xref: /freebsd/sys/kern/sysv_shm.c (revision d34048812292b714a0bf99967270d18fe3097c62)
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 error, 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 	if ((shmflg & SHM_RDONLY) == 0)
431 		prot |= VM_PROT_WRITE;
432 	if (shmaddr != NULL) {
433 		if ((shmflg & SHM_RND) != 0)
434 			attach_va = rounddown2((vm_offset_t)shmaddr, SHMLBA);
435 		else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
436 			attach_va = (vm_offset_t)shmaddr;
437 		else
438 			return (EINVAL);
439 	} else {
440 		/*
441 		 * This is just a hint to vm_map_find() about where to
442 		 * put it.
443 		 */
444 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
445 		    lim_max(td, RLIMIT_DATA));
446 	}
447 
448 	vm_object_reference(shmseg->object);
449 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
450 	    size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
451 	    prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
452 	if (rv != KERN_SUCCESS) {
453 		vm_object_deallocate(shmseg->object);
454 		return (ENOMEM);
455 	}
456 
457 	shmmap_s->va = attach_va;
458 	shmmap_s->shmid = shmid;
459 	shmseg->u.shm_lpid = p->p_pid;
460 	shmseg->u.shm_atime = time_second;
461 	shmseg->u.shm_nattch++;
462 	td->td_retval[0] = attach_va;
463 	return (error);
464 }
465 
466 int
467 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
468 {
469 	int error;
470 
471 	SYSVSHM_LOCK();
472 	error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
473 	SYSVSHM_UNLOCK();
474 	return (error);
475 }
476 
477 #ifndef _SYS_SYSPROTO_H_
478 struct shmat_args {
479 	int shmid;
480 	const void *shmaddr;
481 	int shmflg;
482 };
483 #endif
484 int
485 sys_shmat(struct thread *td, struct shmat_args *uap)
486 {
487 
488 	return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
489 }
490 
491 static int
492 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
493     size_t *bufsz)
494 {
495 	struct prison *rpr;
496 	struct shmid_kernel *shmseg;
497 	struct shmid_ds *shmidp;
498 	struct shm_info shm_info;
499 	int error;
500 
501 	SYSVSHM_ASSERT_LOCKED();
502 
503 	rpr = shm_find_prison(td->td_ucred);
504 	if (rpr == NULL)
505 		return (ENOSYS);
506 
507 	AUDIT_ARG_SVIPC_ID(shmid);
508 	AUDIT_ARG_SVIPC_CMD(cmd);
509 
510 	switch (cmd) {
511 	/*
512 	 * It is possible that kern_shmctl is being called from the Linux ABI
513 	 * layer, in which case, we will need to implement IPC_INFO.  It should
514 	 * be noted that other shmctl calls will be funneled through here for
515 	 * Linix binaries as well.
516 	 *
517 	 * NB: The Linux ABI layer will convert this data to structure(s) more
518 	 * consistent with the Linux ABI.
519 	 */
520 	case IPC_INFO:
521 		memcpy(buf, &shminfo, sizeof(shminfo));
522 		if (bufsz)
523 			*bufsz = sizeof(shminfo);
524 		td->td_retval[0] = shmalloced;
525 		return (0);
526 	case SHM_INFO: {
527 		shm_info.used_ids = shm_nused;
528 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
529 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
530 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
531 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
532 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
533 		memcpy(buf, &shm_info, sizeof(shm_info));
534 		if (bufsz != NULL)
535 			*bufsz = sizeof(shm_info);
536 		td->td_retval[0] = shmalloced;
537 		return (0);
538 	}
539 	}
540 	shmseg = shm_find_segment(rpr, shmid, cmd != SHM_STAT);
541 	if (shmseg == NULL)
542 		return (EINVAL);
543 #ifdef MAC
544 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
545 	if (error != 0)
546 		return (error);
547 #endif
548 	switch (cmd) {
549 	case SHM_STAT:
550 	case IPC_STAT:
551 		shmidp = (struct shmid_ds *)buf;
552 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
553 		if (error != 0)
554 			return (error);
555 		memcpy(shmidp, &shmseg->u, sizeof(struct shmid_ds));
556 		if (td->td_ucred->cr_prison != shmseg->cred->cr_prison)
557 			shmidp->shm_perm.key = IPC_PRIVATE;
558 		if (bufsz != NULL)
559 			*bufsz = sizeof(struct shmid_ds);
560 		if (cmd == SHM_STAT) {
561 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
562 			    shmseg->u.shm_perm);
563 		}
564 		break;
565 	case IPC_SET:
566 		shmidp = (struct shmid_ds *)buf;
567 		AUDIT_ARG_SVIPC_PERM(&shmidp->shm_perm);
568 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
569 		if (error != 0)
570 			return (error);
571 		shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
572 		shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
573 		shmseg->u.shm_perm.mode =
574 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
575 		    (shmidp->shm_perm.mode & ACCESSPERMS);
576 		shmseg->u.shm_ctime = time_second;
577 		break;
578 	case IPC_RMID:
579 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
580 		if (error != 0)
581 			return (error);
582 		shm_remove(shmseg, IPCID_TO_IX(shmid));
583 		break;
584 #if 0
585 	case SHM_LOCK:
586 	case SHM_UNLOCK:
587 #endif
588 	default:
589 		error = EINVAL;
590 		break;
591 	}
592 	return (error);
593 }
594 
595 int
596 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
597 {
598 	int error;
599 
600 	SYSVSHM_LOCK();
601 	error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
602 	SYSVSHM_UNLOCK();
603 	return (error);
604 }
605 
606 
607 #ifndef _SYS_SYSPROTO_H_
608 struct shmctl_args {
609 	int shmid;
610 	int cmd;
611 	struct shmid_ds *buf;
612 };
613 #endif
614 int
615 sys_shmctl(struct thread *td, struct shmctl_args *uap)
616 {
617 	int error;
618 	struct shmid_ds buf;
619 	size_t bufsz;
620 
621 	/*
622 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
623 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
624 	 * return an error back to the user since we do not to support this.
