536
458 /* Read the HTTP request head from FD and return it. The error
459 conditions are the same as with fd_read_hunk.
461 To support HTTP/0.9 responses, this function tries to make sure
462 that the data begins with "HTTP". If this is not the case, no data
463 is read and an empty request is returned, so that the remaining
464 data can be treated as body. */
467 read_http_response_head (int fd)
469 return fd_read_hunk (fd, response_head_terminator, 512,
470 HTTP_RESPONSE_MAX_SIZE);
474 /* The response data. */
477 /* The array of pointers that indicate where each header starts.
478 For example, given this HTTP response:
485 The headers are located like this:
487 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
489 headers[0] headers[1] headers[2] headers[3]
491 I.e. headers[0] points to the beginning of the request,
492 headers[1] points to the end of the first header and the
493 beginning of the second one, etc. */
495 const char **headers;
498 /* Create a new response object from the text of the HTTP response,
499 available in HEAD. That text is automatically split into
500 constituent header lines for fast retrieval using
503 static struct response *
504 resp_new (const char *head)
509 struct response *resp = xnew0 (struct response);
514 /* Empty head means that we're dealing with a headerless
515 (HTTP/0.9) response. In that case, don't set HEADERS at
520 /* Split HEAD into header lines, so that resp_header_* functions
521 don't need to do this over and over again. */
527 DO_REALLOC (resp->headers, size, count + 1, const char *);
528 resp->headers[count++] = hdr;
530 /* Break upon encountering an empty line. */
531 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
534 /* Find the end of HDR, including continuations. */
537 const char *end = strchr (hdr, '\n');
543 while (*hdr == ' ' || *hdr == '\t');
545 DO_REALLOC (resp->headers, size, count + 1, const char *);
546 resp->headers[count] = NULL;
551 /* Locate the header named NAME in the request data, starting with
552 position START. This allows the code to loop through the request
553 data, filtering for all requests of a given name. Returns the
554 found position, or -1 for failure. The code that uses this
555 function typically looks like this:
557 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
558 ... do something with header ...
560 If you only care about one header, use resp_header_get instead of
564 resp_header_locate (const struct response *resp, const char *name, int start,
565 const char **begptr, const char **endptr)
568 const char **headers = resp->headers;
571 if (!headers || !headers[1])
574 name_len = strlen (name);
580 for (; headers[i + 1]; i++)
582 const char *b = headers[i];
583 const char *e = headers[i + 1];
585 && b[name_len] == ':'
586 && 0 == strncasecmp (b, name, name_len))
589 while (b < e && ISSPACE (*b))
591 while (b < e && ISSPACE (e[-1]))
601 /* Find and retrieve the header named NAME in the request data. If
602 found, set *BEGPTR to its starting, and *ENDPTR to its ending
603 position, and return 1. Otherwise return 0.
605 This function is used as a building block for resp_header_copy
606 and resp_header_strdup. */
609 resp_header_get (const struct response *resp, const char *name,
610 const char **begptr, const char **endptr)
612 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
616 /* Copy the response header named NAME to buffer BUF, no longer than
617 BUFSIZE (BUFSIZE includes the terminating 0). If the header
618 exists, 1 is returned, otherwise 0. If there should be no limit on
619 the size of the header, use resp_header_strdup instead.
621 If BUFSIZE is 0, no data is copied, but the boolean indication of
622 whether the header is present is still returned. */
625 resp_header_copy (const struct response *resp, const char *name,
626 char *buf, int bufsize)
629 if (!resp_header_get (resp, name, &b, &e))
633 int len = MIN (e - b, bufsize - 1);
634 memcpy (buf, b, len);
640 /* Return the value of header named NAME in RESP, allocated with
641 malloc. If such a header does not exist in RESP, return NULL. */
644 resp_header_strdup (const struct response *resp, const char *name)
647 if (!resp_header_get (resp, name, &b, &e))
649 return strdupdelim (b, e);
652 /* Parse the HTTP status line, which is of format:
654 HTTP-Version SP Status-Code SP Reason-Phrase
656 The function returns the status-code, or -1 if the status line
657 appears malformed. The pointer to "reason-phrase" message is
658 returned in *MESSAGE. */
661 resp_status (const struct response *resp, char **message)
668 /* For a HTTP/0.9 response, assume status 200. */
670 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
674 p = resp->headers[0];
675 end = resp->headers[1];
681 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
685 /* Match the HTTP version. This is optional because Gnutella
686 servers have been reported to not specify HTTP version. */
687 if (p < end && *p == '/')
690 while (p < end && ISDIGIT (*p))
692 if (p < end && *p == '.')
694 while (p < end && ISDIGIT (*p))
698 while (p < end && ISSPACE (*p))
700 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
703 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
708 while (p < end && ISSPACE (*p))
710 while (p < end && ISSPACE (end[-1]))
712 *message = strdupdelim (p, end);
718 /* Release the resources used by RESP. */
721 resp_free (struct response *resp)
723 xfree_null (resp->headers);
727 /* Print the server response, line by line, omitting the trailing CRLF
728 from individual header lines, and prefixed with PREFIX. */
731 print_server_response (const struct response *resp, const char *prefix)
736 for (i = 0; resp->headers[i + 1]; i++)
738 const char *b = resp->headers[i];
739 const char *e = resp->headers[i + 1];
741 if (b < e && e[-1] == '\n')
743 if (b < e && e[-1] == '\r')
745 /* This is safe even on printfs with broken handling of "%.<n>s"
746 because resp->headers ends with \0. */
747 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
751 /* Parse the `Content-Range' header and extract the information it
752 contains. Returns 1 if successful, -1 otherwise. */
754 parse_content_range (const char *hdr, wgint *first_byte_ptr,
755 wgint *last_byte_ptr, wgint *entity_length_ptr)
759 /* Ancient versions of Netscape proxy server, presumably predating
760 rfc2068, sent out `Content-Range' without the "bytes"
762 if (!strncasecmp (hdr, "bytes", 5))
765 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
769 while (ISSPACE (*hdr))
776 for (num = 0; ISDIGIT (*hdr); hdr++)
777 num = 10 * num + (*hdr - '0');
778 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
780 *first_byte_ptr = num;
782 for (num = 0; ISDIGIT (*hdr); hdr++)
783 num = 10 * num + (*hdr - '0');
784 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
786 *last_byte_ptr = num;
788 for (num = 0; ISDIGIT (*hdr); hdr++)
789 num = 10 * num + (*hdr - '0');
790 *entity_length_ptr = num;
794 /* Read the body of the request, but don't store it anywhere and don't
795 display a progress gauge. This is useful for reading the bodies of
796 administrative responses to which we will soon issue another
797 request. The response is not useful to the user, but reading it
798 allows us to continue using the same connection to the server.
