xref: /freebsd/sys/kern/sysv_shm.c (revision cbd30a72ca196976c1c700400ecd424baa1b9c16)
1 /*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
2 /*-
3  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Adam Glass and Charles
16  *	Hannum.
17  * 4. The names of the authors may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*-
32  * Copyright (c) 2003-2005 McAfee, Inc.
33  * All rights reserved.
34  *
35  * This software was developed for the FreeBSD Project in part by McAfee
36  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
37  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
38  * program.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  */
61 
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64 
65 #include "opt_compat.h"
66 #include "opt_sysvipc.h"
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
71 #include <sys/limits.h>
72 #include <sys/lock.h>
73 #include <sys/sysctl.h>
74 #include <sys/shm.h>
75 #include <sys/proc.h>
76 #include <sys/malloc.h>
77 #include <sys/mman.h>
78 #include <sys/module.h>
79 #include <sys/mutex.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
83 #include <sys/stat.h>
84 #include <sys/syscall.h>
85 #include <sys/syscallsubr.h>
86 #include <sys/sysent.h>
87 #include <sys/sysproto.h>
88 #include <sys/jail.h>
89 
90 #include <security/mac/mac_framework.h>
91 
92 #include <vm/vm.h>
93 #include <vm/vm_param.h>
94 #include <vm/pmap.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_map.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pager.h>
99 
100 FEATURE(sysv_shm, "System V shared memory segments support");
101 
102 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
103 
104 static int shmget_allocate_segment(struct thread *td,
105     struct shmget_args *uap, int mode);
106 static int shmget_existing(struct thread *td, struct shmget_args *uap,
107     int mode, int segnum);
108 
109 #define	SHMSEG_FREE     	0x0200
110 #define	SHMSEG_REMOVED  	0x0400
111 #define	SHMSEG_ALLOCATED	0x0800
112 
113 static int shm_last_free, shm_nused, shmalloced;
114 vm_size_t shm_committed;
115 static struct shmid_kernel *shmsegs;
116 static unsigned shm_prison_slot;
117 
118 struct shmmap_state {
119 	vm_offset_t va;
120 	int shmid;
121 };
122 
123 static void shm_deallocate_segment(struct shmid_kernel *);
124 static int shm_find_segment_by_key(struct prison *, key_t);
125 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
126 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
127 static void shmrealloc(void);
128 static int shminit(void);
129 static int sysvshm_modload(struct module *, int, void *);
130 static int shmunload(void);
131 static void shmexit_myhook(struct vmspace *vm);
132 static void shmfork_myhook(struct proc *p1, struct proc *p2);
133 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
134 static void shm_remove(struct shmid_kernel *, int);
135 static struct prison *shm_find_prison(struct ucred *);
136 static int shm_prison_cansee(struct prison *, struct shmid_kernel *);
137 static int shm_prison_check(void *, void *);
138 static int shm_prison_set(void *, void *);
139 static int shm_prison_get(void *, void *);
140 static int shm_prison_remove(void *, void *);
141 static void shm_prison_cleanup(struct prison *);
142 
143 /*
144  * Tuneable values.
145  */
146 #ifndef SHMMAXPGS
147 #define	SHMMAXPGS	131072	/* Note: sysv shared memory is swap backed. */
148 #endif
149 #ifndef SHMMAX
150 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
151 #endif
152 #ifndef SHMMIN
153 #define	SHMMIN	1
154 #endif
155 #ifndef SHMMNI
156 #define	SHMMNI	192
157 #endif
158 #ifndef SHMSEG
159 #define	SHMSEG	128
160 #endif
161 #ifndef SHMALL
162 #define	SHMALL	(SHMMAXPGS)
163 #endif
164 
165 struct	shminfo shminfo = {
166 	.shmmax = SHMMAX,
167 	.shmmin = SHMMIN,
168 	.shmmni = SHMMNI,
169 	.shmseg = SHMSEG,
170 	.shmall = SHMALL
171 };
172 
173 static int shm_use_phys;
174 static int shm_allow_removed = 1;
175 
176 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
177     "Maximum shared memory segment size");
178 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
179     "Minimum shared memory segment size");
180 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
181     "Number of shared memory identifiers");
182 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
183     "Number of segments per process");
184 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
185     "Maximum number of pages available for shared memory");
186 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
187     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
188 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
189     &shm_allow_removed, 0,
190     "Enable/Disable attachment to attached segments marked for removal");
191 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
192     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
193     "Current number of shared memory segments allocated");
194 
195 static struct sx sysvshmsx;
196 #define	SYSVSHM_LOCK()		sx_xlock(&sysvshmsx)
197 #define	SYSVSHM_UNLOCK()	sx_xunlock(&sysvshmsx)
198 #define	SYSVSHM_ASSERT_LOCKED()	sx_assert(&sysvshmsx, SA_XLOCKED)
199 
200 static int
201 shm_find_segment_by_key(struct prison *pr, key_t key)
202 {
203 	int i;
204 
205 	for (i = 0; i < shmalloced; i++)
206 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
207 		    shmsegs[i].cred != NULL &&
208 		    shmsegs[i].cred->cr_prison == pr &&
209 		    shmsegs[i].u.shm_perm.key == key)
210 			return (i);
211 	return (-1);
212 }
213 
214 /*
215  * Finds segment either by shmid if is_shmid is true, or by segnum if
216  * is_shmid is false.
