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_compat.h"
54 #include "opt_sysvipc.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/priv.h>
61 #include <sys/proc.h>
62 #include <sys/lock.h>
63 #include <sys/mutex.h>
64 #include <sys/module.h>
65 #include <sys/mount.h>
66 #include <sys/msg.h>
67 #include <sys/racct.h>
68 #include <sys/sx.h>
69 #include <sys/syscall.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysent.h>
72 #include <sys/sysctl.h>
73 #include <sys/malloc.h>
74 #include <sys/jail.h>
75
76 #include <security/mac/mac_framework.h>
77
78 FEATURE(sysv_msg, "System V message queues support");
79
80 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues");
81
82 static int msginit(void);
83 static int msgunload(void);
84 static int sysvmsg_modload(struct module *, int, void *);
85 static void msq_remove(struct msqid_kernel *);
86 static struct prison *msg_find_prison(struct ucred *);
87 static int msq_prison_cansee(struct prison *, struct msqid_kernel *);
88 static int msg_prison_check(void *, void *);
89 static int msg_prison_set(void *, void *);
90 static int msg_prison_get(void *, void *);
91 static int msg_prison_remove(void *, void *);
92 static void msg_prison_cleanup(struct prison *);
93
94
95 #ifdef MSG_DEBUG
96 #define DPRINTF(a) printf a
97 #else
98 #define DPRINTF(a) (void)0
99 #endif
100
101 static void msg_freehdr(struct msg *msghdr);
102
103 #ifndef MSGSSZ
104 #define MSGSSZ 8 /* Each segment must be 2^N long */
105 #endif
106 #ifndef MSGSEG
107 #define MSGSEG 2048 /* must be less than 32767 */
108 #endif
109 #define MSGMAX (MSGSSZ*MSGSEG)
110 #ifndef MSGMNB
111 #define MSGMNB 2048 /* max # of bytes in a queue */
112 #endif
113 #ifndef MSGMNI
114 #define MSGMNI 40
115 #endif
116 #ifndef MSGTQL
117 #define MSGTQL 40
118 #endif
119
120 /*
121 * Based on the configuration parameters described in an SVR2 (yes, two)
122 * config(1m) man page.
123 *
124 * Each message is broken up and stored in segments that are msgssz bytes
125 * long. For efficiency reasons, this should be a power of two. Also,
126 * it doesn't make sense if it is less than 8 or greater than about 256.
127 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of
128 * two between 8 and 1024 inclusive (and panic's if it isn't).
129 */
130 struct msginfo msginfo = {
131 MSGMAX, /* max chars in a message */
132 MSGMNI, /* # of message queue identifiers */
133 MSGMNB, /* max chars in a queue */
134 MSGTQL, /* max messages in system */
135 MSGSSZ, /* size of a message segment */
136 /* (must be small power of 2 greater than 4) */
137 MSGSEG /* number of message segments */
138 };
139
140 /*
141 * macros to convert between msqid_ds's and msqid's.
142 * (specific to this implementation)
143 */
144 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000))
145 #define MSQID_IX(id) ((id) & 0xffff)
146 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff)
147
148 /*
149 * The rest of this file is specific to this particular implementation.
150 */
151
152 struct msgmap {
153 short next; /* next segment in buffer */
154 /* -1 -> available */
155 /* 0..(MSGSEG-1) -> index of next segment */
156 };
157
158 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */
159
160 static int nfree_msgmaps; /* # of free map entries */
161 static short free_msgmaps; /* head of linked list of free map entries */
162 static struct msg *free_msghdrs;/* list of free msg headers */
163 static char *msgpool; /* MSGMAX byte long msg buffer pool */
164 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */
165 static struct msg *msghdrs; /* MSGTQL msg headers */
166 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */
167 static struct mtx msq_mtx; /* global mutex for message queues. */
168 static unsigned msg_prison_slot;/* prison OSD slot */
169
170 static struct syscall_helper_data msg_syscalls[] = {
171 SYSCALL_INIT_HELPER(msgctl),
172 SYSCALL_INIT_HELPER(msgget),
173 SYSCALL_INIT_HELPER(msgsnd),
174 SYSCALL_INIT_HELPER(msgrcv),
175 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
176 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
177 SYSCALL_INIT_HELPER(msgsys),
178 SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl),
179 #endif
180 SYSCALL_INIT_LAST
181 };
182
183 #ifdef COMPAT_FREEBSD32
184 #include <compat/freebsd32/freebsd32.h>
185 #include <compat/freebsd32/freebsd32_ipc.h>
186 #include <compat/freebsd32/freebsd32_proto.h>
187 #include <compat/freebsd32/freebsd32_signal.h>
188 #include <compat/freebsd32/freebsd32_syscall.h>
189 #include <compat/freebsd32/freebsd32_util.h>
190
191 static struct syscall_helper_data msg32_syscalls[] = {
192 SYSCALL32_INIT_HELPER(freebsd32_msgctl),
193 SYSCALL32_INIT_HELPER(freebsd32_msgsnd),
194 SYSCALL32_INIT_HELPER(freebsd32_msgrcv),
195 SYSCALL32_INIT_HELPER_COMPAT(msgget),
196 SYSCALL32_INIT_HELPER(freebsd32_msgsys),
197 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
198 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
199 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl),
200 #endif
201 SYSCALL_INIT_LAST
202 };
203 #endif
204
205 static int
206 msginit()
207 {
208 struct prison *pr;
209 void **rsv;
210 int i, error;
211 osd_method_t methods[PR_MAXMETHOD] = {
212 [PR_METHOD_CHECK] = msg_prison_check,
213 [PR_METHOD_SET] = msg_prison_set,
214 [PR_METHOD_GET] = msg_prison_get,
215 [PR_METHOD_REMOVE] = msg_prison_remove,
216 };
217
218 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz;
219 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK);
220 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK);
221 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK);
222 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG,
223 M_WAITOK | M_ZERO);
224
225 /*
226 * msginfo.msgssz should be a power of two for efficiency reasons.
227 * It is also pretty silly if msginfo.msgssz is less than 8
228 * or greater than about 256 so ...
