xref: /freebsd/sys/kern/sysv_shm.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause AND BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Adam Glass and Charles
17  *	Hannum.
18  * 4. The names of the authors may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  *
32  * $NetBSD: sysv_shm.c,v 1.39 1997/10/07 10:02:03 drochner Exp $
33  */
34 /*-
35  * Copyright (c) 2003-2005 McAfee, Inc.
36  * Copyright (c) 2016-2017 Robert N. M. Watson
37  * All rights reserved.
38  *
39  * This software was developed for the FreeBSD Project in part by McAfee
40  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
41  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
42  * program.
43  *
44  * Portions of this software were developed by BAE Systems, the University of
45  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
46  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
47  * Computing (TC) research program.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  */
70 
71 #include <sys/cdefs.h>
72 __FBSDID("$FreeBSD$");
73 
74 #include "opt_sysvipc.h"
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/limits.h>
80 #include <sys/lock.h>
81 #include <sys/sysctl.h>
82 #include <sys/shm.h>
83 #include <sys/proc.h>
84 #include <sys/malloc.h>
85 #include <sys/mman.h>
86 #include <sys/module.h>
87 #include <sys/mutex.h>
88 #include <sys/racct.h>
89 #include <sys/resourcevar.h>
90 #include <sys/rwlock.h>
91 #include <sys/stat.h>
92 #include <sys/syscall.h>
93 #include <sys/syscallsubr.h>
94 #include <sys/sysent.h>
95 #include <sys/sysproto.h>
96 #include <sys/jail.h>
97 
98 #include <security/audit/audit.h>
99 #include <security/mac/mac_framework.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_param.h>
103 #include <vm/pmap.h>
104 #include <vm/vm_object.h>
105 #include <vm/vm_map.h>
106 #include <vm/vm_page.h>
107 #include <vm/vm_pager.h>
108 
109 FEATURE(sysv_shm, "System V shared memory segments support");
110 
111 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
112 
113 static int shmget_allocate_segment(struct thread *td,
114     struct shmget_args *uap, int mode);
115 static int shmget_existing(struct thread *td, struct shmget_args *uap,
116     int mode, int segnum);
117 
118 #define	SHMSEG_FREE     	0x0200
119 #define	SHMSEG_REMOVED  	0x0400
120 #define	SHMSEG_ALLOCATED	0x0800
121 
122 static int shm_last_free, shm_nused, shmalloced;
123 vm_size_t shm_committed;
124 static struct shmid_kernel *shmsegs;
125 static unsigned shm_prison_slot;
126 
127 struct shmmap_state {
128 	vm_offset_t va;
129 	int shmid;
130 };
131 
132 static void shm_deallocate_segment(struct shmid_kernel *);
133 static int shm_find_segment_by_key(struct prison *, key_t);
134 static struct shmid_kernel *shm_find_segment(struct prison *, int, bool);
135 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
136 static void shmrealloc(void);
137 static int shminit(void);
138 static int sysvshm_modload(struct module *, int, void *);
139 static int shmunload(void);
140 static void shmexit_myhook(struct vmspace *vm);
141 static void shmfork_myhook(struct proc *p1, struct proc *p2);
142 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
143 static void shm_remove(struct shmid_kernel *, int);
144 static struct prison *shm_find_prison(struct ucred *);
145 static int shm_prison_cansee(struct prison *, struct shmid_kernel *);
146 static int shm_prison_check(void *, void *);
147 static int shm_prison_set(void *, void *);
148 static int shm_prison_get(void *, void *);
149 static int shm_prison_remove(void *, void *);
150 static void shm_prison_cleanup(struct prison *);
151 
152 /*
153  * Tuneable values.
154  */
155 #ifndef SHMMAXPGS
156 #define	SHMMAXPGS	131072	/* Note: sysv shared memory is swap backed. */
157 #endif
158 #ifndef SHMMAX
159 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
160 #endif
161 #ifndef SHMMIN
162 #define	SHMMIN	1
163 #endif
164 #ifndef SHMMNI
165 #define	SHMMNI	192
166 #endif
167 #ifndef SHMSEG
168 #define	SHMSEG	128
169 #endif
170 #ifndef SHMALL
171 #define	SHMALL	(SHMMAXPGS)
172 #endif
173 
174 struct	shminfo shminfo = {
175 	.shmmax = SHMMAX,
176 	.shmmin = SHMMIN,
177 	.shmmni = SHMMNI,
178 	.shmseg = SHMSEG,
179 	.shmall = SHMALL
180 };
181 
182 static int shm_use_phys;
183 static int shm_allow_removed = 1;
184 
185 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
186     "Maximum shared memory segment size");
187 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
188     "Minimum shared memory segment size");
189 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
190     "Number of shared memory identifiers");
191 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
192     "Number of segments per process");
193 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
194     "Maximum number of pages available for shared memory");
195 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
196     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
197 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
198     &shm_allow_removed, 0,
199     "Enable/Disable attachment to attached segments marked for removal");
200 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
201     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
202     "Array of struct shmid_kernel for each potential shared memory segment");
203 
204 static struct sx sysvshmsx;
205 #define	SYSVSHM_LOCK()		sx_xlock(&sysvshmsx)
206 #define	SYSVSHM_UNLOCK()	sx_xunlock(&sysvshmsx)
207 #define	SYSVSHM_ASSERT_LOCKED()	sx_assert(&sysvshmsx, SA_XLOCKED)
208 
209 static int
210 shm_find_segment_by_key(struct prison *pr, key_t key)
211 {
212 	int i;
213 
214 	for (i = 0; i < shmalloced; i++)
215 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
216 		    shmsegs[i].cred != NULL &&
217 		    shmsegs[i].cred->cr_prison == pr &&
218 		    shmsegs[i].u.shm_perm.key == key)
219 			return (i);
220 	return (-1);
221 }
222 
223 /*
224  * Finds segment either by shmid if is_shmid is true, or by segnum if
225  * is_shmid is false.
