xref: /freebsd/sys/kern/sysv_shm.c (revision f16f8610bbacd15b2f665891a32a36950d1a2b3e)
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 #endif
282 	int error, i;
283 
284 	SYSVSHM_ASSERT_LOCKED();
285 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
286 		return (ENOSYS);
287 	shmmap_s = p->p_vmspace->vm_shm;
288  	if (shmmap_s == NULL)
289 		return (EINVAL);
290 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
291 		if (shmmap_s->shmid != -1 &&
292 		    shmmap_s->va == (vm_offset_t)shmaddr) {
293 			break;
294 		}
295 	}
296 	if (i == shminfo.shmseg)
297 		return (EINVAL);
298 #ifdef MAC
299 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
300 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
301 	if (error != 0)
302 		return (error);
303 #endif
304 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
305 	return (error);
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 		PROC_LOCK(p);
384 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
385 		    lim_max(p, RLIMIT_DATA));
386 		PROC_UNLOCK(p);
387 	}
388 
389 	vm_object_reference(shmseg->object);
390 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
391 	    size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
392 	    prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
393 	if (rv != KERN_SUCCESS) {
394 		vm_object_deallocate(shmseg->object);
395 		return (ENOMEM);
396 	}
397 
398 	shmmap_s->va = attach_va;
399 	shmmap_s->shmid = shmid;
400 	shmseg->u.shm_lpid = p->p_pid;
401 	shmseg->u.shm_atime = time_second;
402 	shmseg->u.shm_nattch++;
403 	td->td_retval[0] = attach_va;
404 	return (error);
405 }
406 
407 int
408 kern_shmat(struct thread *td, int shmid, const void *shmaddr, int shmflg)
409 {
410 	int error;
411 
412 	SYSVSHM_LOCK();
413 	error = kern_shmat_locked(td, shmid, shmaddr, shmflg);
414 	SYSVSHM_UNLOCK();
415 	return (error);
416 }
417 
418 #ifndef _SYS_SYSPROTO_H_
419 struct shmat_args {
420 	int shmid;
421 	const void *shmaddr;
422 	int shmflg;
423 };
424 #endif
425 int
426 sys_shmat(struct thread *td, struct shmat_args *uap)
427 {
428 
429 	return (kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg));
430 }
431 
432 static int
433 kern_shmctl_locked(struct thread *td, int shmid, int cmd, void *buf,
434     size_t *bufsz)
435 {
436 	struct shmid_kernel *shmseg;
437 	struct shmid_ds *shmidp;
438 	struct shm_info shm_info;
439 	int error;
440 
441 	SYSVSHM_ASSERT_LOCKED();
442 
443 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
444 		return (ENOSYS);
445 
446 	error = 0;
447 	switch (cmd) {
448 	/*
449 	 * It is possible that kern_shmctl is being called from the Linux ABI
450 	 * layer, in which case, we will need to implement IPC_INFO.  It should
451 	 * be noted that other shmctl calls will be funneled through here for
452 	 * Linix binaries as well.
453 	 *
454 	 * NB: The Linux ABI layer will convert this data to structure(s) more
455 	 * consistent with the Linux ABI.