229 */
230
231 i = 8;
232 while (i < 1024 && i != msginfo.msgssz)
233 i <<= 1;
234 if (i != msginfo.msgssz) {
235 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz,
236 msginfo.msgssz));
237 panic("msginfo.msgssz not a small power of 2");
238 }
239
240 if (msginfo.msgseg > 32767) {
241 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg));
242 panic("msginfo.msgseg > 32767");
243 }
244
245 for (i = 0; i < msginfo.msgseg; i++) {
246 if (i > 0)
247 msgmaps[i-1].next = i;
248 msgmaps[i].next = -1; /* implies entry is available */
249 }
250 free_msgmaps = 0;
251 nfree_msgmaps = msginfo.msgseg;
252
253 for (i = 0; i < msginfo.msgtql; i++) {
254 msghdrs[i].msg_type = 0;
255 if (i > 0)
256 msghdrs[i-1].msg_next = &msghdrs[i];
257 msghdrs[i].msg_next = NULL;
258 #ifdef MAC
259 mac_sysvmsg_init(&msghdrs[i]);
260 #endif
261 }
262 free_msghdrs = &msghdrs[0];
263
264 for (i = 0; i < msginfo.msgmni; i++) {
265 msqids[i].u.msg_qbytes = 0; /* implies entry is available */
266 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */
267 msqids[i].u.msg_perm.mode = 0;
268 #ifdef MAC
269 mac_sysvmsq_init(&msqids[i]);
270 #endif
271 }
272 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF);
273
274 /* Set current prisons according to their allow.sysvipc. */
275 msg_prison_slot = osd_jail_register(NULL, methods);
276 rsv = osd_reserve(msg_prison_slot);
277 prison_lock(&prison0);
278 (void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0);
279 prison_unlock(&prison0);
280 rsv = NULL;
281 sx_slock(&allprison_lock);
282 TAILQ_FOREACH(pr, &allprison, pr_list) {
283 if (rsv == NULL)
284 rsv = osd_reserve(msg_prison_slot);
285 prison_lock(pr);
286 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
287 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
288 &prison0);
289 rsv = NULL;
290 }
291 prison_unlock(pr);
292 }
293 if (rsv != NULL)
294 osd_free_reserved(rsv);
295 sx_sunlock(&allprison_lock);
296
297 error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD);
298 if (error != 0)
299 return (error);
300 #ifdef COMPAT_FREEBSD32
301 error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD);
302 if (error != 0)
303 return (error);
304 #endif
305 return (0);
306 }
307
308 static int
309 msgunload()
310 {
311 struct msqid_kernel *msqkptr;
312 int msqid;
313 #ifdef MAC
314 int i;
315 #endif
316
317 syscall_helper_unregister(msg_syscalls);
318 #ifdef COMPAT_FREEBSD32
319 syscall32_helper_unregister(msg32_syscalls);
320 #endif
321
322 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
323 msqkptr = &msqids[msqid];
324 if (msqkptr->u.msg_qbytes != 0 ||
325 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
326 break;
327 }
328 if (msqid != msginfo.msgmni)
329 return (EBUSY);
330
331 if (msg_prison_slot != 0)
332 osd_jail_deregister(msg_prison_slot);
333 #ifdef MAC
334 for (i = 0; i < msginfo.msgtql; i++)
335 mac_sysvmsg_destroy(&msghdrs[i]);
336 for (msqid = 0; msqid < msginfo.msgmni; msqid++)
337 mac_sysvmsq_destroy(&msqids[msqid]);
338 #endif
339 free(msgpool, M_MSG);
340 free(msgmaps, M_MSG);
341 free(msghdrs, M_MSG);
342 free(msqids, M_MSG);
343 mtx_destroy(&msq_mtx);
344 return (0);
345 }
346
347
348 static int
349 sysvmsg_modload(struct module *module, int cmd, void *arg)
350 {
351 int error = 0;
352
353 switch (cmd) {
354 case MOD_LOAD:
355 error = msginit();
356 if (error != 0)
357 msgunload();
358 break;
359 case MOD_UNLOAD:
360 error = msgunload();
361 break;
362 case MOD_SHUTDOWN:
363 break;
364 default:
365 error = EINVAL;
366 break;
367 }
368 return (error);
369 }
370
371 static moduledata_t sysvmsg_mod = {
372 "sysvmsg",
373 &sysvmsg_modload,
374 NULL
375 };
376
377 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST);
378 MODULE_VERSION(sysvmsg, 1);
379
380 static void
381 msg_freehdr(msghdr)
382 struct msg *msghdr;
383 {
384 while (msghdr->msg_ts > 0) {
385 short next;
386 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg)
387 panic("msghdr->msg_spot out of range");
388 next = msgmaps[msghdr->msg_spot].next;
389 msgmaps[msghdr->msg_spot].next = free_msgmaps;
390 free_msgmaps = msghdr->msg_spot;
391 nfree_msgmaps++;
392 msghdr->msg_spot = next;
393 if (msghdr->msg_ts >= msginfo.msgssz)
394 msghdr->msg_ts -= msginfo.msgssz;
395 else
396 msghdr->msg_ts = 0;
397 }
398 if (msghdr->msg_spot != -1)
399 panic("msghdr->msg_spot != -1");
400 msghdr->msg_next = free_msghdrs;
401 free_msghdrs = msghdr;
402 #ifdef MAC
403 mac_sysvmsg_cleanup(msghdr);
404 #endif
405 }
406
407 static void
408 msq_remove(struct msqid_kernel *msqkptr)
409 {
410 struct msg *msghdr;
411
412 racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1);
413 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum);
414 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes);
415 crfree(msqkptr->cred);
416 msqkptr->cred = NULL;
417
418 /* Free the message headers */
419 msghdr = msqkptr->u.msg_first;
420 while (msghdr != NULL) {
421 struct msg *msghdr_tmp;
422
423 /* Free the segments of each message */
424 msqkptr->u.msg_cbytes -= msghdr->msg_ts;
425 msqkptr->u.msg_qnum--;
426 msghdr_tmp = msghdr;
427 msghdr = msghdr->msg_next;
428 msg_freehdr(msghdr_tmp);
429 }
430
431 if (msqkptr->u.msg_cbytes != 0)
432 panic("msg_cbytes is screwed up");
433 if (msqkptr->u.msg_qnum != 0)
434 panic("msg_qnum is screwed up");
435
436 msqkptr->u.msg_qbytes = 0; /* Mark it as free */
437
438 #ifdef MAC
439 mac_sysvmsq_cleanup(msqkptr);
440 #endif
441
442 wakeup(msqkptr);
443 }
444
445 static struct prison *
446 msg_find_prison(struct ucred *cred)
447 {
448 struct prison *pr, *rpr;
449
450 pr = cred->cr_prison;
451 prison_lock(pr);
452 rpr = osd_jail_get(pr, msg_prison_slot);
453 prison_unlock(pr);
454 return rpr;
455 }
456
457 static int
458 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr)
459 {
460
461 if (msqkptr->cred == NULL ||
462 !(rpr == msqkptr->cred->cr_prison ||
463 prison_ischild(rpr, msqkptr->cred->cr_prison)))
464 return (EINVAL);
465 return (0);
466 }
467
468 #ifndef _SYS_SYSPROTO_H_
469 struct msgctl_args {
470 int msqid;
471 int cmd;
472 struct msqid_ds *buf;
473 };
474 #endif
475 int
476 sys_msgctl(struct thread *td, struct msgctl_args *uap)
477 {
478 int msqid = uap->msqid;
479 int cmd = uap->cmd;
480 struct msqid_ds msqbuf;
481 int error;
482
483 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf));
484 if (cmd == IPC_SET &&
485 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0)
486 return (error);
487 error = kern_msgctl(td, msqid, cmd, &msqbuf);
488 if (cmd == IPC_STAT && error == 0)
489 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds));
490 return (error);
491 }
492
493 int
494 kern_msgctl(td, msqid, cmd, msqbuf)
495 struct thread *td;
496 int msqid;
497 int cmd;
498 struct msqid_ds *msqbuf;
499 {
500 int rval, error, msqix;
501 struct msqid_kernel *msqkptr;
502 struct prison *rpr;
503
504 rpr = msg_find_prison(td->td_ucred);
505 if (rpr == NULL)
506 return (ENOSYS);
507
508 msqix = IPCID_TO_IX(msqid);
509
510 if (msqix < 0 || msqix >= msginfo.msgmni) {
511 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
512 msginfo.msgmni));
513 return (EINVAL);
514 }
515
516 msqkptr = &msqids[msqix];
517
518 mtx_lock(&msq_mtx);
519 if (msqkptr->u.msg_qbytes == 0) {
520 DPRINTF(("no such msqid\n"));
521 error = EINVAL;
522 goto done2;
523 }
524 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
525 DPRINTF(("wrong sequence number\n"));
526 error = EINVAL;
527 goto done2;
528 }
529
530 error = msq_prison_cansee(rpr, msqkptr);
531 if (error != 0) {
532 DPRINTF(("requester can't see prison\n"));
533 goto done2;
534 }
535
536 #ifdef MAC
537 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd);
538 if (error != 0)
539 goto done2;
540 #endif
541
542 error = 0;
543 rval = 0;
544
545 switch (cmd) {
546
547 case IPC_RMID:
548 {
549 #ifdef MAC
550 struct msg *msghdr;
551 #endif
552 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
553 goto done2;
554
555 #ifdef MAC
556 /*
557 * Check that the thread has MAC access permissions to
558 * individual msghdrs. Note: We need to do this in a
559 * separate loop because the actual loop alters the
560 * msq/msghdr info as it progresses, and there is no going
561 * back if half the way through we discover that the
562 * thread cannot free a certain msghdr. The msq will get
563 * into an inconsistent state.
