FreeBSD/Linux Kernel Cross Reference
sys/kern/sysv_msg.c
1 /*-
2 * Implementation of SVID messages
3 *
4 * Author: Daniel Boulet
5 *
6 * Copyright 1993 Daniel Boulet and RTMX Inc.
7 *
8 * This system call was implemented by Daniel Boulet under contract from RTMX.
9 *
10 * Redistribution and use in source forms, with and without modification,
11 * are permitted provided that this entire comment appears intact.
12 *
13 * Redistribution in binary form may occur without any restrictions.
14 * Obviously, it would be nice if you gave credit where credit is due
15 * but requiring it would be too onerous.
16 *
17 * This software is provided ``AS IS'' without any warranties of any kind.
18 */
19 /*-
20 * Copyright (c) 2003-2005 McAfee, Inc.
21 * All rights reserved.
22 *
23 * This software was developed for the FreeBSD Project in part by McAfee
24 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
25 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
26 * program.
27 *
28 * Redistribution and use in source and binary forms, with or without
29 * modification, are permitted provided that the following conditions
30 * are met:
31 * 1. Redistributions of source code must retain the above copyright
32 * notice, this list of conditions and the following disclaimer.
33 * 2. Redistributions in binary form must reproduce the above copyright
34 * notice, this list of conditions and the following disclaimer in the
35 * documentation and/or other materials provided with the distribution.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * SUCH DAMAGE.
48 */
49
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
52
53 #include "opt_sysvipc.h"
54 #include "opt_mac.h"
55
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/sysproto.h>
59 #include <sys/kernel.h>
60 #include <sys/proc.h>
61 #include <sys/lock.h>
62 #include <sys/mac.h>
63 #include <sys/mutex.h>
64 #include <sys/module.h>
65 #include <sys/msg.h>
66 #include <sys/syscall.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysent.h>
69 #include <sys/sysctl.h>
70 #include <sys/malloc.h>
71 #include <sys/jail.h>
72
73 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
74
75 static void msginit(void);
76 static int msgunload(void);
77 static int sysvmsg_modload(struct module *, int, void *);
78
79 #ifdef MSG_DEBUG
80 #define DPRINTF(a) printf a
81 #else
82 #define DPRINTF(a)
83 #endif
84 #ifdef MAC_DEBUG
85 #define MPRINTF(a) printf a
86 #else
87 #define MPRINTF(a)
88 #endif
89
90 static void msg_freehdr(struct msg *msghdr);
91
92 /* XXX casting to (sy_call_t *) is bogus, as usual. */
93 static sy_call_t *msgcalls[] = {
94 (sy_call_t *)msgctl, (sy_call_t *)msgget,
95 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv
96 };
97
98 #ifndef MSGSSZ
99 #define MSGSSZ 8 /* Each segment must be 2^N long */
100 #endif
101 #ifndef MSGSEG
102 #define MSGSEG 2048 /* must be less than 32767 */
103 #endif
104 #define MSGMAX (MSGSSZ*MSGSEG)
105 #ifndef MSGMNB
106 #define MSGMNB 2048 /* max # of bytes in a queue */
107 #endif
108 #ifndef MSGMNI
109 #define MSGMNI 40
110 #endif
111 #ifndef MSGTQL
112 #define MSGTQL 40
113 #endif
114
115 /*
116 * Based on the configuration parameters described in an SVR2 (yes, two)
117 * config(1m) man page.
118 *
119 * Each message is broken up and stored in segments that are msgssz bytes
120 * long. For efficiency reasons, this should be a power of two. Also,
121 * it doesn't make sense if it is less than 8 or greater than about 256.
122 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
123 * two between 8 and 1024 inclusive (and panic's if it isn't).
124 */
125 struct msginfo msginfo = {
126 MSGMAX, /* max chars in a message */
127 MSGMNI, /* # of message queue identifiers */
128 MSGMNB, /* max chars in a queue */
129 MSGTQL, /* max messages in system */
130 MSGSSZ, /* size of a message segment */
131 /* (must be small power of 2 greater than 4) */
132 MSGSEG /* number of message segments */
133 };
134
135 /*
136 * macros to convert between msqid_ds's and msqid's.
137 * (specific to this implementation)
138 */
139 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
140 #define MSQID_IX(id) ((id) & 0xffff)
141 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff)
142
143 /*
144 * The rest of this file is specific to this particular implementation.
145 */
146
147 struct msgmap {
148 short next; /* next segment in buffer */
149 /* -1 -> available */
150 /* 0..(MSGSEG-1) -> index of next segment */
151 };
152
153 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */
154
155 static int nfree_msgmaps; /* # of free map entries */
156 static short free_msgmaps; /* head of linked list of free map entries */
157 static struct msg *free_msghdrs;/* list of free msg headers */
158 static char *msgpool; /* MSGMAX byte long msg buffer pool */
159 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */
160 static struct msg *msghdrs; /* MSGTQL msg headers */
161 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */
162 static struct mtx msq_mtx; /* global mutex for message queues. */
163
164 static void
165 msginit()
166 {
167 register int i;
168
169 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg);
170 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz);
171 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz;
172 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni);
173 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb);
174 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql);
175
176 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
177 if (msgpool == NULL)
178 panic("msgpool is NULL");
179 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
180 if (msgmaps == NULL)
181 panic("msgmaps is NULL");
182 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
183 if (msghdrs == NULL)
184 panic("msghdrs is NULL");
185 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG,
186 M_WAITOK);
187 if (msqids == NULL)
188 panic("msqids is NULL");
189
190 /*
191 * msginfo.msgssz should be a power of two for efficiency reasons.
