xref: /freebsd/sys/kern/sysv_shm.c (revision 0761549550dad58bfcc0f227ba6ffa78b4d8c017)
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 #define	SHMSEG_WANTED		0x1000
113 
114 static int shm_last_free, shm_nused, shmalloced;
115 vm_size_t shm_committed;
116 static struct shmid_kernel	*shmsegs;
117 
118 struct shmmap_state {
119 	vm_offset_t va;
120 	int shmid;
121 };
122 
123 static void shm_deallocate_segment(struct shmid_kernel *);
124 static int shm_find_segment_by_key(key_t);
125 static struct shmid_kernel *shm_find_segment_by_shmid(int);
126 static struct shmid_kernel *shm_find_segment_by_shmidx(int);
127 static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *);
128 static void shmrealloc(void);
129 static int shminit(void);
130 static int sysvshm_modload(struct module *, int, void *);
131 static int shmunload(void);
132 static void shmexit_myhook(struct vmspace *vm);
133 static void shmfork_myhook(struct proc *p1, struct proc *p2);
134 static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
135 
136 /*
137  * Tuneable values.
138  */
139 #ifndef SHMMAXPGS
140 #define	SHMMAXPGS	131072	/* Note: sysv shared memory is swap backed. */
141 #endif
142 #ifndef SHMMAX
143 #define	SHMMAX	(SHMMAXPGS*PAGE_SIZE)
144 #endif
145 #ifndef SHMMIN
146 #define	SHMMIN	1
147 #endif
148 #ifndef SHMMNI
149 #define	SHMMNI	192
150 #endif
151 #ifndef SHMSEG
152 #define	SHMSEG	128
153 #endif
154 #ifndef SHMALL
155 #define	SHMALL	(SHMMAXPGS)
156 #endif
157 
158 struct	shminfo shminfo = {
159 	.shmmax = SHMMAX,
160 	.shmmin = SHMMIN,
161 	.shmmni = SHMMNI,
162 	.shmseg = SHMSEG,
163 	.shmall = SHMALL
164 };
165 
166 static int shm_use_phys;
167 static int shm_allow_removed;
168 
169 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RWTUN, &shminfo.shmmax, 0,
170     "Maximum shared memory segment size");
171 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RWTUN, &shminfo.shmmin, 0,
172     "Minimum shared memory segment size");
173 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0,
174     "Number of shared memory identifiers");
175 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0,
176     "Number of segments per process");
177 SYSCTL_ULONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RWTUN, &shminfo.shmall, 0,
178     "Maximum number of pages available for shared memory");
179 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RWTUN,
180     &shm_use_phys, 0, "Enable/Disable locking of shared memory pages in core");
181 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RWTUN,
182     &shm_allow_removed, 0,
183     "Enable/Disable attachment to attached segments marked for removal");
184 SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLTYPE_OPAQUE | CTLFLAG_RD,
185     NULL, 0, sysctl_shmsegs, "",
186     "Current number of shared memory segments allocated");
187 
188 static int
189 shm_find_segment_by_key(key)
190 	key_t key;
191 {
192 	int i;
193 
194 	for (i = 0; i < shmalloced; i++)
195 		if ((shmsegs[i].u.shm_perm.mode & SHMSEG_ALLOCATED) &&
196 		    shmsegs[i].u.shm_perm.key == key)
197 			return (i);
198 	return (-1);
199 }
200 
201 static struct shmid_kernel *
202 shm_find_segment_by_shmid(int shmid)
203 {
204 	int segnum;
205 	struct shmid_kernel *shmseg;
206 
207 	segnum = IPCID_TO_IX(shmid);
208 	if (segnum < 0 || segnum >= shmalloced)
209 		return (NULL);
210 	shmseg = &shmsegs[segnum];
211 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
212 	    (!shm_allow_removed &&
213 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0) ||
214 	    shmseg->u.shm_perm.seq != IPCID_TO_SEQ(shmid))
215 		return (NULL);
216 	return (shmseg);
217 }
218 
219 static struct shmid_kernel *
220 shm_find_segment_by_shmidx(int segnum)
221 {
222 	struct shmid_kernel *shmseg;
223 
224 	if (segnum < 0 || segnum >= shmalloced)
225 		return (NULL);
226 	shmseg = &shmsegs[segnum];
227 	if ((shmseg->u.shm_perm.mode & SHMSEG_ALLOCATED) == 0 ||
228 	    (!shm_allow_removed &&
229 	     (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) != 0))
230 		return (NULL);
231 	return (shmseg);
232 }
233 
234 static void
235 shm_deallocate_segment(shmseg)
236 	struct shmid_kernel *shmseg;
237 {
238 	vm_size_t size;
239 
240 	GIANT_REQUIRED;
241 
242 	vm_object_deallocate(shmseg->object);
243 	shmseg->object = NULL;
244 	size = round_page(shmseg->u.shm_segsz);
245 	shm_committed -= btoc(size);
246 	shm_nused--;
247 	shmseg->u.shm_perm.mode = SHMSEG_FREE;
248 #ifdef MAC
249 	mac_sysvshm_cleanup(shmseg);
250 #endif
251 	racct_sub_cred(shmseg->cred, RACCT_NSHM, 1);
252 	racct_sub_cred(shmseg->cred, RACCT_SHMSIZE, size);
253 	crfree(shmseg->cred);
254 	shmseg->cred = NULL;
255 }
256 
257 static int
258 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
259 {
260 	struct shmid_kernel *shmseg;
261 	int segnum, result;
262 	vm_size_t size;
263 
264 	GIANT_REQUIRED;
265 
266 	segnum = IPCID_TO_IX(shmmap_s->shmid);
267 	shmseg = &shmsegs[segnum];
268 	size = round_page(shmseg->u.shm_segsz);
269 	result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
270 	if (result != KERN_SUCCESS)