564 */
565 for (msghdr = msqkptr->u.msg_first; msghdr != NULL;
566 msghdr = msghdr->msg_next) {
567 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr);
568 if (error != 0)
569 goto done2;
570 }
571 #endif
572
573 msq_remove(msqkptr);
574 }
575
576 break;
577
578 case IPC_SET:
579 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M)))
580 goto done2;
581 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) {
582 error = priv_check(td, PRIV_IPC_MSGSIZE);
583 if (error)
584 goto done2;
585 }
586 if (msqbuf->msg_qbytes > msginfo.msgmnb) {
587 DPRINTF(("can't increase msg_qbytes beyond %d"
588 "(truncating)\n", msginfo.msgmnb));
589 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */
590 }
591 if (msqbuf->msg_qbytes == 0) {
592 DPRINTF(("can't reduce msg_qbytes to 0\n"));
593 error = EINVAL; /* non-standard errno! */
594 goto done2;
595 }
596 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */
597 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */
598 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) |
599 (msqbuf->msg_perm.mode & 0777);
600 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes;
601 msqkptr->u.msg_ctime = time_second;
602 break;
603
604 case IPC_STAT:
605 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
606 DPRINTF(("requester doesn't have read access\n"));
607 goto done2;
608 }
609 *msqbuf = msqkptr->u;
610 if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison)
611 msqbuf->msg_perm.key = IPC_PRIVATE;
612 break;
613
614 default:
615 DPRINTF(("invalid command %d\n", cmd));
616 error = EINVAL;
617 goto done2;
618 }
619
620 if (error == 0)
621 td->td_retval[0] = rval;
622 done2:
623 mtx_unlock(&msq_mtx);
624 return (error);
625 }
626
627 #ifndef _SYS_SYSPROTO_H_
628 struct msgget_args {
629 key_t key;
630 int msgflg;
631 };
632 #endif
633
634 int
635 sys_msgget(struct thread *td, struct msgget_args *uap)
636 {
637 int msqid, error = 0;
638 int key = uap->key;
639 int msgflg = uap->msgflg;
640 struct ucred *cred = td->td_ucred;
641 struct msqid_kernel *msqkptr = NULL;
642
643 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg));
644
645 if (msg_find_prison(cred) == NULL)
646 return (ENOSYS);
647
648 mtx_lock(&msq_mtx);
649 if (key != IPC_PRIVATE) {
650 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
651 msqkptr = &msqids[msqid];
652 if (msqkptr->u.msg_qbytes != 0 &&
653 msqkptr->cred != NULL &&
654 msqkptr->cred->cr_prison == cred->cr_prison &&
655 msqkptr->u.msg_perm.key == key)
656 break;
657 }
658 if (msqid < msginfo.msgmni) {
659 DPRINTF(("found public key\n"));
660 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) {
661 DPRINTF(("not exclusive\n"));
662 error = EEXIST;
663 goto done2;
664 }
665 if ((error = ipcperm(td, &msqkptr->u.msg_perm,
666 msgflg & 0700))) {
667 DPRINTF(("requester doesn't have 0%o access\n",
668 msgflg & 0700));
669 goto done2;
670 }
671 #ifdef MAC
672 error = mac_sysvmsq_check_msqget(cred, msqkptr);
673 if (error != 0)
674 goto done2;
675 #endif
676 goto found;
677 }
678 }
679
680 DPRINTF(("need to allocate the msqid_ds\n"));
681 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) {
682 for (msqid = 0; msqid < msginfo.msgmni; msqid++) {
683 /*
684 * Look for an unallocated and unlocked msqid_ds.
685 * msqid_ds's can be locked by msgsnd or msgrcv while
686 * they are copying the message in/out. We can't
687 * re-use the entry until they release it.
688 */
689 msqkptr = &msqids[msqid];
690 if (msqkptr->u.msg_qbytes == 0 &&
691 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0)
692 break;
693 }
694 if (msqid == msginfo.msgmni) {
695 DPRINTF(("no more msqid_ds's available\n"));
696 error = ENOSPC;
697 goto done2;
698 }
699 #ifdef RACCT
700 if (racct_enable) {
701 PROC_LOCK(td->td_proc);
702 error = racct_add(td->td_proc, RACCT_NMSGQ, 1);
703 PROC_UNLOCK(td->td_proc);
704 if (error != 0) {
705 error = ENOSPC;
706 goto done2;
707 }
708 }
709 #endif
710 DPRINTF(("msqid %d is available\n", msqid));
711 msqkptr->u.msg_perm.key = key;
712 msqkptr->u.msg_perm.cuid = cred->cr_uid;
713 msqkptr->u.msg_perm.uid = cred->cr_uid;
714 msqkptr->u.msg_perm.cgid = cred->cr_gid;
715 msqkptr->u.msg_perm.gid = cred->cr_gid;
716 msqkptr->u.msg_perm.mode = (msgflg & 0777);
717 msqkptr->cred = crhold(cred);
718 /* Make sure that the returned msqid is unique */
719 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff;
720 msqkptr->u.msg_first = NULL;
721 msqkptr->u.msg_last = NULL;
722 msqkptr->u.msg_cbytes = 0;
723 msqkptr->u.msg_qnum = 0;
724 msqkptr->u.msg_qbytes = msginfo.msgmnb;
725 msqkptr->u.msg_lspid = 0;
726 msqkptr->u.msg_lrpid = 0;
727 msqkptr->u.msg_stime = 0;
728 msqkptr->u.msg_rtime = 0;
729 msqkptr->u.msg_ctime = time_second;
730 #ifdef MAC
731 mac_sysvmsq_create(cred, msqkptr);
732 #endif
733 } else {
734 DPRINTF(("didn't find it and wasn't asked to create it\n"));
735 error = ENOENT;
736 goto done2;
737 }
738
739 found:
740 /* Construct the unique msqid */
741 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm);
742 done2:
743 mtx_unlock(&msq_mtx);
744 return (error);
745 }
746
747 #ifndef _SYS_SYSPROTO_H_
748 struct msgsnd_args {
749 int msqid;
750 const void *msgp; /* XXX msgp is actually mtext. */
751 size_t msgsz;
752 int msgflg;
753 };
754 #endif
755 int
756 kern_msgsnd(struct thread *td, int msqid, const void *msgp,
757 size_t msgsz, int msgflg, long mtype)
758 {
759 int msqix, segs_needed, error = 0;
760 struct msqid_kernel *msqkptr;
761 struct msg *msghdr;
762 struct prison *rpr;
763 short next;
764 #ifdef RACCT
765 size_t saved_msgsz;
766 #endif
767
768 rpr = msg_find_prison(td->td_ucred);
769 if (rpr == NULL)
770 return (ENOSYS);
771
772 mtx_lock(&msq_mtx);
773 msqix = IPCID_TO_IX(msqid);
774
775 if (msqix < 0 || msqix >= msginfo.msgmni) {
776 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
777 msginfo.msgmni));
778 error = EINVAL;
779 goto done2;
780 }
781
782 msqkptr = &msqids[msqix];
783 if (msqkptr->u.msg_qbytes == 0) {
784 DPRINTF(("no such message queue id\n"));
785 error = EINVAL;
786 goto done2;
787 }
788 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