625 	 */
626 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
627 	    uap->cmd == SHM_STAT)
628 		return (EINVAL);
629 
630 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
631 	if (uap->cmd == IPC_SET) {
632 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
633 			goto done;
634 	}
635 
636 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
637 	if (error)
638 		goto done;
639 
640 	/* Cases in which we need to copyout */
641 	switch (uap->cmd) {
642 	case IPC_STAT:
643 		error = copyout(&buf, uap->buf, bufsz);
644 		break;
645 	}
646 
647 done:
648 	if (error) {
649 		/* Invalidate the return value */
650 		td->td_retval[0] = -1;
651 	}
652 	return (error);
653 }
654 
655 
656 static int
657 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
658     int segnum)
659 {
660 	struct shmid_kernel *shmseg;
661 #ifdef MAC
662 	int error;
663 #endif
664 
665 	SYSVSHM_ASSERT_LOCKED();
666 	KASSERT(segnum >= 0 && segnum < shmalloced,
667 	    ("segnum %d shmalloced %d", segnum, shmalloced));
668 	shmseg = &shmsegs[segnum];
669 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
670 		return (EEXIST);
671 #ifdef MAC
672 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
673 	if (error != 0)
674 		return (error);
675 #endif
676 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
677 		return (EINVAL);
678 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
679 	return (0);
680 }
681 
682 static int
683 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
684 {
685 	struct ucred *cred = td->td_ucred;
686 	struct shmid_kernel *shmseg;
687 	vm_object_t shm_object;
688 	int i, segnum;
689 	size_t size;
690 
691 	SYSVSHM_ASSERT_LOCKED();
692 
693 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
694 		return (EINVAL);
695 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
696 		return (ENOSPC);
697 	size = round_page(uap->size);
698 	if (shm_committed + btoc(size) > shminfo.shmall)
699 		return (ENOMEM);
700 	if (shm_last_free < 0) {
701 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
702 		for (i = 0; i < shmalloced; i++)
703 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
704 				break;
705 		if (i == shmalloced)
706 			return (ENOSPC);
707 		segnum = i;
708 	} else  {
709 		segnum = shm_last_free;
710 		shm_last_free = -1;
711 	}
712 	KASSERT(segnum >= 0 && segnum < shmalloced,
713 	    ("segnum %d shmalloced %d", segnum, shmalloced));
714 	shmseg = &shmsegs[segnum];
715 #ifdef RACCT
716 	if (racct_enable) {
717 		PROC_LOCK(td->td_proc);
718 		if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
719 			PROC_UNLOCK(td->td_proc);
720 			return (ENOSPC);
721 		}
722 		if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
723 			racct_sub(td->td_proc, RACCT_NSHM, 1);
724 			PROC_UNLOCK(td->td_proc);
725 			return (ENOMEM);
726 		}
727 		PROC_UNLOCK(td->td_proc);
728 	}
729 #endif
730 
731 	/*
732 	 * We make sure that we have allocated a pager before we need
733 	 * to.