226  */
227 static struct shmid_kernel *
228 shm_find_segment(struct prison *rpr, int arg, bool is_shmid)
229 {
230 	struct shmid_kernel *shmseg;
231 	int segnum;
232 
233 	segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
234 	if (segnum < 0 || segnum >= shmalloced)
235 		return (NULL);
236 	shmseg = &shmsegs[segnum];
237 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
238 	    (!shm_allow_removed &&
239 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
240 	    (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)) ||
241 	    shm_prison_cansee(rpr, shmseg) != 0)
242 		return (NULL);
243 	return (shmseg);
244 }
245 
246 static void
247 shm_deallocate_segment(struct shmid_kernel *shmseg)
248 {
249 	vm_size_t size;
250 
251 	SYSVSHM_ASSERT_LOCKED();
252 
253 	vm_object_deallocate(shmseg->object);
254 	shmseg->object = NULL;
255 	size = round_page(shmseg->u.shm_segsz);
256 	shm_committed -= btoc(size);
257 	shm_nused--;
258 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
259 #ifdef MAC
260 	mac_sysvshm_cleanup(shmseg);
261 #endif
262 	racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
263 	racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
264 	crfree(shmseg->cred);
265 	shmseg->cred = NULL;
266 }
267 
268 static int
269 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
270 {
271 	struct shmid_kernel *shmseg;
272 	int segnum, result;
273 	vm_size_t size;
274 
275 	SYSVSHM_ASSERT_LOCKED();
276 	segnum = IPCID_TO_IX(shmmap_s->shmid);
277 	KASSERT(segnum >= 0 && segnum < shmalloced,
278 	    ("segnum %d shmalloced %d", segnum, shmalloced));
279 
280 	shmseg = &shmsegs[segnum];
281 	size = round_page(shmseg->u.shm_segsz);
282 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
283 	if (result != KERN_SUCCESS)
284 		return (EINVAL);
285 	shmmap_s->shmid = -1;
286 	shmseg->u.shm_dtime = time_second;
287 	if (--shmseg->u.shm_nattch == 0 &&
288 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
289 		shm_deallocate_segment(shmseg);
290 		shm_last_free = segnum;
291 	}
292 	return (0);
293 }
294 
295 static void
296 shm_remove(struct shmid_kernel *shmseg, int segnum)
297 {
298 
299 	shmseg->u.shm_perm.key = IPC_PRIVATE;
300 	shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
301 	if (shmseg->u.shm_nattch == 0) {
302 		shm_deallocate_segment(shmseg);
303 		shm_last_free = segnum;
304 	}
305 }
306 
307 static struct prison *
308 shm_find_prison(struct ucred *cred)
309 {
310 	struct prison *pr, *rpr;
311 
312 	pr = cred->cr_prison;
313 	prison_lock(pr);
314 	rpr = osd_jail_get(pr, shm_prison_slot);
315 	prison_unlock(pr);
316 	return rpr;
317 }
318 
319 static int
320 shm_prison_cansee(struct prison *rpr, struct shmid_kernel *shmseg)
321 {
322 
323 	if (shmseg->cred == NULL ||
324 	    !(rpr == shmseg->cred->cr_prison ||
325 	      prison_ischild(rpr, shmseg->cred->cr_prison)))
326 		return (EINVAL);
327 	return (0);
328 }
329 
330 static int
331 kern_shmdt_locked(struct thread *td, const void *shmaddr)
332 {
333 	struct proc *p = td->td_proc;
334 	struct shmmap_state *shmmap_s;
335 #ifdef MAC
336 	int error;
337 #endif
338 	int i;
339 
340 	SYSVSHM_ASSERT_LOCKED();
341 	if (shm_find_prison(td->td_ucred) == NULL)
342 		return (ENOSYS);
343 	shmmap_s = p->p_vmspace->vm_shm;
344  	if (shmmap_s == NULL)
345 		return (EINVAL);
346 	AUDIT_ARG_SVIPC_ID(shmmap_s->shmid);
347 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
348 		if (shmmap_s->shmid != -1 &&
349 		    shmmap_s->va == (vm_offset_t)shmaddr) {
350 			break;
351 		}
352 	}
353 	if (i == shminfo.shmseg)
354 		return (EINVAL);
355 #ifdef MAC
356 	error = mac_sysvshm_check_shmdt(td->td_ucred,
357 	    &shmsegs[IPCID_TO_IX(shmmap_s->shmid)]);
358 	if (error != 0)
359 		return (error);
360 #endif
361 	return (shm_delete_mapping(p->p_vmspace, shmmap_s));
362 }
363 
364 #ifndef _SYS_SYSPROTO_H_
365 struct shmdt_args {
366 	const void *shmaddr;
367 };
368 #endif
369 int
370 sys_shmdt(struct thread *td, struct shmdt_args *uap)
371 {
372 	int error;
373 
374 	SYSVSHM_LOCK();
375 	error = kern_shmdt_locked(td, uap->shmaddr);
376 	SYSVSHM_UNLOCK();
377 	return (error);
378 }
379 
380 static int
381 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
382     int shmflg)
383 {
384 	struct prison *rpr;
385 	struct proc *p = td->td_proc;
386 	struct shmid_kernel *shmseg;
387 	struct shmmap_state *shmmap_s;
388 	vm_offset_t attach_va;
389 	vm_prot_t prot;
390 	vm_size_t size;
391 	int cow, error, find_space, i, rv;
392 
393 	AUDIT_ARG_SVIPC_ID(shmid);
394 	AUDIT_ARG_VALUE(shmflg);
395 
396 	SYSVSHM_ASSERT_LOCKED();
397 	rpr = shm_find_prison(td->td_ucred);
398 	if (rpr == NULL)
399 		return (ENOSYS);
400 	shmmap_s = p->p_vmspace->vm_shm;
401 	if (shmmap_s == NULL) {
402 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
403 		    M_SHM, M_WAITOK);
404 		for (i = 0; i < shminfo.shmseg; i++)
405 			shmmap_s[i].shmid = -1;
406 		KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
407 		p->p_vmspace->vm_shm = shmmap_s;
408 	}
409 	shmseg = shm_find_segment(rpr, shmid, true);
410 	if (shmseg == NULL)
411 		return (EINVAL);
412 	error = ipcperm(td, &shmseg->u.shm_perm,
413 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
414 	if (error != 0)
415 		return (error);
416 #ifdef MAC
417 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
418 	if (error != 0)
419 		return (error);
420 #endif
421 	for (i = 0; i < shminfo.shmseg; i++) {
422 		if (shmmap_s->shmid == -1)
423 			break;
424 		shmmap_s++;
425 	}
426 	if (i >= shminfo.shmseg)
427 		return (EMFILE);
428 	size = round_page(shmseg->u.shm_segsz);
429 	prot = VM_PROT_READ;
430 	cow = MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL;
431 	if ((shmflg & SHM_RDONLY) == 0)
432 		prot |= VM_PROT_WRITE;
433 	if (shmaddr != NULL) {
434 		if ((shmflg & SHM_RND) != 0)
435 			attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
436 		else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
437 			attach_va = (vm_offset_t)shmaddr;
438 		else
439 			return (EINVAL);
440 		if ((shmflg & SHM_REMAP) != 0)
441 			cow |= MAP_REMAP;
442 		find_space = VMFS_NO_SPACE;
443 	} else {
444 		/*
445 		 * This is just a hint to vm_map_find() about where to
446 		 * put it.