789 DPRINTF(("wrong sequence number\n"));
790 error = EINVAL;
791 goto done2;
792 }
793
794 if ((error = msq_prison_cansee(rpr, msqkptr))) {
795 DPRINTF(("requester can't see prison\n"));
796 goto done2;
797 }
798
799 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) {
800 DPRINTF(("requester doesn't have write access\n"));
801 goto done2;
802 }
803
804 #ifdef MAC
805 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr);
806 if (error != 0)
807 goto done2;
808 #endif
809
810 #ifdef RACCT
811 if (racct_enable) {
812 PROC_LOCK(td->td_proc);
813 if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) {
814 PROC_UNLOCK(td->td_proc);
815 error = EAGAIN;
816 goto done2;
817 }
818 saved_msgsz = msgsz;
819 if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) {
820 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
821 PROC_UNLOCK(td->td_proc);
822 error = EAGAIN;
823 goto done2;
824 }
825 PROC_UNLOCK(td->td_proc);
826 }
827 #endif
828
829 segs_needed = howmany(msgsz, msginfo.msgssz);
830 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz,
831 msginfo.msgssz, segs_needed));
832 for (;;) {
833 int need_more_resources = 0;
834
835 /*
836 * check msgsz
837 * (inside this loop in case msg_qbytes changes while we sleep)
838 */
839
840 if (msgsz > msqkptr->u.msg_qbytes) {
841 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n"));
842 error = EINVAL;
843 goto done3;
844 }
845
846 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) {
847 DPRINTF(("msqid is locked\n"));
848 need_more_resources = 1;
849 }
850 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) {
851 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n"));
852 need_more_resources = 1;
853 }
854 if (segs_needed > nfree_msgmaps) {
855 DPRINTF(("segs_needed > nfree_msgmaps\n"));
856 need_more_resources = 1;
857 }
858 if (free_msghdrs == NULL) {
859 DPRINTF(("no more msghdrs\n"));
860 need_more_resources = 1;
861 }
862
863 if (need_more_resources) {
864 int we_own_it;
865
866 if ((msgflg & IPC_NOWAIT) != 0) {
867 DPRINTF(("need more resources but caller "
868 "doesn't want to wait\n"));
869 error = EAGAIN;
870 goto done3;
871 }
872
873 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) {
874 DPRINTF(("we don't own the msqid_ds\n"));
875 we_own_it = 0;
876 } else {
877 /* Force later arrivals to wait for our
878 request */
879 DPRINTF(("we own the msqid_ds\n"));
880 msqkptr->u.msg_perm.mode |= MSG_LOCKED;
881 we_own_it = 1;
882 }
883 DPRINTF(("msgsnd: goodnight\n"));
884 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
885 "msgsnd", hz);
886 DPRINTF(("msgsnd: good morning, error=%d\n", error));
887 if (we_own_it)
888 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
889 if (error == EWOULDBLOCK) {
890 DPRINTF(("msgsnd: timed out\n"));
891 continue;
892 }
893 if (error != 0) {
894 DPRINTF(("msgsnd: interrupted system call\n"));
895 error = EINTR;
896 goto done3;
897 }
898
899 /*
900 * Make sure that the msq queue still exists
901 */
902
903 if (msqkptr->u.msg_qbytes == 0) {
904 DPRINTF(("msqid deleted\n"));
905 error = EIDRM;
906 goto done3;
907 }
908
909 } else {
910 DPRINTF(("got all the resources that we need\n"));
911 break;
912 }
913 }
914
915 /*
916 * We have the resources that we need.
917 * Make sure!
918 */
919
920 if (msqkptr->u.msg_perm.mode & MSG_LOCKED)
921 panic("msg_perm.mode & MSG_LOCKED");
922 if (segs_needed > nfree_msgmaps)
923 panic("segs_needed > nfree_msgmaps");
924 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes)
925 panic("msgsz + msg_cbytes > msg_qbytes");
926 if (free_msghdrs == NULL)
927 panic("no more msghdrs");
928
929 /*
930 * Re-lock the msqid_ds in case we page-fault when copying in the
931 * message
932 */
933
934 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0)
935 panic("msqid_ds is already locked");
936 msqkptr->u.msg_perm.mode |= MSG_LOCKED;
937
938 /*
939 * Allocate a message header
940 */
941
942 msghdr = free_msghdrs;
943 free_msghdrs = msghdr->msg_next;
944 msghdr->msg_spot = -1;
945 msghdr->msg_ts = msgsz;
946 msghdr->msg_type = mtype;
947 #ifdef MAC
948 /*
949 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here
950 * immediately? Or, should it be checked just before the msg is
951 * enqueued in the msgq (as it is done now)?
952 */
953 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr);
954 #endif
955
956 /*
957 * Allocate space for the message
958 */
959
960 while (segs_needed > 0) {
961 if (nfree_msgmaps <= 0)
962 panic("not enough msgmaps");
963 if (free_msgmaps == -1)
964 panic("nil free_msgmaps");
965 next = free_msgmaps;
966 if (next <= -1)
967 panic("next too low #1");
968 if (next >= msginfo.msgseg)
969 panic("next out of range #1");
970 DPRINTF(("allocating segment %d to message\n", next));
971 free_msgmaps = msgmaps[next].next;
972 nfree_msgmaps--;
973 msgmaps[next].next = msghdr->msg_spot;
974 msghdr->msg_spot = next;
975 segs_needed--;
976 }
977
978 /*
979 * Validate the message type
980 */
981
982 if (msghdr->msg_type < 1) {
983 msg_freehdr(msghdr);
984 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
985 wakeup(msqkptr);
986 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type));
987 error = EINVAL;
988 goto done3;
989 }
990
991 /*
992 * Copy in the message body
993 */
994
995 next = msghdr->msg_spot;
996 while (msgsz > 0) {
997 size_t tlen;
998 if (msgsz > msginfo.msgssz)
999 tlen = msginfo.msgssz;
1000 else
1001 tlen = msgsz;
1002 if (next <= -1)
1003 panic("next too low #2");
1004 if (next >= msginfo.msgseg)
1005 panic("next out of range #2");
1006 mtx_unlock(&msq_mtx);
1007 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz],
1008 tlen)) != 0) {
1009 mtx_lock(&msq_mtx);
1010 DPRINTF(("error %d copying in message segment\n",
1011 error));
1012 msg_freehdr(msghdr);
1013 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1014 wakeup(msqkptr);
1015 goto done3;
1016 }
1017 mtx_lock(&msq_mtx);
1018 msgsz -= tlen;
1019 msgp = (const char *)msgp + tlen;
1020 next = msgmaps[next].next;
1021 }
1022 if (next != -1)
1023 panic("didn't use all the msg segments");
1024
1025 /*
1026 * We've got the message. Unlock the msqid_ds.