800 If reading fails, 0 is returned, non-zero otherwise. In debug
801 mode, the body is displayed for debugging purposes. */
804 skip_short_body (int fd, wgint contlen)
807 SKIP_SIZE = 512, /* size of the download buffer */
808 SKIP_THRESHOLD = 4096 /* the largest size we read */
810 char dlbuf[SKIP_SIZE + 1];
811 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
813 /* We shouldn't get here with unknown contlen. (This will change
814 with HTTP/1.1, which supports "chunked" transfer.) */
815 assert (contlen != -1);
817 /* If the body is too large, it makes more sense to simply close the
818 connection than to try to read the body. */
819 if (contlen > SKIP_THRESHOLD)
822 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
826 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
829 /* Don't normally report the error since this is an
830 optimization that should be invisible to the user. */
831 DEBUGP (("] aborting (%s).\n",
832 ret < 0 ? strerror (errno) : "EOF received"));
836 /* Safe even if %.*s bogusly expects terminating \0 because
837 we've zero-terminated dlbuf above. */
838 DEBUGP (("%.*s", ret, dlbuf));
841 DEBUGP (("] done.\n"));
845 /* Persistent connections. Currently, we cache the most recently used
846 connection as persistent, provided that the HTTP server agrees to
847 make it such. The persistence data is stored in the variables
848 below. Ideally, it should be possible to cache an arbitrary fixed
849 number of these connections. */
851 /* Whether a persistent connection is active. */
852 static int pconn_active;
855 /* The socket of the connection. */
858 /* Host and port of the currently active persistent connection. */
862 /* Whether a ssl handshake has occoured on this connection. */
865 /* Whether the connection was authorized. This is only done by
866 NTLM, which authorizes *connections* rather than individual
867 requests. (That practice is peculiar for HTTP, but it is a
868 useful optimization.) */
872 /* NTLM data of the current connection. */
873 struct ntlmdata ntlm;
877 /* Mark the persistent connection as invalid and free the resources it
878 uses. This is used by the CLOSE_* macros after they forcefully
879 close a registered persistent connection. */
882 invalidate_persistent (void)
884 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
886 fd_close (pconn.socket);
891 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
892 persistent. This will enable someone to use the same connection
893 later. In the context of HTTP, this must be called only AFTER the
894 response has been received and the server has promised that the
895 connection will remain alive.
897 If a previous connection was persistent, it is closed. */
900 register_persistent (const char *host, int port, int fd, int ssl)
904 if (pconn.socket == fd)
906 /* The connection FD is already registered. */
911 /* The old persistent connection is still active; close it
912 first. This situation arises whenever a persistent
913 connection exists, but we then connect to a different
914 host, and try to register a persistent connection to that
916 invalidate_persistent ();
922 pconn.host = xstrdup (host);
925 pconn.authorized = 0;
927 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
930 /* Return non-zero if a persistent connection is available for
931 connecting to HOST:PORT. */
934 persistent_available_p (const char *host, int port, int ssl,
935 int *host_lookup_failed)
937 /* First, check whether a persistent connection is active at all. */
941 /* If we want SSL and the last connection wasn't or vice versa,
942 don't use it. Checking for host and port is not enough because
943 HTTP and HTTPS can apparently coexist on the same port. */
944 if (ssl != pconn.ssl)
947 /* If we're not connecting to the same port, we're not interested. */
948 if (port != pconn.port)
951 /* If the host is the same, we're in business. If not, there is
952 still hope -- read below. */
953 if (0 != strcasecmp (host, pconn.host))
955 /* Check if pconn.socket is talking to HOST under another name.
956 This happens often when both sites are virtual hosts
957 distinguished only by name and served by the same network
958 interface, and hence the same web server (possibly set up by
959 the ISP and serving many different web sites). This
960 admittedly unconventional optimization does not contradict
961 HTTP and works well with popular server software. */
965 struct address_list *al;
968 /* Don't try to talk to two different SSL sites over the same
969 secure connection! (Besides, it's not clear that
970 name-based virtual hosting is even possible with SSL.) */
973 /* If pconn.socket's peer is one of the IP addresses HOST
974 resolves to, pconn.socket is for all intents and purposes
975 already talking to HOST. */
977 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
979 /* Can't get the peer's address -- something must be very
980 wrong with the connection. */
981 invalidate_persistent ();
984 al = lookup_host (host, 0);
987 *host_lookup_failed = 1;
991 found = address_list_contains (al, &ip);
992 address_list_release (al);
997 /* The persistent connection's peer address was found among the
998 addresses HOST resolved to; therefore, pconn.sock is in fact
999 already talking to HOST -- no need to reconnect. */
1002 /* Finally, check whether the connection is still open. This is
1003 important because most server implement a liberal (short) timeout
1004 on persistent connections. Wget can of course always reconnect
1005 if the connection doesn't work out, but it's nicer to know in
1006 advance. This test is a logical followup of the first test, but
1007 is "expensive" and therefore placed at the end of the list. */
1009 if (!test_socket_open (pconn.socket))
1011 /* Oops, the socket is no longer open. Now that we know that,
1012 let's invalidate the persistent connection before returning
1014 invalidate_persistent ();
1021 /* The idea behind these two CLOSE macros is to distinguish between
1022 two cases: one when the job we've been doing is finished, and we
1023 want to close the connection and leave, and two when something is
1024 seriously wrong and we're closing the connection as part of
1027 In case of keep_alive, CLOSE_FINISH should leave the connection
1028 open, while CLOSE_INVALIDATE should still close it.
1030 Note that the semantics of the flag `keep_alive' is "this
1031 connection *will* be reused (the server has promised not to close
1032 the connection once we're done)", while the semantics of
1033 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1034 active, registered connection". */
1036 #define CLOSE_FINISH(fd) do { \
1039 if (pconn_active && (fd) == pconn.socket) \
1040 invalidate_persistent (); \
1049 #define CLOSE_INVALIDATE(fd) do { \
1050 if (pconn_active && (fd) == pconn.socket) \
1051 invalidate_persistent (); \
1059 wgint len; /* received length */
1060 wgint contlen; /* expected length */
1061 wgint restval; /* the restart value */
1062 int res; /* the result of last read */
1063 char *newloc; /* new location (redirection) */
1064 char *remote_time; /* remote time-stamp string */
1065 char *error; /* textual HTTP error */
1066 int statcode; /* status code */
1067 wgint rd_size; /* amount of data read from socket */
1068 double dltime; /* time it took to download the data */
1069 const char *referer; /* value of the referer header. */
1070 char **local_file; /* local file. */
1074 free_hstat (struct http_stat *hs)
1076 xfree_null (hs->newloc);
1077 xfree_null (hs->remote_time);
1078 xfree_null (hs->error);
1080 /* Guard against being called twice. */
1082 hs->remote_time = NULL;
1086 static char *create_authorization_line (const char *, const char *,
1087 const char *, const char *,
1088 const char *, int *);
1089 static char *basic_authentication_encode (const char *, const char *);
1090 static int known_authentication_scheme_p (const char *, const char *);
1092 time_t http_atotm (const char *);
1094 #define BEGINS_WITH(line, string_constant) \
1095 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1096 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1097 || !line[sizeof (string_constant) - 1]))
1099 #define SET_USER_AGENT(req) do { \
1100 if (!opt.useragent) \
1101 request_set_header (req, "User-Agent", \
1102 aprintf ("Wget/%s", version_string), rel_value); \
1103 else if (*opt.useragent) \
1104 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1107 /* The flags that allow clobbering the file (opening with "wb").
