xref: /freebsd/sys/kern/sysv_shm.c (revision 6ae1554a5d9b318f8ad53ccc39fa5a961403da73)
1 /*	$NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $	*/
2 /*-
3  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Adam Glass and Charles
16  *	Hannum.
17  * 4. The names of the authors may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*-
32  * Copyright (c) 2003-2005 McAfee, Inc.
33  * All rights reserved.
34  *
35  * This software was developed for the FreeBSD Project in part by McAfee
36  * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR
37  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research
38  * program.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  */
61 
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64 
65 #include "opt_compat.h"
66 #include "opt_sysvipc.h"
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
71 #include <sys/limits.h>
72 #include <sys/lock.h>
73 #include <sys/sysctl.h>
74 #include <sys/shm.h>
75 #include <sys/proc.h>
76 #include <sys/malloc.h>
77 #include <sys/mman.h>
78 #include <sys/module.h>
79 #include <sys/mutex.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
83 #include <sys/stat.h>
84 #include <sys/syscall.h>
85 #include <sys/syscallsubr.h>
86 #include <sys/sysent.h>
87 #include <sys/sysproto.h>
88 #include <sys/jail.h>
89 
90 #include <security/mac/mac_framework.h>
91 
92 #include <vm/vm.h>
93 #include <vm/vm_param.h>
94 #include <vm/pmap.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_map.h>
97 #include <vm/vm_page.h>
98 #include <vm/vm_pager.h>
99 
100 FEATURE(sysv_shm, "System V shared memory segments support");
101 
102 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
103 
104 static int shmget_allocate_segment(struct thread *td,
105     struct shmget_args *uap, int mode);
106 static int shmget_existing(struct thread *td, struct shmget_args *uap,
107     int mode, int segnum);
108 
109 #define	SHMSEG_FREE     	0x0200
110 #define	SHMSEG_REMOVED  	0x0400
111 #define	SHMSEG_ALLOCATED	0x0800
112 
113 static int shm_last_free, shm_nused, shmalloced;
114 vm_size_t shm_committed;
115 static struct shmid_kernel	*shmsegs;
116 
117 struct shmmap_state {
118 	vm_offset_t va;
119 	int shmid;
120 };
121 
122 static void shm_deallocate_segment(struct shmid_kernel *);
123 static int shm_find_segment_by_key(key_t);
124 static struct shmid_kernel *shm_find_segment(int, bool);
125 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
126 static void shmrealloc(void);
127 static int shminit(void);
128 static int sysvshm_modload(struct module *, int, void *);
129 static int shmunload(void);
130 static void shmexit_myhook(struct vmspace *vm);
131 static void shmfork_myhook(struct proc *p1, struct proc *p2);
132 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
133 
134 /*
135  * Tuneable values.
136  */
137 #ifndef SHMMAXPGS
138 #define	SHMMAXPGS	131072	/* Note: sysv shared memory is swap backed. */
139 #endif
140 #ifndef SHMMAX
141 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
142 #endif
143 #ifndef SHMMIN
144 #define	SHMMIN	1
145 #endif
146 #ifndef SHMMNI
147 #define	SHMMNI	192
148 #endif
149 #ifndef SHMSEG
150 #define	SHMSEG	128
151 #endif
152 #ifndef SHMALL
153 #define	SHMALL	(SHMMAXPGS)
154 #endif
155 
156 struct	shminfo shminfo = {
157 	.shmmax = SHMMAX,
158 	.shmmin = SHMMIN,
159 	.shmmni = SHMMNI,
160 	.shmseg = SHMSEG,
161 	.shmall = SHMALL
162 };
163 
164 static int shm_use_phys;
165 static int shm_allow_removed;
166 
167 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
168     "Maximum shared memory segment size");
169 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
170     "Minimum shared memory segment size");
171 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
172     "Number of shared memory identifiers");
173 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
174     "Number of segments per process");
175 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
176     "Maximum number of pages available for shared memory");
177 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
178     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
179 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
180     &shm_allow_removed, 0,
181     "Enable/Disable attachment to attached segments marked for removal");
182 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD |
183     CTLFLAG_MPSAFE, NULL, 0, sysctl_shmsegs, "",
184     "Current number of shared memory segments allocated");
185 
186 static struct sx sysvshmsx;
187 #define	SYSVSHM_LOCK()		sx_xlock(&sysvshmsx)
188 #define	SYSVSHM_UNLOCK()	sx_xunlock(&sysvshmsx)
189 #define	SYSVSHM_ASSERT_LOCKED()	sx_assert(&sysvshmsx, SA_XLOCKED)
190 
191 static int
192 shm_find_segment_by_key(key_t key)
193 {
194 	int i;
195 
196 	for (i = 0; i < shmalloced; i++)
197 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
198 		    shmsegs[i].u.shm_perm.key == key)
199 			return (i);
200 	return (-1);
201 }
202 
203 /*
204  * Finds segment either by shmid if is_shmid is true, or by segnum if
205  * is_shmid is false.