1027 */
1028
1029 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED;
1030
1031 /*
1032 * Make sure that the msqid_ds is still allocated.
1033 */
1034
1035 if (msqkptr->u.msg_qbytes == 0) {
1036 msg_freehdr(msghdr);
1037 wakeup(msqkptr);
1038 error = EIDRM;
1039 goto done3;
1040 }
1041
1042 #ifdef MAC
1043 /*
1044 * Note: Since the task/thread allocates the msghdr and usually
1045 * primes it with its own MAC label, for a majority of policies, it
1046 * won't be necessary to check whether the msghdr has access
1047 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would
1048 * suffice in that case. However, this hook may be required where
1049 * individual policies derive a non-identical label for the msghdr
1050 * from the current thread label and may want to check the msghdr
1051 * enqueue permissions, along with read/write permissions to the
1052 * msgq.
1053 */
1054 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr);
1055 if (error != 0) {
1056 msg_freehdr(msghdr);
1057 wakeup(msqkptr);
1058 goto done3;
1059 }
1060 #endif
1061
1062 /*
1063 * Put the message into the queue
1064 */
1065 if (msqkptr->u.msg_first == NULL) {
1066 msqkptr->u.msg_first = msghdr;
1067 msqkptr->u.msg_last = msghdr;
1068 } else {
1069 msqkptr->u.msg_last->msg_next = msghdr;
1070 msqkptr->u.msg_last = msghdr;
1071 }
1072 msqkptr->u.msg_last->msg_next = NULL;
1073
1074 msqkptr->u.msg_cbytes += msghdr->msg_ts;
1075 msqkptr->u.msg_qnum++;
1076 msqkptr->u.msg_lspid = td->td_proc->p_pid;
1077 msqkptr->u.msg_stime = time_second;
1078
1079 wakeup(msqkptr);
1080 td->td_retval[0] = 0;
1081 done3:
1082 #ifdef RACCT
1083 if (racct_enable && error != 0) {
1084 PROC_LOCK(td->td_proc);
1085 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1);
1086 racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz);
1087 PROC_UNLOCK(td->td_proc);
1088 }
1089 #endif
1090 done2:
1091 mtx_unlock(&msq_mtx);
1092 return (error);
1093 }
1094
1095 int
1096 sys_msgsnd(struct thread *td, struct msgsnd_args *uap)
1097 {
1098 int error;
1099 long mtype;
1100
1101 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp,
1102 uap->msgsz, uap->msgflg));
1103
1104 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) {
1105 DPRINTF(("error %d copying the message type\n", error));
1106 return (error);
1107 }
1108 return (kern_msgsnd(td, uap->msqid,
1109 (const char *)uap->msgp + sizeof(mtype),
1110 uap->msgsz, uap->msgflg, mtype));
1111 }
1112
1113 #ifndef _SYS_SYSPROTO_H_
1114 struct msgrcv_args {
1115 int msqid;
1116 void *msgp;
1117 size_t msgsz;
1118 long msgtyp;
1119 int msgflg;
1120 };
1121 #endif
1122 /* XXX msgp is actually mtext. */
1123 int
1124 kern_msgrcv(struct thread *td, int msqid, void *msgp, size_t msgsz, long msgtyp,
1125 int msgflg, long *mtype)
1126 {
1127 size_t len;
1128 struct msqid_kernel *msqkptr;
1129 struct msg *msghdr;
1130 struct prison *rpr;
1131 int msqix, error = 0;
1132 short next;
1133
1134 rpr = msg_find_prison(td->td_ucred);
1135 if (rpr == NULL)
1136 return (ENOSYS);
1137
1138 msqix = IPCID_TO_IX(msqid);
1139
1140 if (msqix < 0 || msqix >= msginfo.msgmni) {
1141 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix,
1142 msginfo.msgmni));
1143 return (EINVAL);
1144 }
1145
1146 msqkptr = &msqids[msqix];
1147 mtx_lock(&msq_mtx);
1148 if (msqkptr->u.msg_qbytes == 0) {
1149 DPRINTF(("no such message queue id\n"));
1150 error = EINVAL;
1151 goto done2;
1152 }
1153 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1154 DPRINTF(("wrong sequence number\n"));
1155 error = EINVAL;
1156 goto done2;
1157 }
1158
1159 if ((error = msq_prison_cansee(rpr, msqkptr))) {
1160 DPRINTF(("requester can't see prison\n"));
1161 goto done2;
1162 }
1163
1164 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) {
1165 DPRINTF(("requester doesn't have read access\n"));
1166 goto done2;
1167 }
1168
1169 #ifdef MAC
1170 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr);
1171 if (error != 0)
1172 goto done2;
1173 #endif
1174
1175 msghdr = NULL;
1176 while (msghdr == NULL) {
1177 if (msgtyp == 0) {
1178 msghdr = msqkptr->u.msg_first;
1179 if (msghdr != NULL) {
1180 if (msgsz < msghdr->msg_ts &&
1181 (msgflg & MSG_NOERROR) == 0) {
1182 DPRINTF(("first message on the queue "
1183 "is too big (want %zu, got %d)\n",
1184 msgsz, msghdr->msg_ts));
1185 error = E2BIG;
1186 goto done2;
1187 }
1188 #ifdef MAC
1189 error = mac_sysvmsq_check_msgrcv(td->td_ucred,
1190 msghdr);
1191 if (error != 0)
1192 goto done2;
1193 #endif
1194 if (msqkptr->u.msg_first == msqkptr->u.msg_last) {
1195 msqkptr->u.msg_first = NULL;
1196 msqkptr->u.msg_last = NULL;
1197 } else {
1198 msqkptr->u.msg_first = msghdr->msg_next;
1199 if (msqkptr->u.msg_first == NULL)
1200 panic("msg_first/last screwed up #1");
1201 }
1202 }
1203 } else {
1204 struct msg *previous;
1205 struct msg **prev;
1206
1207 previous = NULL;
1208 prev = &(msqkptr->u.msg_first);
1209 while ((msghdr = *prev) != NULL) {
1210 /*
1211 * Is this message's type an exact match or is
1212 * this message's type less than or equal to
1213 * the absolute value of a negative msgtyp?
1214 * Note that the second half of this test can
1215 * NEVER be true if msgtyp is positive since
1216 * msg_type is always positive!
1217 */
1218
1219 if (msgtyp == msghdr->msg_type ||
1220 msghdr->msg_type <= -msgtyp) {
1221 DPRINTF(("found message type %ld, "
1222 "requested %ld\n",
1223 msghdr->msg_type, msgtyp));
1224 if (msgsz < msghdr->msg_ts &&
1225 (msgflg & MSG_NOERROR) == 0) {
1226 DPRINTF(("requested message "
1227 "on the queue is too big "
1228 "(want %zu, got %hu)\n",
1229 msgsz, msghdr->msg_ts));
1230 error = E2BIG;
1231 goto done2;
1232 }
1233 #ifdef MAC
1234 error = mac_sysvmsq_check_msgrcv(
1235 td->td_ucred, msghdr);
1236 if (error != 0)
1237 goto done2;
1238 #endif
1239 *prev = msghdr->msg_next;
1240 if (msghdr == msqkptr->u.msg_last) {
1241 if (previous == NULL) {
1242 if (prev !=
1243 &msqkptr->u.msg_first)
1244 panic("msg_first/last screwed up #2");
1245 msqkptr->u.msg_first =
1246 NULL;
1247 msqkptr->u.msg_last =
1248 NULL;
1249 } else {
1250 if (prev ==
1251 &msqkptr->u.msg_first)
1252 panic("msg_first/last screwed up #3");
1253 msqkptr->u.msg_last =
1254 previous;
1255 }
1256 }
1257 break;
1258 }
1259 previous = msghdr;
1260 prev = &(msghdr->msg_next);
1261 }
1262 }
1263
1264 /*
1265 * We've either extracted the msghdr for the appropriate
1266 * message or there isn't one.