1108 Defined here to avoid repetition later. #### This will require
1110 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1111 || opt.dirstruct || opt.output_document)
1113 /* Retrieve a document through HTTP protocol. It recognizes status
1114 code, and correctly handles redirections. It closes the network
1115 socket. If it receives an error from the functions below it, it
1116 will print it if there is enough information to do so (almost
1117 always), returning the error to the caller (i.e. http_loop).
1119 Various HTTP parameters are stored to hs.
1121 If PROXY is non-NULL, the connection will be made to the proxy
1122 server, and u->url will be requested. */
1124 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1126 struct request *req;
1129 char *user, *passwd;
1133 wgint contlen, contrange;
1140 /* Set to 1 when the authorization has failed permanently and should
1141 not be tried again. */
1142 int auth_finished = 0;
1144 /* Whether NTLM authentication is used for this request. */
1147 /* Whether our connection to the remote host is through SSL. */
1150 /* Whether a HEAD request will be issued (as opposed to GET or
1152 int head_only = *dt & HEAD_ONLY;
1155 struct response *resp;
1159 /* Whether this connection will be kept alive after the HTTP request
1163 /* Whether keep-alive should be inhibited.
1165 RFC 2068 requests that 1.0 clients not send keep-alive requests
1166 to proxies. This is because many 1.0 proxies do not interpret
1167 the Connection header and transfer it to the remote server,
1168 causing it to not close the connection and leave both the proxy
1169 and the client hanging. */
1170 int inhibit_keep_alive =
1171 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1173 /* Headers sent when using POST. */
1174 wgint post_data_size = 0;
1176 int host_lookup_failed = 0;
1179 if (u->scheme == SCHEME_HTTPS)
1181 /* Initialize the SSL context. After this has once been done,
1182 it becomes a no-op. */
1185 scheme_disable (SCHEME_HTTPS);
1186 logprintf (LOG_NOTQUIET,
1187 _("Disabling SSL due to encountered errors.\n"));
1188 return SSLINITFAILED;
1191 #endif /* HAVE_SSL */
1194 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1195 know the local filename so we can save to it. */
1196 assert (*hs->local_file != NULL);
1198 /* Initialize certain elements of struct http_stat. */
1203 hs->remote_time = NULL;
1208 /* Prepare the request to send. */
1210 req = request_new ();
1213 const char *meth = "GET";
1216 else if (opt.post_file_name || opt.post_data)
1218 /* Use the full path, i.e. one that includes the leading slash and
1219 the query string. E.g. if u->path is "foo/bar" and u->query is
1220 "param=value", full_path will be "/foo/bar?param=value". */
1223 /* When using SSL over proxy, CONNECT establishes a direct
1224 connection to the HTTPS server. Therefore use the same
1225 argument as when talking to the server directly. */
1226 && u->scheme != SCHEME_HTTPS
1229 meth_arg = xstrdup (u->url);
1231 meth_arg = url_full_path (u);
1232 request_set_method (req, meth, meth_arg);
1235 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1236 if (*dt & SEND_NOCACHE)
1237 request_set_header (req, "Pragma", "no-cache", rel_none);
1239 request_set_header (req, "Range",
1240 aprintf ("bytes=%s-",
1241 number_to_static_string (hs->restval)),
1243 SET_USER_AGENT (req);
1244 request_set_header (req, "Accept", "*/*", rel_none);
1246 /* Find the username and password for authentication. */
1249 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1250 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1251 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1255 /* We have the username and the password, but haven't tried
1256 any authorization yet. Let's see if the "Basic" method
1257 works. If not, we'll come back here and construct a
1258 proper authorization method with the right challenges.
1260 If we didn't employ this kind of logic, every URL that
1261 requires authorization would have to be processed twice,
1262 which is very suboptimal and generates a bunch of false
1263 "unauthorized" errors in the server log.
1265 #### But this logic also has a serious problem when used
1266 with stronger authentications: we *first* transmit the
1267 username and the password in clear text, and *then* attempt a
1268 stronger authentication scheme. That cannot be right! We
1269 are only fortunate that almost everyone still uses the
1270 `Basic' scheme anyway.
1272 There should be an option to prevent this from happening, for
1273 those who use strong authentication schemes and value their
1275 request_set_header (req, "Authorization",
1276 basic_authentication_encode (user, passwd),
1283 char *proxy_user, *proxy_passwd;
1284 /* For normal username and password, URL components override
1285 command-line/wgetrc parameters. With proxy
1286 authentication, it's the reverse, because proxy URLs are
1287 normally the "permanent" ones, so command-line args
1288 should take precedence. */
1289 if (opt.proxy_user && opt.proxy_passwd)
1291 proxy_user = opt.proxy_user;
1292 proxy_passwd = opt.proxy_passwd;
1296 proxy_user = proxy->user;
1297 proxy_passwd = proxy->passwd;
1299 /* #### This does not appear right. Can't the proxy request,
1300 say, `Digest' authentication? */
1301 if (proxy_user && proxy_passwd)
1302 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1304 /* If we're using a proxy, we will be connecting to the proxy
1308 /* Proxy authorization over SSL is handled below. */
1310 if (u->scheme != SCHEME_HTTPS)
1312 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1316 /* Whether we need to print the host header with braces around
1317 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1318 usual "Host: symbolic-name:1234". */
1319 int squares = strchr (u->host, ':') != NULL;
1320 if (u->port == scheme_default_port (u->scheme))
1321 request_set_header (req, "Host",
1322 aprintf (squares ? "[%s]" : "%s", u->host),
1325 request_set_header (req, "Host",
1326 aprintf (squares ? "[%s]:%d" : "%s:%d",
1331 if (!inhibit_keep_alive)
1332 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1335 request_set_header (req, "Cookie",
1336 cookie_header (wget_cookie_jar,
1337 u->host, u->port, u->path,
1339 u->scheme == SCHEME_HTTPS
1346 if (opt.post_data || opt.post_file_name)
1348 request_set_header (req, "Content-Type",
1349 "application/x-www-form-urlencoded", rel_none);
1351 post_data_size = strlen (opt.post_data);
1354 post_data_size = file_size (opt.post_file_name);
1355 if (post_data_size == -1)
1357 logprintf (LOG_NOTQUIET, _("POST data file missing: %s (%s)\n"),
1358 opt.post_file_name, strerror (errno));
1362 request_set_header (req, "Content-Length",
1363 xstrdup (number_to_static_string (post_data_size)),
1367 /* Add the user headers. */
1368 if (opt.user_headers)
1371 for (i = 0; opt.user_headers[i]; i++)
1372 request_set_user_header (req, opt.user_headers[i]);
1376 /* We need to come back here when the initial attempt to retrieve
1377 without authorization header fails. (Expected to happen at least
1378 for the Digest authorization scheme.) */
1382 /* Establish the connection. */
1384 if (!inhibit_keep_alive)
1386 /* Look for a persistent connection to target host, unless a
1387 proxy is used. The exception is when SSL is in use, in which
1388 case the proxy is nothing but a passthrough to the target
1389 host, registered as a connection to the latter. */
1390 struct url *relevant = conn;
1392 if (u->scheme == SCHEME_HTTPS)
1396 if (persistent_available_p (relevant->host, relevant->port,
1398 relevant->scheme == SCHEME_HTTPS,
1402 &host_lookup_failed))
1404 sock = pconn.socket;
1405 using_ssl = pconn.ssl;
1406 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1407 escnonprint (pconn.host), pconn.port);
1408 DEBUGP (("Reusing fd %d.\n", sock));
1409 if (pconn.authorized)
1410 /* If the connection is already authorized, the "Basic"
1411 authorization added by code above is unnecessary and
1413 request_remove_header (req, "Authorization");
1419 /* In its current implementation, persistent_available_p will
1420 look up conn->host in some cases. If that lookup failed, we
1421 don't need to bother with connect_to_host. */
1422 if (host_lookup_failed)
1428 sock = connect_to_host (conn->host, conn->port);
1437 return (retryable_socket_connect_error (errno)
1438 ? CONERROR : CONIMPOSSIBLE);
1442 if (proxy && u->scheme == SCHEME_HTTPS)
1444 /* When requesting SSL URLs through proxies, use the
1445 CONNECT method to request passthrough. */
1446 struct request *connreq = request_new ();
1447 request_set_method (connreq, "CONNECT",
1448 aprintf ("%s:%d", u->host, u->port));
1449 SET_USER_AGENT (connreq);
1452 request_set_header (connreq, "Proxy-Authorization",
1453 proxyauth, rel_value);
1454 /* Now that PROXYAUTH is part of the CONNECT request,
1455 zero it out so we don't send proxy authorization with