447 		 */
448 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
449 		    lim_max(td, RLIMIT_DATA));
450 		find_space = VMFS_OPTIMAL_SPACE;
451 	}
452 
453 	vm_object_reference(shmseg->object);
454 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
455 	    size, 0, find_space, prot, prot, cow);
456 	if (rv != KERN_SUCCESS) {
457 		vm_object_deallocate(shmseg->object);
458 		return (ENOMEM);
459 	}
460 
461 	shmmap_s->va = attach_va;
462 	shmmap_s->shmid = shmid;
463 	shmseg->u.shm_lpid = p->p_pid;
464 	shmseg->u.shm_atime = time_second;
465 	shmseg->u.shm_nattch++;
466 	td->td_retval[0] = attach_va;
467 	return (error);
468 }
469 
470 int
471 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
472 {
473 	int error;
474 
475 	SYSVSHM_LOCK();
476 	error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
477 	SYSVSHM_UNLOCK();
478 	return (error);
479 }
480 
481 #ifndef _SYS_SYSPROTO_H_
482 struct shmat_args {
483 	int shmid;
484 	const void *shmaddr;
485 	int shmflg;
486 };
487 #endif
488 int
489 sys_shmat(struct thread *td, struct shmat_args *uap)
490 {
491 
492 	return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
493 }
494 
495 static int
496 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
497     size_t *bufsz)
498 {
499 	struct prison *rpr;
500 	struct shmid_kernel *shmseg;
501 	struct shmid_ds *shmidp;
502 	struct shm_info shm_info;
503 	int error;
504 
505 	SYSVSHM_ASSERT_LOCKED();
506 
507 	rpr = shm_find_prison(td->td_ucred);
508 	if (rpr == NULL)
509 		return (ENOSYS);
510 
511 	AUDIT_ARG_SVIPC_ID(shmid);
512 	AUDIT_ARG_SVIPC_CMD(cmd);
513 
514 	switch (cmd) {
515 	/*
516 	 * It is possible that kern_shmctl is being called from the Linux ABI
517 	 * layer, in which case, we will need to implement IPC_INFO.  It should
518 	 * be noted that other shmctl calls will be funneled through here for
519 	 * Linix binaries as well.
520 	 *
521 	 * NB: The Linux ABI layer will convert this data to structure(s) more
522 	 * consistent with the Linux ABI.
523 	 */
524 	case IPC_INFO:
525 		memcpy(buf, &shminfo, sizeof(shminfo));
526 		if (bufsz)
527 			*bufsz = sizeof(shminfo);
528 		td->td_retval[0] = shmalloced;
529 		return (0);
530 	case SHM_INFO: {
531 		shm_info.used_ids = shm_nused;
532 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
533 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
534 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
535 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
536 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
537 		memcpy(buf, &shm_info, sizeof(shm_info));
538 		if (bufsz != NULL)
539 			*bufsz = sizeof(shm_info);
540 		td->td_retval[0] = shmalloced;
541 		return (0);
542 	}
543 	}
544 	shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
545 	if (shmseg == NULL)
546 		return (EINVAL);
547 #ifdef MAC
548 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
549 	if (error != 0)
550 		return (error);
551 #endif
552 	switch (cmd) {
553 	case SHM_STAT:
554 	case IPC_STAT:
555 		shmidp = (struct shmid_ds *)buf;
556 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
557 		if (error != 0)
558 			return (error);
559 		memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
560 		if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
561 			shmidp->shm_perm.key = IPC_PRIVATE;
562 		if (bufsz != NULL)
563 			*bufsz = sizeof(struct shmid_ds);
564 		if (cmd == SHM_STAT) {
565 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
566 			    shmseg->u.shm_perm);
567 		}
568 		break;
569 	case IPC_SET:
570 		shmidp = (struct shmid_ds *)buf;
571 		AUDIT_ARG_SVIPC_PERM(&shmidp->shm_perm);
572 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
573 		if (error != 0)
574 			return (error);
575 		shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
576 		shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
577 		shmseg->u.shm_perm.mode =
578 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
579 		    (shmidp->shm_perm.mode & ACCESSPERMS);
580 		shmseg->u.shm_ctime = time_second;
581 		break;
582 	case IPC_RMID:
583 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
584 		if (error != 0)
585 			return (error);
586 		shm_remove(shmseg, IPCID_TO_IX(shmid));
587 		break;
588 #if 0
589 	case SHM_LOCK:
590 	case SHM_UNLOCK:
591 #endif
592 	default:
593 		error = EINVAL;
594 		break;
595 	}
596 	return (error);
597 }
598 
599 int
600 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
601 {
602 	int error;
603 
604 	SYSVSHM_LOCK();
605 	error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
606 	SYSVSHM_UNLOCK();
607 	return (error);
608 }
609 
610 
611 #ifndef _SYS_SYSPROTO_H_
612 struct shmctl_args {
613 	int shmid;
614 	int cmd;
615 	struct shmid_ds *buf;
616 };
617 #endif
618 int
619 sys_shmctl(struct thread *td, struct shmctl_args *uap)
620 {
621 	int error;
622 	struct shmid_ds buf;
623 	size_t bufsz;
624 
625 	/*
626 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
627 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
628 	 * return an error back to the user since we do not to support this.