192 * It is also pretty silly if msginfo.msgssz is less than 8
193 * or greater than about 256 so ...
194 */
195
196 i = 8;
197 while (i < 1024 && i != msginfo.msgssz)
198 i <<= 1;
199 if (i != msginfo.msgssz) {
200 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
201 msginfo.msgssz));
202 panic("msginfo.msgssz not a small power of 2");
203 }
204
205 if (msginfo.msgseg > 32767) {
206 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg));
207 panic("msginfo.msgseg > 32767");
208 }
209
210 if (msgmaps == NULL)
211 panic("msgmaps is NULL");
212
213 for (i = 0; i < msginfo.msgseg; i++) {
214 if (i > 0)
215 msgmaps[i-1].next = i;
216 msgmaps[i].next = -1; /* implies entry is available */
217 }
218 free_msgmaps = 0;
219 nfree_msgmaps = msginfo.msgseg;
220
221 if (msghdrs == NULL)
222 panic("msghdrs is NULL");
223
224 for (i = 0; i < msginfo.msgtql; i++) {
225 msghdrs[i].msg_type = 0;
226 if (i > 0)
227 msghdrs[i-1].msg_next = &msghdrs[i];
228 msghdrs[i].msg_next = NULL;
229 #ifdef MAC
230 mac_init_sysv_msgmsg(&msghdrs[i]);
231 #endif
232 }
233 free_msghdrs = &msghdrs[0];
234
235 if (msqids == NULL)
236 panic("msqids is NULL");
237
238 for (i = 0; i < msginfo.msgmni; i++) {
239 msqids[i].u.msg_qbytes = 0; /* implies entry is available */
240 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */
241 msqids[i].u.msg_perm.mode = 0;
242 #ifdef MAC
243 mac_init_sysv_msgqueue(&msqids[i]);
244 #endif
245 }
246 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF);
247 }
248
249 static int
250 msgunload()
251 {
252 struct msqid_kernel *msqkptr;
253 int msqid;
254 #ifdef MAC
255 int i;
256 #endif
257
258 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
259 /*
260 * Look for an unallocated and unlocked msqid_ds.
261 * msqid_ds's can be locked by msgsnd or msgrcv while
262 * they are copying the message in/out. We can't
263 * re-use the entry until they release it.
264 */
265 msqkptr = &msqids[msqid];
266 if (msqkptr->u.msg_qbytes != 0 ||
267 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
268 break;
269 }
270 if (msqid != msginfo.msgmni)
271 return (EBUSY);
272
273 #ifdef MAC
274 for (i = 0; i < msginfo.msgtql; i++)
275 mac_destroy_sysv_msgmsg(&msghdrs[i]);
276 for (msqid = 0; msqid < msginfo.msgmni; msqid++)
277 mac_destroy_sysv_msgqueue(&msqids[msqid]);
278 #endif
279 free(msgpool, M_MSG);
280 free(msgmaps, M_MSG);
281 free(msghdrs, M_MSG);
282 free(msqids, M_MSG);
283 mtx_destroy(&msq_mtx);
284 return (0);
285 }
286
287
288 static int
289 sysvmsg_modload(struct module *module, int cmd, void *arg)
290 {
291 int error = 0;
292
293 switch (cmd) {
294 case MOD_LOAD:
295 msginit();
296 break;
297 case MOD_UNLOAD:
298 error = msgunload();
299 break;
300 case MOD_SHUTDOWN:
301 break;
302 default:
303 error = EINVAL;
304 break;
305 }
306 return (error);
307 }
308
309 static moduledata_t sysvmsg_mod = {
310 "sysvmsg",
311 &sysvmsg_modload,
312 NULL
313 };
314
315 SYSCALL_MODULE_HELPER(msgsys);
316 SYSCALL_MODULE_HELPER(msgctl);
317 SYSCALL_MODULE_HELPER(msgget);
318 SYSCALL_MODULE_HELPER(msgsnd);
319 SYSCALL_MODULE_HELPER(msgrcv);
320
321 DECLARE_MODULE(sysvmsg, sysvmsg_mod,
322 SI_SUB_SYSV_MSG, SI_ORDER_FIRST);
323 MODULE_VERSION(sysvmsg, 1);
324
325 /*
326 * Entry point for all MSG calls
327 *
328 * MPSAFE
329 */
330 int
331 msgsys(td, uap)
332 struct thread *td;
333 /* XXX actually varargs. */
334 struct msgsys_args /* {