1456 the regular request below. */
1459 /* Examples in rfc2817 use the Host header in CONNECT
1460 requests. I don't see how that gains anything, given
1461 that the contents of Host would be exactly the same as
1462 the contents of CONNECT. */
1464 write_error = request_send (connreq, sock);
1465 request_free (connreq);
1466 if (write_error < 0)
1468 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1470 CLOSE_INVALIDATE (sock);
1474 head = read_http_response_head (sock);
1477 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1479 CLOSE_INVALIDATE (sock);
1488 DEBUGP (("proxy responded with: [%s]\n", head));
1490 resp = resp_new (head);
1491 statcode = resp_status (resp, &message);
1494 if (statcode != 200)
1497 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1498 message ? escnonprint (message) : "?");
1499 xfree_null (message);
1502 xfree_null (message);
1504 /* SOCK is now *really* connected to u->host, so update CONN
1505 to reflect this. That way register_persistent will
1506 register SOCK as being connected to u->host:u->port. */
1510 if (conn->scheme == SCHEME_HTTPS)
1512 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1519 #endif /* HAVE_SSL */
1522 /* Send the request to server. */
1523 write_error = request_send (req, sock);
1525 if (write_error >= 0)
1529 DEBUGP (("[POST data: %s]\n", opt.post_data));
1530 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1532 else if (opt.post_file_name && post_data_size != 0)
1533 write_error = post_file (sock, opt.post_file_name, post_data_size);
1536 if (write_error < 0)
1538 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1540 CLOSE_INVALIDATE (sock);
1544 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1545 proxy ? "Proxy" : "HTTP");
1550 head = read_http_response_head (sock);
1555 logputs (LOG_NOTQUIET, _("No data received.\n"));
1556 CLOSE_INVALIDATE (sock);
1562 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1564 CLOSE_INVALIDATE (sock);
1569 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1571 resp = resp_new (head);
1573 /* Check for status line. */
1575 statcode = resp_status (resp, &message);
1576 if (!opt.server_response)
1577 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1578 message ? escnonprint (message) : "");
1581 logprintf (LOG_VERBOSE, "\n");
1582 print_server_response (resp, " ");
1585 if (!opt.ignore_length
1586 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1590 parsed = str_to_wgint (hdrval, NULL, 10);
1591 if (parsed == WGINT_MAX && errno == ERANGE)
1593 #### If Content-Length is out of range, it most likely
1594 means that the file is larger than 2G and that we're
1595 compiled without LFS. In that case we should probably
1596 refuse to even attempt to download the file. */
1602 /* Check for keep-alive related responses. */
1603 if (!inhibit_keep_alive && contlen != -1)
1605 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1607 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1609 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1614 /* The server has promised that it will not close the connection
1615 when we're done. This means that we can register it. */
1616 register_persistent (conn->host, conn->port, sock, using_ssl);
1618 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1620 /* Authorization is required. */
1621 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1622 CLOSE_FINISH (sock);
1624 CLOSE_INVALIDATE (sock);
1625 pconn.authorized = 0;
1626 if (!auth_finished && (user && passwd))
1628 /* IIS sends multiple copies of WWW-Authenticate, one with
1629 the value "negotiate", and other(s) with data. Loop over
1630 all the occurrences and pick the one we recognize. */
1632 const char *wabeg, *waend;
1633 char *www_authenticate = NULL;
1635 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1636 &wabeg, &waend)) != -1;
1638 if (known_authentication_scheme_p (wabeg, waend))
1640 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1644 if (!www_authenticate)
1645 /* If the authentication header is missing or
1646 unrecognized, there's no sense in retrying. */
1647 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1648 else if (BEGINS_WITH (www_authenticate, "Basic"))
1649 /* If the authentication scheme is "Basic", which we send
1650 by default, there's no sense in retrying either. (This
1651 should be changed when we stop sending "Basic" data by
1657 pth = url_full_path (u);
1658 request_set_header (req, "Authorization",
1659 create_authorization_line (www_authenticate,
1661 request_method (req),
1665 if (BEGINS_WITH (www_authenticate, "NTLM"))
1668 goto retry_with_auth;
1671 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1675 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1677 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1679 pconn.authorized = 1;
1683 hs->statcode = statcode;
1685 hs->error = xstrdup (_("Malformed status line"));
1687 hs->error = xstrdup (_("(no description)"));
1689 hs->error = xstrdup (message);
1692 type = resp_header_strdup (resp, "Content-Type");
1695 char *tmp = strchr (type, ';');
1698 while (tmp > type && ISSPACE (tmp[-1]))
1703 hs->newloc = resp_header_strdup (resp, "Location");
1704 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1706 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1710 const char *scbeg, *scend;
1711 /* The jar should have been created by now. */
1712 assert (wget_cookie_jar != NULL);
1714 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1715 &scbeg, &scend)) != -1;
1718 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1721 /* u->path doesn't begin with /, which cookies.c expects. */
1722 pth = (char *) alloca (1 + strlen (u->path) + 1);
1724 strcpy (pth + 1, u->path);
1726 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, pth,
1731 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1733 wgint first_byte_pos, last_byte_pos, entity_length;
1734 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1736 contrange = first_byte_pos;
1740 /* 20x responses are counted among successful by default. */
1741 if (H_20X (statcode))
1744 /* Return if redirected. */
1745 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1747 /* RFC2068 says that in case of the 300 (multiple choices)
1748 response, the server can output a preferred URL through
1749 `Location' header; otherwise, the request should be treated
1750 like GET. So, if the location is set, it will be a
1751 redirection; otherwise, just proceed normally. */
1752 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1756 logprintf (LOG_VERBOSE,
1757 _("Location: %s%s\n"),
1758 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1759 hs->newloc ? _(" [following]") : "");
1760 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1761 CLOSE_FINISH (sock);
1763 CLOSE_INVALIDATE (sock);
1769 /* If content-type is not given, assume text/html. This is because
1770 of the multitude of broken CGI's that "forget" to generate the
1773 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1774 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1779 if (opt.html_extension && (*dt & TEXTHTML))
1780 /* -E / --html-extension / html_extension = on was specified, and this is a
1781 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1782 already the file's suffix, tack on ".html". */
1784 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1786 if (last_period_in_local_filename == NULL
1787 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1788 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1790 int local_filename_len = strlen (*hs->local_file);
1791 /* Resize the local file, allowing for ".html" preceded by
1792 optional ".NUMBER". */
1793 *hs->local_file = xrealloc (*hs->local_file,
1794 local_filename_len + 24 + sizeof (".html"));
1795 strcpy(*hs->local_file + local_filename_len, ".html");
1796 /* If clobbering is not allowed and the file, as named,
1797 exists, tack on ".NUMBER.html" instead. */
1802 sprintf (*hs->local_file + local_filename_len,
1803 ".%d.html", ext_num++);
1804 while (file_exists_p (*hs->local_file));
1806 *dt |= ADDED_HTML_EXTENSION;
1810 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1812 /* If `-c' is in use and the file has been fully downloaded (or
1813 the remote file has shrunk), Wget effectively requests bytes
1814 after the end of file and the server response with 416. */
1815 logputs (LOG_VERBOSE, _("\
1816 \n The file is already fully retrieved; nothing to do.\n\n"));
1817 /* In case the caller inspects. */
1820 /* Mark as successfully retrieved. */
1823 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1824 might be more bytes in the body. */
1825 return RETRUNNEEDED;
1827 if ((contrange != 0 && contrange != hs->restval)
1828 || (H_PARTIAL (statcode) && !contrange))
1830 /* The Range request was somehow misunderstood by the server.
