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.
51 #include <libio/iolibio.h>
53 static bool_t
xdrrec_getlong (XDR
*, long *);
54 static bool_t
xdrrec_putlong (XDR
*, const long *);
55 static bool_t
xdrrec_getbytes (XDR
*, caddr_t
, u_int
);
56 static bool_t
xdrrec_putbytes (XDR
*, const char *, u_int
);
57 static u_int
xdrrec_getpos (const XDR
*);
58 static bool_t
xdrrec_setpos (XDR
*, u_int
);
59 static int32_t *xdrrec_inline (XDR
*, u_int
);
60 static void xdrrec_destroy (XDR
*);
61 static bool_t
xdrrec_getint32 (XDR
*, int32_t *);
62 static bool_t
xdrrec_putint32 (XDR
*, const int32_t *);
64 static const struct xdr_ops xdrrec_ops
= {
78 * A record is composed of one or more record fragments.
79 * A record fragment is a two-byte header followed by zero to
80 * 2**32-1 bytes. The header is treated as a long unsigned and is
81 * encode/decoded to the network via htonl/ntohl. The low order 31 bits
82 * are a byte count of the fragment. The highest order bit is a boolean:
83 * 1 => this fragment is the last fragment of the record,
84 * 0 => this fragment is followed by more fragment(s).
86 * The fragment/record machinery is not general; it is constructed to
87 * meet the needs of xdr and rpc based on tcp.
90 #define LAST_FRAG (1UL << 31)
92 typedef struct rec_strm
99 int (*writeit
) (char *, char *, int);
100 caddr_t out_base
; /* output buffer (points to frag header) */
101 caddr_t out_finger
; /* next output position */
102 caddr_t out_boundry
; /* data cannot up to this address */
103 u_int32_t
*frag_header
; /* beginning of curren fragment */
104 bool_t frag_sent
; /* true if buffer sent in middle of record */
108 int (*readit
) (char *, char *, int);
109 u_long in_size
; /* fixed size of the input buffer */
111 caddr_t in_finger
; /* location of next byte to be had */
112 caddr_t in_boundry
; /* can read up to this location */
113 long fbtbc
; /* fragment bytes to be consumed */
120 static u_int
fix_buf_size (u_int
) internal_function
;
121 static bool_t
skip_input_bytes (RECSTREAM
*, long) internal_function
;
122 static bool_t
flush_out (RECSTREAM
*, bool_t
) internal_function
;
123 static bool_t
set_input_fragment (RECSTREAM
*) internal_function
;
124 static bool_t
get_input_bytes (RECSTREAM
*, caddr_t
, int) internal_function
;
127 * Create an xdr handle for xdrrec
128 * xdrrec_create fills in xdrs. Sendsize and recvsize are
129 * send and recv buffer sizes (0 => use default).
130 * tcp_handle is an opaque handle that is passed as the first parameter to
131 * the procedures readit and writeit. Readit and writeit are read and
132 * write respectively. They are like the system
133 * calls expect that they take an opaque handle rather than an fd.
136 xdrrec_create (XDR
*xdrs
, u_int sendsize
,
137 u_int recvsize
, caddr_t tcp_handle
,
138 int (*readit
) (char *, char *, int),
139 int (*writeit
) (char *, char *, int))
141 RECSTREAM
*rstrm
= (RECSTREAM
*) mem_alloc (sizeof (RECSTREAM
));
145 sendsize
= fix_buf_size (sendsize
);
146 recvsize
= fix_buf_size (recvsize
);
147 buf
= mem_alloc (sendsize
+ recvsize
+ BYTES_PER_XDR_UNIT
);
149 if (rstrm
== NULL
|| buf
== NULL
)
151 (void) __fxprintf (NULL
, "%s: %s", __func__
, _("out of memory\n"));
152 mem_free (rstrm
, sizeof (RECSTREAM
));
153 mem_free (buf
, sendsize
+ recvsize
+ BYTES_PER_XDR_UNIT
);
155 * This is bad. Should rework xdrrec_create to
156 * return a handle, and in this case return NULL