734 	 */
735 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
736 	    0, size, VM_PROT_DEFAULT, 0, cred);
737 	if (shm_object == NULL) {
738 #ifdef RACCT
739 		if (racct_enable) {
740 			PROC_LOCK(td->td_proc);
741 			racct_sub(td->td_proc, RACCT_NSHM, 1);
742 			racct_sub(td->td_proc, RACCT_SHMSIZE, size);
743 			PROC_UNLOCK(td->td_proc);
744 		}
745 #endif
746 		return (ENOMEM);
747 	}
748 	shm_object->pg_color = 0;
749 	VM_OBJECT_WLOCK(shm_object);
750 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
751 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
752 	VM_OBJECT_WUNLOCK(shm_object);
753 
754 	shmseg->object = shm_object;
755 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
756 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
757 	shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
758 	shmseg->u.shm_perm.key = uap->key;
759 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
760 	shmseg->cred = crhold(cred);
761 	shmseg->u.shm_segsz = uap->size;
762 	shmseg->u.shm_cpid = td->td_proc->p_pid;
763 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
764 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
765 #ifdef MAC
766 	mac_sysvshm_create(cred, shmseg);
767 #endif
768 	shmseg->u.shm_ctime = time_second;
769 	shm_committed += btoc(size);
770 	shm_nused++;
771 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
772 
773 	return (0);
774 }
775 
776 #ifndef _SYS_SYSPROTO_H_
777 struct shmget_args {
778 	key_t key;
779 	size_t size;
780 	int shmflg;
781 };
782 #endif
783 int
784 sys_shmget(struct thread *td, struct shmget_args *uap)
785 {
786 	int segnum, mode;
787 	int error;
788 
789 	if (shm_find_prison(td->td_ucred) == NULL)
790 		return (ENOSYS);
791 	mode = uap->shmflg & ACCESSPERMS;
792 	SYSVSHM_LOCK();
793 	if (uap->key == IPC_PRIVATE) {
794 		error = shmget_allocate_segment(td, uap, mode);
795 	} else {
796 		segnum = shm_find_segment_by_key(td->td_ucred->cr_prison,
797 		    uap->key);
798 		if (segnum >= 0)
799 			error = shmget_existing(td, uap, mode, segnum);
800 		else if ((uap->shmflg & IPC_CREAT) == 0)
801 			error = ENOENT;
802 		else
803 			error = shmget_allocate_segment(td, uap, mode);
804 	}
805 	SYSVSHM_UNLOCK();
806 	return (error);
807 }
808 
809 static void
810 shmfork_myhook(struct proc *p1, struct proc *p2)
811 {
812 	struct shmmap_state *shmmap_s;
813 	size_t size;
814 	int i;
815 
816 	SYSVSHM_LOCK();
817 	size = shminfo.shmseg * sizeof(struct shmmap_state);
818 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
819 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
820 	p2->p_vmspace->vm_shm = shmmap_s;
821 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
822 		if (shmmap_s->shmid != -1) {
823 			KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
824 			    IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
825 			    ("segnum %d shmalloced %d",
826 			    IPCID_TO_IX(shmmap_s->shmid), shmalloced));
827 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
828 		}
829 	}
830 	SYSVSHM_UNLOCK();
831 }
832 
833 static void
834 shmexit_myhook(struct vmspace *vm)
835 {
836 	struct shmmap_state *base, *shm;
837 	int i;
838 
839 	base = vm->vm_shm;
840 	if (base != NULL) {
841 		vm->vm_shm = NULL;
842 		SYSVSHM_LOCK();
843 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
844 			if (shm->shmid != -1)
845 				shm_delete_mapping(vm, shm);
846 		}
847 		SYSVSHM_UNLOCK();
848 		free(base, M_SHM);
849 	}
850 }
851 
852 static void
853 shmrealloc(void)
854 {
855 	struct shmid_kernel *newsegs;
856 	int i;
857 
858 	SYSVSHM_ASSERT_LOCKED();
859 
860 	if (shmalloced >= shminfo.shmmni)
861 		return;
862 
863 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM,
864 	    M_WAITOK | M_ZERO);
865 	for (i = 0; i < shmalloced; i++)
866 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
867 	for (; i < shminfo.shmmni; i++) {
868 		newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
869 		newsegs[i].u.shm_perm.seq = 0;
870 #ifdef MAC
871 		mac_sysvshm_init(&newsegs[i]);
872 #endif
873 	}
874 	free(shmsegs, M_SHM);
875 	shmsegs = newsegs;
876 	shmalloced = shminfo.shmmni;
877 }
878 
879 static struct syscall_helper_data shm_syscalls[] = {
880 	SYSCALL_INIT_HELPER(shmat),
881 	SYSCALL_INIT_HELPER(shmctl),
882 	SYSCALL_INIT_HELPER(shmdt),
883 	SYSCALL_INIT_HELPER(shmget),
884 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
885     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
886 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
887 #endif
888 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
889 	SYSCALL_INIT_HELPER(shmsys),
890 #endif
891 	SYSCALL_INIT_LAST
892 };
893 
894 #ifdef COMPAT_FREEBSD32
895 #include <compat/freebsd32/freebsd32.h>
896 #include <compat/freebsd32/freebsd32_ipc.h>
897 #include <compat/freebsd32/freebsd32_proto.h>
898 #include <compat/freebsd32/freebsd32_signal.h>
899 #include <compat/freebsd32/freebsd32_syscall.h>
900 #include <compat/freebsd32/freebsd32_util.h>
901 
902 static struct syscall_helper_data shm32_syscalls[] = {
903 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
904 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
905 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
906 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
907 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
908 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