271 		return (EINVAL);
272 	shmmap_s->shmid = -1;
273 	shmseg->u.shm_dtime = time_second;
274 	if ((--shmseg->u.shm_nattch <= 0) &&
275 	    (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) {
276 		shm_deallocate_segment(shmseg);
277 		shm_last_free = segnum;
278 	}
279 	return (0);
280 }
281 
282 #ifndef _SYS_SYSPROTO_H_
283 struct shmdt_args {
284 	const void *shmaddr;
285 };
286 #endif
287 int
288 sys_shmdt(td, uap)
289 	struct thread *td;
290 	struct shmdt_args *uap;
291 {
292 	struct proc *p = td->td_proc;
293 	struct shmmap_state *shmmap_s;
294 #ifdef MAC
295 	struct shmid_kernel *shmsegptr;
296 #endif
297 	int i;
298 	int error = 0;
299 
300 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
301 		return (ENOSYS);
302 	mtx_lock(&Giant);
303 	shmmap_s = p->p_vmspace->vm_shm;
304  	if (shmmap_s == NULL) {
305 		error = EINVAL;
306 		goto done2;
307 	}
308 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
309 		if (shmmap_s->shmid != -1 &&
310 		    shmmap_s->va == (vm_offset_t)uap->shmaddr) {
311 			break;
312 		}
313 	}
314 	if (i == shminfo.shmseg) {
315 		error = EINVAL;
316 		goto done2;
317 	}
318 #ifdef MAC
319 	shmsegptr = &shmsegs[IPCID_TO_IX(shmmap_s->shmid)];
320 	error = mac_sysvshm_check_shmdt(td->td_ucred, shmsegptr);
321 	if (error != 0)
322 		goto done2;
323 #endif
324 	error = shm_delete_mapping(p->p_vmspace, shmmap_s);
325 done2:
326 	mtx_unlock(&Giant);
327 	return (error);
328 }
329 
330 #ifndef _SYS_SYSPROTO_H_
331 struct shmat_args {
332 	int shmid;
333 	const void *shmaddr;
334 	int shmflg;
335 };
336 #endif
337 int
338 kern_shmat(td, shmid, shmaddr, shmflg)
339 	struct thread *td;
340 	int shmid;
341 	const void *shmaddr;
342 	int shmflg;
343 {
344 	struct proc *p = td->td_proc;
345 	int i;
346 	struct shmid_kernel *shmseg;
347 	struct shmmap_state *shmmap_s = NULL;
348 	vm_offset_t attach_va;
349 	vm_prot_t prot;
350 	vm_size_t size;
351 	int rv;
352 	int error = 0;
353 
354 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
355 		return (ENOSYS);
356 	mtx_lock(&Giant);
357 	shmmap_s = p->p_vmspace->vm_shm;
358 	if (shmmap_s == NULL) {
359 		shmmap_s = malloc(shminfo.shmseg * sizeof(struct shmmap_state),
360 		    M_SHM, M_WAITOK);
361 		for (i = 0; i < shminfo.shmseg; i++)
362 			shmmap_s[i].shmid = -1;
363 		p->p_vmspace->vm_shm = shmmap_s;
364 	}
365 	shmseg = shm_find_segment_by_shmid(shmid);
366 	if (shmseg == NULL) {
367 		error = EINVAL;
368 		goto done2;
369 	}
370 	error = ipcperm(td, &shmseg->u.shm_perm,
371 	    (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
372 	if (error)
373 		goto done2;
374 #ifdef MAC
375 	error = mac_sysvshm_check_shmat(td->td_ucred, shmseg, shmflg);
376 	if (error != 0)
377 		goto done2;
378 #endif
379 	for (i = 0; i < shminfo.shmseg; i++) {
380 		if (shmmap_s->shmid == -1)
381 			break;
382 		shmmap_s++;
383 	}
384 	if (i >= shminfo.shmseg) {
385 		error = EMFILE;
386 		goto done2;
387 	}
388 	size = round_page(shmseg->u.shm_segsz);
389 	prot = VM_PROT_READ;
390 	if ((shmflg & SHM_RDONLY) == 0)
391 		prot |= VM_PROT_WRITE;
392 	if (shmaddr) {
393 		if (shmflg & SHM_RND) {
394 			attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1);
395 		} else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) {
396 			attach_va = (vm_offset_t)shmaddr;
397 		} else {
398 			error = EINVAL;
399 			goto done2;
400 		}
401 	} else {
402 		/*
403 		 * This is just a hint to vm_map_find() about where to
404 		 * put it.
405 		 */
406 		PROC_LOCK(p);
407 		attach_va = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
408 		    lim_max(p, RLIMIT_DATA));
409 		PROC_UNLOCK(p);
410 	}
411 
412 	vm_object_reference(shmseg->object);
413 	rv = vm_map_find(&p->p_vmspace->vm_map, shmseg->object, 0, &attach_va,
414 	    size, 0, shmaddr != NULL ? VMFS_NO_SPACE : VMFS_OPTIMAL_SPACE,
415 	    prot, prot, MAP_INHERIT_SHARE | MAP_PREFAULT_PARTIAL);
416 	if (rv != KERN_SUCCESS) {
417 		vm_object_deallocate(shmseg->object);
418 		error = ENOMEM;
419 		goto done2;
420 	}
421 
422 	shmmap_s->va = attach_va;
423 	shmmap_s->shmid = shmid;
424 	shmseg->u.shm_lpid = p->p_pid;
425 	shmseg->u.shm_atime = time_second;
426 	shmseg->u.shm_nattch++;
427 	td->td_retval[0] = attach_va;
428 done2:
429 	mtx_unlock(&Giant);
430 	return (error);
431 }
432 
433 int
434 sys_shmat(td, uap)
435 	struct thread *td;
436 	struct shmat_args *uap;
437 {
438 	return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg);
439 }
440 
441 int
442 kern_shmctl(td, shmid, cmd, buf, bufsz)
443 	struct thread *td;
444 	int shmid;
445 	int cmd;
446 	void *buf;
447 	size_t *bufsz;
448 {
449 	int error = 0;
450 	struct shmid_kernel *shmseg;
451 
452 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
453 		return (ENOSYS);
454 
455 	mtx_lock(&Giant);
456 	switch (cmd) {
457 	/*
458 	 * It is possible that kern_shmctl is being called from the Linux ABI
459 	 * layer, in which case, we will need to implement IPC_INFO.  It should
460 	 * be noted that other shmctl calls will be funneled through here for
461 	 * Linix binaries as well.