217  */
218 static struct shmid_kernel *
219 shm_find_segment(struct prison *rpr, int arg, bool is_shmid)
220 {
221 	struct shmid_kernel *shmseg;
222 	int segnum;
223 
224 	segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
225 	if (segnum < 0 || segnum >= shmalloced)
226 		return (NULL);
227 	shmseg = &shmsegs[segnum];
228 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
229 	    (!shm_allow_removed &&
230 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
231 	    (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)) ||
232 	    shm_prison_cansee(rpr, shmseg) != 0)
233 		return (NULL);
234 	return (shmseg);
235 }
236 
237 static void
238 shm_deallocate_segment(struct shmid_kernel *shmseg)
239 {
240 	vm_size_t size;
241 
242 	SYSVSHM_ASSERT_LOCKED();
243 
244 	vm_object_deallocate(shmseg->object);
245 	shmseg->object = NULL;
246 	size = round_page(shmseg->u.shm_segsz);
247 	shm_committed -= btoc(size);
248 	shm_nused--;
249 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
250 #ifdef MAC
251 	mac_sysvshm_cleanup(shmseg);
252 #endif
253 	racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
254 	racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
255 	crfree(shmseg->cred);
256 	shmseg->cred = NULL;
257 }
258 
259 static int
260 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
261 {
262 	struct shmid_kernel *shmseg;
263 	int segnum, result;
264 	vm_size_t size;
265 
266 	SYSVSHM_ASSERT_LOCKED();
267 	segnum = IPCID_TO_IX(shmmap_s->shmid);
268 	KASSERT(segnum >= 0 && segnum < shmalloced,
269 	    ("segnum %d shmalloced %d", segnum, shmalloced));
270 
271 	shmseg = &shmsegs[segnum];
272 	size = round_page(shmseg->u.shm_segsz);
273 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
274 	if (result != KERN_SUCCESS)
275 		return (EINVAL);
276 	shmmap_s->shmid = -1;
277 	shmseg->u.shm_dtime = time_second;
278 	if (--shmseg->u.shm_nattch == 0 &&
279 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
280 		shm_deallocate_segment(shmseg);
281 		shm_last_free = segnum;
282 	}
283 	return (0);
284 }
285 
286 static void
287 shm_remove(struct shmid_kernel *shmseg, int segnum)
288 {
289 
290 	shmseg->u.shm_perm.key = IPC_PRIVATE;
291 	shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
292 	if (shmseg->u.shm_nattch == 0) {
293 		shm_deallocate_segment(shmseg);
294 		shm_last_free = segnum;
295 	}
296 }
297 
298 static struct prison *
299 shm_find_prison(struct ucred *cred)
300 {
301 	struct prison *pr, *rpr;
302 
303 	pr = cred->cr_prison;
304 	prison_lock(pr);
305 	rpr = osd_jail_get(pr, shm_prison_slot);
306 	prison_unlock(pr);
307 	return rpr;
308 }
309 
310 static int
311 shm_prison_cansee(struct prison *rpr, struct shmid_kernel *shmseg)
312 {
313 
314 	if (shmseg->cred == NULL ||
315 	    !(rpr == shmseg->cred->cr_prison ||
316 	      prison_ischild(rpr, shmseg->cred->cr_prison)))
317 		return (EINVAL);
318 	return (0);
319 }
320 
321 static int
322 kern_shmdt_locked(struct thread *td, const void *shmaddr)
323 {
324 	struct proc *p = td->td_proc;
325 	struct shmmap_state *shmmap_s;
326 #ifdef MAC
327 	struct shmid_kernel *shmsegptr;
328 	int error;
329 #endif
330 	int i;
331 
332 	SYSVSHM_ASSERT_LOCKED();
333 	if (shm_find_prison(td->td_ucred) == NULL)
334 		return (ENOSYS);
335 	shmmap_s = p->p_vmspace->vm_shm;
336  	if (shmmap_s == NULL)
337 		return (EINVAL);
338 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
339 		if (shmmap_s->shmid != -1 &&
340 		    shmmap_s->va == (vm_offset_t)shmaddr) {
341 			break;
342 		}
343 	}
344 	if (i == shminfo.shmseg)
345 		return (EINVAL);
346 #ifdef MAC
347 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
348 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
349 	if (error != 0)
350 		return (error);
351 #endif
352 	return (shm_delete_mapping(p->p_vmspace, shmmap_s));
353 }
354 
355 #ifndef _SYS_SYSPROTO_H_
356 struct shmdt_args {
357 	const void *shmaddr;
358 };
359 #endif
360 int
361 sys_shmdt(struct thread *td, struct shmdt_args *uap)
362 {
363 	int error;
364 
365 	SYSVSHM_LOCK();
366 	error = kern_shmdt_locked(td, uap->shmaddr);
367 	SYSVSHM_UNLOCK();
368 	return (error);
369 }
370 
371 static int
372 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
373     int shmflg)
374 {
375 	struct prison *rpr;
376 	struct proc *p = td->td_proc;
377 	struct shmid_kernel *shmseg;
378 	struct shmmap_state *shmmap_s;
379 	vm_offset_t attach_va;
380 	vm_prot_t prot;
381 	vm_size_t size;
382 	int error, i, rv;
383 
384 	SYSVSHM_ASSERT_LOCKED();
385 	rpr = shm_find_prison(td->td_ucred);
386 	if (rpr == NULL)
387 		return (ENOSYS);
388 	shmmap_s = p->p_vmspace->vm_shm;
389 	if (shmmap_s == NULL) {
390 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
391 		    M_SHM, M_WAITOK);
392 		for (i = 0; i < shminfo.shmseg; i++)
393 			shmmap_s[i].shmid = -1;
394 		KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
395 		p->p_vmspace->vm_shm = shmmap_s;
396 	}
397 	shmseg = shm_find_segment(rpr, shmid, true);
398 	if (shmseg == NULL)
399 		return (EINVAL);
400 	error = ipcperm(td, &shmseg->u.shm_perm,
401 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
402 	if (error != 0)
403 		return (error);
404 #ifdef MAC
405 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
406 	if (error != 0)
407 		return (error);
408 #endif
409 	for (i = 0; i < shminfo.shmseg; i++) {
410 		if (shmmap_s->shmid == -1)
411 			break;
412 		shmmap_s++;
413 	}
414 	if (i >= shminfo.shmseg)
415 		return (EMFILE);
416 	size = round_page(shmseg->u.shm_segsz);
417 	prot = VM_PROT_READ;
418 	if ((shmflg & SHM_RDONLY) == 0)
419 		prot |= VM_PROT_WRITE;
420 	if (shmaddr != NULL) {
421 		if ((shmflg & SHM_RND) != 0)
422 			attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
423 		else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
424 			attach_va = (vm_offset_t)shmaddr;
425 		else
426 			return (EINVAL);
427 	} else {
428 		/*
429 		 * This is just a hint to vm_map_find() about where to
430 		 * put it.