909     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
910 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
911 #endif
912 	SYSCALL_INIT_LAST
913 };
914 #endif
915 
916 static int
917 shminit(void)
918 {
919 	struct prison *pr;
920 	void **rsv;
921 	int i, error;
922 	osd_method_t methods[PR_MAXMETHOD] = {
923 	    [PR_METHOD_CHECK] =		shm_prison_check,
924 	    [PR_METHOD_SET] =		shm_prison_set,
925 	    [PR_METHOD_GET] =		shm_prison_get,
926 	    [PR_METHOD_REMOVE] =	shm_prison_remove,
927 	};
928 
929 #ifndef BURN_BRIDGES
930 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
931 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
932 #endif
933 	if (shminfo.shmmax == SHMMAX) {
934 		/* Initialize shmmax dealing with possible overflow. */
935 		for (i = PAGE_SIZE; i != 0; i--) {
936 			shminfo.shmmax = shminfo.shmall * i;
937 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
938 				break;
939 		}
940 	}
941 	shmalloced = shminfo.shmmni;
942 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM,
943 	    M_WAITOK|M_ZERO);
944 	for (i = 0; i < shmalloced; i++) {
945 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
946 		shmsegs[i].u.shm_perm.seq = 0;
947 #ifdef MAC
948 		mac_sysvshm_init(&shmsegs[i]);
949 #endif
950 	}
951 	shm_last_free = 0;
952 	shm_nused = 0;
953 	shm_committed = 0;
954 	sx_init(&sysvshmsx, "sysvshmsx");
955 	shmexit_hook = &shmexit_myhook;
956 	shmfork_hook = &shmfork_myhook;
957 
958 	/* Set current prisons according to their allow.sysvipc. */
959 	shm_prison_slot = osd_jail_register(NULL, methods);
960 	rsv = osd_reserve(shm_prison_slot);
961 	prison_lock(&prison0);
962 	(void)osd_jail_set_reserved(&prison0, shm_prison_slot, rsv, &prison0);
963 	prison_unlock(&prison0);
964 	rsv = NULL;
965 	sx_slock(&allprison_lock);
966 	TAILQ_FOREACH(pr, &allprison, pr_list) {
967 		if (rsv == NULL)
968 			rsv = osd_reserve(shm_prison_slot);
969 		prison_lock(pr);
970 		if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) {
971 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
972 			    &prison0);
973 			rsv = NULL;
974 		}
975 		prison_unlock(pr);
976 	}
977 	if (rsv != NULL)
978 		osd_free_reserved(rsv);
979 	sx_sunlock(&allprison_lock);
980 
981 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
982 	if (error != 0)
983 		return (error);
984 #ifdef COMPAT_FREEBSD32
985 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
986 	if (error != 0)
987 		return (error);
988 #endif
989 	return (0);
990 }
991 
992 static int
993 shmunload(void)
994 {
995 	int i;
996 
997 	if (shm_nused > 0)
998 		return (EBUSY);
999 
1000 #ifdef COMPAT_FREEBSD32
1001 	syscall32_helper_unregister(shm32_syscalls);
1002 #endif
1003 	syscall_helper_unregister(shm_syscalls);
1004 	if (shm_prison_slot != 0)
1005 		osd_jail_deregister(shm_prison_slot);
1006 
1007 	for (i = 0; i < shmalloced; i++) {
1008 #ifdef MAC
1009 		mac_sysvshm_destroy(&shmsegs[i]);
1010 #endif
1011 		/*
1012 		 * Objects might be still mapped into the processes
1013 		 * address spaces.  Actual free would happen on the
1014 		 * last mapping destruction.
1015 		 */
1016 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
1017 			vm_object_deallocate(shmsegs[i].object);
1018 	}
1019 	free(shmsegs, M_SHM);
1020 	shmexit_hook = NULL;
1021 	shmfork_hook = NULL;
1022 	sx_destroy(&sysvshmsx);
1023 	return (0);
1024 }
1025 
1026 static int
1027 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
1028 {
1029 	struct shmid_kernel tshmseg;
1030 #ifdef COMPAT_FREEBSD32
1031 	struct shmid_kernel32 tshmseg32;
1032 #endif
1033 	struct prison *pr, *rpr;
1034 	void *outaddr;
1035 	size_t outsize;
1036 	int error, i;
1037 
1038 	SYSVSHM_LOCK();
1039 	pr = req->td->td_ucred->cr_prison;
1040 	rpr = shm_find_prison(req->td->td_ucred);
1041 	error = 0;
1042 	for (i = 0; i < shmalloced; i++) {
1043 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
1044 		    rpr == NULL || shm_prison_cansee(rpr, &shmsegs[i]) != 0) {
1045 			bzero(&tshmseg, sizeof(tshmseg));
1046 			tshmseg.u.shm_perm.mode = SHMSEG_FREE;
1047 		} else {
1048 			tshmseg = shmsegs[i];
1049 			if (tshmseg.cred->cr_prison != pr)
1050 				tshmseg.u.shm_perm.key = IPC_PRIVATE;
1051 		}
1052 #ifdef COMPAT_FREEBSD32
1053 		if (SV_CURPROC_FLAG(SV_ILP32)) {
1054 			bzero(&tshmseg32, sizeof(tshmseg32));
1055 			freebsd32_ipcperm_out(&tshmseg.u.shm_perm,
1056 			    &tshmseg32.u.shm_perm);
1057 			CP(tshmseg, tshmseg32, u.shm_segsz);
1058 			CP(tshmseg, tshmseg32, u.shm_lpid);
1059 			CP(tshmseg, tshmseg32, u.shm_cpid);
1060 			CP(tshmseg, tshmseg32, u.shm_nattch);
1061 			CP(tshmseg, tshmseg32, u.shm_atime);
1062 			CP(tshmseg, tshmseg32, u.shm_dtime);
1063 			CP(tshmseg, tshmseg32, u.shm_ctime);
1064 			/* Don't copy object, label, or cred */
1065 			outaddr = &tshmseg32;
1066 			outsize = sizeof(tshmseg32);
1067 		} else
1068 #endif
1069 		{
1070 			tshmseg.object = NULL;
1071 			tshmseg.label = NULL;
1072 			tshmseg.cred = NULL;
1073 			outaddr = &tshmseg;
1074 			outsize = sizeof(tshmseg);
1075 		}
1076 		error = SYSCTL_OUT(req, outaddr, outsize);
1077 		if (error != 0)
1078 			break;
1079 	}
1080 	SYSVSHM_UNLOCK();
1081 	return (error);
1082 }
1083 
1084 static int
1085 shm_prison_check(void *obj, void *data)
1086 {
1087 	struct prison *pr = obj;
1088 	struct prison *prpr;
1089 	struct vfsoptlist *opts = data;
1090 	int error, jsys;
1091 
1092 	/*
1093 	 * sysvshm is a jailsys integer.