206  */
207 static struct shmid_kernel *
208 shm_find_segment(int arg, bool is_shmid)
209 {
210 	struct shmid_kernel *shmseg;
211 	int segnum;
212 
213 	segnum = is_shmid ? IPCID_TO_IX(arg) : arg;
214 	if (segnum < 0 || segnum >= shmalloced)
215 		return (NULL);
216 	shmseg = &shmsegs[segnum];
217 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
218 	    (!shm_allow_removed &&
219 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
220 	    (is_shmid && shmseg->u.shm_perm.seq != IPCID_TO_SEQ(arg)))
221 		return (NULL);
222 	return (shmseg);
223 }
224 
225 static void
226 shm_deallocate_segment(struct shmid_kernel *shmseg)
227 {
228 	vm_size_t size;
229 
230 	SYSVSHM_ASSERT_LOCKED();
231 
232 	vm_object_deallocate(shmseg->object);
233 	shmseg->object = NULL;
234 	size = round_page(shmseg->u.shm_segsz);
235 	shm_committed -= btoc(size);
236 	shm_nused--;
237 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
238 #ifdef MAC
239 	mac_sysvshm_cleanup(shmseg);
240 #endif
241 	racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
242 	racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
243 	crfree(shmseg->cred);
244 	shmseg->cred = NULL;
245 }
246 
247 static int
248 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
249 {
250 	struct shmid_kernel *shmseg;
251 	int segnum, result;
252 	vm_size_t size;
253 
254 	SYSVSHM_ASSERT_LOCKED();
255 	segnum = IPCID_TO_IX(shmmap_s->shmid);
256 	KASSERT(segnum >= 0 && segnum < shmalloced,
257 	    ("segnum %d shmalloced %d", segnum, shmalloced));
258 
259 	shmseg = &shmsegs[segnum];
260 	size = round_page(shmseg->u.shm_segsz);
261 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
262 	if (result != KERN_SUCCESS)
263 		return (EINVAL);
264 	shmmap_s->shmid = -1;
265 	shmseg->u.shm_dtime = time_second;
266 	if ((--shmseg->u.shm_nattch <= 0) &&
267 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
268 		shm_deallocate_segment(shmseg);
269 		shm_last_free = segnum;
270 	}
271 	return (0);
272 }
273 
274 static int
275 kern_shmdt_locked(struct thread *td, const void *shmaddr)
276 {
277 	struct proc *p = td->td_proc;
278 	struct shmmap_state *shmmap_s;
279 #ifdef MAC
280 	struct shmid_kernel *shmsegptr;
281 	int error;
282 #endif
283 	int i;
284 
285 	SYSVSHM_ASSERT_LOCKED();
286 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
287 		return (ENOSYS);
288 	shmmap_s = p->p_vmspace->vm_shm;
289  	if (shmmap_s == NULL)
290 		return (EINVAL);
291 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
292 		if (shmmap_s->shmid != -1 &&
293 		    shmmap_s->va == (vm_offset_t)shmaddr) {
294 			break;
295 		}
296 	}
297 	if (i == shminfo.shmseg)
298 		return (EINVAL);
299 #ifdef MAC
300 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
301 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
302 	if (error != 0)
303 		return (error);
304 #endif
305 	return (shm_delete_mapping(p->p_vmspace, shmmap_s));
306 }
307 
308 #ifndef _SYS_SYSPROTO_H_
309 struct shmdt_args {
310 	const void *shmaddr;
311 };
312 #endif
313 int
314 sys_shmdt(struct thread *td, struct shmdt_args *uap)
315 {
316 	int error;
317 
318 	SYSVSHM_LOCK();
319 	error = kern_shmdt_locked(td, uap->shmaddr);
320 	SYSVSHM_UNLOCK();
321 	return (error);
322 }
323 
324 static int
325 kern_shmat_locked(struct thread *td, int shmid, const void *shmaddr,
326     int shmflg)
327 {
328 	struct proc *p = td->td_proc;
329 	struct shmid_kernel *shmseg;
330 	struct shmmap_state *shmmap_s;
331 	vm_offset_t attach_va;
332 	vm_prot_t prot;
333 	vm_size_t size;
334 	int error, i, rv;
335 
336 	SYSVSHM_ASSERT_LOCKED();
337 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
338 		return (ENOSYS);
339 	shmmap_s = p->p_vmspace->vm_shm;
340 	if (shmmap_s == NULL) {
341 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
342 		    M_SHM, M_WAITOK);
343 		for (i = 0; i < shminfo.shmseg; i++)
344 			shmmap_s[i].shmid = -1;
345 		KASSERT(p->p_vmspace->vm_shm == NULL, ("raced"));
346 		p->p_vmspace->vm_shm = shmmap_s;
347 	}
348 	shmseg = shm_find_segment(shmid, true);
349 	if (shmseg == NULL)
350 		return (EINVAL);
351 	error = ipcperm(td, &shmseg->u.shm_perm,
352 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
353 	if (error != 0)
354 		return (error);
355 #ifdef MAC
356 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
357 	if (error != 0)
358 		return (error);
359 #endif
360 	for (i = 0; i < shminfo.shmseg; i++) {
361 		if (shmmap_s->shmid == -1)
362 			break;
363 		shmmap_s++;
364 	}
365 	if (i >= shminfo.shmseg)
366 		return (EMFILE);
367 	size = round_page(shmseg->u.shm_segsz);
368 	prot = VM_PROT_READ;
369 	if ((shmflg & SHM_RDONLY) == 0)
370 		prot |= VM_PROT_WRITE;
371 	if (shmaddr != NULL) {
372 		if ((shmflg & SHM_RND) != 0)
373 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
374 		else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0)
375 			attach_va = (vm_offset_t)shmaddr;
376 		else
377 			return (EINVAL);
378 	} else {
379 		/*
380 		 * This is just a hint to vm_map_find() about where to
381 		 * put it.