335 int which;
336 int a2;
337 int a3;
338 int a4;
339 int a5;
340 int a6;
341 } */ *uap;
342 {
343 int error;
344
345 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
346 return (ENOSYS);
347 if (uap->which < 0 ||
348 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0]))
349 return (EINVAL);
350 error = (*msgcalls[uap->which])(td, &uap->a2);
351 return (error);
352 }
353
354 static void
355 msg_freehdr(msghdr)
356 struct msg *msghdr;
357 {
358 while (msghdr->msg_ts > 0) {
359 short next;
360 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
361 panic("msghdr->msg_spot out of range");
362 next = msgmaps[msghdr->msg_spot].next;
363 msgmaps[msghdr->msg_spot].next = free_msgmaps;
364 free_msgmaps = msghdr->msg_spot;
365 nfree_msgmaps++;
366 msghdr->msg_spot = next;
367 if (msghdr->msg_ts >= msginfo.msgssz)
368 msghdr->msg_ts -= msginfo.msgssz;
369 else
370 msghdr->msg_ts = 0;
371 }
372 if (msghdr->msg_spot != -1)
373 panic("msghdr->msg_spot != -1");
374 msghdr->msg_next = free_msghdrs;
375 free_msghdrs = msghdr;
376 #ifdef MAC
377 mac_cleanup_sysv_msgmsg(msghdr);
378 #endif
379 }
380
381 #ifndef _SYS_SYSPROTO_H_
382 struct msgctl_args {
383 int msqid;
384 int cmd;
385 struct msqid_ds *buf;
386 };
387 #endif
388
389 /*
390 * MPSAFE
391 */
392 int
393 msgctl(td, uap)
394 struct thread *td;
395 register struct msgctl_args *uap;
396 {
397 int msqid = uap->msqid;
398 int cmd = uap->cmd;
399 struct msqid_ds msqbuf;
400 int error;
401
402 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf));
403 if (cmd == IPC_SET &&
404 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0)
405 return (error);
406 error = kern_msgctl(td, msqid, cmd, &msqbuf);
407 if (cmd == IPC_STAT && error == 0)
408 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds));
409 return (error);
410 }
411
412 int
413 kern_msgctl(td, msqid, cmd, msqbuf)
414 struct thread *td;
415 int msqid;
416 int cmd;
417 struct msqid_ds *msqbuf;
418 {
419 int rval, error, msqix;
420 register struct msqid_kernel *msqkptr;
421
422 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
423 return (ENOSYS);
424
425 msqix = IPCID_TO_IX(msqid);
426
427 if (msqix < 0 || msqix >= msginfo.msgmni) {
428 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
429 msginfo.msgmni));
430 return (EINVAL);
431 }
432
433 msqkptr = &msqids[msqix];
434
435 mtx_lock(&msq_mtx);
436 if (msqkptr->u.msg_qbytes == 0) {
437 DPRINTF(("no such msqid\n"));
438 error = EINVAL;
439 goto done2;
440 }
441 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
442 DPRINTF(("wrong sequence number\n"));
443 error = EINVAL;
444 goto done2;
445 }
446 #ifdef MAC
447 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd);
448 if (error != 0) {
449 MPRINTF(("mac_check_sysv_msqctl returned %d\n", error));
450 goto done2;
451 }
452 #endif
453
454 error = 0;
455 rval = 0;
456
457 switch (cmd) {
458
459 case IPC_RMID:
460 {
461 struct msg *msghdr;
462 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
463 goto done2;
464
465 #ifdef MAC
466 /*
467 * Check that the thread has MAC access permissions to
468 * individual msghdrs. Note: We need to do this in a
469 * separate loop because the actual loop alters the
470 * msq/msghdr info as it progresses, and there is no going
471 * back if half the way through we discover that the
472 * thread cannot free a certain msghdr. The msq will get
473 * into an inconsistent state.