431 		 */
432 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
433 		    lim_max(td, RLIMIT_DATA));
434 	}
435 
436 	vm_object_reference(shmseg->object);
437 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
438 	    size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
439 	    prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
440 	if (rv != KERN_SUCCESS) {
441 		vm_object_deallocate(shmseg->object);
442 		return (ENOMEM);
443 	}
444 
445 	shmmap_s->va = attach_va;
446 	shmmap_s->shmid = shmid;
447 	shmseg->u.shm_lpid = p->p_pid;
448 	shmseg->u.shm_atime = time_second;
449 	shmseg->u.shm_nattch++;
450 	td->td_retval[0] = attach_va;
451 	return (error);
452 }
453 
454 int
455 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
456 {
457 	int error;
458 
459 	SYSVSHM_LOCK();
460 	error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
461 	SYSVSHM_UNLOCK();
462 	return (error);
463 }
464 
465 #ifndef _SYS_SYSPROTO_H_
466 struct shmat_args {
467 	int shmid;
468 	const void *shmaddr;
469 	int shmflg;
470 };
471 #endif
472 int
473 sys_shmat(struct thread *td, struct shmat_args *uap)
474 {
475 
476 	return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
477 }
478 
479 static int
480 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
481     size_t *bufsz)
482 {
483 	struct prison *rpr;
484 	struct shmid_kernel *shmseg;
485 	struct shmid_ds *shmidp;
486 	struct shm_info shm_info;
487 	int error;
488 
489 	SYSVSHM_ASSERT_LOCKED();
490 
491 	rpr = shm_find_prison(td->td_ucred);
492 	if (rpr == NULL)
493 		return (ENOSYS);
494 
495 	switch (cmd) {
496 	/*
497 	 * It is possible that kern_shmctl is being called from the Linux ABI
498 	 * layer, in which case, we will need to implement IPC_INFO.  It should
499 	 * be noted that other shmctl calls will be funneled through here for
500 	 * Linix binaries as well.
501 	 *
502 	 * NB: The Linux ABI layer will convert this data to structure(s) more
503 	 * consistent with the Linux ABI.
504 	 */
505 	case IPC_INFO:
506 		memcpy(buf, &shminfo, sizeof(shminfo));
507 		if (bufsz)
508 			*bufsz = sizeof(shminfo);
509 		td->td_retval[0] = shmalloced;
510 		return (0);
511 	case SHM_INFO: {
512 		shm_info.used_ids = shm_nused;
513 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
514 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
515 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
516 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
517 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
518 		memcpy(buf, &shm_info, sizeof(shm_info));
519 		if (bufsz != NULL)
520 			*bufsz = sizeof(shm_info);
521 		td->td_retval[0] = shmalloced;
522 		return (0);
523 	}
524 	}
525 	shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
526 	if (shmseg == NULL)
527 		return (EINVAL);
528 #ifdef MAC
529 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
530 	if (error != 0)
531 		return (error);
532 #endif
533 	switch (cmd) {
534 	case SHM_STAT:
535 	case IPC_STAT:
536 		shmidp = (struct shmid_ds *)buf;
537 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
538 		if (error != 0)
539 			return (error);
540 		memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
541 		if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
542 			shmidp->shm_perm.key = IPC_PRIVATE;
543 		if (bufsz != NULL)
544 			*bufsz = sizeof(struct shmid_ds);
545 		if (cmd == SHM_STAT) {
546 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
547 			    shmseg->u.shm_perm);
548 		}
549 		break;
550 	case IPC_SET:
551 		shmidp = (struct shmid_ds *)buf;
552 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
553 		if (error != 0)
554 			return (error);
555 		shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
556 		shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
557 		shmseg->u.shm_perm.mode =
558 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
559 		    (shmidp->shm_perm.mode & ACCESSPERMS);
560 		shmseg->u.shm_ctime = time_second;
561 		break;
562 	case IPC_RMID:
563 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
564 		if (error != 0)
565 			return (error);
566 		shm_remove(shmseg, IPCID_TO_IX(shmid));
567 		break;
568 #if 0
569 	case SHM_LOCK:
570 	case SHM_UNLOCK:
571 #endif
572 	default:
573 		error = EINVAL;
574 		break;
575 	}
576 	return (error);
577 }
578 
579 int
580 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
581 {
582 	int error;
583 
584 	SYSVSHM_LOCK();
585 	error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
586 	SYSVSHM_UNLOCK();
587 	return (error);
588 }
589 
590 
591 #ifndef _SYS_SYSPROTO_H_
592 struct shmctl_args {
593 	int shmid;
594 	int cmd;
595 	struct shmid_ds *buf;
596 };
597 #endif
598 int
599 sys_shmctl(struct thread *td, struct shmctl_args *uap)
600 {
601 	int error;
602 	struct shmid_ds buf;
603 	size_t bufsz;
604 
605 	/*
606 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
607 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
608 	 * return an error back to the user since we do not to support this.
