2 * xdr_rec.c, Implements TCP/IP based XDR streams with a "record marking"
3 * layer above tcp (for rpc's use).
5 * Copyright (c) 2010, Oracle America, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer in the documentation and/or other materials
16 * provided with the distribution.
17 * * Neither the name of the "Oracle America, Inc." nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
26 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
28 * GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
31 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 * These routines interface XDRSTREAMS to a tcp/ip connection.
35 * There is a record marking layer between the xdr stream
36 * and the tcp transport level. A record is composed on one or more
37 * record fragments. A record fragment is a thirty-two bit header followed
38 * by n bytes of data, where n is contained in the header. The header
39 * is represented as a htonl(u_long). The high order bit encodes
40 * whether or not the fragment is the last fragment of the record
41 * (1 => fragment is last, 0 => more fragments to follow.
42 * The other 31 bits encode the byte length of the fragment.
52 #include <libio/iolibio.h>
54 static bool_t
xdrrec_getlong (XDR
*, long *);
55 static bool_t
xdrrec_putlong (XDR
*, const long *);
56 static bool_t
xdrrec_getbytes (XDR
*, caddr_t
, u_int
);
57 static bool_t
xdrrec_putbytes (XDR
*, const char *, u_int
);
58 static u_int
xdrrec_getpos (const XDR
*);
59 static bool_t
xdrrec_setpos (XDR
*, u_int
);
60 static int32_t *xdrrec_inline (XDR
*, u_int
);
61 static void xdrrec_destroy (XDR
*);
62 static bool_t
xdrrec_getint32 (XDR
*, int32_t *);
63 static bool_t
xdrrec_putint32 (XDR
*, const int32_t *);
65 static const struct xdr_ops xdrrec_ops
= {
79 * A record is composed of one or more record fragments.
80 * A record fragment is a two-byte header followed by zero to
81 * 2**32-1 bytes. The header is treated as a long unsigned and is
82 * encode/decoded to the network via htonl/ntohl. The low order 31 bits
83 * are a byte count of the fragment. The highest order bit is a boolean:
84 * 1 => this fragment is the last fragment of the record,
85 * 0 => this fragment is followed by more fragment(s).
87 * The fragment/record machinery is not general; it is constructed to
88 * meet the needs of xdr and rpc based on tcp.
91 #define LAST_FRAG (1UL << 31)
93 typedef struct rec_strm
100 int (*writeit
) (char *, char *, int);
101 caddr_t out_base
; /* output buffer (points to frag header) */
102 caddr_t out_finger
; /* next output position */
103 caddr_t out_boundry
; /* data cannot up to this address */
104 u_int32_t
*frag_header
; /* beginning of curren fragment */
105 bool_t frag_sent
; /* true if buffer sent in middle of record */
109 int (*readit
) (char *, char *, int);
110 u_long in_size
; /* fixed size of the input buffer */
112 caddr_t in_finger
; /* location of next byte to be had */
113 caddr_t in_boundry
; /* can read up to this location */
114 long fbtbc
; /* fragment bytes to be consumed */
121 static u_int
fix_buf_size (u_int
) internal_function
;
122 static bool_t
skip_input_bytes (RECSTREAM
*, long) internal_function
;
123 static bool_t
flush_out (RECSTREAM
*, bool_t
) internal_function
;
124 static bool_t
set_input_fragment (RECSTREAM
*) internal_function
;
125 static bool_t
get_input_bytes (RECSTREAM
*, caddr_t
, int) internal_function
;
128 * Create an xdr handle for xdrrec
129 * xdrrec_create fills in xdrs. Sendsize and recvsize are
130 * send and recv buffer sizes (0 => use default).
131 * tcp_handle is an opaque handle that is passed as the first parameter to
132 * the procedures readit and writeit. Readit and writeit are read and
133 * write respectively. They are like the system
134 * calls expect that they take an opaque handle rather than an fd.