1833 CLOSE_INVALIDATE (sock);
1836 hs->contlen = contlen + contrange;
1842 /* No need to print this output if the body won't be
1843 downloaded at all, or if the original server response is
1845 logputs (LOG_VERBOSE, _("Length: "));
1848 logputs (LOG_VERBOSE, with_thousand_seps (contlen + contrange));
1849 if (contlen + contrange >= 1024)
1850 logprintf (LOG_VERBOSE, " (%s)",
1851 human_readable (contlen + contrange));
1854 if (contlen >= 1024)
1855 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1856 with_thousand_seps (contlen),
1857 human_readable (contlen));
1859 logprintf (LOG_VERBOSE, _(", %s remaining"),
1860 with_thousand_seps (contlen));
1864 logputs (LOG_VERBOSE,
1865 opt.ignore_length ? _("ignored") : _("unspecified"));
1867 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1869 logputs (LOG_VERBOSE, "\n");
1873 type = NULL; /* We don't need it any more. */
1875 /* Return if we have no intention of further downloading. */
1876 if (!(*dt & RETROKF) || head_only)
1878 /* In case the caller cares to look... */
1882 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1883 servers not to send body in response to a HEAD request. If
1884 you encounter such a server (more likely a broken CGI), use
1885 `--no-http-keep-alive'. */
1886 CLOSE_FINISH (sock);
1887 return RETRFINISHED;
1890 /* Open the local file. */
1893 mkalldirs (*hs->local_file);
1895 rotate_backups (*hs->local_file);
1897 fp = fopen (*hs->local_file, "ab");
1898 else if (ALLOW_CLOBBER)
1899 fp = fopen (*hs->local_file, "wb");
1902 fp = fopen_excl (*hs->local_file, 1);
1903 if (!fp && errno == EEXIST)
1905 /* We cannot just invent a new name and use it (which is
1906 what functions like unique_create typically do)
1907 because we told the user we'd use this name.
1908 Instead, return and retry the download. */
1909 logprintf (LOG_NOTQUIET,
1910 _("%s has sprung into existence.\n"),
1912 CLOSE_INVALIDATE (sock);
1913 return FOPEN_EXCL_ERR;
1918 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1919 CLOSE_INVALIDATE (sock);
1926 /* #### This confuses the timestamping code that checks for file
1927 size. Maybe we should save some additional information? */
1928 if (opt.save_headers)
1929 fwrite (head, 1, strlen (head), fp);
1931 /* Now we no longer need to store the response header. */
1934 /* Download the request body. */
1937 flags |= rb_read_exactly;
1938 if (hs->restval > 0 && contrange == 0)
1939 /* If the server ignored our range request, instruct fd_read_body
1940 to skip the first RESTVAL bytes of body. */
1941 flags |= rb_skip_startpos;
1942 hs->len = hs->restval;
1944 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1945 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1949 CLOSE_FINISH (sock);
1951 CLOSE_INVALIDATE (sock);
1954 /* Close or flush the file. We have to be careful to check for
1955 error here. Checking the result of fwrite() is not enough --
1956 errors could go unnoticed! */
1959 flush_res = fclose (fp);
1961 flush_res = fflush (fp);
1962 if (flush_res == EOF)
1967 return RETRFINISHED;
1970 /* The genuine HTTP loop! This is the part where the retrieval is
1971 retried, and retried, and retried, and... */
1973 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1974 int *dt, struct url *proxy)
1977 int use_ts, got_head = 0; /* time-stamping info */
1978 char *filename_plus_orig_suffix;
1979 char *local_filename = NULL;
1980 char *tms, *locf, *tmrate;
1982 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1983 wgint local_size = 0; /* the size of the local file */
1984 size_t filename_len;
1985 struct http_stat hstat; /* HTTP status */
1989 /* This used to be done in main(), but it's a better idea to do it
1990 here so that we don't go through the hoops if we're just using
1994 if (!wget_cookie_jar)
1995 wget_cookie_jar = cookie_jar_new ();
1996 if (opt.cookies_input && !cookies_loaded_p)
1998 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1999 cookies_loaded_p = 1;
2005 /* Warn on (likely bogus) wildcard usage in HTTP. */
2006 if (has_wildcards_p (u->path))
2007 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2011 /* Determine the local filename. */
2012 if (local_file && *local_file)
2013 hstat.local_file = local_file;
2014 else if (local_file && !opt.output_document)
2016 *local_file = url_file_name (u);
2017 hstat.local_file = local_file;
2021 dummy = url_file_name (u);
2022 hstat.local_file = &dummy;
2023 /* be honest about where we will save the file */
2024 if (local_file && opt.output_document)
2025 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2028 if (!opt.output_document)
2029 locf = *hstat.local_file;
2031 locf = opt.output_document;
2033 hstat.referer = referer;
2035 filename_len = strlen (*hstat.local_file);
2036 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2038 if (opt.noclobber && file_exists_p (*hstat.local_file))
2040 /* If opt.noclobber is turned on and file already exists, do not
2041 retrieve the file */
2042 logprintf (LOG_VERBOSE, _("\
2043 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2044 /* If the file is there, we suppose it's retrieved OK. */
2047 /* #### Bogusness alert. */
2048 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2049 if (has_html_suffix_p (*hstat.local_file))
2057 if (opt.timestamping)
2059 int local_dot_orig_file_exists = 0;
2061 if (opt.backup_converted)
2062 /* If -K is specified, we'll act on the assumption that it was specified
2063 last time these files were downloaded as well, and instead of just
2064 comparing local file X against server file X, we'll compare local
2065 file X.orig (if extant, else X) against server file X. If -K
2066 _wasn't_ specified last time, or the server contains files called
2067 *.orig, -N will be back to not operating correctly with -k. */
2069 /* Would a single s[n]printf() call be faster? --dan
2071 Definitely not. sprintf() is horribly slow. It's a
2072 different question whether the difference between the two
2073 affects a program. Usually I'd say "no", but at one
2074 point I profiled Wget, and found that a measurable and
2075 non-negligible amount of time was lost calling sprintf()
2076 in url.c. Replacing sprintf with inline calls to
2077 strcpy() and number_to_string() made a difference.