609 	 */
610 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
611 	    uap->cmd == SHM_STAT)
612 		return (EINVAL);
613 
614 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
615 	if (uap->cmd == IPC_SET) {
616 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
617 			goto done;
618 	}
619 
620 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
621 	if (error)
622 		goto done;
623 
624 	/* Cases in which we need to copyout */
625 	switch (uap->cmd) {
626 	case IPC_STAT:
627 		error = copyout(&buf, uap->buf, bufsz);
628 		break;
629 	}
630 
631 done:
632 	if (error) {
633 		/* Invalidate the return value */
634 		td->td_retval[0] = -1;
635 	}
636 	return (error);
637 }
638 
639 
640 static int
641 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
642     int segnum)
643 {
644 	struct shmid_kernel *shmseg;
645 #ifdef MAC
646 	int error;
647 #endif
648 
649 	SYSVSHM_ASSERT_LOCKED();
650 	KASSERT(segnum >= 0 && segnum < shmalloced,
651 	    ("segnum %d shmalloced %d", segnum, shmalloced));
652 	shmseg = &shmsegs[segnum];
653 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
654 		return (EEXIST);
655 #ifdef MAC
656 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
657 	if (error != 0)
658 		return (error);
659 #endif
660 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
661 		return (EINVAL);
662 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
663 	return (0);
664 }
665 
666 static int
667 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
668 {
669 	struct ucred *cred = td->td_ucred;
670 	struct shmid_kernel *shmseg;
671 	vm_object_t shm_object;
672 	int i, segnum;
673 	size_t size;
674 
675 	SYSVSHM_ASSERT_LOCKED();
676 
677 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
678 		return (EINVAL);
679 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
680 		return (ENOSPC);
681 	size = round_page(uap->size);
682 	if (shm_committed + btoc(size) > shminfo.shmall)
683 		return (ENOMEM);
684 	if (shm_last_free < 0) {
685 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
686 		for (i = 0; i < shmalloced; i++)
687 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
688 				break;
689 		if (i == shmalloced)
690 			return (ENOSPC);
691 		segnum = i;
692 	} else  {
693 		segnum = shm_last_free;
694 		shm_last_free = -1;
695 	}
696 	KASSERT(segnum >= 0 && segnum < shmalloced,
697 	    ("segnum %d shmalloced %d", segnum, shmalloced));
698 	shmseg = &shmsegs[segnum];
699 #ifdef RACCT
700 	if (racct_enable) {
701 		PROC_LOCK(td->td_proc);
702 		if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
703 			PROC_UNLOCK(td->td_proc);
704 			return (ENOSPC);
705 		}
706 		if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
707 			racct_sub(td->td_proc, RACCT_NSHM, 1);
708 			PROC_UNLOCK(td->td_proc);
709 			return (ENOMEM);
710 		}
711 		PROC_UNLOCK(td->td_proc);
712 	}
713 #endif
714 
715 	/*
716 	 * We make sure that we have allocated a pager before we need
717 	 * to.
718 	 */
719 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
720 	    0, size, VM_PROT_DEFAULT, 0, cred);
721 	if (shm_object == NULL) {
722 #ifdef RACCT
723 		if (racct_enable) {
724 			PROC_LOCK(td->td_proc);
725 			racct_sub(td->td_proc, RACCT_NSHM, 1);
726 			racct_sub(td->td_proc, RACCT_SHMSIZE, size);
727 			PROC_UNLOCK(td->td_proc);
728 		}
729 #endif
730 		return (ENOMEM);
731 	}
732 	shm_object->pg_color = 0;
733 	VM_OBJECT_WLOCK(shm_object);
734 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
735 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
736 	VM_OBJECT_WUNLOCK(shm_object);
737 
738 	shmseg->object = shm_object;
739 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
740 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
741 	shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
742 	shmseg->u.shm_perm.key = uap->key;
743 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
744 	shmseg->cred = crhold(cred);
745 	shmseg->u.shm_segsz = uap->size;
746 	shmseg->u.shm_cpid = td->td_proc->p_pid;
747 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
748 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
749 #ifdef MAC
750 	mac_sysvshm_create(cred, shmseg);
751 #endif
752 	shmseg->u.shm_ctime = time_second;
753 	shm_committed += btoc(size);
754 	shm_nused++;
755 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
756 
757 	return (0);
758 }
759 
760 #ifndef _SYS_SYSPROTO_H_
761 struct shmget_args {
762 	key_t key;
763 	size_t size;
764 	int shmflg;
765 };
766 #endif
767 int
768 sys_shmget(struct thread *td, struct shmget_args *uap)
769 {
770 	int segnum, mode;
771 	int error;
772 
773 	if (shm_find_prison(td->td_ucred) == NULL)
774 		return (ENOSYS);
775 	mode = uap->shmflg & ACCESSPERMS;
776 	SYSVSHM_LOCK();
777 	if (uap->key == IPC_PRIVATE) {
778 		error = shmget_allocate_segment(td, uap, mode);
779 	} else {
780 		segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
781 		    uap->key);
782 		if (segnum >= 0)
783 			error = shmget_existing(td, uap, mode, segnum);
784 		else if ((uap->shmflg & IPC_CREAT) == 0)
785 			error = ENOENT;
786 		else
787 			error = shmget_allocate_segment(td, uap, mode);
788 	}
789 	SYSVSHM_UNLOCK();
790 	return (error);
791 }
792 
793 static void
794 shmfork_myhook(struct proc *p1, struct proc *p2)
795 {
796 	struct shmmap_state *shmmap_s;
797 	size_t size;
798 	int i;
799 
800 	SYSVSHM_LOCK();
801 	size = shminfo.shmseg * sizeof(struct shmmap_state);
802 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
803 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
804 	p2->p_vmspace->vm_shm = shmmap_s;
805 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
806 		if (shmmap_s->shmid != -1) {
807 			KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
808 			    IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
809 			    ("segnum %d shmalloced %d",
810 			    IPCID_TO_IX(shmmap_s->shmid), shmalloced));
811 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
812 		}
813 	}
814 	SYSVSHM_UNLOCK();
815 }
816 
817 static void
818 shmexit_myhook(struct vmspace *vm)
819 {
820 	struct shmmap_state *base, *shm;
821 	int i;
822 
823 	base = vm->vm_shm;
824 	if (base != NULL) {
825 		vm->vm_shm = NULL;
826 		SYSVSHM_LOCK();
827 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
828 			if (shm->shmid != -1)
829 				shm_delete_mapping(vm, shm);
830 		}
831 		SYSVSHM_UNLOCK();
832 		free(base, M_SHM);