456 	 */
457 	case IPC_INFO:
458 		memcpy(buf, &shminfo, sizeof(shminfo));
459 		if (bufsz)
460 			*bufsz = sizeof(shminfo);
461 		td->td_retval[0] = shmalloced;
462 		return (0);
463 	case SHM_INFO: {
464 		shm_info.used_ids = shm_nused;
465 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
466 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
467 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
468 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
469 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
470 		memcpy(buf, &shm_info, sizeof(shm_info));
471 		if (bufsz != NULL)
472 			*bufsz = sizeof(shm_info);
473 		td->td_retval[0] = shmalloced;
474 		return (0);
475 	}
476 	}
477 	shmseg = shm_find_segment(shmid, cmd != SHM_STAT);
478 	if (shmseg == NULL)
479 		return (EINVAL);
480 #ifdef MAC
481 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
482 	if (error != 0)
483 		return (error);
484 #endif
485 	switch (cmd) {
486 	case SHM_STAT:
487 	case IPC_STAT:
488 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
489 		if (error != 0)
490 			return (error);
491 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
492 		if (bufsz != NULL)
493 			*bufsz = sizeof(struct shmid_ds);
494 		if (cmd == SHM_STAT) {
495 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid,
496 			    shmseg->u.shm_perm);
497 		}
498 		break;
499 	case IPC_SET:
500 		shmidp = (struct shmid_ds *)buf;
501 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
502 		if (error != 0)
503 			return (error);
504 		shmseg->u.shm_perm.uid = shmidp->shm_perm.uid;
505 		shmseg->u.shm_perm.gid = shmidp->shm_perm.gid;
506 		shmseg->u.shm_perm.mode =
507 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
508 		    (shmidp->shm_perm.mode & ACCESSPERMS);
509 		shmseg->u.shm_ctime = time_second;
510 		break;
511 	case IPC_RMID:
512 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
513 		if (error != 0)
514 			return (error);
515 		shmseg->u.shm_perm.key = IPC_PRIVATE;
516 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
517 		if (shmseg->u.shm_nattch <= 0) {
518 			shm_deallocate_segment(shmseg);
519 			shm_last_free = IPCID_TO_IX(shmid);
520 		}
521 		break;
522 #if 0
523 	case SHM_LOCK:
524 	case SHM_UNLOCK:
525 #endif
526 	default:
527 		error = EINVAL;
528 		break;
529 	}
530 	return (error);
531 }
532 
533 int
534 kern_shmctl(struct thread *td, int shmid, int cmd, void *buf, size_t *bufsz)
535 {
536 	int error;
537 
538 	SYSVSHM_LOCK();
539 	error = kern_shmctl_locked(td, shmid, cmd, buf, bufsz);
540 	SYSVSHM_UNLOCK();
541 	return (error);
542 }
543 
544 
545 #ifndef _SYS_SYSPROTO_H_
546 struct shmctl_args {
547 	int shmid;
548 	int cmd;
549 	struct shmid_ds *buf;
550 };
551 #endif
552 int
553 sys_shmctl(struct thread *td, struct shmctl_args *uap)
554 {
555 	int error = 0;
556 	struct shmid_ds buf;
557 	size_t bufsz;
558 
559 	/*
560 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
561 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
562 	 * return an error back to the user since we do not to support this.
563 	 */
564 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
565 	    uap->cmd == SHM_STAT)
566 		return (EINVAL);
567 
568 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
569 	if (uap->cmd == IPC_SET) {
570 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
571 			goto done;
572 	}
573 
574 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
575 	if (error)
576 		goto done;
577 
578 	/* Cases in which we need to copyout */
579 	switch (uap->cmd) {
580 	case IPC_STAT:
581 		error = copyout(&buf, uap->buf, bufsz);
582 		break;
583 	}
584 
585 done:
586 	if (error) {
587 		/* Invalidate the return value */
588 		td->td_retval[0] = -1;
589 	}
590 	return (error);
591 }
592 
593 
594 static int
595 shmget_existing(struct thread *td, struct shmget_args *uap, int mode,
596     int segnum)
597 {
598 	struct shmid_kernel *shmseg;
599 #ifdef MAC
600 	int error;
601 #endif
602 
603 	SYSVSHM_ASSERT_LOCKED();
604 	KASSERT(segnum >= 0 && segnum < shmalloced,
605 	    ("segnum %d shmalloced %d", segnum, shmalloced));
606 	shmseg = &shmsegs[segnum];
607 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
608 		return (EEXIST);
609 #ifdef MAC
610 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
611 	if (error != 0)
612 		return (error);
613 #endif
614 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
615 		return (EINVAL);
616 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
617 	return (0);
618 }
619 
620 static int
621 shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode)
622 {
623 	struct ucred *cred = td->td_ucred;
624 	struct shmid_kernel *shmseg;
625 	vm_object_t shm_object;
626 	int i, segnum;
627 	size_t size;
628 
629 	SYSVSHM_ASSERT_LOCKED();
630 
631 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
632 		return (EINVAL);
633 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
634 		return (ENOSPC);
635 	size = round_page(uap->size);
636 	if (shm_committed + btoc(size) > shminfo.shmall)
637 		return (ENOMEM);
638 	if (shm_last_free < 0) {
639 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
640 		for (i = 0; i < shmalloced; i++)
641 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
642 				break;
643 		if (i == shmalloced)
644 			return (ENOSPC);
645 		segnum = i;
646 	} else  {
647 		segnum = shm_last_free;
648 		shm_last_free = -1;
649 	}
650 	KASSERT(segnum >= 0 && segnum < shmalloced,
651 	    ("segnum %d shmalloced %d", segnum, shmalloced));
652 	shmseg = &shmsegs[segnum];
653 #ifdef RACCT
654 	PROC_LOCK(td->td_proc);
655 	if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
656 		PROC_UNLOCK(td->td_proc);
657 		return (ENOSPC);
658 	}
659 	if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
660 		racct_sub(td->td_proc, RACCT_NSHM, 1);
661 		PROC_UNLOCK(td->td_proc);
662 		return (ENOMEM);
663 	}
664 	PROC_UNLOCK(td->td_proc);
665 #endif
666 
667 	/*
668 	 * We make sure that we have allocated a pager before we need
669 	 * to.