137 xdrrec_create (XDR
*xdrs
, u_int sendsize
,
138 u_int recvsize
, caddr_t tcp_handle
,
139 int (*readit
) (char *, char *, int),
140 int (*writeit
) (char *, char *, int))
142 RECSTREAM
*rstrm
= (RECSTREAM
*) mem_alloc (sizeof (RECSTREAM
));
146 sendsize
= fix_buf_size (sendsize
);
147 recvsize
= fix_buf_size (recvsize
);
148 buf
= mem_alloc (sendsize
+ recvsize
+ BYTES_PER_XDR_UNIT
);
150 if (rstrm
== NULL
|| buf
== NULL
)
152 (void) __fxprintf (NULL
, "%s: %s", __func__
, _("out of memory\n"));
153 mem_free (rstrm
, sizeof (RECSTREAM
));
154 mem_free (buf
, sendsize
+ recvsize
+ BYTES_PER_XDR_UNIT
);
156 * This is bad. Should rework xdrrec_create to
157 * return a handle, and in this case return NULL
162 * adjust sizes and allocate buffer quad byte aligned
164 rstrm
->sendsize
= sendsize
;
165 rstrm
->recvsize
= recvsize
;
166 rstrm
->the_buffer
= buf
;
167 tmp
= rstrm
->the_buffer
;
168 if ((size_t)tmp
% BYTES_PER_XDR_UNIT
)
169 tmp
+= BYTES_PER_XDR_UNIT
- (size_t)tmp
% BYTES_PER_XDR_UNIT
;
170 rstrm
->out_base
= tmp
;
171 rstrm
->in_base
= tmp
+ sendsize
;
175 /* We have to add the cast since the `struct xdr_ops' in `struct XDR'
177 xdrs
->x_ops
= (struct xdr_ops
*) &xdrrec_ops
;
178 xdrs
->x_private
= (caddr_t
) rstrm
;
179 rstrm
->tcp_handle
= tcp_handle
;
180 rstrm
->readit
= readit
;
181 rstrm
->writeit
= writeit
;
182 rstrm
->out_finger
= rstrm
->out_boundry
= rstrm
->out_base
;
183 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_base
;
184 rstrm
->out_finger
+= 4;
185 rstrm
->out_boundry
+= sendsize
;
186 rstrm
->frag_sent
= FALSE
;
187 rstrm
->in_size
= recvsize
;
188 rstrm
->in_boundry
= rstrm
->in_base
;
189 rstrm
->in_finger
= (rstrm
->in_boundry
+= recvsize
);
191 rstrm
->last_frag
= TRUE
;
193 libc_hidden_nolink_sunrpc (xdrrec_create
, GLIBC_2_0
)
197 * The routines defined below are the xdr ops which will go into the
198 * xdr handle filled in by xdrrec_create.
202 xdrrec_getlong (XDR
*xdrs
, long *lp
)
204 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
205 int32_t *buflp
= (int32_t *) rstrm
->in_finger
;
208 /* first try the inline, fast case */
209 if (rstrm
->fbtbc
>= BYTES_PER_XDR_UNIT
&&
210 rstrm
->in_boundry
- (char *) buflp
>= BYTES_PER_XDR_UNIT
)
212 *lp
= (int32_t) ntohl (*buflp
);
213 rstrm
->fbtbc
-= BYTES_PER_XDR_UNIT
;
214 rstrm
->in_finger
+= BYTES_PER_XDR_UNIT
;
218 if (!xdrrec_getbytes (xdrs
, (caddr_t
) & mylong
,
221 *lp
= (int32_t) ntohl (mylong
);
227 xdrrec_putlong (XDR
*xdrs
, const long *lp
)
229 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
230 int32_t *dest_lp
= (int32_t *) rstrm
->out_finger
;
232 if ((rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
) > rstrm
->out_boundry
)
235 * this case should almost never happen so the code is
238 rstrm
->out_finger
-= BYTES_PER_XDR_UNIT
;
239 rstrm
->frag_sent
= TRUE
;
240 if (!flush_out (rstrm
, FALSE
))
242 dest_lp
= (int32_t *) rstrm
->out_finger
;
243 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
245 *dest_lp
= htonl (*lp
);
249 static bool_t
/* must manage buffers, fragments, and records */
250 xdrrec_getbytes (XDR
*xdrs
, caddr_t addr
, u_int len
)
252 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
257 current
= rstrm
->fbtbc
;
260 if (rstrm
->last_frag
)
262 if (!set_input_fragment (rstrm
))
266 current
= (len
< current
) ? len
: current
;
267 if (!get_input_bytes (rstrm
, addr
, current
))
270 rstrm
->fbtbc
-= current
;
277 xdrrec_putbytes (XDR
*xdrs
, const char *addr
, u_int len
)
279 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
284 current
= rstrm