1267 * If there is one then bail out of this loop.
1268 */
1269
1270 if (msghdr != NULL)
1271 break;
1272
1273 /*
1274 * Hmph! No message found. Does the user want to wait?
1275 */
1276
1277 if ((msgflg & IPC_NOWAIT) != 0) {
1278 DPRINTF(("no appropriate message found (msgtyp=%ld)\n",
1279 msgtyp));
1280 /* The SVID says to return ENOMSG. */
1281 error = ENOMSG;
1282 goto done2;
1283 }
1284
1285 /*
1286 * Wait for something to happen
1287 */
1288
1289 DPRINTF(("msgrcv: goodnight\n"));
1290 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH,
1291 "msgrcv", 0);
1292 DPRINTF(("msgrcv: good morning (error=%d)\n", error));
1293
1294 if (error != 0) {
1295 DPRINTF(("msgrcv: interrupted system call\n"));
1296 error = EINTR;
1297 goto done2;
1298 }
1299
1300 /*
1301 * Make sure that the msq queue still exists
1302 */
1303
1304 if (msqkptr->u.msg_qbytes == 0 ||
1305 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) {
1306 DPRINTF(("msqid deleted\n"));
1307 error = EIDRM;
1308 goto done2;
1309 }
1310 }
1311
1312 /*
1313 * Return the message to the user.
1314 *
1315 * First, do the bookkeeping (before we risk being interrupted).
1316 */
1317
1318 msqkptr->u.msg_cbytes -= msghdr->msg_ts;
1319 msqkptr->u.msg_qnum--;
1320 msqkptr->u.msg_lrpid = td->td_proc->p_pid;
1321 msqkptr->u.msg_rtime = time_second;
1322
1323 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1);
1324 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts);
1325
1326 /*
1327 * Make msgsz the actual amount that we'll be returning.
1328 * Note that this effectively truncates the message if it is too long
1329 * (since msgsz is never increased).
1330 */
1331
1332 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz,
1333 msghdr->msg_ts));
1334 if (msgsz > msghdr->msg_ts)
1335 msgsz = msghdr->msg_ts;
1336 *mtype = msghdr->msg_type;
1337
1338 /*
1339 * Return the segments to the user
1340 */
1341
1342 next = msghdr->msg_spot;
1343 for (len = 0; len < msgsz; len += msginfo.msgssz) {
1344 size_t tlen;
1345
1346 if (msgsz - len > msginfo.msgssz)
1347 tlen = msginfo.msgssz;
1348 else
1349 tlen = msgsz - len;
1350 if (next <= -1)
1351 panic("next too low #3");
1352 if (next >= msginfo.msgseg)
1353 panic("next out of range #3");
1354 mtx_unlock(&msq_mtx);
1355 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen);
1356 mtx_lock(&msq_mtx);
1357 if (error != 0) {
1358 DPRINTF(("error (%d) copying out message segment\n",
1359 error));
1360 msg_freehdr(msghdr);
1361 wakeup(msqkptr);
1362 goto done2;
1363 }
1364 msgp = (char *)msgp + tlen;
1365 next = msgmaps[next].next;
1366 }
1367
1368 /*
1369 * Done, return the actual number of bytes copied out.
1370 */
1371
1372 msg_freehdr(msghdr);
1373 wakeup(msqkptr);
1374 td->td_retval[0] = msgsz;
1375 done2:
1376 mtx_unlock(&msq_mtx);
1377 return (error);
1378 }
1379
1380 int
1381 sys_msgrcv(struct thread *td, struct msgrcv_args *uap)
1382 {
1383 int error;
1384 long mtype;
1385
1386 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid,
1387 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg));
1388
1389 if ((error = kern_msgrcv(td, uap->msqid,
1390 (char *)uap->msgp + sizeof(mtype), uap->msgsz,
1391 uap->msgtyp, uap->msgflg, &mtype)) != 0)
1392 return (error);
1393 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0)
1394 DPRINTF(("error %d copying the message type\n", error));
1395 return (error);
1396 }
1397
1398 static int
1399 sysctl_msqids(SYSCTL_HANDLER_ARGS)
1400 {
1401 struct msqid_kernel tmsqk;
1402 #ifdef COMPAT_FREEBSD32
1403 struct msqid_kernel32 tmsqk32;
1404 #endif
1405 struct prison *pr, *rpr;
1406 void *outaddr;
1407 size_t outsize;
1408 int error, i;
1409
1410 pr = req->td->td_ucred->cr_prison;
1411 rpr = msg_find_prison(req->td->td_ucred);
1412 error = 0;
1413 for (i = 0; i < msginfo.msgmni; i++) {
1414 mtx_lock(&msq_mtx);
1415 if (msqids[i].u.msg_qbytes == 0 || rpr == NULL ||
1416 msq_prison_cansee(rpr, &msqids[i]) != 0)
1417 bzero(&tmsqk, sizeof(tmsqk));
1418 else {
1419 tmsqk = msqids[i];
1420 if (tmsqk.cred->cr_prison != pr)
1421 tmsqk.u.msg_perm.key = IPC_PRIVATE;
1422 }
1423 mtx_unlock(&msq_mtx);
1424 #ifdef COMPAT_FREEBSD32
1425 if (SV_CURPROC_FLAG(SV_ILP32)) {
1426 bzero(&tmsqk32, sizeof(tmsqk32));
1427 freebsd32_ipcperm_out(&tmsqk.u.msg_perm,
1428 &tmsqk32.u.msg_perm);
1429 /* Don't copy u.msg_first or u.msg_last */
1430 CP(tmsqk, tmsqk32, u.msg_cbytes);
1431 CP(tmsqk, tmsqk32, u.msg_qnum);
1432 CP(tmsqk, tmsqk32, u.msg_qbytes);
1433 CP(tmsqk, tmsqk32, u.msg_lspid);
1434 CP(tmsqk, tmsqk32, u.msg_lrpid);
1435 CP(tmsqk, tmsqk32, u.msg_stime);
1436 CP(tmsqk, tmsqk32, u.msg_rtime);
1437 CP(tmsqk, tmsqk32, u.msg_ctime);
1438 /* Don't copy label or cred */
1439 outaddr = &tmsqk32;
1440 outsize = sizeof(tmsqk32);
1441 } else
1442 #endif
1443 {
1444 /* Don't leak kernel pointers */
1445 tmsqk.u.msg_first = NULL;
1446 tmsqk.u.msg_last = NULL;
1447 tmsqk.label = NULL;
1448 tmsqk.cred = NULL;
1449 /*
1450 * XXX: some padding also exists, but we take care to
1451 * allocate our pool of msqid_kernel structs with
1452 * zeroed memory so this should be OK.