462 	 *
463 	 * NB: The Linux ABI layer will convert this data to structure(s) more
464 	 * consistent with the Linux ABI.
465 	 */
466 	case IPC_INFO:
467 		memcpy(buf, &shminfo, sizeof(shminfo));
468 		if (bufsz)
469 			*bufsz = sizeof(shminfo);
470 		td->td_retval[0] = shmalloced;
471 		goto done2;
472 	case SHM_INFO: {
473 		struct shm_info shm_info;
474 		shm_info.used_ids = shm_nused;
475 		shm_info.shm_rss = 0;	/*XXX where to get from ? */
476 		shm_info.shm_tot = 0;	/*XXX where to get from ? */
477 		shm_info.shm_swp = 0;	/*XXX where to get from ? */
478 		shm_info.swap_attempts = 0;	/*XXX where to get from ? */
479 		shm_info.swap_successes = 0;	/*XXX where to get from ? */
480 		memcpy(buf, &shm_info, sizeof(shm_info));
481 		if (bufsz)
482 			*bufsz = sizeof(shm_info);
483 		td->td_retval[0] = shmalloced;
484 		goto done2;
485 	}
486 	}
487 	if (cmd == SHM_STAT)
488 		shmseg = shm_find_segment_by_shmidx(shmid);
489 	else
490 		shmseg = shm_find_segment_by_shmid(shmid);
491 	if (shmseg == NULL) {
492 		error = EINVAL;
493 		goto done2;
494 	}
495 #ifdef MAC
496 	error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, cmd);
497 	if (error != 0)
498 		goto done2;
499 #endif
500 	switch (cmd) {
501 	case SHM_STAT:
502 	case IPC_STAT:
503 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
504 		if (error)
505 			goto done2;
506 		memcpy(buf, &shmseg->u, sizeof(struct shmid_ds));
507 		if (bufsz)
508 			*bufsz = sizeof(struct shmid_ds);
509 		if (cmd == SHM_STAT)
510 			td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->u.shm_perm);
511 		break;
512 	case IPC_SET: {
513 		struct shmid_ds *shmid;
514 
515 		shmid = (struct shmid_ds *)buf;
516 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
517 		if (error)
518 			goto done2;
519 		shmseg->u.shm_perm.uid = shmid->shm_perm.uid;
520 		shmseg->u.shm_perm.gid = shmid->shm_perm.gid;
521 		shmseg->u.shm_perm.mode =
522 		    (shmseg->u.shm_perm.mode & ~ACCESSPERMS) |
523 		    (shmid->shm_perm.mode & ACCESSPERMS);
524 		shmseg->u.shm_ctime = time_second;
525 		break;
526 	}
527 	case IPC_RMID:
528 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_M);
529 		if (error)
530 			goto done2;
531 		shmseg->u.shm_perm.key = IPC_PRIVATE;
532 		shmseg->u.shm_perm.mode |= SHMSEG_REMOVED;
533 		if (shmseg->u.shm_nattch <= 0) {
534 			shm_deallocate_segment(shmseg);
535 			shm_last_free = IPCID_TO_IX(shmid);
536 		}
537 		break;
538 #if 0
539 	case SHM_LOCK:
540 	case SHM_UNLOCK:
541 #endif
542 	default:
543 		error = EINVAL;
544 		break;
545 	}
546 done2:
547 	mtx_unlock(&Giant);
548 	return (error);
549 }
550 
551 #ifndef _SYS_SYSPROTO_H_
552 struct shmctl_args {
553 	int shmid;
554 	int cmd;
555 	struct shmid_ds *buf;
556 };
557 #endif
558 int
559 sys_shmctl(td, uap)
560 	struct thread *td;
561 	struct shmctl_args *uap;
562 {
563 	int error = 0;
564 	struct shmid_ds buf;
565 	size_t bufsz;
566 
567 	/*
568 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
569 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
570 	 * return an error back to the user since we do not to support this.
571 	 */
572 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
573 	    uap->cmd == SHM_STAT)
574 		return (EINVAL);
575 
576 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
577 	if (uap->cmd == IPC_SET) {
578 		if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds))))
579 			goto done;
580 	}
581 
582 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
583 	if (error)
584 		goto done;
585 
586 	/* Cases in which we need to copyout */
587 	switch (uap->cmd) {
588 	case IPC_STAT:
589 		error = copyout(&buf, uap->buf, bufsz);
590 		break;
591 	}
592 
593 done:
594 	if (error) {
595 		/* Invalidate the return value */
596 		td->td_retval[0] = -1;
597 	}
598 	return (error);
599 }
600 
601 
602 static int
603 shmget_existing(td, uap, mode, segnum)
604 	struct thread *td;
605 	struct shmget_args *uap;
606 	int mode;
607 	int segnum;
608 {
609 	struct shmid_kernel *shmseg;
610 	int error;
611 
612 	shmseg = &shmsegs[segnum];
613 	if (shmseg->u.shm_perm.mode & SHMSEG_REMOVED) {
614 		/*
615 		 * This segment is in the process of being allocated.  Wait
616 		 * until it's done, and look the key up again (in case the
617 		 * allocation failed or it was freed).