1094 	 * It must be "disable" if the parent jail is disabled.
1095 	 */
1096 	error = vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys));
1097 	if (error != ENOENT) {
1098 		if (error != 0)
1099 			return (error);
1100 		switch (jsys) {
1101 		case JAIL_SYS_DISABLE:
1102 			break;
1103 		case JAIL_SYS_NEW:
1104 		case JAIL_SYS_INHERIT:
1105 			prison_lock(pr->pr_parent);
1106 			prpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1107 			prison_unlock(pr->pr_parent);
1108 			if (prpr == NULL)
1109 				return (EPERM);
1110 			break;
1111 		default:
1112 			return (EINVAL);
1113 		}
1114 	}
1115 
1116 	return (0);
1117 }
1118 
1119 static int
1120 shm_prison_set(void *obj, void *data)
1121 {
1122 	struct prison *pr = obj;
1123 	struct prison *tpr, *orpr, *nrpr, *trpr;
1124 	struct vfsoptlist *opts = data;
1125 	void *rsv;
1126 	int jsys, descend;
1127 
1128 	/*
1129 	 * sysvshm controls which jail is the root of the associated segments
1130 	 * (this jail or same as the parent), or if the feature is available
1131 	 * at all.
1132 	 */
1133 	if (vfs_copyopt(opts, "sysvshm", &jsys, sizeof(jsys)) == ENOENT)
1134 		jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0)
1135 		    ? JAIL_SYS_INHERIT
1136 		    : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0)
1137 		    ? JAIL_SYS_DISABLE
1138 		    : -1;
1139 	if (jsys == JAIL_SYS_DISABLE) {
1140 		prison_lock(pr);
1141 		orpr = osd_jail_get(pr, shm_prison_slot);
1142 		if (orpr != NULL)
1143 			osd_jail_del(pr, shm_prison_slot);
1144 		prison_unlock(pr);
1145 		if (orpr != NULL) {
1146 			if (orpr == pr)
1147 				shm_prison_cleanup(pr);
1148 			/* Disable all child jails as well. */
1149 			FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1150 				prison_lock(tpr);
1151 				trpr = osd_jail_get(tpr, shm_prison_slot);
1152 				if (trpr != NULL) {
1153 					osd_jail_del(tpr, shm_prison_slot);
1154 					prison_unlock(tpr);
1155 					if (trpr == tpr)
1156 						shm_prison_cleanup(tpr);
1157 				} else {
1158 					prison_unlock(tpr);
1159 					descend = 0;
1160 				}
1161 			}
1162 		}
1163 	} else if (jsys != -1) {
1164 		if (jsys == JAIL_SYS_NEW)
1165 			nrpr = pr;
1166 		else {
1167 			prison_lock(pr->pr_parent);
1168 			nrpr = osd_jail_get(pr->pr_parent, shm_prison_slot);
1169 			prison_unlock(pr->pr_parent);
1170 		}
1171 		rsv = osd_reserve(shm_prison_slot);
1172 		prison_lock(pr);
1173 		orpr = osd_jail_get(pr, shm_prison_slot);
1174 		if (orpr != nrpr)
1175 			(void)osd_jail_set_reserved(pr, shm_prison_slot, rsv,
1176 			    nrpr);
1177 		else
1178 			osd_free_reserved(rsv);
1179 		prison_unlock(pr);
1180 		if (orpr != nrpr) {
1181 			if (orpr == pr)
1182 				shm_prison_cleanup(pr);
1183 			if (orpr != NULL) {
1184 				/* Change child jails matching the old root, */
1185 				FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1186 					prison_lock(tpr);
1187 					trpr = osd_jail_get(tpr,
1188 					    shm_prison_slot);
1189 					if (trpr == orpr) {
1190 						(void)osd_jail_set(tpr,
1191 						    shm_prison_slot, nrpr);
1192 						prison_unlock(tpr);
1193 						if (trpr == tpr)
1194 							shm_prison_cleanup(tpr);
1195 					} else {
1196 						prison_unlock(tpr);
1197 						descend = 0;
1198 					}
1199 				}
1200 			}
1201 		}
1202 	}
1203 
1204 	return (0);
1205 }
1206 
1207 static int
1208 shm_prison_get(void *obj, void *data)
1209 {
1210 	struct prison *pr = obj;
1211 	struct prison *rpr;
1212 	struct vfsoptlist *opts = data;
1213 	int error, jsys;
1214 
1215 	/* Set sysvshm based on the jail's root prison. */
1216 	prison_lock(pr);
1217 	rpr = osd_jail_get(pr, shm_prison_slot);
1218 	prison_unlock(pr);
1219 	jsys = rpr == NULL ? JAIL_SYS_DISABLE
1220 	    : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT;
1221 	error = vfs_setopt(opts, "sysvshm", &jsys, sizeof(jsys));
1222 	if (error == ENOENT)
1223 		error = 0;
1224 	return (error);
1225 }
1226 
1227 static int
1228 shm_prison_remove(void *obj, void *data __unused)
1229 {
1230 	struct prison *pr = obj;
1231 	struct prison *rpr;
1232 
1233 	SYSVSHM_LOCK();
1234 	prison_lock(pr);
1235 	rpr = osd_jail_get(pr, shm_prison_slot);
1236 	prison_unlock(pr);
1237 	if (rpr == pr)
1238 		shm_prison_cleanup(pr);
1239 	SYSVSHM_UNLOCK();
1240 	return (0);
1241 }
1242 
1243 static void
1244 shm_prison_cleanup(struct prison *pr)
1245 {
1246 	struct shmid_kernel *shmseg;
1247 	int i;
1248 
1249 	/* Remove any segments that belong to this jail. */
1250 	for (i = 0; i < shmalloced; i++) {
1251 		shmseg = &shmsegs[i];