629 	 */
630 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
631 	    uap->cmd == SHM_STAT)
632 		return (EINVAL);
633 
634 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
635 	if (uap->cmd == IPC_SET) {
636 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
637 			goto done;
638 	}
639 
640 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
641 	if (error)
642 		goto done;
643 
644 	/* Cases in which we need to copyout */
645 	switch (uap->cmd) {
646 	case IPC_STAT:
647 		error = copyout(&buf, uap->buf, bufsz);
648 		break;
649 	}
650 
651 done:
652 	if (error) {
653 		/* Invalidate the return value */
654 		td->td_retval[0] = -1;
655 	}
656 	return (error);
657 }
658 
659 
660 static int
661 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
662     int segnum)
663 {
664 	struct shmid_kernel *shmseg;
665 #ifdef MAC
666 	int error;
667 #endif
668 
669 	SYSVSHM_ASSERT_LOCKED();
670 	KASSERT(segnum >= 0 && segnum < shmalloced,
671 	    ("segnum %d shmalloced %d", segnum, shmalloced));
672 	shmseg = &shmsegs[segnum];
673 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
674 		return (EEXIST);
675 #ifdef MAC
676 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
677 	if (error != 0)
678 		return (error);
679 #endif
680 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
681 		return (EINVAL);
682 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
683 	return (0);
684 }
685 
686 static int
687 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
688 {
689 	struct ucred *cred = td->td_ucred;
690 	struct shmid_kernel *shmseg;
691 	vm_object_t shm_object;
692 	int i, segnum;
693 	size_t size;
694 
695 	SYSVSHM_ASSERT_LOCKED();
696 
697 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
698 		return (EINVAL);
699 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
700 		return (ENOSPC);
701 	size = round_page(uap->size);
702 	if (shm_committed + btoc(size) > shminfo.shmall)
703 		return (ENOMEM);
704 	if (shm_last_free < 0) {
705 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
706 		for (i = 0; i < shmalloced; i++)
707 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
708 				break;
709 		if (i == shmalloced)
710 			return (ENOSPC);
711 		segnum = i;
712 	} else  {
713 		segnum = shm_last_free;
714 		shm_last_free = -1;
715 	}
716 	KASSERT(segnum >= 0 && segnum < shmalloced,
717 	    ("segnum %d shmalloced %d", segnum, shmalloced));
718 	shmseg = &shmsegs[segnum];
719 #ifdef RACCT
720 	if (racct_enable) {
721 		PROC_LOCK(td->td_proc);
722 		if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
723 			PROC_UNLOCK(td->td_proc);
724 			return (ENOSPC);
725 		}
726 		if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
727 			racct_sub(td->td_proc, RACCT_NSHM, 1);
728 			PROC_UNLOCK(td->td_proc);
729 			return (ENOMEM);
730 		}
731 		PROC_UNLOCK(td->td_proc);
732 	}
733 #endif
734 
735 	/*
736 	 * We make sure that we have allocated a pager before we need
737 	 * to.
738 	 */
739 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
740 	    0, size, VM_PROT_DEFAULT, 0, cred);
741 	if (shm_object == NULL) {
742 #ifdef RACCT
743 		if (racct_enable) {
744 			PROC_LOCK(td->td_proc);
745 			racct_sub(td->td_proc, RACCT_NSHM, 1);
746 			racct_sub(td->td_proc, RACCT_SHMSIZE, size);
747 			PROC_UNLOCK(td->td_proc);
748 		}
749 #endif
750 		return (ENOMEM);
751 	}
752 	shm_object->pg_color = 0;
753 	VM_OBJECT_WLOCK(shm_object);
754 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
755 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
756 	VM_OBJECT_WUNLOCK(shm_object);
757 
758 	shmseg->object = shm_object;
759 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
760 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
761 	shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
762 	shmseg->u.shm_perm.key = uap->key;
763 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
764 	shmseg->cred = crhold(cred);
765 	shmseg->u.shm_segsz = uap->size;
766 	shmseg->u.shm_cpid = td->td_proc->p_pid;
767 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
768 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
769 #ifdef MAC
770 	mac_sysvshm_create(cred, shmseg);
771 #endif
772 	shmseg->u.shm_ctime = time_second;
773 	shm_committed += btoc(size);
774 	shm_nused++;
775 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
776 
777 	return (0);
778 }
779 
780 #ifndef _SYS_SYSPROTO_H_
781 struct shmget_args {
782 	key_t key;
783 	size_t size;
784 	int shmflg;
785 };
786 #endif
787 int
788 sys_shmget(struct thread *td, struct shmget_args *uap)
789 {
790 	int segnum, mode;
791 	int error;
792 
793 	if (shm_find_prison(td->td_ucred) == NULL)
794 		return (ENOSYS);
795 	mode = uap->shmflg & ACCESSPERMS;
796 	SYSVSHM_LOCK();
797 	if (uap->key == IPC_PRIVATE) {
798 		error = shmget_allocate_segment(td, uap, mode);
799 	} else {
800 		segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
801 		    uap->key);
802 		if (segnum >= 0)
803 			error = shmget_existing(td, uap, mode, segnum);
804 		else if ((uap->shmflg & IPC_CREAT) == 0)
805 			error = ENOENT;
806 		else
807 			error = shmget_allocate_segment(td, uap, mode);
808 	}
809 	SYSVSHM_UNLOCK();
810 	return (error);
811 }
812 
813 static void
814 shmfork_myhook(struct proc *p1, struct proc *p2)
815 {
816 	struct shmmap_state *shmmap_s;
817 	size_t size;
818 	int i;
819 
820 	SYSVSHM_LOCK();
821 	size = shminfo.shmseg * sizeof(struct shmmap_state);
822 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
823 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
824 	p2->p_vmspace->vm_shm = shmmap_s;
825 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
826 		if (shmmap_s->shmid != -1) {
827 			KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
828 			    IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
829 			    ("segnum %d shmalloced %d",
830 			    IPCID_TO_IX(shmmap_s->shmid), shmalloced));
831 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
832 		}
833 	}
834 	SYSVSHM_UNLOCK();
835 }
836 
837 static void
838 shmexit_myhook(struct vmspace *vm)
839 {
840 	struct shmmap_state *base, *shm;
841 	int i;
842 
843 	base = vm->vm_shm;
844 	if (base != NULL) {
845 		vm->vm_shm = NULL;
846 		SYSVSHM_LOCK();
847 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
848 			if (shm->shmid != -1)
849 				shm_delete_mapping(vm, shm);
850 		}
851 		SYSVSHM_UNLOCK();
852 		free(base, M_SHM);
853 	}
854 }
855 
856 static void
857 shmrealloc(void)
858 {
859 	struct shmid_kernel *newsegs;
860 	int i;
861 
862 	SYSVSHM_ASSERT_LOCKED();
863 
864 	if (shmalloced >= shminfo.shmmni)
865 		return;
866 
867 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM,
868 	    M_WAITOK | M_ZERO);
869 	for (i = 0; i < shmalloced; i++)
870 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
871 	for (; i < shminfo.shmmni; i++) {
872 		newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