618 		 */
619 		shmseg->u.shm_perm.mode |= SHMSEG_WANTED;
620 		error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
621 		if (error)
622 			return (error);
623 		return (EAGAIN);
624 	}
625 	if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
626 		return (EEXIST);
627 #ifdef MAC
628 	error = mac_sysvshm_check_shmget(td->td_ucred, shmseg, uap->shmflg);
629 	if (error != 0)
630 		return (error);
631 #endif
632 	if (uap->size != 0 && uap->size > shmseg->u.shm_segsz)
633 		return (EINVAL);
634 	td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
635 	return (0);
636 }
637 
638 static int
639 shmget_allocate_segment(td, uap, mode)
640 	struct thread *td;
641 	struct shmget_args *uap;
642 	int mode;
643 {
644 	int i, segnum, shmid;
645 	size_t size;
646 	struct ucred *cred = td->td_ucred;
647 	struct shmid_kernel *shmseg;
648 	vm_object_t shm_object;
649 
650 	GIANT_REQUIRED;
651 
652 	if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
653 		return (EINVAL);
654 	if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
655 		return (ENOSPC);
656 	size = round_page(uap->size);
657 	if (shm_committed + btoc(size) > shminfo.shmall)
658 		return (ENOMEM);
659 	if (shm_last_free < 0) {
660 		shmrealloc();	/* Maybe expand the shmsegs[] array. */
661 		for (i = 0; i < shmalloced; i++)
662 			if (shmsegs[i].u.shm_perm.mode & SHMSEG_FREE)
663 				break;
664 		if (i == shmalloced)
665 			return (ENOSPC);
666 		segnum = i;
667 	} else  {
668 		segnum = shm_last_free;
669 		shm_last_free = -1;
670 	}
671 	shmseg = &shmsegs[segnum];
672 #ifdef RACCT
673 	PROC_LOCK(td->td_proc);
674 	if (racct_add(td->td_proc, RACCT_NSHM, 1)) {
675 		PROC_UNLOCK(td->td_proc);
676 		return (ENOSPC);
677 	}
678 	if (racct_add(td->td_proc, RACCT_SHMSIZE, size)) {
679 		racct_sub(td->td_proc, RACCT_NSHM, 1);
680 		PROC_UNLOCK(td->td_proc);
681 		return (ENOMEM);
682 	}
683 	PROC_UNLOCK(td->td_proc);
684 #endif
685 	/*
686 	 * In case we sleep in malloc(), mark the segment present but deleted
687 	 * so that noone else tries to create the same key.
688 	 */
689 	shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
690 	shmseg->u.shm_perm.key = uap->key;
691 	shmseg->u.shm_perm.seq = (shmseg->u.shm_perm.seq + 1) & 0x7fff;
692 	shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm);
693 
694 	/*
695 	 * We make sure that we have allocated a pager before we need
696 	 * to.
697 	 */
698 	shm_object = vm_pager_allocate(shm_use_phys ? OBJT_PHYS : OBJT_SWAP,
699 	    0, size, VM_PROT_DEFAULT, 0, cred);
700 	if (shm_object == NULL) {
701 #ifdef RACCT
702 		PROC_LOCK(td->td_proc);
703 		racct_sub(td->td_proc, RACCT_NSHM, 1);
704 		racct_sub(td->td_proc, RACCT_SHMSIZE, size);
705 		PROC_UNLOCK(td->td_proc);
706 #endif
707 		return (ENOMEM);
708 	}
709 	shm_object->pg_color = 0;
710 	VM_OBJECT_WLOCK(shm_object);
711 	vm_object_clear_flag(shm_object, OBJ_ONEMAPPING);
712 	vm_object_set_flag(shm_object, OBJ_COLORED | OBJ_NOSPLIT);
713 	VM_OBJECT_WUNLOCK(shm_object);
714 
715 	shmseg->object = shm_object;
716 	shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = cred->cr_uid;
717 	shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid;
718 	shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) |
719 	    (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
720 	shmseg->cred = crhold(cred);
721 	shmseg->u.shm_segsz = uap->size;
722 	shmseg->u.shm_cpid = td->td_proc->p_pid;
723 	shmseg->u.shm_lpid = shmseg->u.shm_nattch = 0;
724 	shmseg->u.shm_atime = shmseg->u.shm_dtime = 0;
725 #ifdef MAC
726 	mac_sysvshm_create(cred, shmseg);
727 #endif
728 	shmseg->u.shm_ctime = time_second;
729 	shm_committed += btoc(size);
730 	shm_nused++;
731 	if (shmseg->u.shm_perm.mode & SHMSEG_WANTED) {
732 		/*
733 		 * Somebody else wanted this key while we were asleep.  Wake
734 		 * them up now.