2079 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2080 memcpy (filename_plus_orig_suffix + filename_len,
2081 ".orig", sizeof (".orig"));
2083 /* Try to stat() the .orig file. */
2084 if (stat (filename_plus_orig_suffix, &st) == 0)
2086 local_dot_orig_file_exists = 1;
2087 local_filename = filename_plus_orig_suffix;
2091 if (!local_dot_orig_file_exists)
2092 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2093 if (stat (*hstat.local_file, &st) == 0)
2094 local_filename = *hstat.local_file;
2096 if (local_filename != NULL)
2097 /* There was a local file, so we'll check later to see if the version
2098 the server has is the same version we already have, allowing us to
2104 /* Modification time granularity is 2 seconds for Windows, so
2105 increase local time by 1 second for later comparison. */
2108 local_size = st.st_size;
2112 /* Reset the counter. */
2118 /* Increment the pass counter. */
2120 sleep_between_retrievals (count);
2121 /* Get the current time string. */
2122 tms = time_str (NULL);
2123 /* Print fetch message, if opt.verbose. */
2126 char *hurl = url_string (u, 1);
2130 sprintf (tmp, _("(try:%2d)"), count);
2131 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2132 tms, hurl, tmp, locf);
2134 ws_changetitle (hurl);
2139 /* Default document type is empty. However, if spider mode is
2140 on or time-stamping is employed, HEAD_ONLY commands is
2141 encoded within *dt. */
2142 if (opt.spider || (use_ts && !got_head))
2147 /* Decide whether or not to restart. */
2149 && stat (locf, &st) == 0
2150 && S_ISREG (st.st_mode))
2151 /* When -c is used, continue from on-disk size. (Can't use
2152 hstat.len even if count>1 because we don't want a failed
2153 first attempt to clobber existing data.) */
2154 hstat.restval = st.st_size;
2156 /* otherwise, continue where the previous try left off */
2157 hstat.restval = hstat.len;
2161 /* Decide whether to send the no-cache directive. We send it in
2163 a) we're using a proxy, and we're past our first retrieval.
2164 Some proxies are notorious for caching incomplete data, so
2165 we require a fresh get.
2166 b) caching is explicitly inhibited. */
2167 if ((proxy && count > 1) /* a */
2168 || !opt.allow_cache /* b */
2170 *dt |= SEND_NOCACHE;
2172 *dt &= ~SEND_NOCACHE;
2174 /* Try fetching the document, or at least its head. */
2175 err = gethttp (u, &hstat, dt, proxy);
2177 /* It's unfortunate that wget determines the local filename before finding
2178 out the Content-Type of the file. Barring a major restructuring of the
2179 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2180 *hstat.local_file to tack on ".html". */
2181 if (!opt.output_document)
2182 locf = *hstat.local_file;
2185 tms = time_str (NULL);
2186 /* Get the new location (with or without the redirection). */
2188 *newloc = xstrdup (hstat.newloc);
2191 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2192 case CONERROR: case READERR: case WRITEFAILED:
2193 case RANGEERR: case FOPEN_EXCL_ERR:
2194 /* Non-fatal errors continue executing the loop, which will
2195 bring them to "while" statement at the end, to judge
2196 whether the number of tries was exceeded. */
2197 free_hstat (&hstat);
2198 printwhat (count, opt.ntry);
2199 if (err == FOPEN_EXCL_ERR)
2201 /* Re-determine the file name. */
2202 if (local_file && *local_file)
2204 xfree (*local_file);
2205 *local_file = url_file_name (u);
2206 hstat.local_file = local_file;
2211 dummy = url_file_name (u);
2212 hstat.local_file = &dummy;
2214 /* be honest about where we will save the file */
2215 if (local_file && opt.output_document)
2216 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2217 if (!opt.output_document)
2218 locf = *hstat.local_file;
2220 locf = opt.output_document;
2223 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2224 case SSLINITFAILED: case CONTNOTSUPPORTED:
2225 /* Fatal errors just return from the function. */
2226 free_hstat (&hstat);
2229 case FWRITEERR: case FOPENERR:
2230 /* Another fatal error. */
2231 logputs (LOG_VERBOSE, "\n");
2232 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2233 *hstat.local_file, strerror (errno));
2234 free_hstat (&hstat);
2238 /* Another fatal error. */
2239 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2240 free_hstat (&hstat);
2244 /* Return the new location to the caller. */
2247 logprintf (LOG_NOTQUIET,
2248 _("ERROR: Redirection (%d) without location.\n"),
2250 free_hstat (&hstat);
2254 free_hstat (&hstat);
2258 /* The file was already fully retrieved. */
2259 free_hstat (&hstat);
2263 /* Deal with you later. */
2266 /* All possibilities should have been exhausted. */
2269 if (!(*dt & RETROKF))
2273 /* #### Ugly ugly ugly! */
2274 char *hurl = url_string (u, 1);
2275 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2278 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2279 tms, hstat.statcode, escnonprint (hstat.error));
2280 logputs (LOG_VERBOSE, "\n");
2281 free_hstat (&hstat);
2286 /* Did we get the time-stamp? */
2289 if (opt.timestamping && !hstat.remote_time)
2291 logputs (LOG_NOTQUIET, _("\
2292 Last-modified header missing -- time-stamps turned off.\n"));
2294 else if (hstat.remote_time)
2296 /* Convert the date-string into struct tm. */
2297 tmr = http_atotm (hstat.remote_time);
2298 if (tmr == (time_t) (-1))
2299 logputs (LOG_VERBOSE, _("\
2300 Last-modified header invalid -- time-stamp ignored.\n"));
2304 /* The time-stamping section. */
2309 use_ts = 0; /* no more time-stamping */
2310 count = 0; /* the retrieve count for HEAD is
2312 if (hstat.remote_time && tmr != (time_t) (-1))
2314 /* Now time-stamping can be used validly. Time-stamping
2315 means that if the sizes of the local and remote file
2316 match, and local file is newer than the remote file,
2317 it will not be retrieved. Otherwise, the normal
2318 download procedure is resumed. */
2320 (hstat.contlen == -1 || local_size == hstat.contlen))
2322 logprintf (LOG_VERBOSE, _("\
2323 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2325 free_hstat (&hstat);
2329 else if (tml >= tmr)
2330 logprintf (LOG_VERBOSE, _("\
2331 The sizes do not match (local %s) -- retrieving.\n"),
2332 number_to_static_string (local_size));
2334 logputs (LOG_VERBOSE,
2335 _("Remote file is newer, retrieving.\n"));
2337 free_hstat (&hstat);
2340 if ((tmr != (time_t) (-1))
2342 && ((hstat.len == hstat.contlen) ||
2343 ((hstat.res == 0) &&
2344 ((hstat.contlen == -1) ||
2345 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2347 /* #### This code repeats in http.c and ftp.c. Move it to a
2349 const char *fl = NULL;
2350 if (opt.output_document)
2352 if (output_stream_regular)
2353 fl = opt.output_document;
2356 fl = *hstat.local_file;
2360 /* End of time-stamping section. */
2364 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2365 escnonprint (hstat.error));