873 		newsegs[i].u.shm_perm.seq = 0;
874 #ifdef MAC
875 		mac_sysvshm_init(&newsegs[i]);
876 #endif
877 	}
878 	free(shmsegs, M_SHM);
879 	shmsegs = newsegs;
880 	shmalloced = shminfo.shmmni;
881 }
882 
883 static struct syscall_helper_data shm_syscalls[] = {
884 	SYSCALL_INIT_HELPER(shmat),
885 	SYSCALL_INIT_HELPER(shmctl),
886 	SYSCALL_INIT_HELPER(shmdt),
887 	SYSCALL_INIT_HELPER(shmget),
888 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
889     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
890 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
891 #endif
892 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
893 	SYSCALL_INIT_HELPER(shmsys),
894 #endif
895 	SYSCALL_INIT_LAST
896 };
897 
898 #ifdef COMPAT_FREEBSD32
899 #include <compat/freebsd32/freebsd32.h>
900 #include <compat/freebsd32/freebsd32_ipc.h>
901 #include <compat/freebsd32/freebsd32_proto.h>
902 #include <compat/freebsd32/freebsd32_signal.h>
903 #include <compat/freebsd32/freebsd32_syscall.h>
904 #include <compat/freebsd32/freebsd32_util.h>
905 
906 static struct syscall_helper_data shm32_syscalls[] = {
907 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
908 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
909 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
910 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
911 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
912 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
913     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
914 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
915 #endif
916 	SYSCALL_INIT_LAST
917 };
918 #endif
919 
920 static int
921 shminit(void)
922 {
923 	struct prison *pr;
924 	void **rsv;
925 	int i, error;
926 	osd_method_t methods[PR_MAXMETHOD] = {
927 	    [PR_METHOD_CHECK] =		shm_prison_check,
928 	    [PR_METHOD_SET] =		shm_prison_set,
929 	    [PR_METHOD_GET] =		shm_prison_get,
930 	    [PR_METHOD_REMOVE] =	shm_prison_remove,
931 	};
932 
933 #ifndef BURN_BRIDGES
934 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
935 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
936 #endif
937 	if (shminfo.shmmax == SHMMAX) {
938 		/* Initialize shmmax dealing with possible overflow. */
939 		for (i = PAGE_SIZE; i != 0; i--) {
940 			shminfo.shmmax = shminfo.shmall * i;
941 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
942 				break;
943 		}
944 	}
945 	shmalloced = shminfo.shmmni;
946 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM,
947 	    M_WAITOK|M_ZERO);
948 	for (i = 0; i < shmalloced; i++) {
949 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
950 		shmsegs[i].u.shm_perm.seq = 0;
951 #ifdef MAC
952 		mac_sysvshm_init(&shmsegs[i]);
953 #endif
954 	}
955 	shm_last_free = 0;
956 	shm_nused = 0;
957 	shm_committed = 0;
958 	sx_init(&sysvshmsx, "sysvshmsx");
959 	shmexit_hook = &shmexit_myhook;
960 	shmfork_hook = &shmfork_myhook;
961 
962 	/* Set current prisons according to their allow.sysvipc. */
963 	shm_prison_slot = osd_jail_register(NULL, methods);
964 	rsv = osd_reserve(shm_prison_slot);
965 	prison_lock(&prison0);
966 	(void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
967 	prison_unlock(&prison0);
968 	rsv = NULL;
969 	sx_slock(&allprison_lock);
970 	TAILQ_FOREACH(pr, &allprison, pr_list) {
971 		if (rsv == NULL)
972 			rsv = osd_reserve(shm_prison_slot);
973 		prison_lock(pr);
974 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
975 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
976 			    &prison0);
977 			rsv = NULL;
978 		}
979 		prison_unlock(pr);
980 	}
981 	if (rsv != NULL)
982 		osd_free_reserved(rsv);
983 	sx_sunlock(&allprison_lock);
984 
985 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
986 	if (error != 0)
987 		return (error);
988 #ifdef COMPAT_FREEBSD32
989 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
990 	if (error != 0)
991 		return (error);
992 #endif
993 	return (0);
994 }
995 
996 static int
997 shmunload(void)
998 {
999 	int i;
1000 
1001 	if (shm_nused > 0)
1002 		return (EBUSY);
1003 
1004 #ifdef COMPAT_FREEBSD32
1005 	syscall32_helper_unregister(shm32_syscalls);
1006 #endif
1007 	syscall_helper_unregister(shm_syscalls);
1008 	if (shm_prison_slot != 0)
1009 		osd_jail_deregister(shm_prison_slot);
1010 
1011 	for (i = 0; i < shmalloced; i++) {
1012 #ifdef MAC
1013 		mac_sysvshm_destroy(&shmsegs[i]);
1014 #endif
1015 		/*
1016 		 * Objects might be still mapped into the processes
1017 		 * address spaces.  Actual free would happen on the
1018 		 * last mapping destruction.
1019 		 */
1020 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1021 			vm_object_deallocate(shmsegs[i].object);
1022 	}
1023 	free(shmsegs, M_SHM);
1024 	shmexit_hook = NULL;
1025 	shmfork_hook = NULL;
1026 	sx_destroy(&sysvshmsx);
1027 	return (0);
1028 }
1029 
1030 static int
1031 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1032 {
1033 	struct shmid_kernel tshmseg;
1034 #ifdef COMPAT_FREEBSD32
1035 	struct shmid_kernel32 tshmseg32;
1036 #endif
1037 	struct prison *pr, *rpr;
1038 	void *outaddr;
1039 	size_t outsize;
1040 	int error, i;
1041 
1042 	SYSVSHM_LOCK();
1043 	pr = req->td->td_ucred->cr_prison;
1044 	rpr = shm_find_prison(req->td->td_ucred);
1045 	error = 0;
1046 	for (i = 0; i < shmalloced; i++) {
1047 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1048 		    rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1049 			bzero(&tshmseg, sizeof(tshmseg));
1050 			tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1051 		} else {
1052 			tshmseg = shmsegs[i];
1053 			if (tshmseg.cred->cr_prison != pr)
1054 				tshmseg.u.shm_perm.key = IPC_PRIVATE;
1055 		}
1056 #ifdef COMPAT_FREEBSD32
1057 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1058 			bzero(&tshmseg32, sizeof(tshmseg32));
1059 			freebsd32_ipcperm_out(&tshmseg.u.shm_perm,
1060 			    &tshmseg32.u.shm_perm);
1061 			CP(tshmseg, tshmseg32, u.shm_segsz);
1062 			CP(tshmseg, tshmseg32, u.shm_lpid);
1063 			CP(tshmseg, tshmseg32, u.shm_cpid);
1064 			CP(tshmseg, tshmseg32, u.shm_nattch);
1065 			CP(tshmseg, tshmseg32, u.shm_atime);
1066 			CP(tshmseg, tshmseg32, u.shm_dtime);
1067 			CP(tshmseg, tshmseg32, u.shm_ctime);
1068 			/* Don't copy object, label, or cred */
1069 			outaddr = &tshmseg32;
1070 			outsize = sizeof(tshmseg32);
1071 		} else
1072 #endif
1073 		{
1074 			tshmseg.object = NULL;
1075 			tshmseg.label = NULL;
1076 			tshmseg.cred = NULL;
1077 			outaddr = &tshmseg;
1078 			outsize = sizeof(tshmseg);
1079 		}
1080 		error = SYSCTL_OUT(req, outaddr, outsize);
1081 		if (error != 0)
1082 			break;
1083 	}
1084 	SYSVSHM_UNLOCK();
1085 	return (error);
1086 }
1087 
1088 static int
1089 shm_prison_check(void *obj, void *data)
1090 {
1091 	struct prison *pr = obj;
1092 	struct prison *prpr;
1093 	struct vfsoptlist *opts = data;
1094 	int error, jsys;
1095 
1096 	/*
1097 	 * sysvshm is a jailsys integer.