1453 */
1454 outaddr = &tmsqk;
1455 outsize = sizeof(tmsqk);
1456 }
1457 error = SYSCTL_OUT(req, outaddr, outsize);
1458 if (error != 0)
1459 break;
1460 }
1461 return (error);
1462 }
1463
1464 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0,
1465 "Maximum message size");
1466 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0,
1467 "Number of message queue identifiers");
1468 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0,
1469 "Maximum number of bytes in a queue");
1470 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0,
1471 "Maximum number of messages in the system");
1472 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0,
1473 "Size of a message segment");
1474 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0,
1475 "Number of message segments");
1476 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids,
1477 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1478 NULL, 0, sysctl_msqids, "",
1479 "Array of struct msqid_kernel for each potential message queue");
1480
1481 static int
1482 msg_prison_check(void *obj, void *data)
1483 {
1484 struct prison *pr = obj;
1485 struct prison *prpr;
1486 struct vfsoptlist *opts = data;
1487 int error, jsys;
1488
1489 /*
1490 * sysvmsg is a jailsys integer.
1491 * It must be "disable" if the parent jail is disabled.
1492 */
1493 error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1494 if (error != ENOENT) {
1495 if (error != 0)
1496 return (error);
1497 switch (jsys) {
1498 case JAIL_SYS_DISABLE:
1499 break;
1500 case JAIL_SYS_NEW:
1501 case JAIL_SYS_INHERIT:
1502 prison_lock(pr->pr_parent);
1503 prpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1504 prison_unlock(pr->pr_parent);
1505 if (prpr == NULL)
1506 return (EPERM);
1507 break;
1508 default:
1509 return (EINVAL);
1510 }
1511 }
1512
1513 return (0);
1514 }
1515
1516 static int
1517 msg_prison_set(void *obj, void *data)
1518 {
1519 struct prison *pr = obj;
1520 struct prison *tpr, *orpr, *nrpr, *trpr;
1521 struct vfsoptlist *opts = data;
1522 void *rsv;
1523 int jsys, descend;
1524
1525 /*
1526 * sysvmsg controls which jail is the root of the associated msgs (this
1527 * jail or same as the parent), or if the feature is available at all.
1528 */
1529 if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT)
1530 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1531 ? JAIL_SYS_INHERIT
1532 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1533 ? JAIL_SYS_DISABLE
1534 : -1;
1535 if (jsys == JAIL_SYS_DISABLE) {
1536 prison_lock(pr);
1537 orpr = osd_jail_get(pr, msg_prison_slot);
1538 if (orpr != NULL)
1539 osd_jail_del(pr, msg_prison_slot);
1540 prison_unlock(pr);
1541 if (orpr != NULL) {
1542 if (orpr == pr)
1543 msg_prison_cleanup(pr);
1544 /* Disable all child jails as well. */
1545 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1546 prison_lock(tpr);
1547 trpr = osd_jail_get(tpr, msg_prison_slot);
1548 if (trpr != NULL) {
1549 osd_jail_del(tpr, msg_prison_slot);
1550 prison_unlock(tpr);
1551 if (trpr == tpr)
1552 msg_prison_cleanup(tpr);
1553 } else {
1554 prison_unlock(tpr);
1555 descend = 0;
1556 }
1557 }
1558 }
1559 } else if (jsys != -1) {
1560 if (jsys == JAIL_SYS_NEW)
1561 nrpr = pr;
1562 else {
1563 prison_lock(pr->pr_parent);
1564 nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot);
1565 prison_unlock(pr->pr_parent);
1566 }
1567 rsv = osd_reserve(msg_prison_slot);
1568 prison_lock(pr);
1569 orpr = osd_jail_get(pr, msg_prison_slot);
1570 if (orpr != nrpr)
1571 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv,
1572 nrpr);
1573 else
1574 osd_free_reserved(rsv);
1575 prison_unlock(pr);
1576 if (orpr != nrpr) {
1577 if (orpr == pr)
1578 msg_prison_cleanup(pr);
1579 if (orpr != NULL) {
1580 /* Change child jails matching the old root, */
1581 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1582 prison_lock(tpr);
1583 trpr = osd_jail_get(tpr,
1584 msg_prison_slot);
1585 if (trpr == orpr) {
1586 (void)osd_jail_set(tpr,
1587 msg_prison_slot, nrpr);
1588 prison_unlock(tpr);
1589 if (trpr == tpr)
1590 msg_prison_cleanup(tpr);
1591 } else {
1592 prison_unlock(tpr);
1593 descend = 0;
1594 }
1595 }
1596 }
1597 }
1598 }
1599
1600 return (0);
1601 }
1602
1603 static int
1604 msg_prison_get(void *obj, void *data)
1605 {
1606 struct prison *pr = obj;
1607 struct prison *rpr;
1608 struct vfsoptlist *opts = data;
1609 int error, jsys;
1610
1611 /* Set sysvmsg based on the jail's root prison. */
1612 prison_lock(pr);
1613 rpr = osd_jail_get(pr, msg_prison_slot);
1614 prison_unlock(pr);
1615 jsys = rpr == NULL ? JAIL_SYS_DISABLE
1616 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1617 error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys));
1618 if (error == ENOENT)
1619 error = 0;
1620 return (error);
1621 }
1622
1623 static int
1624 msg_prison_remove(void *obj, void *data __unused)
1625 {
1626 struct prison *pr = obj;
1627 struct prison *rpr;
1628
1629 prison_lock(pr);
1630 rpr = osd_jail_get(pr, msg_prison_slot);
1631 prison_unlock(pr);
1632 if (rpr == pr)
1633 msg_prison_cleanup(pr);
1634 return (0);
1635 }
1636
1637 static void
1638 msg_prison_cleanup(struct prison *pr)
1639 {
1640 struct msqid_kernel *msqkptr;
1641 int i;
1642
1643 /* Remove any msqs that belong to this jail. */
1644 mtx_lock(&msq_mtx);
1645 for (i = 0; i < msginfo.msgmni; i++) {
1646 msqkptr = &msqids[i];
1647 if (msqkptr->u.msg_qbytes != 0 &&
1648 msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr)
1649 msq_remove(msqkptr);
1650 }
1651 mtx_unlock(&msq_mtx);
1652 }
1653
1654 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues");
1655
1656 #ifdef COMPAT_FREEBSD32
1657 int
1658 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap)
1659 {
1660
1661 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1662 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1663 switch (uap->which) {
1664 case 0:
1665 return (freebsd7_freebsd32_msgctl(td,
1666 (struct freebsd7_freebsd32_msgctl_args *)&uap->a2));
1667 case 2:
1668 return (freebsd32_msgsnd(td,
1669 (struct freebsd32_msgsnd_args *)&uap->a2));
1670 case 3:
1671 return (freebsd32_msgrcv(td,
1672 (struct freebsd32_msgrcv_args *)&uap->a2));
1673 default:
1674 return (sys_msgsys(td, (struct msgsys_args *)uap));
1675 }
1676 #else
1677 return (nosys(td, NULL));
1678 #endif
1679 }
1680
1681 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1682 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1683 int
1684 freebsd7_freebsd32_msgctl(struct thread *td,
1685 struct freebsd7_freebsd32_msgctl_args *uap)