833 	}
834 }
835 
836 static void
837 shmrealloc(void)
838 {
839 	struct shmid_kernel *newsegs;
840 	int i;
841 
842 	SYSVSHM_ASSERT_LOCKED();
843 
844 	if (shmalloced >= shminfo.shmmni)
845 		return;
846 
847 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
848 	for (i = 0; i < shmalloced; i++)
849 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
850 	for (; i < shminfo.shmmni; i++) {
851 		newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
852 		newsegs[i].u.shm_perm.seq = 0;
853 #ifdef MAC
854 		mac_sysvshm_init(&newsegs[i]);
855 #endif
856 	}
857 	free(shmsegs, M_SHM);
858 	shmsegs = newsegs;
859 	shmalloced = shminfo.shmmni;
860 }
861 
862 static struct syscall_helper_data shm_syscalls[] = {
863 	SYSCALL_INIT_HELPER(shmat),
864 	SYSCALL_INIT_HELPER(shmctl),
865 	SYSCALL_INIT_HELPER(shmdt),
866 	SYSCALL_INIT_HELPER(shmget),
867 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
868     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
869 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
870 #endif
871 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
872 	SYSCALL_INIT_HELPER(shmsys),
873 #endif
874 	SYSCALL_INIT_LAST
875 };
876 
877 #ifdef COMPAT_FREEBSD32
878 #include <compat/freebsd32/freebsd32.h>
879 #include <compat/freebsd32/freebsd32_ipc.h>
880 #include <compat/freebsd32/freebsd32_proto.h>
881 #include <compat/freebsd32/freebsd32_signal.h>
882 #include <compat/freebsd32/freebsd32_syscall.h>
883 #include <compat/freebsd32/freebsd32_util.h>
884 
885 static struct syscall_helper_data shm32_syscalls[] = {
886 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
887 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
888 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
889 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
890 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
891 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
892     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
893 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
894 #endif
895 	SYSCALL_INIT_LAST
896 };
897 #endif
898 
899 static int
900 shminit(void)
901 {
902 	struct prison *pr;
903 	void **rsv;
904 	int i, error;
905 	osd_method_t methods[PR_MAXMETHOD] = {
906 	    [PR_METHOD_CHECK] =		shm_prison_check,
907 	    [PR_METHOD_SET] =		shm_prison_set,
908 	    [PR_METHOD_GET] =		shm_prison_get,
909 	    [PR_METHOD_REMOVE] =	shm_prison_remove,
910 	};
911 
912 #ifndef BURN_BRIDGES
913 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
914 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
915 #endif
916 	if (shminfo.shmmax == SHMMAX) {
917 		/* Initialize shmmax dealing with possible overflow. */
918 		for (i = PAGE_SIZE; i != 0; i--) {
919 			shminfo.shmmax = shminfo.shmall * i;
920 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
921 				break;
922 		}
923 	}
924 	shmalloced = shminfo.shmmni;
925 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
926 	for (i = 0; i < shmalloced; i++) {
927 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
928 		shmsegs[i].u.shm_perm.seq = 0;
929 #ifdef MAC
930 		mac_sysvshm_init(&shmsegs[i]);
931 #endif
932 	}
933 	shm_last_free = 0;
934 	shm_nused = 0;
935 	shm_committed = 0;
936 	sx_init(&sysvshmsx, "sysvshmsx");
937 	shmexit_hook = &shmexit_myhook;
938 	shmfork_hook = &shmfork_myhook;
939 
940 	/* Set current prisons according to their allow.sysvipc. */
941 	shm_prison_slot = osd_jail_register(NULL, methods);
942 	rsv = osd_reserve(shm_prison_slot);
943 	prison_lock(&prison0);
944 	(void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
945 	prison_unlock(&prison0);
946 	rsv = NULL;
947 	sx_slock(&allprison_lock);
948 	TAILQ_FOREACH(pr, &allprison, pr_list) {
949 		if (rsv == NULL)
950 			rsv = osd_reserve(shm_prison_slot);
951 		prison_lock(pr);
952 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
953 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
954 			    &prison0);
955 			rsv = NULL;
956 		}
957 		prison_unlock(pr);
958 	}
959 	if (rsv != NULL)
960 		osd_free_reserved(rsv);
961 	sx_sunlock(&allprison_lock);
962 
963 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
964 	if (error != 0)
965 		return (error);
966 #ifdef COMPAT_FREEBSD32
967 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
968 	if (error != 0)
969 		return (error);
970 #endif
971 	return (0);
972 }
973 
974 static int
975 shmunload(void)
976 {
977 	int i;
978 
979 	if (shm_nused > 0)
980 		return (EBUSY);
981 
982 #ifdef COMPAT_FREEBSD32
983 	syscall32_helper_unregister(shm32_syscalls);
984 #endif
985 	syscall_helper_unregister(shm_syscalls);
986 	if (shm_prison_slot != 0)
987 		osd_jail_deregister(shm_prison_slot);
988 
989 	for (i = 0; i < shmalloced; i++) {
990 #ifdef MAC
991 		mac_sysvshm_destroy(&shmsegs[i]);
992 #endif
993 		/*
994 		 * Objects might be still mapped into the processes
995 		 * address spaces.  Actual free would happen on the
996 		 * last mapping destruction.
997 		 */
998 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
999 			vm_object_deallocate(shmsegs[i].object);
1000 	}
1001 	free(shmsegs, M_SHM);
1002 	shmexit_hook = NULL;
1003 	shmfork_hook = NULL;
1004 	sx_destroy(&sysvshmsx);
1005 	return (0);
1006 }
1007 
1008 static int
1009 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1010 {
1011 	struct shmid_kernel tshmseg;
1012 	struct prison *pr, *rpr;
1013 	int error, i;
1014 
1015 	SYSVSHM_LOCK();
1016 	pr = req->td->td_ucred->cr_prison;
1017 	rpr = shm_find_prison(req->td->td_ucred);
1018 	error = 0;
1019 	for (i = 0; i < shmalloced; i++) {
1020 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1021 		    rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1022 			bzero(&tshmseg, sizeof(tshmseg));
1023 			tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1024 		} else {
1025 			tshmseg = shmsegs[i];
1026 			if (tshmseg.cred->cr_prison != pr)
1027 				tshmseg.u.shm_perm.key = IPC_PRIVATE;
1028 		}
1029 		error = SYSCTL_OUT(req, &tshmseg, sizeof(tshmseg));
1030 		if (error != 0)
1031 			break;
1032 	}
1033 	SYSVSHM_UNLOCK();
1034 	return (error);
1035 }
1036 
1037 static int
1038 shm_prison_check(void *obj, void *data)
1039 {
1040 	struct prison *pr = obj;
1041 	struct prison *prpr;
1042 	struct vfsoptlist *opts = data;
1043 	int error, jsys;
1044 
1045 	/*
1046 	 * sysvshm is a jailsys integer.
1047 	 * It must be "disable" if the parent jail is disabled.