735 		 */
736 		shmseg->u.shm_perm.mode &= ~SHMSEG_WANTED;
737 		wakeup(shmseg);
738 	}
739 	td->td_retval[0] = shmid;
740 	return (0);
741 }
742 
743 #ifndef _SYS_SYSPROTO_H_
744 struct shmget_args {
745 	key_t key;
746 	size_t size;
747 	int shmflg;
748 };
749 #endif
750 int
751 sys_shmget(td, uap)
752 	struct thread *td;
753 	struct shmget_args *uap;
754 {
755 	int segnum, mode;
756 	int error;
757 
758 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
759 		return (ENOSYS);
760 	mtx_lock(&Giant);
761 	mode = uap->shmflg & ACCESSPERMS;
762 	if (uap->key != IPC_PRIVATE) {
763 	again:
764 		segnum = shm_find_segment_by_key(uap->key);
765 		if (segnum >= 0) {
766 			error = shmget_existing(td, uap, mode, segnum);
767 			if (error == EAGAIN)
768 				goto again;
769 			goto done2;
770 		}
771 		if ((uap->shmflg & IPC_CREAT) == 0) {
772 			error = ENOENT;
773 			goto done2;
774 		}
775 	}
776 	error = shmget_allocate_segment(td, uap, mode);
777 done2:
778 	mtx_unlock(&Giant);
779 	return (error);
780 }
781 
782 static void
783 shmfork_myhook(p1, p2)
784 	struct proc *p1, *p2;
785 {
786 	struct shmmap_state *shmmap_s;
787 	size_t size;
788 	int i;
789 
790 	mtx_lock(&Giant);
791 	size = shminfo.shmseg * sizeof(struct shmmap_state);
792 	shmmap_s = malloc(size, M_SHM, M_WAITOK);
793 	bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
794 	p2->p_vmspace->vm_shm = shmmap_s;
795 	for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
796 		if (shmmap_s->shmid != -1)
797 			shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++;
798 	mtx_unlock(&Giant);
799 }
800 
801 static void
802 shmexit_myhook(struct vmspace *vm)
803 {
804 	struct shmmap_state *base, *shm;
805 	int i;
806 
807 	if ((base = vm->vm_shm) != NULL) {
808 		vm->vm_shm = NULL;
809 		mtx_lock(&Giant);
810 		for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
811 			if (shm->shmid != -1)
812 				shm_delete_mapping(vm, shm);
813 		}
814 		mtx_unlock(&Giant);
815 		free(base, M_SHM);
816 	}
817 }
818 
819 static void
820 shmrealloc(void)
821 {
822 	int i;
823 	struct shmid_kernel *newsegs;
824 
825 	if (shmalloced >= shminfo.shmmni)
826 		return;
827 
828 	newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
829 	if (newsegs == NULL)
830 		return;
831 	for (i = 0; i < shmalloced; i++)
832 		bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
833 	for (; i < shminfo.shmmni; i++) {
834 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
835 		shmsegs[i].u.shm_perm.seq = 0;
836 #ifdef MAC
837 		mac_sysvshm_init(&shmsegs[i]);
838 #endif
839 	}
840 	free(shmsegs, M_SHM);
841 	shmsegs = newsegs;
842 	shmalloced = shminfo.shmmni;
843 }
844 
845 static struct syscall_helper_data shm_syscalls[] = {
846 	SYSCALL_INIT_HELPER(shmat),
847 	SYSCALL_INIT_HELPER(shmctl),
848 	SYSCALL_INIT_HELPER(shmdt),
849 	SYSCALL_INIT_HELPER(shmget),
850 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
851     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
852 	SYSCALL_INIT_HELPER_COMPAT(freebsd7_shmctl),
853 #endif
854 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
855 	SYSCALL_INIT_HELPER(shmsys),
856 #endif
857 	SYSCALL_INIT_LAST
858 };
859 
860 #ifdef COMPAT_FREEBSD32
861 #include <compat/freebsd32/freebsd32.h>
862 #include <compat/freebsd32/freebsd32_ipc.h>
863 #include <compat/freebsd32/freebsd32_proto.h>
864 #include <compat/freebsd32/freebsd32_signal.h>
865 #include <compat/freebsd32/freebsd32_syscall.h>
866 #include <compat/freebsd32/freebsd32_util.h>
867 
868 static struct syscall_helper_data shm32_syscalls[] = {
869 	SYSCALL32_INIT_HELPER_COMPAT(shmat),
870 	SYSCALL32_INIT_HELPER_COMPAT(shmdt),
871 	SYSCALL32_INIT_HELPER_COMPAT(shmget),
872 	SYSCALL32_INIT_HELPER(freebsd32_shmsys),
873 	SYSCALL32_INIT_HELPER(freebsd32_shmctl),
874 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
875     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
876 	SYSCALL32_INIT_HELPER(freebsd7_freebsd32_shmctl),
877 #endif
878 	SYSCALL_INIT_LAST
879 };
880 #endif
881 
882 static int
883 shminit()
884 {
885 	int i, error;
886 
887 #ifndef BURN_BRIDGES
888 	if (TUNABLE_ULONG_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall) != 0)
889 		printf("kern.ipc.shmmaxpgs is now called kern.ipc.shmall!\n");
890 #endif
891 	if (shminfo.shmmax == SHMMAX) {
892 		/* Initialize shmmax dealing with possible overflow. */
893 		for (i = PAGE_SIZE; i != 0; i--) {
894 			shminfo.shmmax = shminfo.shmall * i;
895 			if ((shminfo.shmmax / shminfo.shmall) == (u_long)i)
896 				break;
897 		}
898 	}
899 	shmalloced = shminfo.shmmni;
900 	shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
901 	for (i = 0; i < shmalloced; i++) {
902 		shmsegs[i].u.shm_perm.mode = SHMSEG_FREE;
903 		shmsegs[i].u.shm_perm.seq = 0;
904 #ifdef MAC
905 		mac_sysvshm_init(&shmsegs[i]);
906 #endif
907 	}
908 	shm_last_free = 0;
909 	shm_nused = 0;
910 	shm_committed = 0;
911 	shmexit_hook = &shmexit_myhook;
912 	shmfork_hook = &shmfork_myhook;
913 
914 	error = syscall_helper_register(shm_syscalls, SY_THR_STATIC_KLD);
915 	if (error != 0)
916 		return (error);
917 #ifdef COMPAT_FREEBSD32
918 	error = syscall32_helper_register(shm32_syscalls, SY_THR_STATIC_KLD);
919 	if (error != 0)
920 		return (error);
921 #endif
922 	return (0);
923 }
924 
925 static int
926 shmunload()
927 {
928 	int i;
929 
930 	if (shm_nused > 0)
931 		return (EBUSY);
932 
933 #ifdef COMPAT_FREEBSD32
934 	syscall32_helper_unregister(shm32_syscalls);
935 #endif
936 	syscall_helper_unregister(shm_syscalls);
937 
938 	for (i = 0; i < shmalloced; i++) {
939 #ifdef MAC
940 		mac_sysvshm_destroy(&shmsegs[i]);
941 #endif
942 		/*
943 		 * Objects might be still mapped into the processes
944 		 * address spaces.  Actual free would happen on the
945 		 * last mapping destruction.