382 		 */
383 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
384 		    lim_max(td, RLIMIT_DATA));
385 	}
386 
387 	vm_object_reference(shmseg->object);
388 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
389 	    size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
390 	    prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
391 	if (rv != KERN_SUCCESS) {
392 		vm_object_deallocate(shmseg->object);
393 		return (ENOMEM);
394 	}
395 
396 	shmmap_s->va = attach_va;
397 	shmmap_s->shmid = shmid;
398 	shmseg->u.shm_lpid = p->p_pid;
399 	shmseg->u.shm_atime = time_second;
400 	shmseg->u.shm_nattch++;
401 	td->td_retval[0] = attach_va;
402 	return (error);
403 }
404 
405 int
406 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
407 {
408 	int error;
409 
410 	SYSVSHM_LOCK();
411 	error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
412 	SYSVSHM_UNLOCK();
413 	return (error);
414 }
415 
416 #ifndef _SYS_SYSPROTO_H_
417 struct shmat_args {
418 	int shmid;
419 	const void *shmaddr;
420 	int shmflg;
421 };
422 #endif
423 int
424 sys_shmat(struct thread *td, struct shmat_args *uap)
425 {
426 
427 	return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
428 }
429 
430 static int
431 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
432     size_t *bufsz)
433 {
434 	struct shmid_kernel *shmseg;
435 	struct shmid_ds *shmidp;
436 	struct shm_info shm_info;
437 	int error;
438 
439 	SYSVSHM_ASSERT_LOCKED();
440 
441 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
442 		return (ENOSYS);
443 
444 	switch (cmd) {
445 	/*
446 	 * It is possible that kern_shmctl is being called from the Linux ABI
447 	 * layer, in which case, we will need to implement IPC_INFO.  It should
448 	 * be noted that other shmctl calls will be funneled through here for
449 	 * Linix binaries as well.
450 	 *
451 	 * NB: The Linux ABI layer will convert this data to structure(s) more
452 	 * consistent with the Linux ABI.
453 	 */
454 	case IPC_INFO:
455 		memcpy(buf, &shminfo, sizeof(shminfo));
456 		if (bufsz)
457 			*bufsz = sizeof(shminfo);
458 		td->td_retval[0] = shmalloced;
459 		return (0);
460 	case SHM_INFO: {
461 		shm_info.used_ids = shm_nused;
462 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
463 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
464 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
465 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
466 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
467 		memcpy(buf, &shm_info, sizeof(shm_info));
468 		if (bufsz != NULL)
469 			*bufsz = sizeof(shm_info);
470 		td->td_retval[0] = shmalloced;
471 		return (0);
472 	}
473 	}
474 	shmseg = shm_find_segment(shmid, cmd != SHM_STAT);
475 	if (shmseg == NULL)
476 		return (EINVAL);
477 #ifdef MAC
478 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
479 	if (error != 0)
480 		return (error);
481 #endif
482 	switch (cmd) {
483 	case SHM_STAT:
484 	case IPC_STAT:
485 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
486 		if (error != 0)
487 			return (error);
488 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
489 		if (bufsz != NULL)
490 			*bufsz = sizeof(struct shmid_ds);
491 		if (cmd == SHM_STAT) {
492 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
493 			    shmseg->u.shm_perm);
494 		}
495 		break;
496 	case IPC_SET:
497 		shmidp = (struct shmid_ds *)buf;
498 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
499 		if (error != 0)
500 			return (error);
501 		shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
502 		shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
503 		shmseg->u.shm_perm.mode =
504 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
505 		    (shmidp->shm_perm.mode & ACCESSPERMS);
506 		shmseg->u.shm_ctime = time_second;
507 		break;
508 	case IPC_RMID:
509 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
510 		if (error != 0)
511 			return (error);
512 		shmseg->u.shm_perm.key = IPC_PRIVATE;
513 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
514 		if (shmseg->u.shm_nattch <= 0) {
515 			shm_deallocate_segment(shmseg);
516 			shm_last_free = IPCID_TO_IX(shmid);
517 		}
518 		break;
519 #if 0
520 	case SHM_LOCK:
521 	case SHM_UNLOCK:
522 #endif
523 	default:
524 		error = EINVAL;
525 		break;
526 	}
527 	return (error);
528 }
529 
530 int
531 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
532 {
533 	int error;
534 
535 	SYSVSHM_LOCK();
536 	error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
537 	SYSVSHM_UNLOCK();
538 	return (error);
539 }
540 
541 
542 #ifndef _SYS_SYSPROTO_H_
543 struct shmctl_args {
544 	int shmid;
545 	int cmd;
546 	struct shmid_ds *buf;
547 };
548 #endif
549 int
550 sys_shmctl(struct thread *td, struct shmctl_args *uap)
551 {
552 	int error;
553 	struct shmid_ds buf;
554 	size_t bufsz;
555 
556 	/*
557 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
558 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
559 	 * return an error back to the user since we do not to support this.