670 	 */
671 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
672 	    0, size, VM_PROT_DEFAULT, 0, cred);
673 	if (shm_object == NULL) {
674 #ifdef RACCT
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 #endif
680 		return (ENOMEM);
681 	}
682 	shm_object->pg_color = 0;
683 	VM_OBJECT_WLOCK(shm_object);
684 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
685 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
686 	VM_OBJECT_WUNLOCK(shm_object);
687 
688 	shmseg->object = shm_object;
689 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
690 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
691 	shmseg->u.shm_perm.mode = (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
692 	shmseg->u.shm_perm.key = uap->key;
693 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
694 	shmseg->cred = crhold(cred);
695 	shmseg->u.shm_segsz = uap->size;
696 	shmseg->u.shm_cpid = td->td_proc->p_pid;
697 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
698 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
699 #ifdef MAC
700 	mac_sysvshm_create(cred, shmseg);
701 #endif
702 	shmseg->u.shm_ctime = time_second;
703 	shm_committed += btoc(size);
704 	shm_nused++;
705 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
706 
707 	return (0);
708 }
709 
710 #ifndef _SYS_SYSPROTO_H_
711 struct shmget_args {
712 	key_t key;
713 	size_t size;
714 	int shmflg;
715 };
716 #endif
717 int
718 sys_shmget(struct thread *td, struct shmget_args *uap)
719 {
720 	int segnum, mode;
721 	int error;
722 
723 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
724 		return (ENOSYS);
725 	mode = uap->shmflg & ACCESSPERMS;
726 	SYSVSHM_LOCK();
727 	if (uap->key == IPC_PRIVATE) {
728 		error = shmget_allocate_segment(td, uap, mode);
729 	} else {
730 		segnum = shm_find_segment_by_key(uap->key);
731 		if (segnum >= 0)
732 			error = shmget_existing(td, uap, mode, segnum);
733 		else if ((uap->shmflg & IPC_CREAT) == 0)
734 			error = ENOENT;
735 		else
736 			error = shmget_allocate_segment(td, uap, mode);
737 	}
738 	SYSVSHM_UNLOCK();
739 	return (error);
740 }
741 
742 static void
743 shmfork_myhook(struct proc *p1, struct proc *p2)
744 {
745 	struct shmmap_state *shmmap_s;
746 	size_t size;
747 	int i;
748 
749 	SYSVSHM_LOCK();
750 	size = shminfo.shmseg * sizeof(struct shmmap_state);
751 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
752 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
753 	p2->p_vmspace->vm_shm = shmmap_s;
754 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
755 		if (shmmap_s->shmid != -1) {
756 			KASSERT(IPCID_TO_IX(shmmap_s->shmid) >= 0 &&
757 			    IPCID_TO_IX(shmmap_s->shmid) < shmalloced,
758 			    ("segnum %d shmalloced %d",
759 			    IPCID_TO_IX(shmmap_s->shmid), shmalloced));
760 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
761 		}
762 	}
763 	SYSVSHM_UNLOCK();
764 }
765 
766 static void
767 shmexit_myhook(struct vmspace *vm)
768 {
769 	struct shmmap_state *base, *shm;
770 	int i;
771 
772 	base = vm->vm_shm;
773 	if (base != NULL) {
774 		vm->vm_shm = NULL;
775 		SYSVSHM_LOCK();
776 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
777 			if (shm->shmid != -1)
778 				shm_delete_mapping(vm, shm);
779 		}
780 		SYSVSHM_UNLOCK();
781 		free(base, M_SHM);
782 	}
783 }
784 
785 static void
786 shmrealloc(void)
787 {
788 	struct shmid_kernel *newsegs;
789 	int i;
790 
791 	SYSVSHM_ASSERT_LOCKED();
792 
793 	if (shmalloced >= shminfo.shmmni)
794 		return;
795 
796 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
797 	for (i = 0; i < shmalloced; i++)