1252 		if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) &&
1253 		    shmseg->cred != NULL && shmseg->cred->cr_prison == pr) {
1254 			shm_remove(shmseg, i);
1255 		}
1256 	}
1257 }
1258 
1259 SYSCTL_JAIL_PARAM_SYS_NODE(sysvshm, CTLFLAG_RW, "SYSV shared memory");
1260 
1261 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
1262 struct oshmid_ds {
1263 	struct	ipc_perm_old shm_perm;	/* operation perms */
1264 	int	shm_segsz;		/* size of segment (bytes) */
1265 	u_short	shm_cpid;		/* pid, creator */
1266 	u_short	shm_lpid;		/* pid, last operation */
1267 	short	shm_nattch;		/* no. of current attaches */
1268 	time_t	shm_atime;		/* last attach time */
1269 	time_t	shm_dtime;		/* last detach time */
1270 	time_t	shm_ctime;		/* last change time */
1271 	void	*shm_handle;		/* internal handle for shm segment */
1272 };
1273 
1274 struct oshmctl_args {
1275 	int shmid;
1276 	int cmd;
1277 	struct oshmid_ds *ubuf;
1278 };
1279 
1280 static int
1281 oshmctl(struct thread *td, struct oshmctl_args *uap)
1282 {
1283 #ifdef COMPAT_43
1284 	int error = 0;
1285 	struct prison *rpr;
1286 	struct shmid_kernel *shmseg;
1287 	struct oshmid_ds outbuf;
1288 
1289 	rpr = shm_find_prison(td->td_ucred);
1290 	if (rpr == NULL)
1291 		return (ENOSYS);
1292 	if (uap->cmd != IPC_STAT) {
1293 		return (freebsd7_shmctl(td,
1294 		    (struct freebsd7_shmctl_args *)uap));
1295 	}
1296 	SYSVSHM_LOCK();
1297 	shmseg = shm_find_segment(rpr, uap->shmid, true);
1298 	if (shmseg == NULL) {
1299 		SYSVSHM_UNLOCK();
1300 		return (EINVAL);
1301 	}
1302 	error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1303 	if (error != 0) {
1304 		SYSVSHM_UNLOCK();
1305 		return (error);
1306 	}
1307 #ifdef MAC
1308 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1309 	if (error != 0) {
1310 		SYSVSHM_UNLOCK();
1311 		return (error);
1312 	}
1313 #endif
1314 	ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1315 	outbuf.shm_segsz = shmseg->u.shm_segsz;
1316 	outbuf.shm_cpid = shmseg->u.shm_cpid;
1317 	outbuf.shm_lpid = shmseg->u.shm_lpid;
1318 	outbuf.shm_nattch = shmseg->u.shm_nattch;
1319 	outbuf.shm_atime = shmseg->u.shm_atime;
1320 	outbuf.shm_dtime = shmseg->u.shm_dtime;
1321 	outbuf.shm_ctime = shmseg->u.shm_ctime;
1322 	outbuf.shm_handle = shmseg->object;
1323 	SYSVSHM_UNLOCK();
1324 	return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
1325 #else
1326 	return (EINVAL);
1327 #endif
1328 }
1329 
1330 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1331 static sy_call_t *shmcalls[] = {
1332 	(sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1333 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1334 	(sy_call_t *)freebsd7_shmctl
1335 };
1336 
1337 #ifndef _SYS_SYSPROTO_H_
1338 /* XXX actually varargs. */
1339 struct shmsys_args {
1340 	int	which;
1341 	int	a2;
1342 	int	a3;
1343 	int	a4;
1344 };
1345 #endif
1346 int
1347 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1348 {
1349 
1350 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1351 	if (uap->which < 0 || uap->which >= nitems(shmcalls))
1352 		return (EINVAL);
1353 	return ((*shmcalls[uap->which])(td, &uap->a2));
1354 }
1355 
1356 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1357 
1358 #ifdef COMPAT_FREEBSD32
1359 
1360 int
1361 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1362 {
1363 
1364 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1365     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1366 	AUDIT_ARG_SVIPC_WHICH(uap->which);
1367 	switch (uap->which) {
1368 	case 0:	{	/* shmat */
1369 		struct shmat_args ap;
1370 
1371 		ap.shmid = uap->a2;
1372 		ap.shmaddr = PTRIN(uap->a3);
1373 		ap.shmflg = uap->a4;
1374 		return (sysent[SYS_shmat].sy_call(td, &ap));
1375 	}
1376 	case 2: {	/* shmdt */
1377 		struct shmdt_args ap;
1378 
1379 		ap.shmaddr = PTRIN(uap->a2);
1380 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1381 	}
1382 	case 3: {	/* shmget */
1383 		struct shmget_args ap;
1384 
1385 		ap.key = uap->a2;
1386 		ap.size = uap->a3;
1387 		ap.shmflg = uap->a4;
1388 		return (sysent[SYS_shmget].sy_call(td, &ap));
1389 	}
1390 	case 4: {	/* shmctl */
1391 		struct freebsd7_freebsd32_shmctl_args ap;
1392 
1393 		ap.shmid = uap->a2;
1394 		ap.cmd = uap->a3;
1395 		ap.buf = PTRIN(uap->a4);
1396 		return (freebsd7_freebsd32_shmctl(td, &ap));
1397 	}
1398 	case 1:		/* oshmctl */
1399 	default:
1400 		return (EINVAL);
1401 	}
1402 #else
1403 	return (nosys(td, NULL));
1404 #endif
1405 }
1406 