474 */
475 for (msghdr = msqkptr->u.msg_first; msghdr != NULL;
476 msghdr = msghdr->msg_next) {
477 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr);
478 if (error != 0) {
479 MPRINTF(("mac_check_sysv_msgrmid returned %d\n",
480 error));
481 goto done2;
482 }
483 }
484 #endif
485
486 /* Free the message headers */
487 msghdr = msqkptr->u.msg_first;
488 while (msghdr != NULL) {
489 struct msg *msghdr_tmp;
490
491 /* Free the segments of each message */
492 msqkptr->u.msg_cbytes -= msghdr->msg_ts;
493 msqkptr->u.msg_qnum--;
494 msghdr_tmp = msghdr;
495 msghdr = msghdr->msg_next;
496 msg_freehdr(msghdr_tmp);
497 }
498
499 if (msqkptr->u.msg_cbytes != 0)
500 panic("msg_cbytes is screwed up");
501 if (msqkptr->u.msg_qnum != 0)
502 panic("msg_qnum is screwed up");
503
504 msqkptr->u.msg_qbytes = 0; /* Mark it as free */
505
506 #ifdef MAC
507 mac_cleanup_sysv_msgqueue(msqkptr);
508 #endif
509
510 wakeup(msqkptr);
511 }
512
513 break;
514
515 case IPC_SET:
516 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
517 goto done2;
518 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) {
519 error = suser(td);
520 if (error)
521 goto done2;
522 }
523 if (msqbuf->msg_qbytes > msginfo.msgmnb) {
524 DPRINTF(("can't increase msg_qbytes beyond %d"
525 "(truncating)\n", msginfo.msgmnb));
526 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */
527 }
528 if (msqbuf->msg_qbytes == 0) {
529 DPRINTF(("can't reduce msg_qbytes to 0\n"));
530 error = EINVAL; /* non-standard errno! */
531 goto done2;
532 }
533 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */
534 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */
535 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) |
536 (msqbuf->msg_perm.mode & 0777);
537 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes;
538 msqkptr->u.msg_ctime = time_second;
539 break;
540
541 case IPC_STAT:
542 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
543 DPRINTF(("requester doesn't have read access\n"));
544 goto done2;
545 }
546 *msqbuf = msqkptr->u;
547 break;
548
549 default:
550 DPRINTF(("invalid command %d\n", cmd));
551 error = EINVAL;
552 goto done2;
553 }
554
555 if (error == 0)
556 td->td_retval[0] = rval;
557 done2:
558 mtx_unlock(&msq_mtx);
559 return (error);
560 }
561
562 #ifndef _SYS_SYSPROTO_H_
563 struct msgget_args {
564 key_t key;
565 int msgflg;
566 };
567 #endif
568
569 /*
570 * MPSAFE
571 */
572 int
573 msgget(td, uap)
574 struct thread *td;
575 register struct msgget_args *uap;
576 {
577 int msqid, error = 0;
578 int key = uap->key;
579 int msgflg = uap->msgflg;
580 struct ucred *cred = td->td_ucred;
581 register struct msqid_kernel *msqkptr = NULL;
582
583 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg));
584
585 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
586 return (ENOSYS);
587
588 mtx_lock(&msq_mtx);
589 if (key != IPC_PRIVATE) {
590 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
591 msqkptr = &msqids[msqid];
592 if (msqkptr->u.msg_qbytes != 0 &&
593 msqkptr->u.msg_perm.key == key)
594 break;
595 }
596 if (msqid < msginfo.msgmni) {
597 DPRINTF(("found public key\n"));
598 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
599 DPRINTF(("not exclusive\n"));
600 error = EEXIST;
601 goto done2;
602 }
603 if ((error = ipcperm(td, &msqkptr->u.msg_perm,
604 msgflg & 0700))) {
605 DPRINTF(("requester doesn't have 0%o access\n",
606 msgflg & 0700));
607 goto done2;
608 }
609 #ifdef MAC
610 error = mac_check_sysv_msqget(cred, msqkptr);
611 if (error != 0) {
612 MPRINTF(("mac_check_sysv_msqget returned %d\n",
613 error));
614 goto done2;
615 }
616 #endif
617 goto found;
618 }
619 }
620
621 DPRINTF(("need to allocate the msqid_ds\n"));
622 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
623 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
624 /*
625 * Look for an unallocated and unlocked msqid_ds.
626 * msqid_ds's can be locked by msgsnd or msgrcv while
627 * they are copying the message in/out. We can't
628 * re-use the entry until they release it.
629 */
630 msqkptr = &msqids[msqid];
631 if (msqkptr->u.msg_qbytes == 0 &&
632 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0)
633 break;
634 }
635 if (msqid == msginfo.msgmni) {
636 DPRINTF(("no more msqid_ds's available\n"));
637 error = ENOSPC;
638 goto done2;
639 }
640 DPRINTF(("msqid %d is available\n", msqid));
641 msqkptr->u.msg_perm.key = key;
642 msqkptr->u.msg_perm.cuid = cred->cr_uid;
643 msqkptr->u.msg_perm.uid = cred->cr_uid;
644 msqkptr->u.msg_perm.cgid = cred->cr_gid;
645 msqkptr->u.msg_perm.gid = cred->cr_gid;
646 msqkptr->u.msg_perm.mode = (msgflg & 0777);
647 /* Make sure that the returned msqid is unique */
648 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff;
649 msqkptr->u.msg_first = NULL;
650 msqkptr->u.msg_last = NULL;
651 msqkptr->u.msg_cbytes = 0;
652 msqkptr->u.msg_qnum = 0;
653 msqkptr->u.msg_qbytes = msginfo.msgmnb;
654 msqkptr->u.msg_lspid = 0;
655 msqkptr->u.msg_lrpid = 0;
656 msqkptr->u.msg_stime = 0;
657 msqkptr->u.msg_rtime = 0;
658 msqkptr->u.msg_ctime = time_second;