1048 	 */
1049 	error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1050 	if (error != ENOENT) {
1051 		if (error != 0)
1052 			return (error);
1053 		switch (jsys) {
1054 		case JAIL_SYS_DISABLE:
1055 			break;
1056 		case JAIL_SYS_NEW:
1057 		case JAIL_SYS_INHERIT:
1058 			prison_lock(pr->pr_parent);
1059 			prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1060 			prison_unlock(pr->pr_parent);
1061 			if (prpr == NULL)
1062 				return (EPERM);
1063 			break;
1064 		default:
1065 			return (EINVAL);
1066 		}
1067 	}
1068 
1069 	return (0);
1070 }
1071 
1072 static int
1073 shm_prison_set(void *obj, void *data)
1074 {
1075 	struct prison *pr = obj;
1076 	struct prison *tpr, *orpr, *nrpr, *trpr;
1077 	struct vfsoptlist *opts = data;
1078 	void *rsv;
1079 	int jsys, descend;
1080 
1081 	/*
1082 	 * sysvshm controls which jail is the root of the associated segments
1083 	 * (this jail or same as the parent), or if the feature is available
1084 	 * at all.
1085 	 */
1086 	if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1087 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1088 		    ? JAIL_SYS_INHERIT
1089 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1090 		    ? JAIL_SYS_DISABLE
1091 		    : -1;
1092 	if (jsys == JAIL_SYS_DISABLE) {
1093 		prison_lock(pr);
1094 		orpr = osd_jail_get(pr, shm_prison_slot);
1095 		if (orpr != NULL)
1096 			osd_jail_del(pr, shm_prison_slot);
1097 		prison_unlock(pr);
1098 		if (orpr != NULL) {
1099 			if (orpr == pr)
1100 				shm_prison_cleanup(pr);
1101 			/* Disable all child jails as well. */
1102 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1103 				prison_lock(tpr);
1104 				trpr = osd_jail_get(tpr, shm_prison_slot);
1105 				if (trpr != NULL) {
1106 					osd_jail_del(tpr, shm_prison_slot);
1107 					prison_unlock(tpr);
1108 					if (trpr == tpr)
1109 						shm_prison_cleanup(tpr);
1110 				} else {
1111 					prison_unlock(tpr);
1112 					descend = 0;
1113 				}
1114 			}
1115 		}
1116 	} else if (jsys != -1) {
1117 		if (jsys == JAIL_SYS_NEW)
1118 			nrpr = pr;
1119 		else {
1120 			prison_lock(pr->pr_parent);
1121 			nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1122 			prison_unlock(pr->pr_parent);
1123 		}
1124 		rsv = osd_reserve(shm_prison_slot);
1125 		prison_lock(pr);
1126 		orpr = osd_jail_get(pr, shm_prison_slot);
1127 		if (orpr != nrpr)
1128 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1129 			    nrpr);
1130 		else
1131 			osd_free_reserved(rsv);
1132 		prison_unlock(pr);
1133 		if (orpr != nrpr) {
1134 			if (orpr == pr)
1135 				shm_prison_cleanup(pr);
1136 			if (orpr != NULL) {
1137 				/* Change child jails matching the old root, */
1138 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1139 					prison_lock(tpr);
1140 					trpr = osd_jail_get(tpr,
1141 					    shm_prison_slot);
1142 					if (trpr == orpr) {
1143 						(void)osd_jail_set(tpr,
1144 						    shm_prison_slot, nrpr);
1145 						prison_unlock(tpr);
1146 						if (trpr == tpr)
1147 							shm_prison_cleanup(tpr);
1148 					} else {
1149 						prison_unlock(tpr);
1150 						descend = 0;
1151 					}
1152 				}
1153 			}
1154 		}
1155 	}
1156 
1157 	return (0);
1158 }
1159 
1160 static int
1161 shm_prison_get(void *obj, void *data)
1162 {
1163 	struct prison *pr = obj;
1164 	struct prison *rpr;
1165 	struct vfsoptlist *opts = data;
1166 	int error, jsys;
1167 
1168 	/* Set sysvshm based on the jail's root prison. */
1169 	prison_lock(pr);
1170 	rpr = osd_jail_get(pr, shm_prison_slot);
1171 	prison_unlock(pr);
1172 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1173 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1174 	error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1175 	if (error == ENOENT)
1176 		error = 0;
1177 	return (error);
1178 }
1179 
1180 static int
1181 shm_prison_remove(void *obj, void *data __unused)
1182 {
1183 	struct prison *pr = obj;
1184 	struct prison *rpr;
1185 
1186 	SYSVSHM_LOCK();
1187 	prison_lock(pr);
1188 	rpr = osd_jail_get(pr, shm_prison_slot);
1189 	prison_unlock(pr);
1190 	if (rpr == pr)
1191 		shm_prison_cleanup(pr);
1192 	SYSVSHM_UNLOCK();
1193 	return (0);
1194 }
1195 
1196 static void
1197 shm_prison_cleanup(struct prison *pr)
1198 {
1199 	struct shmid_kernel *shmseg;
1200 	int i;
1201 
1202 	/* Remove any segments that belong to this jail. */
1203 	for (i = 0; i < shmalloced; i++) {
1204 		shmseg = &shmsegs[i];
1205 		if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1206 		    shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1207 			shm_remove(shmseg, i);
1208 		}
1209 	}
1210 }
1211 
1212 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1213 
1214 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1215 struct oshmid_ds {
1216 	struct	ipc_perm_old shm_perm;	/* operation perms */
1217 	int	shm_segsz;		/* size of segment (bytes) */
1218 	u_short	shm_cpid;		/* pid, creator */
1219 	u_short	shm_lpid;		/* pid, last operation */
1220 	short	shm_nattch;		/* no. of current attaches */
1221 	time_t	shm_atime;		/* last attach time */
1222 	time_t	shm_dtime;		/* last detach time */
1223 	time_t	shm_ctime;		/* last change time */
1224 	void	*shm_handle;		/* internal handle for shm segment */
1225 };
1226 
1227 struct oshmctl_args {
1228 	int shmid;
1229 	int cmd;
1230 	struct oshmid_ds *ubuf;
1231 };
1232 
1233 static int
1234 oshmctl(struct thread *td, struct oshmctl_args *uap)
1235 {
1236 #ifdef COMPAT_43
1237 	int error = 0;
1238 	struct prison *rpr;
1239 	struct shmid_kernel *shmseg;
1240 	struct oshmid_ds outbuf;
1241 
1242 	rpr = shm_find_prison(td->td_ucred);
1243 	if (rpr == NULL)
1244 		return (ENOSYS);
1245 	if (uap->cmd != IPC_STAT) {
1246 		return (freebsd7_shmctl(td,
1247 		    (struct freebsd7_shmctl_args *)uap));
1248 	}
1249 	SYSVSHM_LOCK();
1250 	shmseg = shm_find_segment(rpr, uap->shmid, true);
1251 	if (shmseg == NULL) {
1252 		SYSVSHM_UNLOCK();
1253 		return (EINVAL);