560 	 */
561 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
562 	    uap->cmd == SHM_STAT)
563 		return (EINVAL);
564 
565 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
566 	if (uap->cmd == IPC_SET) {
567 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
568 			goto done;
569 	}
570 
571 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
572 	if (error)
573 		goto done;
574 
575 	/* Cases in which we need to copyout */
576 	switch (uap->cmd) {
577 	case IPC_STAT:
578 		error = copyout(&buf, uap->buf, bufsz);
579 		break;
580 	}
581 
582 done:
583 	if (error) {
584 		/* Invalidate the return value */
585 		td->td_retval[0] = -1;
586 	}
587 	return (error);
588 }
589 
590 
591 static int
592 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
593     int segnum)
594 {
595 	struct shmid_kernel *shmseg;
596 #ifdef MAC
597 	int error;
598 #endif
599 
600 	SYSVSHM_ASSERT_LOCKED();
601 	KASSERT(segnum >= 0 && segnum < shmalloced,
602 	    ("segnum %d shmalloced %d", segnum, shmalloced));
603 	shmseg = &shmsegs[segnum];
604 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
605 		return (EEXIST);
606 #ifdef MAC
607 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
608 	if (error != 0)
609 		return (error);
610 #endif
611 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
612 		return (EINVAL);
613 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
614 	return (0);
615 }
616 
617 static int
618 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
619 {
620 	struct ucred *cred = td->td_ucred;
621 	struct shmid_kernel *shmseg;
622 	vm_object_t shm_object;
623 	int i, segnum;
624 	size_t size;
625 
626 	SYSVSHM_ASSERT_LOCKED();
627 
628 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
629 		return (EINVAL);
630 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
631 		return (ENOSPC);
632 	size = round_page(uap->size);
633 	if (shm_committed + btoc(size) > shminfo.shmall)
634 		return (ENOMEM);
635 	if (shm_last_free < 0) {
636 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
637 		for (i = 0; i < shmalloced; i++)
638 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
639 				break;
640 		if (i == shmalloced)
641 			return (ENOSPC);
642 		segnum = i;
643 	} else  {
644 		segnum = shm_last_free;
645 		shm_last_free = -1;
646 	}
647 	KASSERT(segnum >= 0 && segnum < shmalloced,
648 	    ("segnum %d shmalloced %d", segnum, shmalloced));
649 	shmseg = &shmsegs[segnum];
650 #ifdef RACCT
651 	if (racct_enable) {
652 		PROC_LOCK(td->td_proc);
653 		if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
654 			PROC_UNLOCK(td->td_proc);
655 			return (ENOSPC);
656 		}
657 		if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
658 			racct_sub(td->td_proc, RACCT_NSHM, 1);
659 			PROC_UNLOCK(td->td_proc);
660 			return (ENOMEM);
661 		}
662 		PROC_UNLOCK(td->td_proc);
663 	}
664 #endif
665 
666 	/*
667 	 * We make sure that we have allocated a pager before we need
668 	 * to.