659 #ifdef MAC
660 mac_create_sysv_msgqueue(cred, msqkptr);
661 #endif
662 } else {
663 DPRINTF(("didn't find it and wasn't asked to create it\n"));
664 error = ENOENT;
665 goto done2;
666 }
667
668 found:
669 /* Construct the unique msqid */
670 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm);
671 done2:
672 mtx_unlock(&msq_mtx);
673 return (error);
674 }
675
676 #ifndef _SYS_SYSPROTO_H_
677 struct msgsnd_args {
678 int msqid;
679 const void *msgp;
680 size_t msgsz;
681 int msgflg;
682 };
683 #endif
684
685 int
686 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype)
687 struct thread *td;
688 int msqid;
689 const void *msgp; /* XXX msgp is actually mtext. */
690 size_t msgsz;
691 int msgflg;
692 long mtype;
693 {
694 int msqix, segs_needed, error = 0;
695 register struct msqid_kernel *msqkptr;
696 register struct msg *msghdr;
697 short next;
698
699 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
700 return (ENOSYS);
701
702 mtx_lock(&msq_mtx);
703 msqix = IPCID_TO_IX(msqid);
704
705 if (msqix < 0 || msqix >= msginfo.msgmni) {
706 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
707 msginfo.msgmni));
708 error = EINVAL;
709 goto done2;
710 }
711
712 msqkptr = &msqids[msqix];
713 if (msqkptr->u.msg_qbytes == 0) {
714 DPRINTF(("no such message queue id\n"));
715 error = EINVAL;
716 goto done2;
717 }
718 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
719 DPRINTF(("wrong sequence number\n"));
720 error = EINVAL;
721 goto done2;
722 }
723
724 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) {
725 DPRINTF(("requester doesn't have write access\n"));
726 goto done2;
727 }
728
729 #ifdef MAC
730 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr);
731 if (error != 0) {
732 MPRINTF(("mac_check_sysv_msqsnd returned %d\n", error));
733 goto done2;
734 }
735 #endif
736
737 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz;
738 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz,
739 msginfo.msgssz, segs_needed));
740 for (;;) {
741 int need_more_resources = 0;
742
743 /*
744 * check msgsz
745 * (inside this loop in case msg_qbytes changes while we sleep)
746 */
747
748 if (msgsz > msqkptr->u.msg_qbytes) {
749 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n"));
750 error = EINVAL;
751 goto done2;
752 }
753
754 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) {
755 DPRINTF(("msqid is locked\n"));
756 need_more_resources = 1;
757 }
758 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) {
759 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n"));
760 need_more_resources = 1;
761 }
762 if (segs_needed > nfree_msgmaps) {
763 DPRINTF(("segs_needed > nfree_msgmaps\n"));
764 need_more_resources = 1;
765 }
766 if (free_msghdrs == NULL) {
767 DPRINTF(("no more msghdrs\n"));
768 need_more_resources = 1;
769 }
770
771 if (need_more_resources) {
772 int we_own_it;
773
774 if ((msgflg & IPC_NOWAIT) != 0) {
775 DPRINTF(("need more resources but caller "
776 "doesn't want to wait\n"));
777 error = EAGAIN;
778 goto done2;
779 }
780
781 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) {
782 DPRINTF(("we don't own the msqid_ds\n"));
783 we_own_it = 0;
784 } else {
785 /* Force later arrivals to wait for our
786 request */
787 DPRINTF(("we own the msqid_ds\n"));
788 msqkptr->u.msg_perm.mode |= MSG_LOCKED;
789 we_own_it = 1;
790 }
791 DPRINTF(("goodnight\n"));
792 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
793 "msgwait", 0);
794 DPRINTF(("good morning, error=%d\n", error));
795 if (we_own_it)
796 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
797 if (error != 0) {
798 DPRINTF(("msgsnd: interrupted system call\n"));
799 error = EINTR;
800 goto done2;
801 }
802
803 /*
804 * Make sure that the msq queue still exists
805 */
806
807 if (msqkptr->u.msg_qbytes == 0) {
808 DPRINTF(("msqid deleted\n"));
809 error = EIDRM;
810 goto done2;
811 }
812
813 } else {
814 DPRINTF(("got all the resources that we need\n"));
815 break;
816 }
817 }
818
819 /*
820 * We have the resources that we need.
821 * Make sure!
822 */
823
824 if (msqkptr->u.msg_perm.mode & MSG_LOCKED)
825 panic("msg_perm.mode & MSG_LOCKED");
826 if (segs_needed > nfree_msgmaps)
827 panic("segs_needed > nfree_msgmaps");
828 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes)
829 panic("msgsz + msg_cbytes > msg_qbytes");
830 if (free_msghdrs == NULL)
831 panic("no more msghdrs");
832
833 /*
834 * Re-lock the msqid_ds in case we page-fault when copying in the
835 * message
836 */
837
838 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
839 panic("msqid_ds is already locked");
840 msqkptr->u.msg_perm.mode |= MSG_LOCKED;
841
842 /*
843 * Allocate a message header
844 */
845
846 msghdr = free_msghdrs;
847 free_msghdrs = msghdr->msg_next;
848 msghdr->msg_spot = -1;
849 msghdr->msg_ts = msgsz;
850 msghdr->msg_type = mtype;
851 #ifdef MAC
852 /*
853 * XXXMAC: Should the mac_check_sysv_msgmsq check follow here
854 * immediately? Or, should it be checked just before the msg is
855 * enqueued in the msgq (as it is done now)?