->out_boundry
- rstrm
->out_finger
;
285 current
= (len
< current
) ? len
: current
;
286 memcpy (rstrm
->out_finger
, addr
, current
);
287 rstrm
->out_finger
+= current
;
290 if (rstrm
->out_finger
== rstrm
->out_boundry
&& len
> 0)
292 rstrm
->frag_sent
= TRUE
;
293 if (!flush_out (rstrm
, FALSE
))
301 xdrrec_getpos (const XDR
*xdrs
)
303 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
306 pos
= __lseek ((int) (long) rstrm
->tcp_handle
, (long) 0, 1);
312 pos
+= rstrm
->out_finger
- rstrm
->out_base
;
316 pos
-= rstrm
->in_boundry
- rstrm
->in_finger
;
327 xdrrec_setpos (XDR
*xdrs
, u_int pos
)
329 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
330 u_int currpos
= xdrrec_getpos (xdrs
);
331 int delta
= currpos
- pos
;
334 if ((int) currpos
!= -1)
339 newpos
= rstrm
->out_finger
- delta
;
340 if (newpos
> (caddr_t
) rstrm
->frag_header
&&
341 newpos
< rstrm
->out_boundry
)
343 rstrm
->out_finger
= newpos
;
349 newpos
= rstrm
->in_finger
- delta
;
350 if ((delta
< (int) (rstrm
->fbtbc
)) &&
351 (newpos
<= rstrm
->in_boundry
) &&
352 (newpos
>= rstrm
->in_base
))
354 rstrm
->in_finger
= newpos
;
355 rstrm
->fbtbc
-= delta
;
367 xdrrec_inline (XDR
*xdrs
, u_int len
)
369 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
376 if ((rstrm
->out_finger
+ len
) <= rstrm
->out_boundry
)
378 buf
= (int32_t *) rstrm
->out_finger
;
379 rstrm
->out_finger
+= len
;
384 if ((len
<= rstrm
->fbtbc
) &&
385 ((rstrm
->in_finger
+ len
) <= rstrm
->in_boundry
))
387 buf
= (int32_t *) rstrm
->in_finger
;
389 rstrm
->in_finger
+= len
;
400 xdrrec_destroy (XDR
*xdrs
)
402 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
404 mem_free (rstrm
->the_buffer
,
405 rstrm
->sendsize
+ rstrm
->recvsize
+ BYTES_PER_XDR_UNIT
);
406 mem_free ((caddr_t
) rstrm
, sizeof (RECSTREAM
));
410 xdrrec_getint32 (XDR
*xdrs
, int32_t *ip
)
412 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
413 int32_t *bufip
= (int32_t *) rstrm
->in_finger
;
416 /* first try the inline, fast case */
417 if (rstrm
->fbtbc
>= BYTES_PER_XDR_UNIT
&&
418 rstrm
->in_boundry
- (char *) bufip
>= BYTES_PER_XDR_UNIT
)
420 *ip
= ntohl (*bufip
);
421 rstrm
->fbtbc
-= BYTES_PER_XDR_UNIT
;
422 rstrm
->in_finger
+= BYTES_PER_XDR_UNIT
;
426 if (!xdrrec_getbytes (xdrs
, (caddr_t
) &mylong
,
429 *ip
= ntohl (mylong
);
435 xdrrec_putint32 (XDR
*xdrs
, const int32_t *ip
)
437 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
438 int32_t *dest_ip
= (int32_t *) rstrm
->out_finger
;
440 if ((rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
) > rstrm
->out_boundry
)
443 * this case should almost never happen so the code is
446 rstrm
->out_finger
-= BYTES_PER_XDR_UNIT
;
447 rstrm
->frag_sent
= TRUE
;
448 if (!flush_out (rstrm
, FALSE
))
450 dest_ip
= (int32_t *) rstrm
->out_finger
;
451 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
453 *dest_ip
= htonl (*ip
);
458 * Exported routines to manage xdr records
462 * Before reading (deserializing from the stream, one should always call
463 * this procedure to guarantee proper record alignment.
466 xdrrec_skiprecord (XDR
*xdrs
)
468 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
470 while (rstrm
->fbtbc
> 0 || (!rstrm
->last_frag
))
472 if (!skip_input_bytes (rstrm
, rstrm
->fbtbc
))
475 if ((!rstrm
->last_frag
) && (!set_input_fragment (rstrm
)))
478 rstrm
->last_frag
= FALSE
;
481 libc_hidden_nolink_sunrpc (xdrrec_skiprecord
, GLIBC_2_0
)
484 * Lookahead function.
485 * Returns TRUE iff there is no more input in the buffer
486 * after consuming the rest of the current record.