798 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
799 	for (; i < shminfo.shmmni; i++) {
800 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
801 		shmsegs[i].u.shm_perm.seq = 0;
802 #ifdef MAC
803 		mac_sysvshm_init(&shmsegs[i]);
804 #endif
805 	}
806 	free(shmsegs, M_SHM);
807 	shmsegs = newsegs;
808 	shmalloced = shminfo.shmmni;
809 }
810 
811 static struct syscall_helper_data shm_syscalls[] = {
812 	SYSCALL_INIT_HELPER(shmat),
813 	SYSCALL_INIT_HELPER(shmctl),
814 	SYSCALL_INIT_HELPER(shmdt),
815 	SYSCALL_INIT_HELPER(shmget),
816 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
817     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
818 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
819 #endif
820 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
821 	SYSCALL_INIT_HELPER(shmsys),
822 #endif
823 	SYSCALL_INIT_LAST
824 };
825 
826 #ifdef COMPAT_FREEBSD32
827 #include <compat/freebsd32/freebsd32.h>
828 #include <compat/freebsd32/freebsd32_ipc.h>
829 #include <compat/freebsd32/freebsd32_proto.h>
830 #include <compat/freebsd32/freebsd32_signal.h>
831 #include <compat/freebsd32/freebsd32_syscall.h>
832 #include <compat/freebsd32/freebsd32_util.h>
833 
834 static struct syscall_helper_data shm32_syscalls[] = {
835 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
836 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
837 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
838 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
839 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
840 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
841     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
842 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
843 #endif
844 	SYSCALL_INIT_LAST
845 };
846 #endif
847 
848 static int
849 shminit(void)
850 {
851 	int i, error;
852 
853 #ifndef BURN_BRIDGES
854 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
855 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
856 #endif
857 	if (shminfo.shmmax == SHMMAX) {
858 		/* Initialize shmmax dealing with possible overflow. */
859 		for (i = PAGE_SIZE; i != 0; i--) {
860 			shminfo.shmmax = shminfo.shmall * i;
861 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
862 				break;
863 		}
864 	}
865 	shmalloced = shminfo.shmmni;
866 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
867 	for (i = 0; i < shmalloced; i++) {
868 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
869 		shmsegs[i].u.shm_perm.seq = 0;
870 #ifdef MAC
871 		mac_sysvshm_init(&shmsegs[i]);
872 #endif
873 	}
874 	shm_last_free = 0;
875 	shm_nused = 0;
876 	shm_committed = 0;
877 	sx_init(&sysvshmsx, "sysvshmsx");
878 	shmexit_hook = &shmexit_myhook;
879 	shmfork_hook = &shmfork_myhook;
880 
881 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
882 	if (error != 0)
883 		return (error);
884 #ifdef COMPAT_FREEBSD32
885 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
886 	if (error != 0)
887 		return (error);
888 #endif
889 	return (0);
890 }
891 
892 static int
893 shmunload(void)
894 {
895 	int i;
896 
897 	if (shm_nused > 0)
898 		return (EBUSY);
899 
900 #ifdef COMPAT_FREEBSD32
901 	syscall32_helper_unregister(shm32_syscalls);
902 #endif
903 	syscall_helper_unregister(shm_syscalls);
904 
905 	for (i = 0; i < shmalloced; i++) {
906 #ifdef MAC
907 		mac_sysvshm_destroy(&shmsegs[i]);
908 #endif
909 		/*
910 		 * Objects might be still mapped into the processes
911 		 * address spaces.  Actual free would happen on the
912 		 * last mapping destruction.