946 		 */
947 		if (shmsegs[i].u.shm_perm.mode != SHMSEG_FREE)
948 			vm_object_deallocate(shmsegs[i].object);
949 	}
950 	free(shmsegs, M_SHM);
951 	shmexit_hook = NULL;
952 	shmfork_hook = NULL;
953 	return (0);
954 }
955 
956 static int
957 sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
958 {
959 
960 	return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
961 }
962 
963 #if defined(__i386__) && (defined(COMPAT_FREEBSD4) || defined(COMPAT_43))
964 struct oshmid_ds {
965 	struct	ipc_perm_old shm_perm;	/* operation perms */
966 	int	shm_segsz;		/* size of segment (bytes) */
967 	u_short	shm_cpid;		/* pid, creator */
968 	u_short	shm_lpid;		/* pid, last operation */
969 	short	shm_nattch;		/* no. of current attaches */
970 	time_t	shm_atime;		/* last attach time */
971 	time_t	shm_dtime;		/* last detach time */
972 	time_t	shm_ctime;		/* last change time */
973 	void	*shm_handle;		/* internal handle for shm segment */
974 };
975 
976 struct oshmctl_args {
977 	int shmid;
978 	int cmd;
979 	struct oshmid_ds *ubuf;
980 };
981 
982 static int
983 oshmctl(struct thread *td, struct oshmctl_args *uap)
984 {
985 #ifdef COMPAT_43
986 	int error = 0;
987 	struct shmid_kernel *shmseg;
988 	struct oshmid_ds outbuf;
989 
990 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
991 		return (ENOSYS);
992 	mtx_lock(&Giant);
993 	shmseg = shm_find_segment_by_shmid(uap->shmid);
994 	if (shmseg == NULL) {
995 		error = EINVAL;
996 		goto done2;
997 	}
998 	switch (uap->cmd) {
999 	case IPC_STAT:
1000 		error = ipcperm(td, &shmseg->u.shm_perm, IPC_R);
1001 		if (error)
1002 			goto done2;
1003 #ifdef MAC
1004 		error = mac_sysvshm_check_shmctl(td->td_ucred, shmseg, uap->cmd);
1005 		if (error != 0)
1006 			goto done2;
1007 #endif
1008 		ipcperm_new2old(&shmseg->u.shm_perm, &outbuf.shm_perm);
1009 		outbuf.shm_segsz = shmseg->u.shm_segsz;
1010 		outbuf.shm_cpid = shmseg->u.shm_cpid;
1011 		outbuf.shm_lpid = shmseg->u.shm_lpid;
1012 		outbuf.shm_nattch = shmseg->u.shm_nattch;
1013 		outbuf.shm_atime = shmseg->u.shm_atime;
1014 		outbuf.shm_dtime = shmseg->u.shm_dtime;
1015 		outbuf.shm_ctime = shmseg->u.shm_ctime;
1016 		outbuf.shm_handle = shmseg->object;
1017 		error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
1018 		if (error)
1019 			goto done2;
1020 		break;
1021 	default:
1022 		error = freebsd7_shmctl(td, (struct freebsd7_shmctl_args *)uap);
1023 		break;
1024 	}
1025 done2:
1026 	mtx_unlock(&Giant);
1027 	return (error);
1028 #else
1029 	return (EINVAL);
1030 #endif
1031 }
1032 
1033 /* XXX casting to (sy_call_t *) is bogus, as usual. */
1034 static sy_call_t *shmcalls[] = {
1035 	(sy_call_t *)sys_shmat, (sy_call_t *)oshmctl,
1036 	(sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
1037 	(sy_call_t *)freebsd7_shmctl
1038 };
1039 
1040 int
1041 sys_shmsys(td, uap)
1042 	struct thread *td;
1043 	/* XXX actually varargs. */
1044 	struct shmsys_args /* {
1045 		int	which;
1046 		int	a2;
1047 		int	a3;
1048 		int	a4;
1049 	} */ *uap;
1050 {
1051 	int error;
1052 
1053 	if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC))
1054 		return (ENOSYS);
1055 	if (uap->which < 0 ||
1056 	    uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
1057 		return (EINVAL);
1058 	mtx_lock(&Giant);
1059 	error = (*shmcalls[uap->which])(td, &uap->a2);
1060 	mtx_unlock(&Giant);
1061 	return (error);
1062 }
1063 
1064 #endif	/* i386 && (COMPAT_FREEBSD4 || COMPAT_43) */
1065 
1066 #ifdef COMPAT_FREEBSD32
1067 
1068 int
1069 freebsd32_shmsys(struct thread *td, struct freebsd32_shmsys_args *uap)
1070 {
1071 
1072 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1073     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1074 	switch (uap->which) {
1075 	case 0:	{	/* shmat */
1076 		struct shmat_args ap;
1077 
1078 		ap.shmid = uap->a2;
1079 		ap.shmaddr = PTRIN(uap->a3);
1080 		ap.shmflg = uap->a4;
1081 		return (sysent[SYS_shmat].sy_call(td, &ap));
1082 	}
1083 	case 2: {	/* shmdt */
1084 		struct shmdt_args ap;
1085 
1086 		ap.shmaddr = PTRIN(uap->a2);
1087 		return (sysent[SYS_shmdt].sy_call(td, &ap));
1088 	}
1089 	case 3: {	/* shmget */
1090 		struct shmget_args ap;
1091 
1092 		ap.key = uap->a2;
1093 		ap.size = uap->a3;
1094 		ap.shmflg = uap->a4;
1095 		return (sysent[SYS_shmget].sy_call(td, &ap));
1096 	}
1097 	case 4: {	/* shmctl */
1098 		struct freebsd7_freebsd32_shmctl_args ap;
1099 
1100 		ap.shmid = uap->a2;
1101 		ap.cmd = uap->a3;
1102 		ap.buf = PTRIN(uap->a4);
1103 		return (freebsd7_freebsd32_shmctl(td, &ap));
1104 	}
1105 	case 1:		/* oshmctl */