1098 	 * It must be "disable" if the parent jail is disabled.
1099 	 */
1100 	error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1101 	if (error != ENOENT) {
1102 		if (error != 0)
1103 			return (error);
1104 		switch (jsys) {
1105 		case JAIL_SYS_DISABLE:
1106 			break;
1107 		case JAIL_SYS_NEW:
1108 		case JAIL_SYS_INHERIT:
1109 			prison_lock(pr->pr_parent);
1110 			prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1111 			prison_unlock(pr->pr_parent);
1112 			if (prpr == NULL)
1113 				return (EPERM);
1114 			break;
1115 		default:
1116 			return (EINVAL);
1117 		}
1118 	}
1119 
1120 	return (0);
1121 }
1122 
1123 static int
1124 shm_prison_set(void *obj, void *data)
1125 {
1126 	struct prison *pr = obj;
1127 	struct prison *tpr, *orpr, *nrpr, *trpr;
1128 	struct vfsoptlist *opts = data;
1129 	void *rsv;
1130 	int jsys, descend;
1131 
1132 	/*
1133 	 * sysvshm controls which jail is the root of the associated segments
1134 	 * (this jail or same as the parent), or if the feature is available
1135 	 * at all.
1136 	 */
1137 	if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1138 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1139 		    ? JAIL_SYS_INHERIT
1140 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1141 		    ? JAIL_SYS_DISABLE
1142 		    : -1;
1143 	if (jsys == JAIL_SYS_DISABLE) {
1144 		prison_lock(pr);
1145 		orpr = osd_jail_get(pr, shm_prison_slot);
1146 		if (orpr != NULL)
1147 			osd_jail_del(pr, shm_prison_slot);
1148 		prison_unlock(pr);
1149 		if (orpr != NULL) {
1150 			if (orpr == pr)
1151 				shm_prison_cleanup(pr);
1152 			/* Disable all child jails as well. */
1153 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1154 				prison_lock(tpr);
1155 				trpr = osd_jail_get(tpr, shm_prison_slot);
1156 				if (trpr != NULL) {
1157 					osd_jail_del(tpr, shm_prison_slot);
1158 					prison_unlock(tpr);
1159 					if (trpr == tpr)
1160 						shm_prison_cleanup(tpr);
1161 				} else {
1162 					prison_unlock(tpr);
1163 					descend = 0;
1164 				}
1165 			}
1166 		}
1167 	} else if (jsys != -1) {
1168 		if (jsys == JAIL_SYS_NEW)
1169 			nrpr = pr;
1170 		else {
1171 			prison_lock(pr->pr_parent);
1172 			nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1173 			prison_unlock(pr->pr_parent);
1174 		}
1175 		rsv = osd_reserve(shm_prison_slot);
1176 		prison_lock(pr);
1177 		orpr = osd_jail_get(pr, shm_prison_slot);
1178 		if (orpr != nrpr)
1179 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1180 			    nrpr);
1181 		else
1182 			osd_free_reserved(rsv);
1183 		prison_unlock(pr);
1184 		if (orpr != nrpr) {
1185 			if (orpr == pr)
1186 				shm_prison_cleanup(pr);
1187 			if (orpr != NULL) {
1188 				/* Change child jails matching the old root, */
1189 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1190 					prison_lock(tpr);
1191 					trpr = osd_jail_get(tpr,
1192 					    shm_prison_slot);
1193 					if (trpr == orpr) {
1194 						(void)osd_jail_set(tpr,
1195 						    shm_prison_slot, nrpr);
1196 						prison_unlock(tpr);
1197 						if (trpr == tpr)
1198 							shm_prison_cleanup(tpr);
1199 					} else {
1200 						prison_unlock(tpr);
1201 						descend = 0;
1202 					}
1203 				}
1204 			}
1205 		}
1206 	}
1207 
1208 	return (0);
1209 }
1210 
1211 static int
1212 shm_prison_get(void *obj, void *data)
1213 {
1214 	struct prison *pr = obj;
1215 	struct prison *rpr;
1216 	struct vfsoptlist *opts = data;
1217 	int error, jsys;
1218 
1219 	/* Set sysvshm based on the jail's root prison. */
1220 	prison_lock(pr);
1221 	rpr = osd_jail_get(pr, shm_prison_slot);
1222 	prison_unlock(pr);
1223 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1224 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1225 	error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1226 	if (error == ENOENT)
1227 		error = 0;
1228 	return (error);
1229 }
1230 
1231 static int
1232 shm_prison_remove(void *obj, void *data __unused)
1233 {
1234 	struct prison *pr = obj;
1235 	struct prison *rpr;
1236 
1237 	SYSVSHM_LOCK();
1238 	prison_lock(pr);
1239 	rpr = osd_jail_get(pr, shm_prison_slot);
1240 	prison_unlock(pr);
1241 	if (rpr == pr)
1242 		shm_prison_cleanup(pr);
1243 	SYSVSHM_UNLOCK();
1244 	return (0);
1245 }
1246 
1247 static void
1248 shm_prison_cleanup(struct prison *pr)
1249 {
1250 	struct shmid_kernel *shmseg;
1251 	int i;
1252 
1253 	/* Remove any segments that belong to this jail. */
1254 	for (i = 0; i < shmalloced; i++) {
1255 		shmseg = &shmsegs[i];
1256 		if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1257 		    shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1258 			shm_remove(shmseg, i);
1259 		}
1260 	}
1261 }
1262 
1263 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1264 
1265 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1266 struct oshmid_ds {
1267 	struct	ipc_perm_old shm_perm;	/* operation perms */
1268 	int	shm_segsz;		/* size of segment (bytes) */
1269 	u_short	shm_cpid;		/* pid, creator */
1270 	u_short	shm_lpid;		/* pid, last operation */
1271 	short	shm_nattch;		/* no. of current attaches */
1272 	time_t	shm_atime;		/* last attach time */
1273 	time_t	shm_dtime;		/* last detach time */
1274 	time_t	shm_ctime;		/* last change time */
1275 	void	*shm_handle;		/* internal handle for shm segment */
1276 };
1277 
1278 struct oshmctl_args {
1279 	int shmid;
1280 	int cmd;
1281 	struct oshmid_ds *ubuf;
1282 };
1283 
1284 static int
1285 oshmctl(struct thread *td, struct oshmctl_args *uap)
1286 {
1287 #ifdef COMPAT_43
1288 	int error = 0;
1289 	struct prison *rpr;
1290 	struct shmid_kernel *shmseg;
1291 	struct oshmid_ds outbuf;
1292 
1293 	rpr = shm_find_prison(td->td_ucred);
1294 	if (rpr == NULL)
1295 		return (ENOSYS);
1296 	if (uap->cmd != IPC_STAT) {
1297 		return (freebsd7_shmctl(td,
1298 		    (struct freebsd7_shmctl_args *)uap));
1299 	}
1300 	SYSVSHM_LOCK();
1301 	shmseg = shm_find_segment(rpr, uap->shmid, true);
1302 	if (shmseg == NULL) {
1303 		SYSVSHM_UNLOCK();
1304 		return (EINVAL);
1305 	}