913 		 */
914 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
915 			vm_object_deallocate(shmsegs[i].object);
916 	}
917 	free(shmsegs, M_SHM);
918 	shmexit_hook = NULL;
919 	shmfork_hook = NULL;
920 	sx_destroy(&sysvshmsx);
921 	return (0);
922 }
923 
924 static int
925 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
926 {
927 	int error;
928 
929 	SYSVSHM_LOCK();
930 	error = SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0]));
931 	SYSVSHM_UNLOCK();
932 	return (error);
933 }
934 
935 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
936 struct oshmid_ds {
937 	struct	ipc_perm_old shm_perm;	/* operation perms */
938 	int	shm_segsz;		/* size of segment (bytes) */
939 	u_short	shm_cpid;		/* pid, creator */
940 	u_short	shm_lpid;		/* pid, last operation */
941 	short	shm_nattch;		/* no. of current attaches */
942 	time_t	shm_atime;		/* last attach time */
943 	time_t	shm_dtime;		/* last detach time */
944 	time_t	shm_ctime;		/* last change time */
945 	void	*shm_handle;		/* internal handle for shm segment */
946 };
947 
948 struct oshmctl_args {
949 	int shmid;
950 	int cmd;
951 	struct oshmid_ds *ubuf;
952 };
953 
954 static int
955 oshmctl(struct thread *td, struct oshmctl_args *uap)
956 {
957 #ifdef COMPAT_43
958 	int error = 0;
959 	struct shmid_kernel *shmseg;
960 	struct oshmid_ds outbuf;
961 
962 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
963 		return (ENOSYS);
964 	if (uap->cmd != IPC_STAT) {
965 		return (freebsd7_shmctl(td,
966 		    (struct freebsd7_shmctl_args *)uap));
967 	}
968 	SYSVSHM_LOCK();
969 	shmseg = shm_find_segment(uap->shmid, true);
970 	if (shmseg == NULL) {
971 		SYSVSHM_UNLOCK();
972 		return (EINVAL);
973 	}
974 	error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
975 	if (error != 0) {
976 		SYSVSHM_UNLOCK();
977 		return (error);
978 	}
979 #ifdef MAC
980 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
981 	if (error != 0) {
982 		SYSVSHM_UNLOCK();
983 		return (error);
984 	}
985 #endif
986 	ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
987 	outbuf.shm_segsz = shmseg->u.shm_segsz;
988 	outbuf.shm_cpid = shmseg->u.shm_cpid;
989 	outbuf.shm_lpid = shmseg->u.shm_lpid;
990 	outbuf.shm_nattch = shmseg->u.shm_nattch;
991 	outbuf.shm_atime = shmseg->u.shm_atime;
992 	outbuf.shm_dtime = shmseg->u.shm_dtime;
993 	outbuf.shm_ctime = shmseg->u.shm_ctime;
994 	outbuf.shm_handle = shmseg->object;
995 	SYSVSHM_UNLOCK();
996 	error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
997 	return (error);
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 	int error;
1023 
1024 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1025 		return (ENOSYS);
1026 	if (uap->which < 0 || uap->which >= nitems(shmcalls))
1027 		return (EINVAL);
1028 	error = (*shmcalls[uap->which])(td, &uap->a2);
1029 	return (error);
1030 }
1031 
1032 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1033 
1034 #ifdef COMPAT_FREEBSD32
1035 
1036 int
1037 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1038 {
1039 
1040 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1041     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1042 	switch (uap->which) {
1043 	case 0:	{	/* shmat */
1044 		struct shmat_args ap;
1045 
1046 		ap.shmid = uap->a2;
1047 		ap.shmaddr = PTRIN(uap->a3);
1048 		ap.shmflg = uap->a4;
1049 		return (sysent[SYS_shmat].sy_call(td, &ap));
1050 	}
1051 	case 2: {	/* shmdt */
1052 		struct shmdt_args ap;
1053 
1054 		ap.shmaddr = PTRIN(uap->a2);
1055 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1056 	}
1057 	case 3: {	/* shmget */
1058 		struct shmget_args ap;
1059 
1060 		ap.key = uap->a2;
1061 		ap.size = uap->a3;
1062 		ap.shmflg = uap->a4;
1063 		return (sysent[SYS_shmget].sy_call(td, &ap));
1064 	}
1065 	case 4: {	/* shmctl */
1066 		struct freebsd7_freebsd32_shmctl_args ap;
1067 
1068 		ap.shmid = uap->a2;
1069 		ap.cmd = uap->a3;
1070 		ap.buf = PTRIN(uap->a4);
1071 		return (freebsd7_freebsd32_shmctl(td, &ap));
1072 	}
1073 	case 1:		/* oshmctl */
1074 	default:
1075 		return (EINVAL);
1076 	}
1077 #else
1078 	return (nosys(td, NULL));
1079 #endif
1080 }
1081 
1082 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1083     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1084 int
1085 freebsd7_freebsd32_shmctl(struct thread *td,
1086     struct freebsd7_freebsd32_shmctl_args *uap)
1087 {
1088 	int error = 0;