2370 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2372 if (hstat.len == hstat.contlen)
2376 logprintf (LOG_VERBOSE,
2377 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2379 number_to_static_string (hstat.len),
2380 number_to_static_string (hstat.contlen));
2381 logprintf (LOG_NONVERBOSE,
2382 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2384 number_to_static_string (hstat.len),
2385 number_to_static_string (hstat.contlen),
2389 total_downloaded_bytes += hstat.len;
2391 /* Remember that we downloaded the file for later ".orig" code. */
2392 if (*dt & ADDED_HTML_EXTENSION)
2393 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2395 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2397 free_hstat (&hstat);
2401 else if (hstat.res == 0) /* No read error */
2403 if (hstat.contlen == -1) /* We don't know how much we were supposed
2404 to get, so assume we succeeded. */
2408 logprintf (LOG_VERBOSE,
2409 _("%s (%s) - `%s' saved [%s]\n\n"),
2411 number_to_static_string (hstat.len));
2412 logprintf (LOG_NONVERBOSE,
2413 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2414 tms, u->url, number_to_static_string (hstat.len),
2418 total_downloaded_bytes += hstat.len;
2420 /* Remember that we downloaded the file for later ".orig" code. */
2421 if (*dt & ADDED_HTML_EXTENSION)
2422 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2424 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2426 free_hstat (&hstat);
2430 else if (hstat.len < hstat.contlen) /* meaning we lost the
2431 connection too soon */
2433 logprintf (LOG_VERBOSE,
2434 _("%s (%s) - Connection closed at byte %s. "),
2435 tms, tmrate, number_to_static_string (hstat.len));
2436 printwhat (count, opt.ntry);
2437 free_hstat (&hstat);
2440 else if (!opt.kill_longer) /* meaning we got more than expected */
2442 logprintf (LOG_VERBOSE,
2443 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2445 number_to_static_string (hstat.len),
2446 number_to_static_string (hstat.contlen));
2447 logprintf (LOG_NONVERBOSE,
2448 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2450 number_to_static_string (hstat.len),
2451 number_to_static_string (hstat.contlen),
2454 total_downloaded_bytes += hstat.len;
2456 /* Remember that we downloaded the file for later ".orig" code. */
2457 if (*dt & ADDED_HTML_EXTENSION)
2458 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2460 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2462 free_hstat (&hstat);
2466 else /* the same, but not accepted */
2468 logprintf (LOG_VERBOSE,
2469 _("%s (%s) - Connection closed at byte %s/%s. "),
2471 number_to_static_string (hstat.len),
2472 number_to_static_string (hstat.contlen));
2473 printwhat (count, opt.ntry);
2474 free_hstat (&hstat);
2478 else /* now hstat.res can only be -1 */
2480 if (hstat.contlen == -1)
2482 logprintf (LOG_VERBOSE,
2483 _("%s (%s) - Read error at byte %s (%s)."),
2484 tms, tmrate, number_to_static_string (hstat.len),
2486 printwhat (count, opt.ntry);
2487 free_hstat (&hstat);
2490 else /* hstat.res == -1 and contlen is given */
2492 logprintf (LOG_VERBOSE,
2493 _("%s (%s) - Read error at byte %s/%s (%s). "),
2495 number_to_static_string (hstat.len),
2496 number_to_static_string (hstat.contlen),
2498 printwhat (count, opt.ntry);
2499 free_hstat (&hstat);
2505 while (!opt.ntry || (count < opt.ntry));
2509 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2510 than local timezone.
2512 mktime is similar but assumes struct tm, also known as the
2513 "broken-down" form of time, is in local time zone. mktime_from_utc
2514 uses mktime to make the conversion understanding that an offset
2515 will be introduced by the local time assumption.
2517 mktime_from_utc then measures the introduced offset by applying
2518 gmtime to the initial result and applying mktime to the resulting
2519 "broken-down" form. The difference between the two mktime results
2520 is the measured offset which is then subtracted from the initial
2521 mktime result to yield a calendar time which is the value returned.
2523 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2524 consistent offset (the non DST offset) since tm and tm+o might be
2525 on opposite sides of a DST change.
2527 Some implementations of mktime return -1 for the nonexistent
2528 localtime hour at the beginning of DST. In this event, use
2529 mktime(tm - 1hr) + 3600.
2533 gmtime(t+o) --> tm+o
2534 mktime(tm+o) --> t+2o
2535 t+o - (t+2o - t+o) = t
2537 Note that glibc contains a function of the same purpose named
2538 `timegm' (reverse of gmtime). But obviously, it is not universally
2539 available, and unfortunately it is not straightforwardly
2540 extractable for use here. Perhaps configure should detect timegm
2541 and use it where available.
2543 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2544 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2545 Further improved by Roger with assistance from Edward J. Sabol
2546 based on input by Jamie Zawinski. */
2549 mktime_from_utc (struct tm *t)
2560 return -1; /* can't deal with output from strptime */
2571 return -1; /* can't deal with output from gmtime */
2574 return (tl - (tb - tl));
2577 /* Check whether the result of strptime() indicates success.
2578 strptime() returns the pointer to how far it got to in the string.
2579 The processing has been successful if the string is at `GMT' or
2580 `+X', or at the end of the string.
2582 In extended regexp parlance, the function returns 1 if P matches
2583 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2584 can return) is considered a failure and 0 is returned. */
2586 check_end (const char *p)
2590 while (ISSPACE (*p))
2593 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2594 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2600 /* Convert the textual specification of time in TIME_STRING to the
2601 number of seconds since the Epoch.
2603 TIME_STRING can be in any of the three formats RFC2616 allows the
2604 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2605 as well as the time format used in the Set-Cookie header.
2606 Timezones are ignored, and should be GMT.
2608 Return the computed time_t representation, or -1 if the conversion
2611 This function uses strptime with various string formats for parsing
2612 TIME_STRING. This results in a parser that is not as lenient in
2613 interpreting TIME_STRING as I would like it to be. Being based on
2614 strptime, it always allows shortened months, one-digit days, etc.,
2615 but due to the multitude of formats in which time can be
2616 represented, an ideal HTTP time parser would be even more
2617 forgiving. It should completely ignore things like week days and
2618 concentrate only on the various forms of representing years,
2619 months, days, hours, minutes, and seconds. For example, it would
2620 be nice if it accepted ISO 8601 out of the box.