1106 	default:
1107 		return (EINVAL);
1108 	}
1109 #else
1110 	return (nosys(td, NULL));
1111 #endif
1112 }
1113 
1114 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1115     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1116 int
1117 freebsd7_freebsd32_shmctl(struct thread *td,
1118     struct freebsd7_freebsd32_shmctl_args *uap)
1119 {
1120 	int error = 0;
1121 	union {
1122 		struct shmid_ds shmid_ds;
1123 		struct shm_info shm_info;
1124 		struct shminfo shminfo;
1125 	} u;
1126 	union {
1127 		struct shmid_ds32_old shmid_ds32;
1128 		struct shm_info32 shm_info32;
1129 		struct shminfo32 shminfo32;
1130 	} u32;
1131 	size_t sz;
1132 
1133 	if (uap->cmd == IPC_SET) {
1134 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1135 		    sizeof(u32.shmid_ds32))))
1136 			goto done;
1137 		freebsd32_ipcperm_old_in(&u32.shmid_ds32.shm_perm,
1138 		    &u.shmid_ds.shm_perm);
1139 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1140 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1141 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1142 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1143 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1144 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1145 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1146 	}
1147 
1148 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1149 	if (error)
1150 		goto done;
1151 
1152 	/* Cases in which we need to copyout */
1153 	switch (uap->cmd) {
1154 	case IPC_INFO:
1155 		CP(u.shminfo, u32.shminfo32, shmmax);
1156 		CP(u.shminfo, u32.shminfo32, shmmin);
1157 		CP(u.shminfo, u32.shminfo32, shmmni);
1158 		CP(u.shminfo, u32.shminfo32, shmseg);
1159 		CP(u.shminfo, u32.shminfo32, shmall);
1160 		error = copyout(&u32.shminfo32, uap->buf,
1161 		    sizeof(u32.shminfo32));
1162 		break;
1163 	case SHM_INFO:
1164 		CP(u.shm_info, u32.shm_info32, used_ids);
1165 		CP(u.shm_info, u32.shm_info32, shm_rss);
1166 		CP(u.shm_info, u32.shm_info32, shm_tot);
1167 		CP(u.shm_info, u32.shm_info32, shm_swp);
1168 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1169 		CP(u.shm_info, u32.shm_info32, swap_successes);
1170 		error = copyout(&u32.shm_info32, uap->buf,
1171 		    sizeof(u32.shm_info32));
1172 		break;
1173 	case SHM_STAT:
1174 	case IPC_STAT:
1175 		freebsd32_ipcperm_old_out(&u.shmid_ds.shm_perm,
1176 		    &u32.shmid_ds32.shm_perm);
1177 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1178 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1179 		else
1180 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1181 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1182 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1183 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1184 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1185 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1186 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1187 		u32.shmid_ds32.shm_internal = 0;
1188 		error = copyout(&u32.shmid_ds32, uap->buf,
1189 		    sizeof(u32.shmid_ds32));
1190 		break;
1191 	}
1192 
1193 done:
1194 	if (error) {
1195 		/* Invalidate the return value */
1196 		td->td_retval[0] = -1;
1197 	}
1198 	return (error);
1199 }
1200 #endif
1201 
1202 int
1203 freebsd32_shmctl(struct thread *td, struct freebsd32_shmctl_args *uap)
1204 {
1205 	int error = 0;
1206 	union {
1207 		struct shmid_ds shmid_ds;
1208 		struct shm_info shm_info;
1209 		struct shminfo shminfo;
1210 	} u;
1211 	union {
1212 		struct shmid_ds32 shmid_ds32;
1213 		struct shm_info32 shm_info32;
1214 		struct shminfo32 shminfo32;
1215 	} u32;
1216 	size_t sz;
1217 
1218 	if (uap->cmd == IPC_SET) {
1219 		if ((error = copyin(uap->buf, &u32.shmid_ds32,
1220 		    sizeof(u32.shmid_ds32))))
1221 			goto done;
1222 		freebsd32_ipcperm_in(&u32.shmid_ds32.shm_perm,
1223 		    &u.shmid_ds.shm_perm);
1224 		CP(u32.shmid_ds32, u.shmid_ds, shm_segsz);
1225 		CP(u32.shmid_ds32, u.shmid_ds, shm_lpid);
1226 		CP(u32.shmid_ds32, u.shmid_ds, shm_cpid);
1227 		CP(u32.shmid_ds32, u.shmid_ds, shm_nattch);
1228 		CP(u32.shmid_ds32, u.shmid_ds, shm_atime);
1229 		CP(u32.shmid_ds32, u.shmid_ds, shm_dtime);
1230 		CP(u32.shmid_ds32, u.shmid_ds, shm_ctime);
1231 	}
1232 
1233 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&u, &sz);
1234 	if (error)
1235 		goto done;
1236 
1237 	/* Cases in which we need to copyout */
1238 	switch (uap->cmd) {
1239 	case IPC_INFO:
1240 		CP(u.shminfo, u32.shminfo32, shmmax);
1241 		CP(u.shminfo, u32.shminfo32, shmmin);
1242 		CP(u.shminfo, u32.shminfo32, shmmni);