1089 	union {
1090 		struct shmid_ds shmid_ds;
1091 		struct shm_info shm_info;
1092 		struct shminfo shminfo;
1093 	} u;
1094 	union {
1095 		struct shmid_ds32_old shmid_ds32;
1096 		struct shm_info32 shm_info32;
1097 		struct shminfo32 shminfo32;
1098 	} u32;
1099 	size_t sz;
1100 
1101 	if (uap->cmd == IPC_SET) {
1102 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1103 		    sizeof(u32.shmid_ds32))))
1104 			goto done;
1105 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1106 		    &u.shmid_ds.shm_perm);
1107 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1108 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1109 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1110 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1111 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1112 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1113 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1114 	}
1115 
1116 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1117 	if (error)
1118 		goto done;
1119 
1120 	/* Cases in which we need to copyout */
1121 	switch (uap->cmd) {
1122 	case IPC_INFO:
1123 		CP(u.shminfo, u32.shminfo32, shmmax);
1124 		CP(u.shminfo, u32.shminfo32, shmmin);
1125 		CP(u.shminfo, u32.shminfo32, shmmni);
1126 		CP(u.shminfo, u32.shminfo32, shmseg);
1127 		CP(u.shminfo, u32.shminfo32, shmall);
1128 		error = copyout(&u32.shminfo32, uap->buf,
1129 		    sizeof(u32.shminfo32));
1130 		break;
1131 	case SHM_INFO:
1132 		CP(u.shm_info, u32.shm_info32, used_ids);
1133 		CP(u.shm_info, u32.shm_info32, shm_rss);
1134 		CP(u.shm_info, u32.shm_info32, shm_tot);
1135 		CP(u.shm_info, u32.shm_info32, shm_swp);
1136 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1137 		CP(u.shm_info, u32.shm_info32, swap_successes);
1138 		error = copyout(&u32.shm_info32, uap->buf,
1139 		    sizeof(u32.shm_info32));
1140 		break;
1141 	case SHM_STAT:
1142 	case IPC_STAT:
1143 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1144 		    &u32.shmid_ds32.shm_perm);
1145 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1146 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1147 		else
1148 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1149 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1150 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1151 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1152 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1153 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1154 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1155 		u32.shmid_ds32.shm_internal = 0;
1156 		error = copyout(&u32.shmid_ds32, uap->buf,
1157 		    sizeof(u32.shmid_ds32));
1158 		break;
1159 	}
1160 
1161 done:
1162 	if (error) {
1163 		/* Invalidate the return value */
1164 		td->td_retval[0] = -1;
1165 	}
1166 	return (error);
1167 }
1168 #endif
1169 
1170 int
1171 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1172 {
1173 	int error = 0;
1174 	union {
1175 		struct shmid_ds shmid_ds;
1176 		struct shm_info shm_info;
1177 		struct shminfo shminfo;
1178 	} u;
1179 	union {
1180 		struct shmid_ds32 shmid_ds32;
1181 		struct shm_info32 shm_info32;
1182 		struct shminfo32 shminfo32;
1183 	} u32;
1184 	size_t sz;
1185 
1186 	if (uap->cmd == IPC_SET) {
1187 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1188 		    sizeof(u32.shmid_ds32))))
1189 			goto done;
1190 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1191 		    &u.shmid_ds.shm_perm);
1192 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1193 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1194 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1195 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1196 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1197 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1198 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1199 	}
1200 
1201 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1202 	if (error)
1203 		goto done;
1204 
1205 	/* Cases in which we need to copyout */
1206 	switch (uap->cmd) {
1207 	case IPC_INFO:
1208 		CP(u.shminfo, u32.shminfo32, shmmax);
1209 		CP(u.shminfo, u32.shminfo32, shmmin);
1210 		CP(u.shminfo, u32.shminfo32, shmmni);
1211 		CP(u.shminfo, u32.shminfo32, shmseg);
1212 		CP(u.shminfo, u32.shminfo32, shmall);
1213 		error = copyout(&u32.shminfo32, uap->buf,