161 * adjust sizes and allocate buffer quad byte aligned
163 rstrm
->sendsize
= sendsize
;
164 rstrm
->recvsize
= recvsize
;
165 rstrm
->the_buffer
= buf
;
166 tmp
= rstrm
->the_buffer
;
167 if ((size_t)tmp
% BYTES_PER_XDR_UNIT
)
168 tmp
+= BYTES_PER_XDR_UNIT
- (size_t)tmp
% BYTES_PER_XDR_UNIT
;
169 rstrm
->out_base
= tmp
;
170 rstrm
->in_base
= tmp
+ sendsize
;
174 /* We have to add the cast since the `struct xdr_ops' in `struct XDR'
176 xdrs
->x_ops
= (struct xdr_ops
*) &xdrrec_ops
;
177 xdrs
->x_private
= (caddr_t
) rstrm
;
178 rstrm
->tcp_handle
= tcp_handle
;
179 rstrm
->readit
= readit
;
180 rstrm
->writeit
= writeit
;
181 rstrm
->out_finger
= rstrm
->out_boundry
= rstrm
->out_base
;
182 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_base
;
183 rstrm
->out_finger
+= 4;
184 rstrm
->out_boundry
+= sendsize
;
185 rstrm
->frag_sent
= FALSE
;
186 rstrm
->in_size
= recvsize
;
187 rstrm
->in_boundry
= rstrm
->in_base
;
188 rstrm
->in_finger
= (rstrm
->in_boundry
+= recvsize
);
190 rstrm
->last_frag
= TRUE
;
192 libc_hidden_nolink_sunrpc (xdrrec_create
, GLIBC_2_0
)
196 * The routines defined below are the xdr ops which will go into the
197 * xdr handle filled in by xdrrec_create.
201 xdrrec_getlong (XDR
*xdrs
, long *lp
)
203 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
204 int32_t *buflp
= (int32_t *) rstrm
->in_finger
;
207 /* first try the inline, fast case */
208 if (rstrm
->fbtbc
>= BYTES_PER_XDR_UNIT
&&
209 rstrm
->in_boundry
- (char *) buflp
>= BYTES_PER_XDR_UNIT
)
211 *lp
= (int32_t) ntohl (*buflp
);
212 rstrm
->fbtbc
-= BYTES_PER_XDR_UNIT
;
213 rstrm
->in_finger
+= BYTES_PER_XDR_UNIT
;
217 if (!xdrrec_getbytes (xdrs
, (caddr_t
) & mylong
,
220 *lp
= (int32_t) ntohl (mylong
);
226 xdrrec_putlong (XDR
*xdrs
, const long *lp
)
228 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
229 int32_t *dest_lp
= (int32_t *) rstrm
->out_finger
;
231 if ((rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
) > rstrm
->out_boundry
)
234 * this case should almost never happen so the code is
237 rstrm
->out_finger
-= BYTES_PER_XDR_UNIT
;
238 rstrm
->frag_sent
= TRUE
;
239 if (!flush_out (rstrm
, FALSE
))
241 dest_lp
= (int32_t *) rstrm
->out_finger
;
242 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
244 *dest_lp
= htonl (*lp
);
248 static bool_t
/* must manage buffers, fragments, and records */
249 xdrrec_getbytes (XDR
*xdrs
, caddr_t addr
, u_int len
)
251 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
256 current
= rstrm
->fbtbc
;
259 if (rstrm
->last_frag
)
261 if (!set_input_fragment (rstrm
))
265 current
= (len
< current
) ? len
: current
;
266 if (!get_input_bytes (rstrm
, addr
, current
))
269 rstrm
->fbtbc
-= current
;
276 xdrrec_putbytes (XDR
*xdrs
, const char *addr
, u_int len
)
278 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
283 current
= rstrm
->out_boundry
- rstrm
->out_finger
;
284 current
= (len
< current
) ? len
: current
;
285 memcpy (rstrm
->out_finger
, addr
, current
);
286 rstrm
->out_finger
+= current
;
289 if (rstrm
->out_finger
== rstrm
->out_boundry
&& len
> 0)
291 rstrm
->frag_sent
= TRUE
;
292 if (!flush_out (rstrm
, FALSE
))
300 xdrrec_getpos (const XDR
*xdrs
)
302 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
305 pos
= __lseek ((int) (long) rstrm
->tcp_handle
, (long) 0, 1);
311 pos
+= rstrm
->out_finger
- rstrm
->out_base
;
315 pos
-= rstrm
->in_boundry
- rstrm
->in_finger
;