669 	 */
670 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
671 	    0, size, VM_PROT_DEFAULT, 0, cred);
672 	if (shm_object == NULL) {
673 #ifdef RACCT
674 		if (racct_enable) {
675 			PROC_LOCK(td->td_proc);
676 			racct_sub(td->td_proc, RACCT_NSHM, 1);
677 			racct_sub(td->td_proc, RACCT_SHMSIZE, size);
678 			PROC_UNLOCK(td->td_proc);
679 		}
680 #endif
681 		return (ENOMEM);
682 	}
683 	shm_object->pg_color = 0;
684 	VM_OBJECT_WLOCK(shm_object);
685 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
686 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
687 	VM_OBJECT_WUNLOCK(shm_object);
688 
689 	shmseg->object = shm_object;
690 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
691 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
692 	shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
693 	shmseg->u.shm_perm.key = uap->key;
694 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
695 	shmseg->cred = crhold(cred);
696 	shmseg->u.shm_segsz = uap->size;
697 	shmseg->u.shm_cpid = td->td_proc->p_pid;
698 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
699 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
700 #ifdef MAC
701 	mac_sysvshm_create(cred, shmseg);
702 #endif
703 	shmseg->u.shm_ctime = time_second;
704 	shm_committed += btoc(size);
705 	shm_nused++;
706 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
707 
708 	return (0);
709 }
710 
711 #ifndef _SYS_SYSPROTO_H_
712 struct shmget_args {
713 	key_t key;
714 	size_t size;
715 	int shmflg;
716 };
717 #endif
718 int
719 sys_shmget(struct thread *td, struct shmget_args *uap)
720 {
721 	int segnum, mode;
722 	int error;
723 
724 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
725 		return (ENOSYS);
726 	mode = uap->shmflg & ACCESSPERMS;
727 	SYSVSHM_LOCK();
728 	if (uap->key == IPC_PRIVATE) {
729 		error = shmget_allocate_segment(td, uap, mode);
730 	} else {
731 		segnum = shm_find_segment_by_key(uap->key);
732 		if (segnum >= 0)
733 			error = shmget_existing(td, uap, mode, segnum);
734 		else if ((uap->shmflg & IPC_CREAT) == 0)
735 			error = ENOENT;
736 		else
737 			error = shmget_allocate_segment(td, uap, mode);
738 	}
739 	SYSVSHM_UNLOCK();
740 	return (error);
741 }
742 
743 static void
744 shmfork_myhook(struct proc *p1, struct proc *p2)
745 {
746 	struct shmmap_state *shmmap_s;
747 	size_t size;
748 	int i;
749 
750 	SYSVSHM_LOCK();
751 	size = shminfo.shmseg * sizeof(struct shmmap_state);
752 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
753 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
754 	p2->p_vmspace->vm_shm = shmmap_s;
755 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
756 		if (shmmap_s->shmid != -1) {
757 			KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
758 			    IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
759 			    ("segnum %d shmalloced %d",
760 			    IPCID_TO_IX(shmmap_s->shmid), shmalloced));
761 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
762 		}
763 	}
764 	SYSVSHM_UNLOCK();
765 }
766 
767 static void
768 shmexit_myhook(struct vmspace *vm)
769 {
770 	struct shmmap_state *base, *shm;
771 	int i;
772 
773 	base = vm->vm_shm;
774 	if (base != NULL) {
775 		vm->vm_shm = NULL;
776 		SYSVSHM_LOCK();
777 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
778 			if (shm->shmid != -1)
779 				shm_delete_mapping(vm, shm);
780 		}
781 		SYSVSHM_UNLOCK();
782 		free(base, M_SHM);
783 	}
784 }
785 
786 static void
787 shmrealloc(void)
788 {
789 	struct shmid_kernel *newsegs;
790 	int i;
791 
792 	SYSVSHM_ASSERT_LOCKED();
793 
794 	if (shmalloced >= shminfo.shmmni)
795 		return;
796 
797 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
798 	for (i = 0; i < shmalloced; i++)
799 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
800 	for (; i < shminfo.shmmni; i++) {
801 		newsegs[i].u.shm_perm.mode = SHMSEG_FREE;
802 		newsegs[i].u.shm_perm.seq = 0;
803 #ifdef MAC
804 		mac_sysvshm_init(&newsegs[i]);
805 #endif
806 	}
807 	free(shmsegs, M_SHM);
808 	shmsegs = newsegs;
809 	shmalloced = shminfo.shmmni;
810 }
811 
812 static struct syscall_helper_data shm_syscalls[] = {
813 	SYSCALL_INIT_HELPER(shmat),
814 	SYSCALL_INIT_HELPER(shmctl),
815 	SYSCALL_INIT_HELPER(shmdt),
816 	SYSCALL_INIT_HELPER(shmget),
817 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
818     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
819 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
820 #endif
821 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
822 	SYSCALL_INIT_HELPER(shmsys),
823 #endif
824 	SYSCALL_INIT_LAST
825 };
826 
827 #ifdef COMPAT_FREEBSD32
828 #include <compat/freebsd32/freebsd32.h>
829 #include <compat/freebsd32/freebsd32_ipc.h>
830 #include <compat/freebsd32/freebsd32_proto.h>
831 #include <compat/freebsd32/freebsd32_signal.h>
832 #include <compat/freebsd32/freebsd32_syscall.h>
833 #include <compat/freebsd32/freebsd32_util.h>
834 
835 static struct syscall_helper_data shm32_syscalls[] = {
836 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
837 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
838 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
839 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
840 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
841 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
842     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
843 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
844 #endif
845 	SYSCALL_INIT_LAST
846 };
847 #endif
848 
849 static int
850 shminit(void)
851 {
852 	int i, error;
853 
854 #ifndef BURN_BRIDGES
855 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
856 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
857 #endif
858 	if (shminfo.shmmax == SHMMAX) {
859 		/* Initialize shmmax dealing with possible overflow. */
860 		for (i = PAGE_SIZE; i != 0; i--) {
861 			shminfo.shmmax = shminfo.shmall * i;
862 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
863 				break;
864 		}
865 	}
866 	shmalloced = shminfo.shmmni;
867 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
868 	for (i = 0; i < shmalloced; i++) {
869 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
870 		shmsegs[i].u.shm_perm.seq = 0;
871 #ifdef MAC
872 		mac_sysvshm_init(&shmsegs[i]);
873 #endif
874 	}
875 	shm_last_free = 0;
876 	shm_nused = 0;
877 	shm_committed = 0;
878 	sx_init(&sysvshmsx, "sysvshmsx");
879 	shmexit_hook = &shmexit_myhook;
880 	shmfork_hook = &shmfork_myhook;
881 
882 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
883 	if (error != 0)
884 		return (error);
885 #ifdef COMPAT_FREEBSD32
886 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
887 	if (error != 0)
888 		return (error);
889 #endif
890 	return (0);
891 }
892 
893 static int
894 shmunload(void)
895 {
896 	int i;
897 
898 	if (shm_nused > 0)
899 		return (EBUSY);
900 
901 #ifdef COMPAT_FREEBSD32
902 	syscall32_helper_unregister(shm32_syscalls);
903 #endif
904 	syscall_helper_unregister(shm_syscalls);
905 
906 	for (i = 0; i < shmalloced; i++) {
907 #ifdef MAC
908 		mac_sysvshm_destroy(&shmsegs[i]);
909 #endif
910 		/*
911 		 * Objects might be still mapped into the processes
912 		 * address spaces.  Actual free would happen on the
913 		 * last mapping destruction.