1214 		    sizeof(u32.shminfo32));
1215 		break;
1216 	case SHM_INFO:
1217 		CP(u.shm_info, u32.shm_info32, used_ids);
1218 		CP(u.shm_info, u32.shm_info32, shm_rss);
1219 		CP(u.shm_info, u32.shm_info32, shm_tot);
1220 		CP(u.shm_info, u32.shm_info32, shm_swp);
1221 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1222 		CP(u.shm_info, u32.shm_info32, swap_successes);
1223 		error = copyout(&u32.shm_info32, uap->buf,
1224 		    sizeof(u32.shm_info32));
1225 		break;
1226 	case SHM_STAT:
1227 	case IPC_STAT:
1228 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1229 		    &u32.shmid_ds32.shm_perm);
1230 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1231 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1232 		else
1233 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1234 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1235 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1236 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1237 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1238 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1239 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1240 		error = copyout(&u32.shmid_ds32, uap->buf,
1241 		    sizeof(u32.shmid_ds32));
1242 		break;
1243 	}
1244 
1245 done:
1246 	if (error) {
1247 		/* Invalidate the return value */
1248 		td->td_retval[0] = -1;
1249 	}
1250 	return (error);
1251 }
1252 #endif
1253 
1254 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1255     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1256 
1257 #ifndef CP
1258 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1259 #endif
1260 
1261 #ifndef _SYS_SYSPROTO_H_
1262 struct freebsd7_shmctl_args {
1263 	int shmid;
1264 	int cmd;
1265 	struct shmid_ds_old *buf;
1266 };
1267 #endif
1268 int
1269 freebsd7_shmctl(struct thread *td, struct freebsd7_shmctl_args *uap)
1270 {
1271 	int error = 0;
1272 	struct shmid_ds_old old;
1273 	struct shmid_ds buf;
1274 	size_t bufsz;
1275 
1276 	/*
1277 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1278 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1279 	 * return an error back to the user since we do not to support this.
1280 	 */
1281 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1282 	    uap->cmd == SHM_STAT)
1283 		return (EINVAL);
1284 
1285 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1286 	if (uap->cmd == IPC_SET) {
1287 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1288 			goto done;
1289 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1290 		CP(old, buf, shm_segsz);
1291 		CP(old, buf, shm_lpid);
1292 		CP(old, buf, shm_cpid);
1293 		CP(old, buf, shm_nattch);
1294 		CP(old, buf, shm_atime);
1295 		CP(old, buf, shm_dtime);
1296 		CP(old, buf, shm_ctime);
1297 	}
1298 
1299 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1300 	if (error)
1301 		goto done;
1302 
1303 	/* Cases in which we need to copyout */
1304 	switch (uap->cmd) {
1305 	case IPC_STAT:
1306 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1307 		if (buf.shm_segsz > INT_MAX)
1308 			old.shm_segsz = INT_MAX;
1309 		else
1310 			CP(buf, old, shm_segsz);
1311 		CP(buf, old, shm_lpid);
1312 		CP(buf, old, shm_cpid);
1313 		if (buf.shm_nattch > SHRT_MAX)
1314 			old.shm_nattch = SHRT_MAX;
1315 		else
1316 			CP(buf, old, shm_nattch);
1317 		CP(buf, old, shm_atime);
1318 		CP(buf, old, shm_dtime);
1319 		CP(buf, old, shm_ctime);
1320 		old.shm_internal = NULL;
1321 		error = copyout(&old, uap->buf, sizeof(old));
1322 		break;
1323 	}
1324 
1325 done:
1326 	if (error) {
1327 		/* Invalidate the return value */
1328 		td->td_retval[0] = -1;
1329 	}
1330 	return (error);
1331 }
1332 
1333 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1334 	   COMPAT_FREEBSD7 */
1335 
1336 static int
1337 sysvshm_modload(struct module *module, int cmd, void *arg)
1338 {
1339 	int error = 0;
1340 
1341 	switch (cmd) {
1342 	case MOD_LOAD:
1343 		error = shminit();
1344 		if (error != 0)
1345 			shmunload();
1346 		break;
1347 	case MOD_UNLOAD:
1348 		error = shmunload();
1349 		break;
1350 	case MOD_SHUTDOWN:
1351 		break;
1352 	default:
1353 		error = EINVAL;
1354 		break;
1355 	}
1356 	return (error);
1357 }
1358 
1359 static moduledata_t sysvshm_mod = {
1360 	"sysvshm",
1361 	&sysvshm_modload,
1362 	NULL
1363 };
1364 
1365 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1366 MODULE_VERSION(sysvshm, 1);
1367