489 xdrrec_eof (XDR
*xdrs
)
491 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
493 while (rstrm
->fbtbc
> 0 || (!rstrm
->last_frag
))
495 if (!skip_input_bytes (rstrm
, rstrm
->fbtbc
))
498 if ((!rstrm
->last_frag
) && (!set_input_fragment (rstrm
)))
501 if (rstrm
->in_finger
== rstrm
->in_boundry
)
505 libc_hidden_nolink_sunrpc (xdrrec_eof
, GLIBC_2_0
)
508 * The client must tell the package when an end-of-record has occurred.
509 * The second parameter tells whether the record should be flushed to the
510 * (output) tcp stream. (This lets the package support batched or
511 * pipelined procedure calls.) TRUE => immediate flush to tcp connection.
514 xdrrec_endofrecord (XDR
*xdrs
, bool_t sendnow
)
516 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
517 u_long len
; /* fragment length */
519 if (sendnow
|| rstrm
->frag_sent
520 || rstrm
->out_finger
+ BYTES_PER_XDR_UNIT
>= rstrm
->out_boundry
)
522 rstrm
->frag_sent
= FALSE
;
523 return flush_out (rstrm
, TRUE
);
525 len
= (rstrm
->out_finger
- (char *) rstrm
->frag_header
526 - BYTES_PER_XDR_UNIT
);
527 *rstrm
->frag_header
= htonl ((u_long
) len
| LAST_FRAG
);
528 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_finger
;
529 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
532 libc_hidden_nolink_sunrpc (xdrrec_endofrecord
, GLIBC_2_0
)
536 * Internal useful routines
540 flush_out (RECSTREAM
*rstrm
, bool_t eor
)
542 u_long eormask
= (eor
== TRUE
) ? LAST_FRAG
: 0;
543 u_long len
= (rstrm
->out_finger
- (char *) rstrm
->frag_header
544 - BYTES_PER_XDR_UNIT
);
546 *rstrm
->frag_header
= htonl (len
| eormask
);
547 len
= rstrm
->out_finger
- rstrm
->out_base
;
548 if ((*(rstrm
->writeit
)) (rstrm
->tcp_handle
, rstrm
->out_base
, (int) len
)
551 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_base
;
552 rstrm
->out_finger
= (caddr_t
) rstrm
->out_base
+ BYTES_PER_XDR_UNIT
;
556 static bool_t
/* knows nothing about records! Only about input buffers */
557 fill_input_buf (RECSTREAM
*rstrm
)
563 where
= rstrm
->in_base
;
564 i
= (size_t) rstrm
->in_boundry
% BYTES_PER_XDR_UNIT
;
566 len
= rstrm
->in_size
- i
;
567 if ((len
= (*(rstrm
->readit
)) (rstrm
->tcp_handle
, where
, len
)) == -1)
569 rstrm
->in_finger
= where
;
571 rstrm
->in_boundry
= where
;
575 static bool_t
/* knows nothing about records! Only about input buffers */
577 get_input_bytes (RECSTREAM
*rstrm
, caddr_t addr
, int len
)
583 current
= rstrm
->in_boundry
- rstrm
->in_finger
;
586 if (!fill_input_buf (rstrm
))
590 current
= (len
< current
) ? len
: current
;
591 memcpy (addr
, rstrm
->in_finger
, current
);
592 rstrm
->in_finger
+= current
;
599 static bool_t
/* next two bytes of the input stream are treated as a header */
601 set_input_fragment (RECSTREAM
*rstrm
)
605 if (! get_input_bytes (rstrm
, (caddr_t
)&header
, BYTES_PER_XDR_UNIT
))
607 header
= ntohl (header
);
608 rstrm
->last_frag
= ((header
& LAST_FRAG
) == 0) ? FALSE
: TRUE
;
610 * Sanity check. Try not to accept wildly incorrect fragment
611 * sizes. Unfortunately, only a size of zero can be identified as
612 * 'wildely incorrect', and this only, if it is not the last
613 * fragment of a message. Ridiculously large fragment sizes may look
614 * wrong, but we don't have any way to be certain that they aren't
615 * what the client actually intended to send us. Many existing RPC
616 * implementations may sent a fragment of size zero as the last
617 * fragment of a message.
621 rstrm
->fbtbc
= header
& ~LAST_FRAG
;
625 static bool_t
/* consumes input bytes; knows nothing about records! */
627 skip_input_bytes (RECSTREAM
*rstrm
, long cnt
)
633 current
= rstrm
->in_boundry
- rstrm
->in_finger
;
636 if (!fill_input_buf (rstrm
))
640 current
= (cnt
< current
) ? cnt
: current
;
641 rstrm
->in_finger
+= current
;
649 fix_buf_size (u_int s
)