1407 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1408     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1409 int
1410 freebsd7_freebsd32_shmctl(struct thread *td,
1411     struct freebsd7_freebsd32_shmctl_args *uap)
1412 {
1413 	int error;
1414 	union {
1415 		struct shmid_ds shmid_ds;
1416 		struct shm_info shm_info;
1417 		struct shminfo shminfo;
1418 	} u;
1419 	union {
1420 		struct shmid_ds32_old shmid_ds32;
1421 		struct shm_info32 shm_info32;
1422 		struct shminfo32 shminfo32;
1423 	} u32;
1424 	size_t sz;
1425 
1426 	if (uap->cmd == IPC_SET) {
1427 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1428 		    sizeof(u32.shmid_ds32))))
1429 			goto done;
1430 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1431 		    &u.shmid_ds.shm_perm);
1432 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1433 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1434 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1435 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1436 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1437 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1438 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1439 	}
1440 
1441 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1442 	if (error)
1443 		goto done;
1444 
1445 	/* Cases in which we need to copyout */
1446 	switch (uap->cmd) {
1447 	case IPC_INFO:
1448 		CP(u.shminfo, u32.shminfo32, shmmax);
1449 		CP(u.shminfo, u32.shminfo32, shmmin);
1450 		CP(u.shminfo, u32.shminfo32, shmmni);
1451 		CP(u.shminfo, u32.shminfo32, shmseg);
1452 		CP(u.shminfo, u32.shminfo32, shmall);
1453 		error = copyout(&u32.shminfo32, uap->buf,
1454 		    sizeof(u32.shminfo32));
1455 		break;
1456 	case SHM_INFO:
1457 		CP(u.shm_info, u32.shm_info32, used_ids);
1458 		CP(u.shm_info, u32.shm_info32, shm_rss);
1459 		CP(u.shm_info, u32.shm_info32, shm_tot);
1460 		CP(u.shm_info, u32.shm_info32, shm_swp);
1461 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1462 		CP(u.shm_info, u32.shm_info32, swap_successes);
1463 		error = copyout(&u32.shm_info32, uap->buf,
1464 		    sizeof(u32.shm_info32));
1465 		break;
1466 	case SHM_STAT:
1467 	case IPC_STAT:
1468 		memset(&u32.shmid_ds32, 0, sizeof(u32.shmid_ds32));
1469 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1470 		    &u32.shmid_ds32.shm_perm);
1471 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1472 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1473 		else
1474 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1475 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1476 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1477 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1478 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1479 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1480 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1481 		u32.shmid_ds32.shm_internal = 0;
1482 		error = copyout(&u32.shmid_ds32, uap->buf,
1483 		    sizeof(u32.shmid_ds32));
1484 		break;
1485 	}
1486 
1487 done:
1488 	if (error) {
1489 		/* Invalidate the return value */
1490 		td->td_retval[0] = -1;
1491 	}
1492 	return (error);
1493 }
1494 #endif
1495 
1496 int
1497 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1498 {
1499 	int error;
1500 	union {
1501 		struct shmid_ds shmid_ds;
1502 		struct shm_info shm_info;
1503 		struct shminfo shminfo;
1504 	} u;
1505 	union {
1506 		struct shmid_ds32 shmid_ds32;
1507 		struct shm_info32 shm_info32;
1508 		struct shminfo32 shminfo32;
1509 	} u32;
1510 	size_t sz;
1511 
1512 	if (uap->cmd == IPC_SET) {
1513 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1514 		    sizeof(u32.shmid_ds32))))
1515 			goto done;
1516 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1517 		    &u.shmid_ds.shm_perm);
1518 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1519 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1520 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1521 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1522 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1523 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1524 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1525 	}
1526 
1527 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1528 	if (error)
1529 		goto done;
1530 
1531 	/* Cases in which we need to copyout */
1532 	switch (uap->cmd) {
1533 	case IPC_INFO:
1534 		CP(u.shminfo, u32.shminfo32, shmmax);
1535 		CP(u.shminfo, u32.shminfo32, shmmin);
1536 		CP(u.shminfo, u32.shminfo32, shmmni);
1537 		CP(u.shminfo, u32.shminfo32, shmseg);