1686 {
1687 struct msqid_ds msqbuf;
1688 struct msqid_ds32_old msqbuf32;
1689 int error;
1690
1691 if (uap->cmd == IPC_SET) {
1692 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1693 if (error)
1694 return (error);
1695 freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1696 PTRIN_CP(msqbuf32, msqbuf, msg_first);
1697 PTRIN_CP(msqbuf32, msqbuf, msg_last);
1698 CP(msqbuf32, msqbuf, msg_cbytes);
1699 CP(msqbuf32, msqbuf, msg_qnum);
1700 CP(msqbuf32, msqbuf, msg_qbytes);
1701 CP(msqbuf32, msqbuf, msg_lspid);
1702 CP(msqbuf32, msqbuf, msg_lrpid);
1703 CP(msqbuf32, msqbuf, msg_stime);
1704 CP(msqbuf32, msqbuf, msg_rtime);
1705 CP(msqbuf32, msqbuf, msg_ctime);
1706 }
1707 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1708 if (error)
1709 return (error);
1710 if (uap->cmd == IPC_STAT) {
1711 bzero(&msqbuf32, sizeof(msqbuf32));
1712 freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1713 PTROUT_CP(msqbuf, msqbuf32, msg_first);
1714 PTROUT_CP(msqbuf, msqbuf32, msg_last);
1715 CP(msqbuf, msqbuf32, msg_cbytes);
1716 CP(msqbuf, msqbuf32, msg_qnum);
1717 CP(msqbuf, msqbuf32, msg_qbytes);
1718 CP(msqbuf, msqbuf32, msg_lspid);
1719 CP(msqbuf, msqbuf32, msg_lrpid);
1720 CP(msqbuf, msqbuf32, msg_stime);
1721 CP(msqbuf, msqbuf32, msg_rtime);
1722 CP(msqbuf, msqbuf32, msg_ctime);
1723 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1724 }
1725 return (error);
1726 }
1727 #endif
1728
1729 int
1730 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap)
1731 {
1732 struct msqid_ds msqbuf;
1733 struct msqid_ds32 msqbuf32;
1734 int error;
1735
1736 if (uap->cmd == IPC_SET) {
1737 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32));
1738 if (error)
1739 return (error);
1740 freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm);
1741 PTRIN_CP(msqbuf32, msqbuf, msg_first);
1742 PTRIN_CP(msqbuf32, msqbuf, msg_last);
1743 CP(msqbuf32, msqbuf, msg_cbytes);
1744 CP(msqbuf32, msqbuf, msg_qnum);
1745 CP(msqbuf32, msqbuf, msg_qbytes);
1746 CP(msqbuf32, msqbuf, msg_lspid);
1747 CP(msqbuf32, msqbuf, msg_lrpid);
1748 CP(msqbuf32, msqbuf, msg_stime);
1749 CP(msqbuf32, msqbuf, msg_rtime);
1750 CP(msqbuf32, msqbuf, msg_ctime);
1751 }
1752 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1753 if (error)
1754 return (error);
1755 if (uap->cmd == IPC_STAT) {
1756 freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm);
1757 PTROUT_CP(msqbuf, msqbuf32, msg_first);
1758 PTROUT_CP(msqbuf, msqbuf32, msg_last);
1759 CP(msqbuf, msqbuf32, msg_cbytes);
1760 CP(msqbuf, msqbuf32, msg_qnum);
1761 CP(msqbuf, msqbuf32, msg_qbytes);
1762 CP(msqbuf, msqbuf32, msg_lspid);
1763 CP(msqbuf, msqbuf32, msg_lrpid);
1764 CP(msqbuf, msqbuf32, msg_stime);
1765 CP(msqbuf, msqbuf32, msg_rtime);
1766 CP(msqbuf, msqbuf32, msg_ctime);
1767 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32));
1768 }
1769 return (error);
1770 }
1771
1772 int
1773 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap)
1774 {
1775 const void *msgp;
1776 long mtype;
1777 int32_t mtype32;
1778 int error;
1779
1780 msgp = PTRIN(uap->msgp);
1781 if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0)
1782 return (error);
1783 mtype = mtype32;
1784 return (kern_msgsnd(td, uap->msqid,
1785 (const char *)msgp + sizeof(mtype32),
1786 uap->msgsz, uap->msgflg, mtype));
1787 }
1788
1789 int
1790 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap)
1791 {
1792 void *msgp;
1793 long mtype;
1794 int32_t mtype32;
1795 int error;
1796
1797 msgp = PTRIN(uap->msgp);
1798 if ((error = kern_msgrcv(td, uap->msqid,
1799 (char *)msgp + sizeof(mtype32), uap->msgsz,
1800 uap->msgtyp, uap->msgflg, &mtype)) != 0)
1801 return (error);
1802 mtype32 = (int32_t)mtype;
1803 return (copyout(&mtype32, msgp, sizeof(mtype32)));
1804 }
1805 #endif
1806
1807 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1808 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1809
1810 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1811 static sy_call_t *msgcalls[] = {
1812 (sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget,
1813 (sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv
1814 };
1815
1816 /*
1817 * Entry point for all MSG calls.
1818 *
1819 * XXX actually varargs.
1820 * struct msgsys_args {
1821 * int which;
1822 * int a2;
1823 * int a3;
1824 * int a4;
1825 * int a5;
1826 * int a6;
1827 * } *uap;
1828 */
1829 int
1830 sys_msgsys(struct thread *td, struct msgsys_args *uap)
1831 {
1832 int error;
1833
1834 if (uap->which < 0 || uap->which >= nitems(msgcalls))
1835 return (EINVAL);
1836 error = (*msgcalls[uap->which])(td, &uap->a2);
1837 return (error);
1838 }
1839
1840 #ifndef CP
1841 #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0)
1842 #endif
1843
1844 #ifndef _SYS_SYSPROTO_H_
1845 struct freebsd7_msgctl_args {
1846 int msqid;
1847 int cmd;
1848 struct msqid_ds_old *buf;
1849 };
1850 #endif
1851 int
1852 freebsd7_msgctl(struct thread *td, struct freebsd7_msgctl_args *uap)
1853 {
1854 struct msqid_ds_old msqold;
1855 struct msqid_ds msqbuf;
1856 int error;
1857
1858 DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd,
1859 uap->buf));
1860 if (uap->cmd == IPC_SET) {
1861 error = copyin(uap->buf, &msqold, sizeof(msqold));
1862 if (error)
1863 return (error);
1864 ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm);
1865 CP(msqold, msqbuf, msg_first);
1866 CP(msqold, msqbuf, msg_last);
1867 CP(msqold, msqbuf, msg_cbytes);
1868 CP(msqold, msqbuf, msg_qnum);
1869 CP(msqold, msqbuf, msg_qbytes);
1870 CP(msqold, msqbuf, msg_lspid);
1871 CP(msqold, msqbuf, msg_lrpid);
1872 CP(msqold, msqbuf, msg_stime);
1873 CP(msqold, msqbuf, msg_rtime);
1874 CP(msqold, msqbuf, msg_ctime);
1875 }
1876 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf);
1877 if (error)
1878 return (error);
1879 if (uap->cmd == IPC_STAT) {
1880 bzero(&msqold, sizeof(msqold));
1881 ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm);
1882 CP(msqbuf, msqold, msg_first);
1883 CP(msqbuf, msqold, msg_last);
1884 CP(msqbuf, msqold, msg_cbytes);
1885 CP(msqbuf, msqold, msg_qnum);
1886 CP(msqbuf, msqold, msg_qbytes);
1887 CP(msqbuf, msqold, msg_lspid);
1888 CP(msqbuf, msqold, msg_lrpid);
1889 CP(msqbuf, msqold, msg_stime);
1890 CP(msqbuf, msqold, msg_rtime);
1891 CP(msqbuf, msqold, msg_ctime);
1892 error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old));
1893 }
1894 return (error);
1895 }
1896
1897 #undef CP
1898
1899 #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1900 COMPAT_FREEBSD7 */
Cache object: 655d37c70b695d3e4df0fe0abb8be80d
|