326 xdrrec_setpos (XDR
*xdrs
, u_int pos
)
328 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
329 u_int currpos
= xdrrec_getpos (xdrs
);
330 int delta
= currpos
- pos
;
333 if ((int) currpos
!= -1)
338 newpos
= rstrm
->out_finger
- delta
;
339 if (newpos
> (caddr_t
) rstrm
->frag_header
&&
340 newpos
< rstrm
->out_boundry
)
342 rstrm
->out_finger
= newpos
;
348 newpos
= rstrm
->in_finger
- delta
;
349 if ((delta
< (int) (rstrm
->fbtbc
)) &&
350 (newpos
<= rstrm
->in_boundry
) &&
351 (newpos
>= rstrm
->in_base
))
353 rstrm
->in_finger
= newpos
;
354 rstrm
->fbtbc
-= delta
;
366 xdrrec_inline (XDR
*xdrs
, u_int len
)
368 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
375 if ((rstrm
->out_finger
+ len
) <= rstrm
->out_boundry
)
377 buf
= (int32_t *) rstrm
->out_finger
;
378 rstrm
->out_finger
+= len
;
383 if ((len
<= rstrm
->fbtbc
) &&
384 ((rstrm
->in_finger
+ len
) <= rstrm
->in_boundry
))
386 buf
= (int32_t *) rstrm
->in_finger
;
388 rstrm
->in_finger
+= len
;
399 xdrrec_destroy (XDR
*xdrs
)
401 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
403 mem_free (rstrm
->the_buffer
,
404 rstrm
->sendsize
+ rstrm
->recvsize
+ BYTES_PER_XDR_UNIT
);
405 mem_free ((caddr_t
) rstrm
, sizeof (RECSTREAM
));
409 xdrrec_getint32 (XDR
*xdrs
, int32_t *ip
)
411 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
412 int32_t *bufip
= (int32_t *) rstrm
->in_finger
;
415 /* first try the inline, fast case */
416 if (rstrm
->fbtbc
>= BYTES_PER_XDR_UNIT
&&
417 rstrm
->in_boundry
- (char *) bufip
>= BYTES_PER_XDR_UNIT
)
419 *ip
= ntohl (*bufip
);
420 rstrm
->fbtbc
-= BYTES_PER_XDR_UNIT
;
421 rstrm
->in_finger
+= BYTES_PER_XDR_UNIT
;
425 if (!xdrrec_getbytes (xdrs
, (caddr_t
) &mylong
,
428 *ip
= ntohl (mylong
);
434 xdrrec_putint32 (XDR
*xdrs
, const int32_t *ip
)
436 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
437 int32_t *dest_ip
= (int32_t *) rstrm
->out_finger
;
439 if ((rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
) > rstrm
->out_boundry
)
442 * this case should almost never happen so the code is
445 rstrm
->out_finger
-= BYTES_PER_XDR_UNIT
;
446 rstrm
->frag_sent
= TRUE
;
447 if (!flush_out (rstrm
, FALSE
))
449 dest_ip
= (int32_t *) rstrm
->out_finger
;
450 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
452 *dest_ip
= htonl (*ip
);
457 * Exported routines to manage xdr records
461 * Before reading (deserializing from the stream, one should always call
462 * this procedure to guarantee proper record alignment.
465 xdrrec_skiprecord (XDR
*xdrs
)
467 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
469 while (rstrm
->fbtbc
> 0 || (!rstrm
->last_frag
))
471 if (!skip_input_bytes (rstrm
, rstrm
->fbtbc
))
474 if ((!rstrm
->last_frag
) && (!set_input_fragment (rstrm
)))
477 rstrm
->last_frag
= FALSE
;
480 libc_hidden_nolink_sunrpc (xdrrec_skiprecord
, GLIBC_2_0
)
483 * Lookahead function.
484 * Returns TRUE iff there is no more input in the buffer
485 * after consuming the rest of the current record.
488 xdrrec_eof (XDR
*xdrs
)
490 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
492 while (rstrm
->fbtbc
> 0 || (!rstrm
->last_frag
))
494 if (!skip_input_bytes (rstrm
, rstrm
->fbtbc
))
497 if ((!rstrm
->last_frag
) && (!set_input_fragment (rstrm
)))
500 if (rstrm
->in_finger
== rstrm
->in_boundry
)
504 libc_hidden_nolink_sunrpc (xdrrec_eof
, GLIBC_2_0
)
507 * The client must tell the package when an end-of-record has occurred.