1538 		CP(u.shminfo, u32.shminfo32, shmall);
1539 		error = copyout(&u32.shminfo32, uap->buf,
1540 		    sizeof(u32.shminfo32));
1541 		break;
1542 	case SHM_INFO:
1543 		CP(u.shm_info, u32.shm_info32, used_ids);
1544 		CP(u.shm_info, u32.shm_info32, shm_rss);
1545 		CP(u.shm_info, u32.shm_info32, shm_tot);
1546 		CP(u.shm_info, u32.shm_info32, shm_swp);
1547 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1548 		CP(u.shm_info, u32.shm_info32, swap_successes);
1549 		error = copyout(&u32.shm_info32, uap->buf,
1550 		    sizeof(u32.shm_info32));
1551 		break;
1552 	case SHM_STAT:
1553 	case IPC_STAT:
1554 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1555 		    &u32.shmid_ds32.shm_perm);
1556 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1557 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1558 		else
1559 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1560 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1561 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1562 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1563 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1564 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1565 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1566 		error = copyout(&u32.shmid_ds32, uap->buf,
1567 		    sizeof(u32.shmid_ds32));
1568 		break;
1569 	}
1570 
1571 done:
1572 	if (error) {
1573 		/* Invalidate the return value */
1574 		td->td_retval[0] = -1;
1575 	}
1576 	return (error);
1577 }
1578 #endif
1579 
1580 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1581     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1582 
1583 #ifndef CP
1584 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1585 #endif
1586 
1587 #ifndef _SYS_SYSPROTO_H_
1588 struct freebsd7_shmctl_args {
1589 	int shmid;
1590 	int cmd;
1591 	struct shmid_ds_old *buf;
1592 };
1593 #endif
1594 int
1595 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1596 {
1597 	int error;
1598 	struct shmid_ds_old old;
1599 	struct shmid_ds buf;
1600 	size_t bufsz;
1601 
1602 	/*
1603 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1604 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1605 	 * return an error back to the user since we do not to support this.
1606 	 */
1607 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1608 	    uap->cmd == SHM_STAT)
1609 		return (EINVAL);
1610 
1611 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1612 	if (uap->cmd == IPC_SET) {
1613 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1614 			goto done;
1615 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1616 		CP(old, buf, shm_segsz);
1617 		CP(old, buf, shm_lpid);
1618 		CP(old, buf, shm_cpid);
1619 		CP(old, buf, shm_nattch);
1620 		CP(old, buf, shm_atime);
1621 		CP(old, buf, shm_dtime);
1622 		CP(old, buf, shm_ctime);
1623 	}
1624 
1625 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1626 	if (error)
1627 		goto done;
1628 
1629 	/* Cases in which we need to copyout */
1630 	switch (uap->cmd) {
1631 	case IPC_STAT:
1632 		memset(&old, 0, sizeof(old));
1633 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1634 		if (buf.shm_segsz > INT_MAX)
1635 			old.shm_segsz = INT_MAX;
1636 		else
1637 			CP(buf, old, shm_segsz);
1638 		CP(buf, old, shm_lpid);
1639 		CP(buf, old, shm_cpid);
1640 		if (buf.shm_nattch > SHRT_MAX)
1641 			old.shm_nattch = SHRT_MAX;
1642 		else
1643 			CP(buf, old, shm_nattch);
1644 		CP(buf, old, shm_atime);
1645 		CP(buf, old, shm_dtime);
1646 		CP(buf, old, shm_ctime);
1647 		old.shm_internal = NULL;
1648 		error = copyout(&old, uap->buf, sizeof(old));
1649 		break;
1650 	}
1651 
1652 done:
1653 	if (error) {
1654 		/* Invalidate the return value */
1655 		td->td_retval[0] = -1;
1656 	}
1657 	return (error);
1658 }
1659 
1660 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1661 	   COMPAT_FREEBSD7 */
1662 
1663 static int
1664 sysvshm_modload(struct module *module, int cmd, void *arg)
1665 {
1666 	int error = 0;
1667 
1668 	switch (cmd) {
1669 	case MOD_LOAD:
1670 		error = shminit();
1671 		if (error != 0)
1672 			shmunload();
1673 		break;
1674 	case MOD_UNLOAD:
1675 		error = shmunload();
1676 		break;
1677 	case MOD_SHUTDOWN:
1678 		break;
1679 	default:
1680 		error = EINVAL;
1681 		break;
1682 	}
1683 	return (error);
1684 }
1685 
1686 static moduledata_t sysvshm_mod = {
1687 	"sysvshm",
1688 	&sysvshm_modload,
1689 	NULL
1690 };
1691 
1692 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1693 MODULE_VERSION(sysvshm, 1);
1694