914 		 */
915 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
916 			vm_object_deallocate(shmsegs[i].object);
917 	}
918 	free(shmsegs, M_SHM);
919 	shmexit_hook = NULL;
920 	shmfork_hook = NULL;
921 	sx_destroy(&sysvshmsx);
922 	return (0);
923 }
924 
925 static int
926 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
927 {
928 	int error;
929 
930 	SYSVSHM_LOCK();
931 	error = SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0]));
932 	SYSVSHM_UNLOCK();
933 	return (error);
934 }
935 
936 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
937 struct oshmid_ds {
938 	struct	ipc_perm_old shm_perm;	/* operation perms */
939 	int	shm_segsz;		/* size of segment (bytes) */
940 	u_short	shm_cpid;		/* pid, creator */
941 	u_short	shm_lpid;		/* pid, last operation */
942 	short	shm_nattch;		/* no. of current attaches */
943 	time_t	shm_atime;		/* last attach time */
944 	time_t	shm_dtime;		/* last detach time */
945 	time_t	shm_ctime;		/* last change time */
946 	void	*shm_handle;		/* internal handle for shm segment */
947 };
948 
949 struct oshmctl_args {
950 	int shmid;
951 	int cmd;
952 	struct oshmid_ds *ubuf;
953 };
954 
955 static int
956 oshmctl(struct thread *td, struct oshmctl_args *uap)
957 {
958 #ifdef COMPAT_43
959 	int error = 0;
960 	struct shmid_kernel *shmseg;
961 	struct oshmid_ds outbuf;
962 
963 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
964 		return (ENOSYS);
965 	if (uap->cmd != IPC_STAT) {
966 		return (freebsd7_shmctl(td,
967 		    (struct freebsd7_shmctl_args *)uap));
968 	}
969 	SYSVSHM_LOCK();
970 	shmseg = shm_find_segment(uap->shmid, true);
971 	if (shmseg == NULL) {
972 		SYSVSHM_UNLOCK();
973 		return (EINVAL);
974 	}
975 	error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
976 	if (error != 0) {
977 		SYSVSHM_UNLOCK();
978 		return (error);
979 	}
980 #ifdef MAC
981 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
982 	if (error != 0) {
983 		SYSVSHM_UNLOCK();
984 		return (error);
985 	}
986 #endif
987 	ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
988 	outbuf.shm_segsz = shmseg->u.shm_segsz;
989 	outbuf.shm_cpid = shmseg->u.shm_cpid;
990 	outbuf.shm_lpid = shmseg->u.shm_lpid;
991 	outbuf.shm_nattch = shmseg->u.shm_nattch;
992 	outbuf.shm_atime = shmseg->u.shm_atime;
993 	outbuf.shm_dtime = shmseg->u.shm_dtime;
994 	outbuf.shm_ctime = shmseg->u.shm_ctime;
995 	outbuf.shm_handle = shmseg->object;
996 	SYSVSHM_UNLOCK();
997 	return (copyout(&outbuf, uap->ubuf, sizeof(outbuf)));
998 #else
999 	return (EINVAL);
1000 #endif
1001 }
1002 
1003 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1004 static sy_call_t *shmcalls[] = {
1005 	(sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1006 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1007 	(sy_call_t *)freebsd7_shmctl
1008 };
1009 
1010 #ifndef _SYS_SYSPROTO_H_
1011 /* XXX actually varargs. */
1012 struct shmsys_args {
1013 	int	which;
1014 	int	a2;
1015 	int	a3;
1016 	int	a4;
1017 };
1018 #endif
1019 int
1020 sys_shmsys(struct thread *td, struct shmsys_args *uap)
1021 {
1022 
1023 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1024 		return (ENOSYS);
1025 	if (uap->which < 0 || uap->which >= nitems(shmcalls))
1026 		return (EINVAL);
1027 	return ((*shmcalls[uap->which])(td, &uap->a2));
1028 }
1029 
1030 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1031 
1032 #ifdef COMPAT_FREEBSD32
1033 
1034 int
1035 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1036 {
1037 
1038 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1039     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1040 	switch (uap->which) {
1041 	case 0:	{	/* shmat */
1042 		struct shmat_args ap;
1043 
1044 		ap.shmid = uap->a2;
1045 		ap.shmaddr = PTRIN(uap->a3);
1046 		ap.shmflg = uap->a4;
1047 		return (sysent[SYS_shmat].sy_call(td, &ap));
1048 	}
1049 	case 2: {	/* shmdt */
1050 		struct shmdt_args ap;
1051 
1052 		ap.shmaddr = PTRIN(uap->a2);
1053 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1054 	}
1055 	case 3: {	/* shmget */
1056 		struct shmget_args ap;
1057 
1058 		ap.key = uap->a2;
1059 		ap.size = uap->a3;
1060 		ap.shmflg = uap->a4;
1061 		return (sysent[SYS_shmget].sy_call(td, &ap));
1062 	}
1063 	case 4: {	/* shmctl */
1064 		struct freebsd7_freebsd32_shmctl_args ap;
1065 
1066 		ap.shmid = uap->a2;
1067 		ap.cmd = uap->a3;
1068 		ap.buf = PTRIN(uap->a4);
1069 		return (freebsd7_freebsd32_shmctl(td, &ap));
1070 	}
1071 	case 1:		/* oshmctl */
1072 	default:
1073 		return (EINVAL);
1074 	}
1075 #else