856 */
857 mac_create_sysv_msgmsg(td->td_ucred, msqkptr, msghdr);
858 #endif
859
860 /*
861 * Allocate space for the message
862 */
863
864 while (segs_needed > 0) {
865 if (nfree_msgmaps <= 0)
866 panic("not enough msgmaps");
867 if (free_msgmaps == -1)
868 panic("nil free_msgmaps");
869 next = free_msgmaps;
870 if (next <= -1)
871 panic("next too low #1");
872 if (next >= msginfo.msgseg)
873 panic("next out of range #1");
874 DPRINTF(("allocating segment %d to message\n", next));
875 free_msgmaps = msgmaps[next].next;
876 nfree_msgmaps--;
877 msgmaps[next].next = msghdr->msg_spot;
878 msghdr->msg_spot = next;
879 segs_needed--;
880 }
881
882 /*
883 * Validate the message type
884 */
885
886 if (msghdr->msg_type < 1) {
887 msg_freehdr(msghdr);
888 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
889 wakeup(msqkptr);
890 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type));
891 error = EINVAL;
892 goto done2;
893 }
894
895 /*
896 * Copy in the message body
897 */
898
899 next = msghdr->msg_spot;
900 while (msgsz > 0) {
901 size_t tlen;
902 if (msgsz > msginfo.msgssz)
903 tlen = msginfo.msgssz;
904 else
905 tlen = msgsz;
906 if (next <= -1)
907 panic("next too low #2");
908 if (next >= msginfo.msgseg)
909 panic("next out of range #2");
910 mtx_unlock(&msq_mtx);
911 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz],
912 tlen)) != 0) {
913 mtx_lock(&msq_mtx);
914 DPRINTF(("error %d copying in message segment\n",
915 error));
916 msg_freehdr(msghdr);
917 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
918 wakeup(msqkptr);
919 goto done2;
920 }
921 mtx_lock(&msq_mtx);
922 msgsz -= tlen;
923 msgp = (const char *)msgp + tlen;
924 next = msgmaps[next].next;
925 }
926 if (next != -1)
927 panic("didn't use all the msg segments");
928
929 /*
930 * We've got the message. Unlock the msqid_ds.
931 */
932
933 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
934
935 /*
936 * Make sure that the msqid_ds is still allocated.
937 */
938
939 if (msqkptr->u.msg_qbytes == 0) {
940 msg_freehdr(msghdr);
941 wakeup(msqkptr);
942 error = EIDRM;
943 goto done2;
944 }
945
946 #ifdef MAC
947 /*
948 * Note: Since the task/thread allocates the msghdr and usually
949 * primes it with its own MAC label, for a majority of policies, it
950 * won't be necessary to check whether the msghdr has access
951 * permissions to the msgq. The mac_check_sysv_msqsnd check would
952 * suffice in that case. However, this hook may be required where
953 * individual policies derive a non-identical label for the msghdr
954 * from the current thread label and may want to check the msghdr
955 * enqueue permissions, along with read/write permissions to the
956 * msgq.
957 */
958 error = mac_check_sysv_msgmsq(td->td_ucred, msghdr, msqkptr);
959 if (error != 0) {
960 MPRINTF(("mac_check_sysv_msqmsq returned %d\n", error));
961 msg_freehdr(msghdr);
962 wakeup(msqkptr);
963 goto done2;
964 }
965 #endif
966
967 /*
968 * Put the message into the queue
969 */
970 if (msqkptr->u.msg_first == NULL) {
971 msqkptr->u.msg_first = msghdr;
972 msqkptr->u.msg_last = msghdr;
973 } else {
974 msqkptr->u.msg_last->msg_next = msghdr;
975 msqkptr->u.msg_last = msghdr;
976 }
977 msqkptr->u.msg_last->msg_next = NULL;
978
979 msqkptr->u.msg_cbytes += msghdr->msg_ts;
980 msqkptr->u.msg_qnum++;
981 msqkptr->u.msg_lspid = td->td_proc->p_pid;
982 msqkptr->u.msg_stime = time_second;
983
984 wakeup(msqkptr);
985 td->td_retval[0] = 0;
986 done2:
987 mtx_unlock(&msq_mtx);
988 return (error);
989 }
990
991 /*
992 * MPSAFE
993 */
994 int
995 msgsnd(td, uap)
996 struct thread *td;
997 register struct msgsnd_args *uap;
998 {
999 int error;
1000 long mtype;
1001
1002 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp,
1003 uap->msgsz, uap->msgflg));
1004
1005 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) {
1006 DPRINTF(("error %d copying the message type\n", error));
1007 return (error);
1008 }
1009 return (kern_msgsnd(td, uap->msqid,
1010 (const char *)uap->msgp + sizeof(mtype),
1011 uap->msgsz, uap->msgflg, mtype));
1012 }
1013
1014 #ifndef _SYS_SYSPROTO_H_