2622 I've investigated free and PD code for this purpose, but none was
2623 usable. getdate was big and unwieldy, and had potential copyright
2624 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2625 distributed with phttpd, is excellent, but we cannot use it because
2626 it is not assigned to the FSF. So I stuck it with strptime. */
2629 http_atotm (const char *time_string)
2631 /* NOTE: Solaris strptime man page claims that %n and %t match white
2632 space, but that's not universally available. Instead, we simply
2633 use ` ' to mean "skip all WS", which works under all strptime
2634 implementations I've tested. */
2636 static const char *time_formats[] = {
2637 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2638 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2639 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2640 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2641 (used in Set-Cookie, defined in the
2642 Netscape cookie specification.) */
2646 for (i = 0; i < countof (time_formats); i++)
2650 /* Some versions of strptime use the existing contents of struct
2651 tm to recalculate the date according to format. Zero it out
2652 to prevent garbage from the stack influencing strptime. */
2655 /* Solaris strptime fails to recognize English month names in
2656 non-English locales, which we work around by not setting the
2657 LC_TIME category. Another way would be to temporarily set
2658 locale to C before invoking strptime, but that's slow and
2659 messy. GNU strptime does not have this problem because it
2660 recognizes English month names along with the local ones. */
2662 if (check_end (strptime (time_string, time_formats[i], &t)))
2663 return mktime_from_utc (&t);
2666 /* All formats have failed. */
2670 /* Authorization support: We support three authorization schemes:
2672 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2674 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2675 consisting of answering to the server's challenge with the proper
2678 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2679 Stenberg for libcurl. Like digest, NTLM is based on a
2680 challenge-response mechanism, but unlike digest, it is non-standard
2681 (authenticates TCP connections rather than requests), undocumented
2682 and Microsoft-specific. */
2684 /* Create the authentication header contents for the `Basic' scheme.
2685 This is done by encoding the string "USER:PASS" to base64 and
2686 prepending the string "Basic " in front of it. */
2689 basic_authentication_encode (const char *user, const char *passwd)
2692 int len1 = strlen (user) + 1 + strlen (passwd);
2694 t1 = (char *)alloca (len1 + 1);
2695 sprintf (t1, "%s:%s", user, passwd);
2697 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2698 base64_encode (t1, len1, t2);
2700 return concat_strings ("Basic ", t2, (char *) 0);
2703 #define SKIP_WS(x) do { \
2704 while (ISSPACE (*(x))) \
2708 #ifdef ENABLE_DIGEST
2709 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2710 of a field in such a header. If the field is the one specified by
2711 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2712 digest authorization code), extract its value in the (char*)
2713 variable pointed by RET. Returns negative on a malformed header,
2714 or number of bytes that have been parsed by this call. */
2716 extract_header_attr (const char *au, const char *attr_name, char **ret)
2719 const char *cp = au;
2721 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2723 cp += strlen (attr_name);
2736 for (ep = cp; *ep && *ep != '\"'; ep++)
2741 *ret = strdupdelim (cp, ep);
2748 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2749 an array of 16 bytes containing the hash keys, and BUF should be a
2750 buffer of 33 writable characters (32 for hex digits plus one for
2751 zero termination). */
2753 dump_hash (unsigned char *buf, const unsigned char *hash)
2757 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2759 *buf++ = XNUM_TO_digit (*hash >> 4);
2760 *buf++ = XNUM_TO_digit (*hash & 0xf);
2765 /* Take the line apart to find the challenge, and compose a digest
2766 authorization header. See RFC2069 section 2.1.2. */
2768 digest_authentication_encode (const char *au, const char *user,
2769 const char *passwd, const char *method,
2772 static char *realm, *opaque, *nonce;
2777 { "realm", &realm },
2778 { "opaque", &opaque },
2783 realm = opaque = nonce = NULL;
2785 au += 6; /* skip over `Digest' */
2791 for (i = 0; i < countof (options); i++)
2793 int skip = extract_header_attr (au, options[i].name,
2794 options[i].variable);
2798 xfree_null (opaque);
2808 if (i == countof (options))
2810 while (*au && *au != '=')
2818 while (*au && *au != '\"')
2825 while (*au && *au != ',')
2830 if (!realm || !nonce || !user || !passwd || !path || !method)
2833 xfree_null (opaque);
2838 /* Calculate the digest value. */
2840 ALLOCA_MD5_CONTEXT (ctx);
2841 unsigned char hash[MD5_HASHLEN];
2842 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2843 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2845 /* A1BUF = H(user ":" realm ":" password) */
2847 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2848 gen_md5_update ((unsigned char *)":", 1, ctx);
2849 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2850 gen_md5_update ((unsigned char *)":", 1, ctx);
2851 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2852 gen_md5_finish (ctx, hash);
2853 dump_hash (a1buf, hash);
2855 /* A2BUF = H(method ":" path) */
2857 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2858 gen_md5_update ((unsigned char *)":", 1, ctx);
2859 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2860 gen_md5_finish (ctx, hash);
2861 dump_hash (a2buf, hash);
2863 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2865 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2866 gen_md5_update ((unsigned char *)":", 1, ctx);
2867 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2868 gen_md5_update ((unsigned char *)":", 1, ctx);
2869 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2870 gen_md5_finish (ctx, hash);
2871 dump_hash (response_digest, hash);
2873 res = xmalloc (strlen (user)
2878 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2879 + (opaque ? strlen (opaque) : 0)
2881 sprintf (res, "Digest \
2882 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2883 user, realm, nonce, path, response_digest);
2886 char *p = res + strlen (res);
2887 strcat (p, ", opaque=\"");
2894 #endif /* ENABLE_DIGEST */
2896 /* Computing the size of a string literal must take into account that
2897 value returned by sizeof includes the terminating \0. */
2898 #define STRSIZE(literal) (sizeof (literal) - 1)
2900 /* Whether chars in [b, e) begin with the literal string provided as
2901 first argument and are followed by whitespace or terminating \0.
2902 The comparison is case-insensitive. */
2903 #define STARTS(literal, b, e) \
2904 ((e) - (b) >= STRSIZE (literal) \
2905 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2906 && ((e) - (b) == STRSIZE (literal) \
2907 || ISSPACE (b[STRSIZE (literal)])))
2910 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2912 return STARTS ("Basic", hdrbeg, hdrend)
2913 #ifdef ENABLE_DIGEST
2914 || STARTS ("Digest", hdrbeg, hdrend)
2917 || STARTS ("NTLM", hdrbeg, hdrend)
2924 /* Create the HTTP authorization request header. When the
2925 `WWW-Authenticate' response header is seen, according to the
2926 authorization scheme specified in that header (`Basic' and `Digest'
2927 are supported by the current implementation), produce an
2928 appropriate HTTP authorization request header. */
2930 create_authorization_line (const char *au, const char *user,
2931 const char *passwd, const char *method,
2932 const char *path, int *finished)
2934 /* We are called only with known schemes, so we can dispatch on the
2936 switch (TOUPPER (*au))
2938 case 'B': /* Basic */
2940 return basic_authentication_encode (user, passwd);
2941 #ifdef ENABLE_DIGEST
2942 case 'D': /* Digest */
2944 return digest_authentication_encode (au, user, passwd, method, path);
2947 case 'N': /* NTLM */
2948 if (!ntlm_input (&pconn.ntlm, au))
2953 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2956 /* We shouldn't get here -- this function should be only called
2957 with values approved by known_authentication_scheme_p. */
2965 if (wget_cookie_jar)
2966 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2972 xfree_null (pconn.host);
2973 if (wget_cookie_jar)
2974 cookie_jar_delete (wget_cookie_jar);