1306 	error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1307 	if (error != 0) {
1308 		SYSVSHM_UNLOCK();
1309 		return (error);
1310 	}
1311 #ifdef MAC
1312 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1313 	if (error != 0) {
1314 		SYSVSHM_UNLOCK();
1315 		return (error);
1316 	}
1317 #endif
1318 	ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1319 	outbuf.shm_segsz = shmseg->u.shm_segsz;
1320 	outbuf.shm_cpid = shmseg->u.shm_cpid;
1321 	outbuf.shm_lpid = shmseg->u.shm_lpid;
1322 	outbuf.shm_nattch = shmseg->u.shm_nattch;
1323 	outbuf.shm_atime = shmseg->u.shm_atime;
1324 	outbuf.shm_dtime = shmseg->u.shm_dtime;
1325 	outbuf.shm_ctime = shmseg->u.shm_ctime;
1326 	outbuf.shm_handle = shmseg->object;
1327 	SYSVSHM_UNLOCK();
1328 	return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1329 #else
1330 	return (EINVAL);
1331 #endif
1332 }
1333 
1334 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1335 static sy_call_t *shmcalls[] = {
1336 	(sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1337 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1338 	(sy_call_t *)freebsd7_shmctl
1339 };
1340 
1341 #ifndef _SYS_SYSPROTO_H_
1342 /* XXX actually varargs. */
1343 struct shmsys_args {
1344 	int	which;
1345 	int	a2;
1346 	int	a3;
1347 	int	a4;
1348 };
1349 #endif
1350 int
1351 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1352 {
1353 
1354 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1355 	if (uap->which < 0 || uap->which >= nitems(shmcalls))
1356 		return (EINVAL);
1357 	return ((*shmcalls[uap->which])(td, &uap->a2));
1358 }
1359 
1360 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1361 
1362 #ifdef COMPAT_FREEBSD32
1363 
1364 int
1365 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1366 {
1367 
1368 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1369     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1370 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1371 	switch (uap->which) {
1372 	case 0:	{	/* shmat */
1373 		struct shmat_args ap;
1374 
1375 		ap.shmid = uap->a2;
1376 		ap.shmaddr = PTRIN(uap->a3);
1377 		ap.shmflg = uap->a4;
1378 		return (sysent[SYS_shmat].sy_call(td, &ap));
1379 	}
1380 	case 2: {	/* shmdt */
1381 		struct shmdt_args ap;
1382 
1383 		ap.shmaddr = PTRIN(uap->a2);
1384 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1385 	}
1386 	case 3: {	/* shmget */
1387 		struct shmget_args ap;
1388 
1389 		ap.key = uap->a2;
1390 		ap.size = uap->a3;
1391 		ap.shmflg = uap->a4;
1392 		return (sysent[SYS_shmget].sy_call(td, &ap));
1393 	}
1394 	case 4: {	/* shmctl */
1395 		struct freebsd7_freebsd32_shmctl_args ap;
1396 
1397 		ap.shmid = uap->a2;
1398 		ap.cmd = uap->a3;
1399 		ap.buf = PTRIN(uap->a4);
1400 		return (freebsd7_freebsd32_shmctl(td, &ap));
1401 	}
1402 	case 1:		/* oshmctl */
1403 	default:
1404 		return (EINVAL);
1405 	}
1406 #else
1407 	return (nosys(td, NULL));
1408 #endif
1409 }
1410 
1411 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1412     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1413 int
1414 freebsd7_freebsd32_shmctl(struct thread *td,
1415     struct freebsd7_freebsd32_shmctl_args *uap)
1416 {
1417 	int error;
1418 	union {
1419 		struct shmid_ds shmid_ds;
1420 		struct shm_info shm_info;
1421 		struct shminfo shminfo;
1422 	} u;
1423 	union {
1424 		struct shmid_ds32_old shmid_ds32;
1425 		struct shm_info32 shm_info32;
1426 		struct shminfo32 shminfo32;
1427 	} u32;
1428 	size_t sz;
1429 
1430 	if (uap->cmd == IPC_SET) {
1431 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1432 		    sizeof(u32.shmid_ds32))))
1433 			goto done;
1434 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1435 		    &u.shmid_ds.shm_perm);
1436 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1437 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1438 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1439 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1440 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1441 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1442 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1443 	}
1444 
1445 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1446 	if (error)
1447 		goto done;
1448 
1449 	/* Cases in which we need to copyout */
1450 	switch (uap->cmd) {
1451 	case IPC_INFO:
1452 		CP(u.shminfo, u32.shminfo32, shmmax);
1453 		CP(u.shminfo, u32.shminfo32, shmmin);
1454 		CP(u.shminfo, u32.shminfo32, shmmni);
1455 		CP(u.shminfo, u32.shminfo32, shmseg);
1456 		CP(u.shminfo, u32.shminfo32, shmall);
1457 		error = copyout(&u32.shminfo32, uap->buf,
1458 		    sizeof(u32.shminfo32));
1459 		break;
1460 	case SHM_INFO:
1461 		CP(u.shm_info, u32.shm_info32, used_ids);
1462 		CP(u.shm_info, u32.shm_info32, shm_rss);
1463 		CP(u.shm_info, u32.shm_info32, shm_tot);
1464 		CP(u.shm_info, u32.shm_info32, shm_swp);
1465 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1466 		CP(u.shm_info, u32.shm_info32, swap_successes);
1467 		error = copyout(&u32.shm_info32, uap->buf,
1468 		    sizeof(u32.shm_info32));
1469 		break;
1470 	case SHM_STAT:
1471 	case IPC_STAT:
1472 		memset(&u32.shmid_ds32, 0, sizeof(u32.shmid_ds32));
1473 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1474 		    &u32.shmid_ds32.shm_perm);
1475 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1476 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1477 		else
1478 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1479 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1480 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1481 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1482 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1483 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1484 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1485 		u32.shmid_ds32.shm_internal = 0;
1486 		error = copyout(&u32.shmid_ds32, uap->buf,
1487 		    sizeof(u32.shmid_ds32));
1488 		break;
1489 	}
1490 
1491 done:
1492 	if (error) {
1493 		/* Invalidate the return value */