1015 struct msgrcv_args {
1016 int msqid;
1017 void *msgp;
1018 size_t msgsz;
1019 long msgtyp;
1020 int msgflg;
1021 };
1022 #endif
1023
1024 int
1025 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype)
1026 struct thread *td;
1027 int msqid;
1028 void *msgp; /* XXX msgp is actually mtext. */
1029 size_t msgsz;
1030 long msgtyp;
1031 int msgflg;
1032 long *mtype;
1033 {
1034 size_t len;
1035 register struct msqid_kernel *msqkptr;
1036 register struct msg *msghdr;
1037 int msqix, error = 0;
1038 short next;
1039
1040 if (!jail_sysvipc_allowed && jailed(td->td_ucred))
1041 return (ENOSYS);
1042
1043 msqix = IPCID_TO_IX(msqid);
1044
1045 if (msqix < 0 || msqix >= msginfo.msgmni) {
1046 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
1047 msginfo.msgmni));
1048 return (EINVAL);
1049 }
1050
1051 msqkptr = &msqids[msqix];
1052 mtx_lock(&msq_mtx);
1053 if (msqkptr->u.msg_qbytes == 0) {
1054 DPRINTF(("no such message queue id\n"));
1055 error = EINVAL;
1056 goto done2;
1057 }
1058 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1059 DPRINTF(("wrong sequence number\n"));
1060 error = EINVAL;
1061 goto done2;
1062 }
1063
1064 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
1065 DPRINTF(("requester doesn't have read access\n"));
1066 goto done2;
1067 }
1068
1069 #ifdef MAC
1070 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr);
1071 if (error != 0) {
1072 MPRINTF(("mac_check_sysv_msqrcv returned %d\n", error));
1073 goto done2;
1074 }
1075 #endif
1076
1077 msghdr = NULL;
1078 while (msghdr == NULL) {
1079 if (msgtyp == 0) {
1080 msghdr = msqkptr->u.msg_first;
1081 if (msghdr != NULL) {
1082 if (msgsz < msghdr->msg_ts &&
1083 (msgflg & MSG_NOERROR) == 0) {
1084 DPRINTF(("first message on the queue "
1085 "is too big (want %zu, got %d)\n",
1086 msgsz, msghdr->msg_ts));
1087 error = E2BIG;
1088 goto done2;
1089 }
1090 #ifdef MAC
1091 error = mac_check_sysv_msgrcv(td->td_ucred,
1092 msghdr);
1093 if (error != 0) {
1094 MPRINTF(("mac_check_sysv_msgrcv "
1095 "returned %d\n", error));
1096 goto done2;
1097 }
1098 #endif
1099 if (msqkptr->u.msg_first == msqkptr->u.msg_last) {
1100 msqkptr->u.msg_first = NULL;
1101 msqkptr->u.msg_last = NULL;
1102 } else {
1103 msqkptr->u.msg_first = msghdr->msg_next;
1104 if (msqkptr->u.msg_first == NULL)
1105 panic("msg_first/last screwed up #1");
1106 }
1107 }
1108 } else {
1109 struct msg *previous;
1110 struct msg **prev;
1111
1112 previous = NULL;
1113 prev = &(msqkptr->u.msg_first);
1114 while ((msghdr = *prev) != NULL) {
1115 /*
1116 * Is this message's type an exact match or is
1117 * this message's type less than or equal to
1118 * the absolute value of a negative msgtyp?
1119 * Note that the second half of this test can
1120 * NEVER be true if msgtyp is positive since
1121 * msg_type is always positive!
1122 */
1123
1124 if (msgtyp == msghdr->msg_type ||
1125 msghdr->msg_type <= -msgtyp) {
1126 DPRINTF(("found message type %ld, "
1127 "requested %ld\n",
1128 msghdr->msg_type, msgtyp));
1129 if (msgsz < msghdr->msg_ts &&
1130 (msgflg & MSG_NOERROR) == 0) {
1131 DPRINTF(("requested message "
1132 "on the queue is too big "
1133 "(want %zu, got %hu)\n",
1134 msgsz, msghdr->msg_ts));
1135 error = E2BIG;
1136 goto done2;
1137 }
1138 #ifdef MAC
1139 error = mac_check_sysv_msgrcv(
1140 td->td_ucred, msghdr);
1141 if (error != 0) {
1142 MPRINTF(("mac_check_sysv_"
1143 "msgrcv returned %d\n",
1144 error));
1145 goto done2;
1146 }
1147 #endif
1148 *prev = msghdr->msg_next;
1149 if (msghdr == msqkptr->u.msg_last) {
1150 if (previous == NULL) {
1151 if (prev !=
1152 &msqkptr->u.msg_first)
1153 panic("msg_first/last screwed up #2");
1154 msqkptr->u.msg_first =
1155 NULL;
1156 msqkptr->u.msg_last =
1157 NULL;
1158 } else {
1159 if (prev ==
1160 &msqkptr->u.msg_first)
1161 panic("msg_first/last screwed up #3");
1162 msqkptr->u.msg_last =
1163 previous;
1164 }
1165 }
1166 break;
1167 }
1168 previous = msghdr;
1169 prev = &(msghdr->msg_next);
1170 }
1171 }
1172
1173 /*
1174 * We've either extracted the msghdr for the appropriate
1175 * message or there isn't one.