508 * The second parameter tells whether the record should be flushed to the
509 * (output) tcp stream. (This lets the package support batched or
510 * pipelined procedure calls.) TRUE => immediate flush to tcp connection.
513 xdrrec_endofrecord (XDR
*xdrs
, bool_t sendnow
)
515 RECSTREAM
*rstrm
= (RECSTREAM
*) xdrs
->x_private
;
516 u_long len
; /* fragment length */
518 if (sendnow
|| rstrm
->frag_sent
519 || rstrm
->out_finger
+ BYTES_PER_XDR_UNIT
>= rstrm
->out_boundry
)
521 rstrm
->frag_sent
= FALSE
;
522 return flush_out (rstrm
, TRUE
);
524 len
= (rstrm
->out_finger
- (char *) rstrm
->frag_header
525 - BYTES_PER_XDR_UNIT
);
526 *rstrm
->frag_header
= htonl ((u_long
) len
| LAST_FRAG
);
527 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_finger
;
528 rstrm
->out_finger
+= BYTES_PER_XDR_UNIT
;
531 libc_hidden_nolink_sunrpc (xdrrec_endofrecord
, GLIBC_2_0
)
535 * Internal useful routines
539 flush_out (RECSTREAM
*rstrm
, bool_t eor
)
541 u_long eormask
= (eor
== TRUE
) ? LAST_FRAG
: 0;
542 u_long len
= (rstrm
->out_finger
- (char *) rstrm
->frag_header
543 - BYTES_PER_XDR_UNIT
);
545 *rstrm
->frag_header
= htonl (len
| eormask
);
546 len
= rstrm
->out_finger
- rstrm
->out_base
;
547 if ((*(rstrm
->writeit
)) (rstrm
->tcp_handle
, rstrm
->out_base
, (int) len
)
550 rstrm
->frag_header
= (u_int32_t
*) rstrm
->out_base
;
551 rstrm
->out_finger
= (caddr_t
) rstrm
->out_base
+ BYTES_PER_XDR_UNIT
;
555 static bool_t
/* knows nothing about records! Only about input buffers */
556 fill_input_buf (RECSTREAM
*rstrm
)
562 where
= rstrm
->in_base
;
563 i
= (size_t) rstrm
->in_boundry
% BYTES_PER_XDR_UNIT
;
565 len
= rstrm
->in_size
- i
;
566 if ((len
= (*(rstrm
->readit
)) (rstrm
->tcp_handle
, where
, len
)) == -1)
568 rstrm
->in_finger
= where
;
570 rstrm
->in_boundry
= where
;
574 static bool_t
/* knows nothing about records! Only about input buffers */
576 get_input_bytes (RECSTREAM
*rstrm
, caddr_t addr
, int len
)
582 current
= rstrm
->in_boundry
- rstrm
->in_finger
;
585 if (!fill_input_buf (rstrm
))
589 current
= (len
< current
) ? len
: current
;
590 memcpy (addr
, rstrm
->in_finger
, current
);
591 rstrm
->in_finger
+= current
;
598 static bool_t
/* next two bytes of the input stream are treated as a header */
600 set_input_fragment (RECSTREAM
*rstrm
)
604 if (! get_input_bytes (rstrm
, (caddr_t
)&header
, BYTES_PER_XDR_UNIT
))
606 header
= ntohl (header
);
607 rstrm
->last_frag
= ((header
& LAST_FRAG
) == 0) ? FALSE
: TRUE
;
609 * Sanity check. Try not to accept wildly incorrect fragment
610 * sizes. Unfortunately, only a size of zero can be identified as
611 * 'wildely incorrect', and this only, if it is not the last
612 * fragment of a message. Ridiculously large fragment sizes may look
613 * wrong, but we don't have any way to be certain that they aren't
614 * what the client actually intended to send us. Many existing RPC
615 * implementations may sent a fragment of size zero as the last
616 * fragment of a message.
620 rstrm
->fbtbc
= header
& ~LAST_FRAG
;
624 static bool_t
/* consumes input bytes; knows nothing about records! */
626 skip_input_bytes (RECSTREAM
*rstrm
, long cnt
)
632 current
= rstrm
->in_boundry
- rstrm
->in_finger
;
635 if (!fill_input_buf (rstrm
))
639 current
= (cnt
< current
) ? cnt
: current
;
640 rstrm
->in_finger
+= current
;
648 fix_buf_size (u_int s
)