1243 		CP(u.shminfo, u32.shminfo32, shmseg);
1244 		CP(u.shminfo, u32.shminfo32, shmall);
1245 		error = copyout(&u32.shminfo32, uap->buf,
1246 		    sizeof(u32.shminfo32));
1247 		break;
1248 	case SHM_INFO:
1249 		CP(u.shm_info, u32.shm_info32, used_ids);
1250 		CP(u.shm_info, u32.shm_info32, shm_rss);
1251 		CP(u.shm_info, u32.shm_info32, shm_tot);
1252 		CP(u.shm_info, u32.shm_info32, shm_swp);
1253 		CP(u.shm_info, u32.shm_info32, swap_attempts);
1254 		CP(u.shm_info, u32.shm_info32, swap_successes);
1255 		error = copyout(&u32.shm_info32, uap->buf,
1256 		    sizeof(u32.shm_info32));
1257 		break;
1258 	case SHM_STAT:
1259 	case IPC_STAT:
1260 		freebsd32_ipcperm_out(&u.shmid_ds.shm_perm,
1261 		    &u32.shmid_ds32.shm_perm);
1262 		if (u.shmid_ds.shm_segsz > INT32_MAX)
1263 			u32.shmid_ds32.shm_segsz = INT32_MAX;
1264 		else
1265 			CP(u.shmid_ds, u32.shmid_ds32, shm_segsz);
1266 		CP(u.shmid_ds, u32.shmid_ds32, shm_lpid);
1267 		CP(u.shmid_ds, u32.shmid_ds32, shm_cpid);
1268 		CP(u.shmid_ds, u32.shmid_ds32, shm_nattch);
1269 		CP(u.shmid_ds, u32.shmid_ds32, shm_atime);
1270 		CP(u.shmid_ds, u32.shmid_ds32, shm_dtime);
1271 		CP(u.shmid_ds, u32.shmid_ds32, shm_ctime);
1272 		error = copyout(&u32.shmid_ds32, uap->buf,
1273 		    sizeof(u32.shmid_ds32));
1274 		break;
1275 	}
1276 
1277 done:
1278 	if (error) {
1279 		/* Invalidate the return value */
1280 		td->td_retval[0] = -1;
1281 	}
1282 	return (error);
1283 }
1284 #endif
1285 
1286 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
1287     defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
1288 
1289 #ifndef CP
1290 #define CP(src, dst, fld)	do { (dst).fld = (src).fld; } while (0)
1291 #endif
1292 
1293 #ifndef _SYS_SYSPROTO_H_
1294 struct freebsd7_shmctl_args {
1295 	int shmid;
1296 	int cmd;
1297 	struct shmid_ds_old *buf;
1298 };
1299 #endif
1300 int
1301 freebsd7_shmctl(td, uap)
1302 	struct thread *td;
1303 	struct freebsd7_shmctl_args *uap;
1304 {
1305 	int error = 0;
1306 	struct shmid_ds_old old;
1307 	struct shmid_ds buf;
1308 	size_t bufsz;
1309 
1310 	/*
1311 	 * The only reason IPC_INFO, SHM_INFO, SHM_STAT exists is to support
1312 	 * Linux binaries.  If we see the call come through the FreeBSD ABI,
1313 	 * return an error back to the user since we do not to support this.
1314 	 */
1315 	if (uap->cmd == IPC_INFO || uap->cmd == SHM_INFO ||
1316 	    uap->cmd == SHM_STAT)
1317 		return (EINVAL);
1318 
1319 	/* IPC_SET needs to copyin the buffer before calling kern_shmctl */
1320 	if (uap->cmd == IPC_SET) {
1321 		if ((error = copyin(uap->buf, &old, sizeof(old))))
1322 			goto done;
1323 		ipcperm_old2new(&old.shm_perm, &buf.shm_perm);
1324 		CP(old, buf, shm_segsz);
1325 		CP(old, buf, shm_lpid);
1326 		CP(old, buf, shm_cpid);
1327 		CP(old, buf, shm_nattch);
1328 		CP(old, buf, shm_atime);
1329 		CP(old, buf, shm_dtime);
1330 		CP(old, buf, shm_ctime);
1331 	}
1332 
1333 	error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz);
1334 	if (error)
1335 		goto done;
1336 
1337 	/* Cases in which we need to copyout */
1338 	switch (uap->cmd) {
1339 	case IPC_STAT:
1340 		ipcperm_new2old(&buf.shm_perm, &old.shm_perm);
1341 		if (buf.shm_segsz > INT_MAX)
1342 			old.shm_segsz = INT_MAX;
1343 		else
1344 			CP(buf, old, shm_segsz);
1345 		CP(buf, old, shm_lpid);
1346 		CP(buf, old, shm_cpid);
1347 		if (buf.shm_nattch > SHRT_MAX)
1348 			old.shm_nattch = SHRT_MAX;
1349 		else
1350 			CP(buf, old, shm_nattch);
1351 		CP(buf, old, shm_atime);
1352 		CP(buf, old, shm_dtime);
1353 		CP(buf, old, shm_ctime);
1354 		old.shm_internal = NULL;
1355 		error = copyout(&old, uap->buf, sizeof(old));
1356 		break;
1357 	}
1358 
1359 done:
1360 	if (error) {
1361 		/* Invalidate the return value */
1362 		td->td_retval[0] = -1;
1363 	}
1364 	return (error);
1365 }
1366 
1367 #endif	/* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 ||
1368 	   COMPAT_FREEBSD7 */
1369 
1370 static int
1371 sysvshm_modload(struct module *module, int cmd, void *arg)
1372 {
1373 	int error = 0;
1374 
1375 	switch (cmd) {
1376 	case MOD_LOAD:
1377 		error = shminit();
1378 		if (error != 0)
1379 			shmunload();
1380 		break;
1381 	case MOD_UNLOAD:
1382 		error = shmunload();
1383 		break;
1384 	case MOD_SHUTDOWN:
1385 		break;
1386 	default:
1387 		error = EINVAL;
1388 		break;
1389 	}
1390 	return (error);
1391 }
1392 
1393 static moduledata_t sysvshm_mod = {
1394 	"sysvshm",
1395 	&sysvshm_modload,
1396 	NULL
1397 };
1398 
1399 DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
1400 MODULE_VERSION(sysvshm, 1);
1401