1176 * If there is one then bail out of this loop.
1177 */
1178
1179 if (msghdr != NULL)
1180 break;
1181
1182 /*
1183 * Hmph! No message found. Does the user want to wait?
1184 */
1185
1186 if ((msgflg & IPC_NOWAIT) != 0) {
1187 DPRINTF(("no appropriate message found (msgtyp=%ld)\n",
1188 msgtyp));
1189 /* The SVID says to return ENOMSG. */
1190 error = ENOMSG;
1191 goto done2;
1192 }
1193
1194 /*
1195 * Wait for something to happen
1196 */
1197
1198 DPRINTF(("msgrcv: goodnight\n"));
1199 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
1200 "msgwait", 0);
1201 DPRINTF(("msgrcv: good morning (error=%d)\n", error));
1202
1203 if (error != 0) {
1204 DPRINTF(("msgsnd: interrupted system call\n"));
1205 error = EINTR;
1206 goto done2;
1207 }
1208
1209 /*
1210 * Make sure that the msq queue still exists
1211 */
1212
1213 if (msqkptr->u.msg_qbytes == 0 ||
1214 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1215 DPRINTF(("msqid deleted\n"));
1216 error = EIDRM;
1217 goto done2;
1218 }
1219 }
1220
1221 /*
1222 * Return the message to the user.
1223 *
1224 * First, do the bookkeeping (before we risk being interrupted).
1225 */
1226
1227 msqkptr->u.msg_cbytes -= msghdr->msg_ts;
1228 msqkptr->u.msg_qnum--;
1229 msqkptr->u.msg_lrpid = td->td_proc->p_pid;
1230 msqkptr->u.msg_rtime = time_second;
1231
1232 /*
1233 * Make msgsz the actual amount that we'll be returning.
1234 * Note that this effectively truncates the message if it is too long
1235 * (since msgsz is never increased).
1236 */
1237
1238 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz,
1239 msghdr->msg_ts));
1240 if (msgsz > msghdr->msg_ts)
1241 msgsz = msghdr->msg_ts;
1242 *mtype = msghdr->msg_type;
1243
1244 /*
1245 * Return the segments to the user
1246 */
1247
1248 next = msghdr->msg_spot;
1249 for (len = 0; len < msgsz; len += msginfo.msgssz) {
1250 size_t tlen;
1251
1252 if (msgsz - len > msginfo.msgssz)
1253 tlen = msginfo.msgssz;
1254 else
1255 tlen = msgsz - len;
1256 if (next <= -1)
1257 panic("next too low #3");
1258 if (next >= msginfo.msgseg)
1259 panic("next out of range #3");
1260 mtx_unlock(&msq_mtx);
1261 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen);
1262 mtx_lock(&msq_mtx);
1263 if (error != 0) {
1264 DPRINTF(("error (%d) copying out message segment\n",
1265 error));
1266 msg_freehdr(msghdr);
1267 wakeup(msqkptr);
1268 goto done2;
1269 }
1270 msgp = (char *)msgp + tlen;
1271 next = msgmaps[next].next;
1272 }
1273
1274 /*
1275 * Done, return the actual number of bytes copied out.
1276 */
1277
1278 msg_freehdr(msghdr);
1279 wakeup(msqkptr);
1280 td->td_retval[0] = msgsz;
1281 done2:
1282 mtx_unlock(&msq_mtx);
1283 return (error);
1284 }
1285
1286 /*
1287 * MPSAFE
1288 */
1289 int
1290 msgrcv(td, uap)
1291 struct thread *td;
1292 register struct msgrcv_args *uap;
1293 {
1294 int error;
1295 long mtype;
1296
1297 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid,
1298 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg));
1299
1300 if ((error = kern_msgrcv(td, uap->msqid,
1301 (char *)uap->msgp + sizeof(mtype), uap->msgsz,
1302 uap->msgtyp, uap->msgflg, &mtype)) != 0)
1303 return (error);
1304 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0)
1305 DPRINTF(("error %d copying the message type\n", error));
1306 return (error);
1307 }
1308
1309 static int
1310 sysctl_msqids(SYSCTL_HANDLER_ARGS)
1311 {
1312
1313 return (SYSCTL_OUT(req, msqids,
1314 sizeof(struct msqid_kernel) * msginfo.msgmni));
1315 }
1316
1317 SYSCTL_DECL(_kern_ipc);
1318 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0,
1319 "Maximum message size");
1320 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0,
1321 "Number of message queue identifiers");
1322 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0,
1323 "Maximum number of bytes in a queue");
1324 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0,
1325 "Maximum number of messages in the system");
1326 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0,
1327 "Size of a message segment");
1328 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0,
1329 "Number of message segments");
1330 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD,
1331 NULL, 0, sysctl_msqids, "", "Message queue IDs");
Cache object: 302f17e3b730cd9c0d18b7a5e01fefb1
|