1494 		td->td_retval[0] = -1;
1495 	}
1496 	return (error);
1497 }
1498 #endif
1499 
1500 int
1501 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1502 {
1503 	int error;
1504 	union {
1505 		struct shmid_ds shmid_ds;
1506 		struct shm_info shm_info;
1507 		struct shminfo shminfo;
1508 	} u;
1509 	union {
1510 		struct shmid_ds32 shmid_ds32;
1511 		struct shm_info32 shm_info32;
1512 		struct shminfo32 shminfo32;
1513 	} u32;
1514 	size_t sz;
1515 
1516 	if (uap->cmd == IPC_SET) {
1517 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1518 		    sizeof(u32.shmid_ds32))))
1519 			goto done;
1520 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1521 		    &u.shmid_ds.shm_perm);
1522 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1523 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1524 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1525 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1526 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1527 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1528 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1529 	}
1530 
1531 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1532 	if (error)
1533 		goto done;
1534 
1535 	/* Cases in which we need to copyout */
1536 	switch (uap->cmd) {
1537 	case IPC_INFO:
1538 		CP(u.shminfo, u32.shminfo32, shmmax);
1539 		CP(u.shminfo, u32.shminfo32, shmmin);
1540 		CP(u.shminfo, u32.shminfo32, shmmni);
1541 		CP(u.shminfo, u32.shminfo32, shmseg);
1542 		CP(u.shminfo, u32.shminfo32, shmall);
1543 		error = copyout(&u32.shminfo32, uap->buf,
1544 		    sizeof(u32.shminfo32));
1545 		break;
1546 	case SHM_INFO:
1547 		CP(u.shm_info, u32.shm_info32, used_ids);
1548 		CP(u.shm_info, u32.shm_info32, shm_rss);
1549 		CP(u.shm_info, u32.shm_info32, shm_tot);
1550 		CP(u.shm_info, u32.shm_info32, shm_swp);
1551 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1552 		CP(u.shm_info, u32.shm_info32, swap_successes);
1553 		error = copyout(&u32.shm_info32, uap->buf,
1554 		    sizeof(u32.shm_info32));
1555 		break;
1556 	case SHM_STAT:
1557 	case IPC_STAT:
1558 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1559 		    &u32.shmid_ds32.shm_perm);
1560 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1561 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1562 		else
1563 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1564 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1565 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1566 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1567 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1568 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1569 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1570 		error = copyout(&u32.shmid_ds32, uap->buf,
1571 		    sizeof(u32.shmid_ds32));
1572 		break;
1573 	}
1574 
1575 done:
1576 	if (error) {
1577 		/* Invalidate the return value */
1578 		td->td_retval[0] = -1;
1579 	}
1580 	return (error);
1581 }
1582 #endif
1583 
1584 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1585     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1586 
1587 #ifndef CP
1588 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1589 #endif
1590 
1591 #ifndef _SYS_SYSPROTO_H_
1592 struct freebsd7_shmctl_args {
1593 	int shmid;
1594 	int cmd;
1595 	struct shmid_ds_old *buf;
1596 };
1597 #endif
1598 int
1599 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1600 {
1601 	int error;
1602 	struct shmid_ds_old old;
1603 	struct shmid_ds buf;
1604 	size_t bufsz;
1605 
1606 	/*
1607 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1608 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1609 	 * return an error back to the user since we do not to support this.
1610 	 */
1611 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1612 	    uap->cmd == SHM_STAT)
1613 		return (EINVAL);
1614 
1615 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1616 	if (uap->cmd == IPC_SET) {
1617 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1618 			goto done;
1619 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1620 		CP(old, buf, shm_segsz);
1621 		CP(old, buf, shm_lpid);
1622 		CP(old, buf, shm_cpid);
1623 		CP(old, buf, shm_nattch);
1624 		CP(old, buf, shm_atime);
1625 		CP(old, buf, shm_dtime);
1626 		CP(old, buf, shm_ctime);
1627 	}
1628 
1629 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1630 	if (error)
1631 		goto done;
1632 
1633 	/* Cases in which we need to copyout */
1634 	switch (uap->cmd) {
1635 	case IPC_STAT:
1636 		memset(&old, 0, sizeof(old));
1637 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1638 		if (buf.shm_segsz > INT_MAX)
1639 			old.shm_segsz = INT_MAX;
1640 		else
1641 			CP(buf, old, shm_segsz);
1642 		CP(buf, old, shm_lpid);
1643 		CP(buf, old, shm_cpid);
1644 		if (buf.shm_nattch > SHRT_MAX)
1645 			old.shm_nattch = SHRT_MAX;
1646 		else
1647 			CP(buf, old, shm_nattch);
1648 		CP(buf, old, shm_atime);
1649 		CP(buf, old, shm_dtime);
1650 		CP(buf, old, shm_ctime);
1651 		old.shm_internal = NULL;
1652 		error = copyout(&old, uap->buf, sizeof(old));
1653 		break;
1654 	}
1655 
1656 done:
1657 	if (error) {
1658 		/* Invalidate the return value */
1659 		td->td_retval[0] = -1;
1660 	}
1661 	return (error);
1662 }
1663 
1664 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1665 	   COMPAT_FREEBSD7 */
1666 
1667 static int
1668 sysvshm_modload(struct module *module, int cmd, void *arg)
1669 {
1670 	int error = 0;
1671 
1672 	switch (cmd) {
1673 	case MOD_LOAD:
1674 		error = shminit();
1675 		if (error != 0)
1676 			shmunload();
1677 		break;
1678 	case MOD_UNLOAD:
1679 		error = shmunload();
1680 		break;
1681 	case MOD_SHUTDOWN:
1682 		break;
1683 	default:
1684 		error = EINVAL;
1685 		break;
1686 	}
1687 	return (error);
1688 }
1689 
1690 static moduledata_t sysvshm_mod = {
1691 	"sysvshm",
1692 	&sysvshm_modload,
1693 	NULL
1694 };
1695 
1696 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1697 MODULE_VERSION(sysvshm, 1);
1698