1254 	}
1255 	error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1256 	if (error != 0) {
1257 		SYSVSHM_UNLOCK();
1258 		return (error);
1259 	}
1260 #ifdef MAC
1261 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1262 	if (error != 0) {
1263 		SYSVSHM_UNLOCK();
1264 		return (error);
1265 	}
1266 #endif
1267 	ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1268 	outbuf.shm_segsz = shmseg->u.shm_segsz;
1269 	outbuf.shm_cpid = shmseg->u.shm_cpid;
1270 	outbuf.shm_lpid = shmseg->u.shm_lpid;
1271 	outbuf.shm_nattch = shmseg->u.shm_nattch;
1272 	outbuf.shm_atime = shmseg->u.shm_atime;
1273 	outbuf.shm_dtime = shmseg->u.shm_dtime;
1274 	outbuf.shm_ctime = shmseg->u.shm_ctime;
1275 	outbuf.shm_handle = shmseg->object;
1276 	SYSVSHM_UNLOCK();
1277 	return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1278 #else
1279 	return (EINVAL);
1280 #endif
1281 }
1282 
1283 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1284 static sy_call_t *shmcalls[] = {
1285 	(sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1286 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1287 	(sy_call_t *)freebsd7_shmctl
1288 };
1289 
1290 #ifndef _SYS_SYSPROTO_H_
1291 /* XXX actually varargs. */
1292 struct shmsys_args {
1293 	int	which;
1294 	int	a2;
1295 	int	a3;
1296 	int	a4;
1297 };
1298 #endif
1299 int
1300 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1301 {
1302 
1303 	if (uap->which < 0 || uap->which >= nitems(shmcalls))
1304 		return (EINVAL);
1305 	return ((*shmcalls[uap->which])(td, &uap->a2));
1306 }
1307 
1308 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1309 
1310 #ifdef COMPAT_FREEBSD32
1311 
1312 int
1313 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1314 {
1315 
1316 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1317     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1318 	switch (uap->which) {
1319 	case 0:	{	/* shmat */
1320 		struct shmat_args ap;
1321 
1322 		ap.shmid = uap->a2;
1323 		ap.shmaddr = PTRIN(uap->a3);
1324 		ap.shmflg = uap->a4;
1325 		return (sysent[SYS_shmat].sy_call(td, &ap));
1326 	}
1327 	case 2: {	/* shmdt */
1328 		struct shmdt_args ap;
1329 
1330 		ap.shmaddr = PTRIN(uap->a2);
1331 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1332 	}
1333 	case 3: {	/* shmget */
1334 		struct shmget_args ap;
1335 
1336 		ap.key = uap->a2;
1337 		ap.size = uap->a3;
1338 		ap.shmflg = uap->a4;
1339 		return (sysent[SYS_shmget].sy_call(td, &ap));
1340 	}
1341 	case 4: {	/* shmctl */
1342 		struct freebsd7_freebsd32_shmctl_args ap;
1343 
1344 		ap.shmid = uap->a2;
1345 		ap.cmd = uap->a3;
1346 		ap.buf = PTRIN(uap->a4);
1347 		return (freebsd7_freebsd32_shmctl(td, &ap));
1348 	}
1349 	case 1:		/* oshmctl */
1350 	default:
1351 		return (EINVAL);
1352 	}
1353 #else
1354 	return (nosys(td, NULL));
1355 #endif
1356 }
1357 
1358 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1359     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1360 int
1361 freebsd7_freebsd32_shmctl(struct thread *td,
1362     struct freebsd7_freebsd32_shmctl_args *uap)
1363 {
1364 	int error;
1365 	union {
1366 		struct shmid_ds shmid_ds;
1367 		struct shm_info shm_info;
1368 		struct shminfo shminfo;
1369 	} u;
1370 	union {
1371 		struct shmid_ds32_old shmid_ds32;
1372 		struct shm_info32 shm_info32;
1373 		struct shminfo32 shminfo32;
1374 	} u32;
1375 	size_t sz;
1376 
1377 	if (uap->cmd == IPC_SET) {
1378 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1379 		    sizeof(u32.shmid_ds32))))
1380 			goto done;
1381 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1382 		    &u.shmid_ds.shm_perm);
1383 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1384 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1385 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1386 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1387 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1388 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1389 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1390 	}
1391 
1392 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1393 	if (error)
1394 		goto done;
1395 
1396 	/* Cases in which we need to copyout */
1397 	switch (uap->cmd) {
1398 	case IPC_INFO:
1399 		CP(u.shminfo, u32.shminfo32, shmmax);
1400 		CP(u.shminfo, u32.shminfo32, shmmin);
1401 		CP(u.shminfo, u32.shminfo32, shmmni);
1402 		CP(u.shminfo, u32.shminfo32, shmseg);
1403 		CP(u.shminfo, u32.shminfo32, shmall);
1404 		error = copyout(&u32.shminfo32, uap->buf,
1405 		    sizeof(u32.shminfo32));
1406 		break;
1407 	case SHM_INFO:
1408 		CP(u.shm_info, u32.shm_info32, used_ids);
1409 		CP(u.shm_info, u32.shm_info32, shm_rss);
1410 		CP(u.shm_info, u32.shm_info32, shm_tot);
1411 		CP(u.shm_info, u32.shm_info32, shm_swp);
1412 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1413 		CP(u.shm_info, u32.shm_info32, swap_successes);
1414 		error = copyout(&u32.shm_info32, uap->buf,
1415 		    sizeof(u32.shm_info32));
1416 		break;
1417 	case SHM_STAT:
1418 	case IPC_STAT:
1419 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1420 		    &u32.shmid_ds32.shm_perm);
1421 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1422 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1423 		else
1424 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1425 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1426 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1427 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1428 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1429 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1430 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1431 		u32.shmid_ds32.shm_internal = 0;
1432 		error = copyout(&u32.shmid_ds32, uap->buf,
1433 		    sizeof(u32.shmid_ds32));
1434 		break;
1435 	}
1436 
1437 done:
1438 	if (error) {
1439 		/* Invalidate the return value */
1440 		td->td_retval[0] = -1;
1441 	}