1076 	return (nosys(td, NULL));
1077 #endif
1078 }
1079 
1080 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1081     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1082 int
1083 freebsd7_freebsd32_shmctl(struct thread *td,
1084     struct freebsd7_freebsd32_shmctl_args *uap)
1085 {
1086 	int error;
1087 	union {
1088 		struct shmid_ds shmid_ds;
1089 		struct shm_info shm_info;
1090 		struct shminfo shminfo;
1091 	} u;
1092 	union {
1093 		struct shmid_ds32_old shmid_ds32;
1094 		struct shm_info32 shm_info32;
1095 		struct shminfo32 shminfo32;
1096 	} u32;
1097 	size_t sz;
1098 
1099 	if (uap->cmd == IPC_SET) {
1100 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1101 		    sizeof(u32.shmid_ds32))))
1102 			goto done;
1103 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1104 		    &u.shmid_ds.shm_perm);
1105 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1106 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1107 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1108 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1109 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1110 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1111 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1112 	}
1113 
1114 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1115 	if (error)
1116 		goto done;
1117 
1118 	/* Cases in which we need to copyout */
1119 	switch (uap->cmd) {
1120 	case IPC_INFO:
1121 		CP(u.shminfo, u32.shminfo32, shmmax);
1122 		CP(u.shminfo, u32.shminfo32, shmmin);
1123 		CP(u.shminfo, u32.shminfo32, shmmni);
1124 		CP(u.shminfo, u32.shminfo32, shmseg);
1125 		CP(u.shminfo, u32.shminfo32, shmall);
1126 		error = copyout(&u32.shminfo32, uap->buf,
1127 		    sizeof(u32.shminfo32));
1128 		break;
1129 	case SHM_INFO:
1130 		CP(u.shm_info, u32.shm_info32, used_ids);
1131 		CP(u.shm_info, u32.shm_info32, shm_rss);
1132 		CP(u.shm_info, u32.shm_info32, shm_tot);
1133 		CP(u.shm_info, u32.shm_info32, shm_swp);
1134 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1135 		CP(u.shm_info, u32.shm_info32, swap_successes);
1136 		error = copyout(&u32.shm_info32, uap->buf,
1137 		    sizeof(u32.shm_info32));
1138 		break;
1139 	case SHM_STAT:
1140 	case IPC_STAT:
1141 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1142 		    &u32.shmid_ds32.shm_perm);
1143 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1144 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1145 		else
1146 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1147 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1148 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1149 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1150 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1151 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1152 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1153 		u32.shmid_ds32.shm_internal = 0;
1154 		error = copyout(&u32.shmid_ds32, uap->buf,
1155 		    sizeof(u32.shmid_ds32));
1156 		break;
1157 	}
1158 
1159 done:
1160 	if (error) {
1161 		/* Invalidate the return value */
1162 		td->td_retval[0] = -1;
1163 	}
1164 	return (error);
1165 }
1166 #endif
1167 
1168 int
1169 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1170 {
1171 	int error;
1172 	union {
1173 		struct shmid_ds shmid_ds;
1174 		struct shm_info shm_info;
1175 		struct shminfo shminfo;
1176 	} u;
1177 	union {
1178 		struct shmid_ds32 shmid_ds32;
1179 		struct shm_info32 shm_info32;
1180 		struct shminfo32 shminfo32;
1181 	} u32;
1182 	size_t sz;
1183 
1184 	if (uap->cmd == IPC_SET) {
1185 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1186 		    sizeof(u32.shmid_ds32))))
1187 			goto done;
1188 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1189 		    &u.shmid_ds.shm_perm);
1190 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1191 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1192 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1193 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1194 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1195 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1196 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1197 	}
1198 
1199 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1200 	if (error)
1201 		goto done;
1202 
1203 	/* Cases in which we need to copyout */
1204 	switch (uap->cmd) {
1205 	case IPC_INFO:
1206 		CP(u.shminfo, u32.shminfo32, shmmax);
1207 		CP(u.shminfo, u32.shminfo32, shmmin);
1208 		CP(u.shminfo, u32.shminfo32, shmmni);