1442 	return (error);
1443 }
1444 #endif
1445 
1446 int
1447 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1448 {
1449 	int error;
1450 	union {
1451 		struct shmid_ds shmid_ds;
1452 		struct shm_info shm_info;
1453 		struct shminfo shminfo;
1454 	} u;
1455 	union {
1456 		struct shmid_ds32 shmid_ds32;
1457 		struct shm_info32 shm_info32;
1458 		struct shminfo32 shminfo32;
1459 	} u32;
1460 	size_t sz;
1461 
1462 	if (uap->cmd == IPC_SET) {
1463 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1464 		    sizeof(u32.shmid_ds32))))
1465 			goto done;
1466 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1467 		    &u.shmid_ds.shm_perm);
1468 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1469 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1470 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1471 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1472 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1473 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1474 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1475 	}
1476 
1477 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1478 	if (error)
1479 		goto done;
1480 
1481 	/* Cases in which we need to copyout */
1482 	switch (uap->cmd) {
1483 	case IPC_INFO:
1484 		CP(u.shminfo, u32.shminfo32, shmmax);
1485 		CP(u.shminfo, u32.shminfo32, shmmin);
1486 		CP(u.shminfo, u32.shminfo32, shmmni);
1487 		CP(u.shminfo, u32.shminfo32, shmseg);
1488 		CP(u.shminfo, u32.shminfo32, shmall);
1489 		error = copyout(&u32.shminfo32, uap->buf,
1490 		    sizeof(u32.shminfo32));
1491 		break;
1492 	case SHM_INFO:
1493 		CP(u.shm_info, u32.shm_info32, used_ids);
1494 		CP(u.shm_info, u32.shm_info32, shm_rss);
1495 		CP(u.shm_info, u32.shm_info32, shm_tot);
1496 		CP(u.shm_info, u32.shm_info32, shm_swp);
1497 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1498 		CP(u.shm_info, u32.shm_info32, swap_successes);
1499 		error = copyout(&u32.shm_info32, uap->buf,
1500 		    sizeof(u32.shm_info32));
1501 		break;
1502 	case SHM_STAT:
1503 	case IPC_STAT:
1504 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1505 		    &u32.shmid_ds32.shm_perm);
1506 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1507 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1508 		else
1509 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1510 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1511 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1512 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1513 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1514 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1515 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1516 		error = copyout(&u32.shmid_ds32, uap->buf,
1517 		    sizeof(u32.shmid_ds32));
1518 		break;
1519 	}
1520 
1521 done:
1522 	if (error) {
1523 		/* Invalidate the return value */
1524 		td->td_retval[0] = -1;
1525 	}
1526 	return (error);
1527 }
1528 #endif
1529 
1530 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1531     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1532 
1533 #ifndef CP
1534 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1535 #endif
1536 
1537 #ifndef _SYS_SYSPROTO_H_
1538 struct freebsd7_shmctl_args {
1539 	int shmid;
1540 	int cmd;
1541 	struct shmid_ds_old *buf;
1542 };
1543 #endif
1544 int
1545 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1546 {
1547 	int error;
1548 	struct shmid_ds_old old;
1549 	struct shmid_ds buf;
1550 	size_t bufsz;
1551 
1552 	/*
1553 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1554 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1555 	 * return an error back to the user since we do not to support this.
1556 	 */
1557 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1558 	    uap->cmd == SHM_STAT)
1559 		return (EINVAL);
1560 
1561 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1562 	if (uap->cmd == IPC_SET) {
1563 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1564 			goto done;
1565 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1566 		CP(old, buf, shm_segsz);
1567 		CP(old, buf, shm_lpid);
1568 		CP(old, buf, shm_cpid);
1569 		CP(old, buf, shm_nattch);
1570 		CP(old, buf, shm_atime);
1571 		CP(old, buf, shm_dtime);
1572 		CP(old, buf, shm_ctime);
1573 	}
1574 
1575 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1576 	if (error)
1577 		goto done;
1578 
1579 	/* Cases in which we need to copyout */
1580 	switch (uap->cmd) {
1581 	case IPC_STAT:
1582 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1583 		if (buf.shm_segsz > INT_MAX)
1584 			old.shm_segsz = INT_MAX;
1585 		else
1586 			CP(buf, old, shm_segsz);
1587 		CP(buf, old, shm_lpid);
1588 		CP(buf, old, shm_cpid);
1589 		if (buf.shm_nattch > SHRT_MAX)
1590 			old.shm_nattch = SHRT_MAX;
1591 		else
1592 			CP(buf, old, shm_nattch);
1593 		CP(buf, old, shm_atime);
1594 		CP(buf, old, shm_dtime);
1595 		CP(buf, old, shm_ctime);
1596 		old.shm_internal = NULL;
1597 		error = copyout(&old, uap->buf, sizeof(old));
1598 		break;
1599 	}
1600 
1601 done:
1602 	if (error) {
1603 		/* Invalidate the return value */
1604 		td->td_retval[0] = -1;
1605 	}
1606 	return (error);
1607 }
1608 
1609 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1610 	   COMPAT_FREEBSD7 */
1611 
1612 static int
1613 sysvshm_modload(struct module *module, int cmd, void *arg)
1614 {
1615 	int error = 0;
1616 
1617 	switch (cmd) {
1618 	case MOD_LOAD:
1619 		error = shminit();
1620 		if (error != 0)
1621 			shmunload();
1622 		break;
1623 	case MOD_UNLOAD:
1624 		error = shmunload();
1625 		break;
1626 	case MOD_SHUTDOWN:
1627 		break;
1628 	default:
1629 		error = EINVAL;
1630 		break;
1631 	}
1632 	return (error);
1633 }
1634 
1635 static moduledata_t sysvshm_mod = {
1636 	"sysvshm",
1637 	&sysvshm_modload,
1638 	NULL
1639 };
1640 
1641 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1642 MODULE_VERSION(sysvshm, 1);
1643