1209 		CP(u.shminfo, u32.shminfo32, shmseg);
1210 		CP(u.shminfo, u32.shminfo32, shmall);
1211 		error = copyout(&u32.shminfo32, uap->buf,
1212 		    sizeof(u32.shminfo32));
1213 		break;
1214 	case SHM_INFO:
1215 		CP(u.shm_info, u32.shm_info32, used_ids);
1216 		CP(u.shm_info, u32.shm_info32, shm_rss);
1217 		CP(u.shm_info, u32.shm_info32, shm_tot);
1218 		CP(u.shm_info, u32.shm_info32, shm_swp);
1219 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1220 		CP(u.shm_info, u32.shm_info32, swap_successes);
1221 		error = copyout(&u32.shm_info32, uap->buf,
1222 		    sizeof(u32.shm_info32));
1223 		break;
1224 	case SHM_STAT:
1225 	case IPC_STAT:
1226 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1227 		    &u32.shmid_ds32.shm_perm);
1228 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1229 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1230 		else
1231 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1232 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1233 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1234 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1235 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1236 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1237 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1238 		error = copyout(&u32.shmid_ds32, uap->buf,
1239 		    sizeof(u32.shmid_ds32));
1240 		break;
1241 	}
1242 
1243 done:
1244 	if (error) {
1245 		/* Invalidate the return value */
1246 		td->td_retval[0] = -1;
1247 	}
1248 	return (error);
1249 }
1250 #endif
1251 
1252 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1253     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1254 
1255 #ifndef CP
1256 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1257 #endif
1258 
1259 #ifndef _SYS_SYSPROTO_H_
1260 struct freebsd7_shmctl_args {
1261 	int shmid;
1262 	int cmd;
1263 	struct shmid_ds_old *buf;
1264 };
1265 #endif
1266 int
1267 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1268 {
1269 	int error;
1270 	struct shmid_ds_old old;
1271 	struct shmid_ds buf;
1272 	size_t bufsz;
1273 
1274 	/*
1275 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1276 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1277 	 * return an error back to the user since we do not to support this.
1278 	 */
1279 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1280 	    uap->cmd == SHM_STAT)
1281 		return (EINVAL);
1282 
1283 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1284 	if (uap->cmd == IPC_SET) {
1285 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1286 			goto done;
1287 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1288 		CP(old, buf, shm_segsz);
1289 		CP(old, buf, shm_lpid);
1290 		CP(old, buf, shm_cpid);
1291 		CP(old, buf, shm_nattch);
1292 		CP(old, buf, shm_atime);
1293 		CP(old, buf, shm_dtime);
1294 		CP(old, buf, shm_ctime);
1295 	}
1296 
1297 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1298 	if (error)
1299 		goto done;
1300 
1301 	/* Cases in which we need to copyout */
1302 	switch (uap->cmd) {
1303 	case IPC_STAT:
1304 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1305 		if (buf.shm_segsz > INT_MAX)
1306 			old.shm_segsz = INT_MAX;
1307 		else
1308 			CP(buf, old, shm_segsz);
1309 		CP(buf, old, shm_lpid);
1310 		CP(buf, old, shm_cpid);
1311 		if (buf.shm_nattch > SHRT_MAX)
1312 			old.shm_nattch = SHRT_MAX;
1313 		else
1314 			CP(buf, old, shm_nattch);
1315 		CP(buf, old, shm_atime);
1316 		CP(buf, old, shm_dtime);
1317 		CP(buf, old, shm_ctime);
1318 		old.shm_internal = NULL;
1319 		error = copyout(&old, uap->buf, sizeof(old));
1320 		break;
1321 	}
1322 
1323 done:
1324 	if (error) {
1325 		/* Invalidate the return value */
1326 		td->td_retval[0] = -1;
1327 	}
1328 	return (error);
1329 }
1330 
1331 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1332 	   COMPAT_FREEBSD7 */
1333 
1334 static int
1335 sysvshm_modload(struct module *module, int cmd, void *arg)
1336 {
1337 	int error = 0;
1338 
1339 	switch (cmd) {
1340 	case MOD_LOAD:
1341 		error = shminit();
1342 		if (error != 0)
1343 			shmunload();
1344 		break;
1345 	case MOD_UNLOAD:
1346 		error = shmunload();
1347 		break;
1348 	case MOD_SHUTDOWN:
1349 		break;
1350 	default:
1351 		error = EINVAL;
1352 		break;
1353 	}
1354 	return (error);
1355 }
1356 
1357 static moduledata_t sysvshm_mod = {
1358 	"sysvshm",
1359 	&sysvshm_modload,
1360 	NULL
1361 };
1362 
1363 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1364 MODULE_VERSION(sysvshm, 1);
1365